Original Research

Take-Home Points

• When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.

• Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months

• CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used.

• Clinical response to CSI may not depend on injection accuracy.

• Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Shoulder pain, a common clinical problem, occurs in 7% to 34% of the general population and in 21% of people older than 70 years.¹ Subacromial impingement refers to shoulder pain resulting from mechanical impingement of the rotator cuff underneath the coracoacromial arch between the acromion and the humeral head.^2,3 Subacromial impingement syndrome (SIS) is the most common cause of shoulder pain, resulting in significant functional deficits and disability.³

Treatment options for SIS include conservative modalities such as use of nonsteroidal anti-inflammatory drugs, physical therapy (PT), and subacromial corticosteroid injections (CSIs). Studies have found short- and long-term improvement in pain, function, and range of motion after CSI.^4-8 Subacromial CSI can be administered through an anterior or a posterior route.^4,9 There have been several studies of the accuracy of anterior and posterior CSIs,^10-12 with 2 studies finding similar accuracy for these routes.^10,11 However, there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹²

Although the accuracy of anterior and posterior routes has been studied, their effect on clinical outcomes has not. We conducted a study to understand the effects of anterior and posterior CSIs on SIS. As one of the accuracy studies suggested anterior CSI is more accurate—the anterior route was theorized to provide easier access to the subacromial space¹²—we hypothesized patients treated with anterior CSI would have superior clinical outcomes 6 months after injection.^12,13

Materials and Methods

Study Participants and Randomization

After this study received Institutional Review Board approval, patients with shoulder pain of more than 3 months’ duration and consistent with SIS were screened for inclusion. Eligible patients had pain in the anterior biceps and over the top of the shoulder with overhead activities as well as one or more clinical findings on physical examination: Hawkins-Kennedy sign, Neer sign, painful arc, and infraspinatus pain (pain with external rotation).

Patients were excluded if their history included prior subacromial CSI, adhesive capsulitis (inability to passively abduct shoulder to 90° with scapular stabilization), calcific tendonitis, radiographic evidence of os acromiale, cervical radiculopathy, Spurling sign, neck pain, radiating arm pain or numbness, sensory deficits, or neck and upper extremity motor dysfunction. Also excluded were patients with full-thickness rotator cuff tear, weakness on arm elevation, positive "drop arm sign," or high-riding humerus on standing shoulder radiograph. Patients who had work-related injuries or who were involved in worker compensation were excluded as well.

Enrolled patients were randomly assigned (with use of a computer-based random number generator) to receive either anterior CSI or posterior CSI.

Injection Procedures

All patients were administered 5 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone by 2 board-certified orthopedic surgeons using a 22-gauge 1½-inch needle. For patients who received their subacromial CSI by the anterior route, the arm was held in 0° of abduction and 20° of external rotation. The needle was inserted medial to the humeral head, lateral to the coracoid process, beginning 1 cm inferior to the clavicle with the needle directed posteriorly and laterally toward the acromion.¹⁰ For patients who received their CSI by the posterior route, the arm was held in 0° of abduction, the posterolateral corner of the acromion was identified by palpation, and the needle was inserted 1 cm inferior and medial to this point with the needle directed anteriorly and laterally toward the acromion.^10,12 In both groups, the subacromial space was identified when a drop in pressure was felt during needle insertion. Accuracy was assessed post hoc by asking patients to grade their response to the injection on a visual analog scale (VAS); VAS score was used as a surrogate for improvement. All patients had a positive Neer test: Pain decreased with impingement maneuvers immediately after injection.

All patients were referred for PT provided according to an evidence-based rehabilitation protocol.¹⁴ This program emphasized range of motion with shoulder shrugs, scapular retraction, and pendulum exercises; flexibility with stretching exercises targeting the anterior and posterior aspects of the shoulder and cane stretching for forward elevation and external rotation; and strength with strengthening exercises involving the rotator cuff and scapular stabilizers.

Outcome Measures

Pain was measured with VAS scores and function with Single Assessment Numeric Evaluation (SANE) scores. The VAS is a validated outcome measure of pain intensity. A numeric version of the VAS was used: Patients selected the whole number, from 0 (no pain) to 10 (worst possible pain), that best reflected their pain intensity. On SANE, another validated outcome measure, patients rated their shoulder function as a percentage of normal, from 0% (no function possible) to 100% (perfect).¹⁵ Before injection, all patients were administered the VAS and SANE questionnaires to establish their baseline pain level and opinion of shoulder function. These measures were repeated 1, 3, and 6 months after injection. Telephone interviews were conducted at 1 month and 6 months. Patients were asked to return to clinic 3 months after injection as part of the standard of care. At 3 months, 47 (86%) of the 55 patients returned for follow-up and were administered the VAS and SANE questionnaires; the other 8 completed the questionnaires by telephone. At each time point, patients were asked to report on their participation in PT and/or adherence to their home exercise program.

Statistical Analysis

Power analysis performed with Student t test and a 2-sided level of P = .05 compared VAS pain scores between the anterior and posterior injection routes and found a mean (SD) difference of 1.4 (1.7).¹⁶ Power calculations made with nQuery Advisor Version 7.0 (Statistical Solutions) indicated a total sample size of 60 patients (30/group) would provide 80% power for detecting a significant difference assuming a 20% dropout rate.

Two-way mixed-model analysis of variance (ANOVA) was used to compare the anterior and posterior routes for statistical differences in both VAS pain scores and SANE function scores at baseline and 1, 3, and 6 months after injection. Likewise, time at baseline (just before injection)was compared with follow-up (1, 3, 6 months) with 2-way mixed-model ANOVA adjusting for anterior or posterior route. Multivariate analysis was performed to evaluate differences between baseline and 6-month follow-up with respect to anterior and posterior injection routes, controlling for age, sex, and body mass index (BMI) for VAS and SANE scores. Parametric testing methods were used for statistical analysis, which was performed with IBM SPSS Statistics Version 21.0 (IBM Corp). Significance was set at P < .05.

Results

Patient Characteristics

Of the 55 patients enrolled, 25 (46%) received anterior subacromial CSI and 30 (54%) received posterior CSI. All enrolled patients had a positive Neer impingement test immediately after injection. Mean (SD) age was 48 (9.3) years for anterior group patients and 48 (9.0) years for posterior group patients. There was no significant difference in age or BMI between the 2 groups (Table).

Five patients (9%) were excluded from the study after randomization and CSI: 2 for a full-thickness rotator cuff tear, 1 for a Bankart lesion, 1 for adhesive capsulitis, and 1 for a worker compensation claim.

One month after injection, 41 patients (75%) reported having engaged in PT as prescribed. Of the 47 patients (86%) who returned for the 3-month follow-up, 25 (46%) reported having engaged in PT between 1 month and 3 months after injection. Fourteen patients (26%) reported attending PT between 3 and 6 months post-injection.

Outcome Measures

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in VAS scores between the anterior and posterior groups at any time point (P = .45). Both groups had highly significant pain reductions (anterior, F = 9.71, P < .001; posterior, F = 13.46, P < .001). Figure 1 shows mean VAS scores and significant reductions in pain 1, 3, and 6 months after injection (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of pain reduction over time, as indicated by a nonsignificant (P = .50) difference in slopes. These pain score reductions were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in SANE scores between the anterior and posterior groups, except for a higher mean score in the anterior group at 3 months
(P = .02). There were no other group differences (P > .10 for all). Both groups had highly significant improvements in function (anterior, F = 17.34,
P < .001; posterior, F = 13.57, P < .001). Figure 2 shows mean SANE scores and significant improvement at 1, 3, and 6 months (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of improved function over time, as indicated by a nonsignificant (P = .51) difference in slopes. These function score improvements were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

From the results of this prospective randomized study, we concluded subacromial CSI significantly reduces pain and improves function regardless of route used. In addition, age, sex, and BMI do not significantly affect the efficacy of either anterior CSI or posterior CSI.

Discussion

In patients with SIS, anterior CSI and posterior CSI provided significant improvements in pain and function 1, 3, and 6 months after injection. These effects were independent of age, sex, BMI, and PT participation. There were no significant differences in outcomes between injection routes.

When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.^4-8 Although clinical outcomes are inconsistent, CSI can be used to address SIS symptoms in appropriate patients. Specifically, Blair and colleagues⁶ found that, CSI consisting of 4 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone was effective in alleviating shoulder pain and improving shoulder range of motion. Other authors have similarly reported improved outcomes after subacromial injection and short-term follow-up with PT.^4,7,8 Our findings are consistent with these reports: CSI coupled with a structured rehabilitation program is effective in alleviating symptoms associated with acute or subacute SIS.

Numerous studies have found improved clinical outcomes after anterior CSI and after posterior CSI,^6-8 but no study has directly compared the clinical impact of anterior CSI with that of posterior CSI—which suggests injection route may not affect ultimate clinical outcomes.

CSI accuracy has been studied extensively.^10-12,17-20 Although 2 studies found similar accuracy for anterior and posterior routes,^10,11 there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹² Collectively, these studies expose the inherent difficulty in treating shoulder pain with localized subacromial injection. Therapy may fail because of errant needle positioning. Two prospective studies found improved clinical outcomes with successful delivery of medication into the subacromial space.^17,18 Poor clinical outcomes may result from inaccurate CSI.

In contrast to other clinical studies, our study found that injection route was not associated with differences in clinical response. In a prospective randomized clinical trial in which 75 patients received a subacromial injection, Marder and colleagues¹² found anterior routes 84% accurate and posterior routes 56% accurate; they concluded acromion anatomy and subacromial bursa anatomy make posterior injections more difficult. As theorized by Gruson and colleagues,¹³ with use of an anterior route, the needle enters inferior to the concavity of the acromion and provides easier access to the subacromial space. This idea is in line with Marder and colleagues’¹² conclusion that subacromial bursa anatomy provides a favorable environment for accurate CSI.

If accuracy is positively correlated with clinical improvement and anterior routes are more accurate, there should be a difference in response to posterior injections. Our results provide evidence that clinical response to CSI may not depend on injection accuracy. Perhaps merely placing the corticosteroid near the bursa is adequate for improving symptoms or perhaps some of the clinical improvement is due to the systemic effect of corticosteroids. These possibilities require further analysis.

Establishing the efficacy of CSI in SIS is difficult. The literature includes various study designs, different CSI indications and medication formulations, and varying emphasis on the role of organized PT. Rehabilitation has been found to alleviate joint pain by reducing inflammation,¹⁴ but data do not universally support this finding.^21,22 Nevertheless, use of PT might explain the divergence in clinical outcomes reported by Marder and colleagues,¹² who found anterior CSI more accurate than posterior CSI. In our practice, PT is recommended for all SIS patients, not only those who have CSI. Thus, our findings are framed within the context of successful CSI but may include patients who improved with PT alone. This issue raises the question of whether subacromial CSI should be guided by ultrasound. Ultrasound guidance can improve CSI accuracy and clinical outcomes,^23-25 though the value of this benefit is debated.²⁶

This study had several limitations. First, pain relief was patient reported. Second, the treatment plan involved CSI with PT but did not control for CSI used alone. PT, which is part of the standard of care for patients with SIS, added another degree of complexity to the study. Third, there may have been some variability in SIS severity (stage 1, 2, or 3). Fourth, although the study design controlled for various shoulder pathologies, advanced imaging, which could have provided diagnosis confirmation, was not available for all patients. Therefore, concurrent conditions may have confounded results. However, randomization was used to try to minimize this effect. Fifth, although injection routes were randomly assigned, the trial was not blinded. Sixth, the study was underpowered by 1 patient, as there was an estimated 20% dropout rate over 3 and 6 months of follow-up. However, we do not think our results were significantly affected.

Although more research is needed to fully describe the role of subacromial CSI in SIS, our study findings suggested that CSI using either an anterior or a posterior route creates a window of symptomatic relief in which patients may be able to engage in PT.

Conclusion

Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months. No differences were found between anterior and posterior CSIs. In the context of this study, CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used. Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Take-Home Points

• When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.

• Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months

• CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used.

• Clinical response to CSI may not depend on injection accuracy.

• Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Shoulder pain, a common clinical problem, occurs in 7% to 34% of the general population and in 21% of people older than 70 years.¹ Subacromial impingement refers to shoulder pain resulting from mechanical impingement of the rotator cuff underneath the coracoacromial arch between the acromion and the humeral head.^2,3 Subacromial impingement syndrome (SIS) is the most common cause of shoulder pain, resulting in significant functional deficits and disability.³

Treatment options for SIS include conservative modalities such as use of nonsteroidal anti-inflammatory drugs, physical therapy (PT), and subacromial corticosteroid injections (CSIs). Studies have found short- and long-term improvement in pain, function, and range of motion after CSI.^4-8 Subacromial CSI can be administered through an anterior or a posterior route.^4,9 There have been several studies of the accuracy of anterior and posterior CSIs,^10-12 with 2 studies finding similar accuracy for these routes.^10,11 However, there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹²

Although the accuracy of anterior and posterior routes has been studied, their effect on clinical outcomes has not. We conducted a study to understand the effects of anterior and posterior CSIs on SIS. As one of the accuracy studies suggested anterior CSI is more accurate—the anterior route was theorized to provide easier access to the subacromial space¹²—we hypothesized patients treated with anterior CSI would have superior clinical outcomes 6 months after injection.^12,13

Materials and Methods

Study Participants and Randomization

After this study received Institutional Review Board approval, patients with shoulder pain of more than 3 months’ duration and consistent with SIS were screened for inclusion. Eligible patients had pain in the anterior biceps and over the top of the shoulder with overhead activities as well as one or more clinical findings on physical examination: Hawkins-Kennedy sign, Neer sign, painful arc, and infraspinatus pain (pain with external rotation).

Patients were excluded if their history included prior subacromial CSI, adhesive capsulitis (inability to passively abduct shoulder to 90° with scapular stabilization), calcific tendonitis, radiographic evidence of os acromiale, cervical radiculopathy, Spurling sign, neck pain, radiating arm pain or numbness, sensory deficits, or neck and upper extremity motor dysfunction. Also excluded were patients with full-thickness rotator cuff tear, weakness on arm elevation, positive "drop arm sign," or high-riding humerus on standing shoulder radiograph. Patients who had work-related injuries or who were involved in worker compensation were excluded as well.

Enrolled patients were randomly assigned (with use of a computer-based random number generator) to receive either anterior CSI or posterior CSI.

Injection Procedures

All patients were administered 5 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone by 2 board-certified orthopedic surgeons using a 22-gauge 1½-inch needle. For patients who received their subacromial CSI by the anterior route, the arm was held in 0° of abduction and 20° of external rotation. The needle was inserted medial to the humeral head, lateral to the coracoid process, beginning 1 cm inferior to the clavicle with the needle directed posteriorly and laterally toward the acromion.¹⁰ For patients who received their CSI by the posterior route, the arm was held in 0° of abduction, the posterolateral corner of the acromion was identified by palpation, and the needle was inserted 1 cm inferior and medial to this point with the needle directed anteriorly and laterally toward the acromion.^10,12 In both groups, the subacromial space was identified when a drop in pressure was felt during needle insertion. Accuracy was assessed post hoc by asking patients to grade their response to the injection on a visual analog scale (VAS); VAS score was used as a surrogate for improvement. All patients had a positive Neer test: Pain decreased with impingement maneuvers immediately after injection.

All patients were referred for PT provided according to an evidence-based rehabilitation protocol.¹⁴ This program emphasized range of motion with shoulder shrugs, scapular retraction, and pendulum exercises; flexibility with stretching exercises targeting the anterior and posterior aspects of the shoulder and cane stretching for forward elevation and external rotation; and strength with strengthening exercises involving the rotator cuff and scapular stabilizers.

Outcome Measures

Pain was measured with VAS scores and function with Single Assessment Numeric Evaluation (SANE) scores. The VAS is a validated outcome measure of pain intensity. A numeric version of the VAS was used: Patients selected the whole number, from 0 (no pain) to 10 (worst possible pain), that best reflected their pain intensity. On SANE, another validated outcome measure, patients rated their shoulder function as a percentage of normal, from 0% (no function possible) to 100% (perfect).¹⁵ Before injection, all patients were administered the VAS and SANE questionnaires to establish their baseline pain level and opinion of shoulder function. These measures were repeated 1, 3, and 6 months after injection. Telephone interviews were conducted at 1 month and 6 months. Patients were asked to return to clinic 3 months after injection as part of the standard of care. At 3 months, 47 (86%) of the 55 patients returned for follow-up and were administered the VAS and SANE questionnaires; the other 8 completed the questionnaires by telephone. At each time point, patients were asked to report on their participation in PT and/or adherence to their home exercise program.

Statistical Analysis

Power analysis performed with Student t test and a 2-sided level of P = .05 compared VAS pain scores between the anterior and posterior injection routes and found a mean (SD) difference of 1.4 (1.7).¹⁶ Power calculations made with nQuery Advisor Version 7.0 (Statistical Solutions) indicated a total sample size of 60 patients (30/group) would provide 80% power for detecting a significant difference assuming a 20% dropout rate.

Two-way mixed-model analysis of variance (ANOVA) was used to compare the anterior and posterior routes for statistical differences in both VAS pain scores and SANE function scores at baseline and 1, 3, and 6 months after injection. Likewise, time at baseline (just before injection)was compared with follow-up (1, 3, 6 months) with 2-way mixed-model ANOVA adjusting for anterior or posterior route. Multivariate analysis was performed to evaluate differences between baseline and 6-month follow-up with respect to anterior and posterior injection routes, controlling for age, sex, and body mass index (BMI) for VAS and SANE scores. Parametric testing methods were used for statistical analysis, which was performed with IBM SPSS Statistics Version 21.0 (IBM Corp). Significance was set at P < .05.

Results

Patient Characteristics

Of the 55 patients enrolled, 25 (46%) received anterior subacromial CSI and 30 (54%) received posterior CSI. All enrolled patients had a positive Neer impingement test immediately after injection. Mean (SD) age was 48 (9.3) years for anterior group patients and 48 (9.0) years for posterior group patients. There was no significant difference in age or BMI between the 2 groups (Table).

Five patients (9%) were excluded from the study after randomization and CSI: 2 for a full-thickness rotator cuff tear, 1 for a Bankart lesion, 1 for adhesive capsulitis, and 1 for a worker compensation claim.

One month after injection, 41 patients (75%) reported having engaged in PT as prescribed. Of the 47 patients (86%) who returned for the 3-month follow-up, 25 (46%) reported having engaged in PT between 1 month and 3 months after injection. Fourteen patients (26%) reported attending PT between 3 and 6 months post-injection.

Outcome Measures

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in VAS scores between the anterior and posterior groups at any time point (P = .45). Both groups had highly significant pain reductions (anterior, F = 9.71, P < .001; posterior, F = 13.46, P < .001). Figure 1 shows mean VAS scores and significant reductions in pain 1, 3, and 6 months after injection (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of pain reduction over time, as indicated by a nonsignificant (P = .50) difference in slopes. These pain score reductions were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in SANE scores between the anterior and posterior groups, except for a higher mean score in the anterior group at 3 months
(P = .02). There were no other group differences (P > .10 for all). Both groups had highly significant improvements in function (anterior, F = 17.34,
P < .001; posterior, F = 13.57, P < .001). Figure 2 shows mean SANE scores and significant improvement at 1, 3, and 6 months (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of improved function over time, as indicated by a nonsignificant (P = .51) difference in slopes. These function score improvements were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

From the results of this prospective randomized study, we concluded subacromial CSI significantly reduces pain and improves function regardless of route used. In addition, age, sex, and BMI do not significantly affect the efficacy of either anterior CSI or posterior CSI.

Discussion

In patients with SIS, anterior CSI and posterior CSI provided significant improvements in pain and function 1, 3, and 6 months after injection. These effects were independent of age, sex, BMI, and PT participation. There were no significant differences in outcomes between injection routes.

When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.^4-8 Although clinical outcomes are inconsistent, CSI can be used to address SIS symptoms in appropriate patients. Specifically, Blair and colleagues⁶ found that, CSI consisting of 4 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone was effective in alleviating shoulder pain and improving shoulder range of motion. Other authors have similarly reported improved outcomes after subacromial injection and short-term follow-up with PT.^4,7,8 Our findings are consistent with these reports: CSI coupled with a structured rehabilitation program is effective in alleviating symptoms associated with acute or subacute SIS.

Numerous studies have found improved clinical outcomes after anterior CSI and after posterior CSI,^6-8 but no study has directly compared the clinical impact of anterior CSI with that of posterior CSI—which suggests injection route may not affect ultimate clinical outcomes.

CSI accuracy has been studied extensively.^10-12,17-20 Although 2 studies found similar accuracy for anterior and posterior routes,^10,11 there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹² Collectively, these studies expose the inherent difficulty in treating shoulder pain with localized subacromial injection. Therapy may fail because of errant needle positioning. Two prospective studies found improved clinical outcomes with successful delivery of medication into the subacromial space.^17,18 Poor clinical outcomes may result from inaccurate CSI.

In contrast to other clinical studies, our study found that injection route was not associated with differences in clinical response. In a prospective randomized clinical trial in which 75 patients received a subacromial injection, Marder and colleagues¹² found anterior routes 84% accurate and posterior routes 56% accurate; they concluded acromion anatomy and subacromial bursa anatomy make posterior injections more difficult. As theorized by Gruson and colleagues,¹³ with use of an anterior route, the needle enters inferior to the concavity of the acromion and provides easier access to the subacromial space. This idea is in line with Marder and colleagues’¹² conclusion that subacromial bursa anatomy provides a favorable environment for accurate CSI.

If accuracy is positively correlated with clinical improvement and anterior routes are more accurate, there should be a difference in response to posterior injections. Our results provide evidence that clinical response to CSI may not depend on injection accuracy. Perhaps merely placing the corticosteroid near the bursa is adequate for improving symptoms or perhaps some of the clinical improvement is due to the systemic effect of corticosteroids. These possibilities require further analysis.

Establishing the efficacy of CSI in SIS is difficult. The literature includes various study designs, different CSI indications and medication formulations, and varying emphasis on the role of organized PT. Rehabilitation has been found to alleviate joint pain by reducing inflammation,¹⁴ but data do not universally support this finding.^21,22 Nevertheless, use of PT might explain the divergence in clinical outcomes reported by Marder and colleagues,¹² who found anterior CSI more accurate than posterior CSI. In our practice, PT is recommended for all SIS patients, not only those who have CSI. Thus, our findings are framed within the context of successful CSI but may include patients who improved with PT alone. This issue raises the question of whether subacromial CSI should be guided by ultrasound. Ultrasound guidance can improve CSI accuracy and clinical outcomes,^23-25 though the value of this benefit is debated.²⁶

This study had several limitations. First, pain relief was patient reported. Second, the treatment plan involved CSI with PT but did not control for CSI used alone. PT, which is part of the standard of care for patients with SIS, added another degree of complexity to the study. Third, there may have been some variability in SIS severity (stage 1, 2, or 3). Fourth, although the study design controlled for various shoulder pathologies, advanced imaging, which could have provided diagnosis confirmation, was not available for all patients. Therefore, concurrent conditions may have confounded results. However, randomization was used to try to minimize this effect. Fifth, although injection routes were randomly assigned, the trial was not blinded. Sixth, the study was underpowered by 1 patient, as there was an estimated 20% dropout rate over 3 and 6 months of follow-up. However, we do not think our results were significantly affected.

Although more research is needed to fully describe the role of subacromial CSI in SIS, our study findings suggested that CSI using either an anterior or a posterior route creates a window of symptomatic relief in which patients may be able to engage in PT.

Conclusion

Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months. No differences were found between anterior and posterior CSIs. In the context of this study, CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used. Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Take-Home Points

• When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.

• Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months

• CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used.

• Clinical response to CSI may not depend on injection accuracy.

• Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Shoulder pain, a common clinical problem, occurs in 7% to 34% of the general population and in 21% of people older than 70 years.¹ Subacromial impingement refers to shoulder pain resulting from mechanical impingement of the rotator cuff underneath the coracoacromial arch between the acromion and the humeral head.^2,3 Subacromial impingement syndrome (SIS) is the most common cause of shoulder pain, resulting in significant functional deficits and disability.³

Treatment options for SIS include conservative modalities such as use of nonsteroidal anti-inflammatory drugs, physical therapy (PT), and subacromial corticosteroid injections (CSIs). Studies have found short- and long-term improvement in pain, function, and range of motion after CSI.^4-8 Subacromial CSI can be administered through an anterior or a posterior route.^4,9 There have been several studies of the accuracy of anterior and posterior CSIs,^10-12 with 2 studies finding similar accuracy for these routes.^10,11 However, there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹²

Although the accuracy of anterior and posterior routes has been studied, their effect on clinical outcomes has not. We conducted a study to understand the effects of anterior and posterior CSIs on SIS. As one of the accuracy studies suggested anterior CSI is more accurate—the anterior route was theorized to provide easier access to the subacromial space¹²—we hypothesized patients treated with anterior CSI would have superior clinical outcomes 6 months after injection.^12,13

Materials and Methods

Study Participants and Randomization

After this study received Institutional Review Board approval, patients with shoulder pain of more than 3 months’ duration and consistent with SIS were screened for inclusion. Eligible patients had pain in the anterior biceps and over the top of the shoulder with overhead activities as well as one or more clinical findings on physical examination: Hawkins-Kennedy sign, Neer sign, painful arc, and infraspinatus pain (pain with external rotation).

Patients were excluded if their history included prior subacromial CSI, adhesive capsulitis (inability to passively abduct shoulder to 90° with scapular stabilization), calcific tendonitis, radiographic evidence of os acromiale, cervical radiculopathy, Spurling sign, neck pain, radiating arm pain or numbness, sensory deficits, or neck and upper extremity motor dysfunction. Also excluded were patients with full-thickness rotator cuff tear, weakness on arm elevation, positive "drop arm sign," or high-riding humerus on standing shoulder radiograph. Patients who had work-related injuries or who were involved in worker compensation were excluded as well.

Enrolled patients were randomly assigned (with use of a computer-based random number generator) to receive either anterior CSI or posterior CSI.

Injection Procedures

All patients were administered 5 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone by 2 board-certified orthopedic surgeons using a 22-gauge 1½-inch needle. For patients who received their subacromial CSI by the anterior route, the arm was held in 0° of abduction and 20° of external rotation. The needle was inserted medial to the humeral head, lateral to the coracoid process, beginning 1 cm inferior to the clavicle with the needle directed posteriorly and laterally toward the acromion.¹⁰ For patients who received their CSI by the posterior route, the arm was held in 0° of abduction, the posterolateral corner of the acromion was identified by palpation, and the needle was inserted 1 cm inferior and medial to this point with the needle directed anteriorly and laterally toward the acromion.^10,12 In both groups, the subacromial space was identified when a drop in pressure was felt during needle insertion. Accuracy was assessed post hoc by asking patients to grade their response to the injection on a visual analog scale (VAS); VAS score was used as a surrogate for improvement. All patients had a positive Neer test: Pain decreased with impingement maneuvers immediately after injection.

All patients were referred for PT provided according to an evidence-based rehabilitation protocol.¹⁴ This program emphasized range of motion with shoulder shrugs, scapular retraction, and pendulum exercises; flexibility with stretching exercises targeting the anterior and posterior aspects of the shoulder and cane stretching for forward elevation and external rotation; and strength with strengthening exercises involving the rotator cuff and scapular stabilizers.

Outcome Measures

Pain was measured with VAS scores and function with Single Assessment Numeric Evaluation (SANE) scores. The VAS is a validated outcome measure of pain intensity. A numeric version of the VAS was used: Patients selected the whole number, from 0 (no pain) to 10 (worst possible pain), that best reflected their pain intensity. On SANE, another validated outcome measure, patients rated their shoulder function as a percentage of normal, from 0% (no function possible) to 100% (perfect).¹⁵ Before injection, all patients were administered the VAS and SANE questionnaires to establish their baseline pain level and opinion of shoulder function. These measures were repeated 1, 3, and 6 months after injection. Telephone interviews were conducted at 1 month and 6 months. Patients were asked to return to clinic 3 months after injection as part of the standard of care. At 3 months, 47 (86%) of the 55 patients returned for follow-up and were administered the VAS and SANE questionnaires; the other 8 completed the questionnaires by telephone. At each time point, patients were asked to report on their participation in PT and/or adherence to their home exercise program.

Statistical Analysis

Power analysis performed with Student t test and a 2-sided level of P = .05 compared VAS pain scores between the anterior and posterior injection routes and found a mean (SD) difference of 1.4 (1.7).¹⁶ Power calculations made with nQuery Advisor Version 7.0 (Statistical Solutions) indicated a total sample size of 60 patients (30/group) would provide 80% power for detecting a significant difference assuming a 20% dropout rate.

Two-way mixed-model analysis of variance (ANOVA) was used to compare the anterior and posterior routes for statistical differences in both VAS pain scores and SANE function scores at baseline and 1, 3, and 6 months after injection. Likewise, time at baseline (just before injection)was compared with follow-up (1, 3, 6 months) with 2-way mixed-model ANOVA adjusting for anterior or posterior route. Multivariate analysis was performed to evaluate differences between baseline and 6-month follow-up with respect to anterior and posterior injection routes, controlling for age, sex, and body mass index (BMI) for VAS and SANE scores. Parametric testing methods were used for statistical analysis, which was performed with IBM SPSS Statistics Version 21.0 (IBM Corp). Significance was set at P < .05.

Results

Patient Characteristics

Of the 55 patients enrolled, 25 (46%) received anterior subacromial CSI and 30 (54%) received posterior CSI. All enrolled patients had a positive Neer impingement test immediately after injection. Mean (SD) age was 48 (9.3) years for anterior group patients and 48 (9.0) years for posterior group patients. There was no significant difference in age or BMI between the 2 groups (Table).

Five patients (9%) were excluded from the study after randomization and CSI: 2 for a full-thickness rotator cuff tear, 1 for a Bankart lesion, 1 for adhesive capsulitis, and 1 for a worker compensation claim.

One month after injection, 41 patients (75%) reported having engaged in PT as prescribed. Of the 47 patients (86%) who returned for the 3-month follow-up, 25 (46%) reported having engaged in PT between 1 month and 3 months after injection. Fourteen patients (26%) reported attending PT between 3 and 6 months post-injection.

Outcome Measures

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in VAS scores between the anterior and posterior groups at any time point (P = .45). Both groups had highly significant pain reductions (anterior, F = 9.71, P < .001; posterior, F = 13.46, P < .001). Figure 1 shows mean VAS scores and significant reductions in pain 1, 3, and 6 months after injection (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of pain reduction over time, as indicated by a nonsignificant (P = .50) difference in slopes. These pain score reductions were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

Two-way repeated-measures ANOVA with age, sex, and BMI included as covariates revealed no significant differences in SANE scores between the anterior and posterior groups, except for a higher mean score in the anterior group at 3 months
(P = .02). There were no other group differences (P > .10 for all). Both groups had highly significant improvements in function (anterior, F = 17.34,
P < .001; posterior, F = 13.57, P < .001). Figure 2 shows mean SANE scores and significant improvement at 1, 3, and 6 months (see asterisks for anterior and posterior groups; P < .001 for all). The groups had parallel rates of improved function over time, as indicated by a nonsignificant (P = .51) difference in slopes. These function score improvements were significant for both injection routes and were independent of age, sex, and BMI (P > .05 for all).

From the results of this prospective randomized study, we concluded subacromial CSI significantly reduces pain and improves function regardless of route used. In addition, age, sex, and BMI do not significantly affect the efficacy of either anterior CSI or posterior CSI.

Discussion

In patients with SIS, anterior CSI and posterior CSI provided significant improvements in pain and function 1, 3, and 6 months after injection. These effects were independent of age, sex, BMI, and PT participation. There were no significant differences in outcomes between injection routes.

When conservative treatments for SIS do not resolve symptoms, inflammation and pain can be reduced with use of subacromial CSI.^4-8 Although clinical outcomes are inconsistent, CSI can be used to address SIS symptoms in appropriate patients. Specifically, Blair and colleagues⁶ found that, CSI consisting of 4 mL of lidocaine 1% (without epinephrine) and 2 mL (80 mg) of triamcinolone was effective in alleviating shoulder pain and improving shoulder range of motion. Other authors have similarly reported improved outcomes after subacromial injection and short-term follow-up with PT.^4,7,8 Our findings are consistent with these reports: CSI coupled with a structured rehabilitation program is effective in alleviating symptoms associated with acute or subacute SIS.

Numerous studies have found improved clinical outcomes after anterior CSI and after posterior CSI,^6-8 but no study has directly compared the clinical impact of anterior CSI with that of posterior CSI—which suggests injection route may not affect ultimate clinical outcomes.

CSI accuracy has been studied extensively.^10-12,17-20 Although 2 studies found similar accuracy for anterior and posterior routes,^10,11 there may be a sex difference: In women, a posterior route may be less accurate than an anterior or a lateral route.¹² Collectively, these studies expose the inherent difficulty in treating shoulder pain with localized subacromial injection. Therapy may fail because of errant needle positioning. Two prospective studies found improved clinical outcomes with successful delivery of medication into the subacromial space.^17,18 Poor clinical outcomes may result from inaccurate CSI.

In contrast to other clinical studies, our study found that injection route was not associated with differences in clinical response. In a prospective randomized clinical trial in which 75 patients received a subacromial injection, Marder and colleagues¹² found anterior routes 84% accurate and posterior routes 56% accurate; they concluded acromion anatomy and subacromial bursa anatomy make posterior injections more difficult. As theorized by Gruson and colleagues,¹³ with use of an anterior route, the needle enters inferior to the concavity of the acromion and provides easier access to the subacromial space. This idea is in line with Marder and colleagues’¹² conclusion that subacromial bursa anatomy provides a favorable environment for accurate CSI.

If accuracy is positively correlated with clinical improvement and anterior routes are more accurate, there should be a difference in response to posterior injections. Our results provide evidence that clinical response to CSI may not depend on injection accuracy. Perhaps merely placing the corticosteroid near the bursa is adequate for improving symptoms or perhaps some of the clinical improvement is due to the systemic effect of corticosteroids. These possibilities require further analysis.

Establishing the efficacy of CSI in SIS is difficult. The literature includes various study designs, different CSI indications and medication formulations, and varying emphasis on the role of organized PT. Rehabilitation has been found to alleviate joint pain by reducing inflammation,¹⁴ but data do not universally support this finding.^21,22 Nevertheless, use of PT might explain the divergence in clinical outcomes reported by Marder and colleagues,¹² who found anterior CSI more accurate than posterior CSI. In our practice, PT is recommended for all SIS patients, not only those who have CSI. Thus, our findings are framed within the context of successful CSI but may include patients who improved with PT alone. This issue raises the question of whether subacromial CSI should be guided by ultrasound. Ultrasound guidance can improve CSI accuracy and clinical outcomes,^23-25 though the value of this benefit is debated.²⁶

This study had several limitations. First, pain relief was patient reported. Second, the treatment plan involved CSI with PT but did not control for CSI used alone. PT, which is part of the standard of care for patients with SIS, added another degree of complexity to the study. Third, there may have been some variability in SIS severity (stage 1, 2, or 3). Fourth, although the study design controlled for various shoulder pathologies, advanced imaging, which could have provided diagnosis confirmation, was not available for all patients. Therefore, concurrent conditions may have confounded results. However, randomization was used to try to minimize this effect. Fifth, although injection routes were randomly assigned, the trial was not blinded. Sixth, the study was underpowered by 1 patient, as there was an estimated 20% dropout rate over 3 and 6 months of follow-up. However, we do not think our results were significantly affected.

Although more research is needed to fully describe the role of subacromial CSI in SIS, our study findings suggested that CSI using either an anterior or a posterior route creates a window of symptomatic relief in which patients may be able to engage in PT.

Conclusion

Both anterior CSI and posterior CSI significantly improved pain and function for up to 6 months. No differences were found between anterior and posterior CSIs. In the context of this study, CSI combined with structured PT produced significant improvement in pain and function in patients with SIS, regardless of injection route used. Clinicians should rely on their clinical acumen when selecting injection routes, as anterior and posterior are both beneficial.

Clostridium difficile infection (CDI) is the most common infectious cause of healthcare-associated diarrhea.¹ Development of a CDI during hospitalization is associated with increases in morbidity, mortality, length of stay (LOS), and cost.^2-5 The prevalence of CDI in hospitalized patients has increased dramatically from the mid-1990s to the mid-2000s to almost 9 cases per 1000 discharges; however, the CDI rate since 2007 appears to have plateaued.^6,7 Antibiotic use has historically been the most important risk factor for acquiring CDI; however, use of acid-suppressing agents, chemotherapy, chronic comorbidities, and healthcare exposure all also increase the risk of CDI.^7-10 The elderly (> 65 years of age) are particularly at risk for developing CDI and having worse clinical outcomes with CDI.^6,7

Patients with septic shock (SS) often have multiple CDI risk factors (in particular, extensive antibiotic exposure) and thus, represent a population at a particularly high risk for acquiring a CDI during hospitalization. However, little data are available on the prevalence of CDI acquired in patients hospitalized with SS. We sought to determine the national-level temporal trends in the prevalence of CDI in patients with SS and the impact of CDI complicating SS on clinical outcomes between 2007 and 2013.

METHODS

Data Source

We used the National Inpatient Sample (NIS) and Nationwide Readmissions Database (NRD) for this study. The NIS is a database developed by the Agency of Healthcare Research and Quality for the Healthcare Cost and Utilization Project (HCUP).¹¹ It is the largest all-payer inpatient database in the United States and has been used by researchers and policy makers to analyze national trends in outcomes and healthcare utilization. The NIS database now approximates a 20% stratified sample of all discharges from all participating US hospitals. Sampling weights are provided by the manufacturer and can be used to produce national-level estimates. Following the redesign of the NIS in 2012, new sampling weights were provided for trend analysis for the years prior to 2012 to account for the new design. Every hospitalization is deidentified and converted into one unique entry that provides information on demographics, hospital characteristics, 1 primary and up to 24 secondary discharge diagnoses, comorbidities, LOS, in-hospital mortality, and procedures performed during stay. The discharge diagnoses are provided in the form of the International Classification of Diseases, 9th Revision-Clinical Modification (ICD-9-CM) codes.

The NRD is a database developed for HCUP that contains about 35 million discharges each year and supports readmission data analyses. In 2013, the NRD contained data from 21 geographically diverse states, accounting for 49.1% of all US hospitalizations. Diagnosis, comorbidities, and outcomes are presented in a similar manner to NIS.

Study Design

This was a retrospective cohort study. Data from the NIS between 2007 and 2013 were used for the analysis. Demographic data obtained included age, gender, race, Charlson-Deyo Comorbidity Index,¹² hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Cases with information missing on key demographic variables (age, gender, and race) were excluded. Only adults (>18 years of age) were included for the analysis.

SS was identified by either (1) ICD-9-CM diagnosis code for SS (785.52) or (2) presence of vasopressor use (00.17) along with ICD-9-CM codes of sepsis, severe sepsis, septicemia, bacteremia, or fungemia. This approach is consistent with what has been utilized in other studies to identify cases of sepsis or SS from administrative databases.^13-15 The appendix provides a complete list of ICD-9-CM codes used in the study. CDI was identified by ICD-9-CM code 008.45 among the secondary diagnosis. This code has been shown to have good accuracy for identifying CDI using administrative data.¹⁶ To minimize the inclusion of cases in which a CDI was present at admission, hospitalizations with a primary diagnosis of CDI were not included as cases of CDI complicating SS.

We used NRD 2013 for estimating the effect of CDI on 30-day readmission after initial hospitalizations with SS. We used the criteria for index admissions and 30-day readmissions as defined by the Centers for Medicare and Medicaid Services. We excluded patients who died during their index admission, patients with index discharges in December due to a lack of sufficient time to capture 30-day readmissions, and patients with missing information on key variables. We also excluded patients who were not a resident of the state of index hospitalization since readmission across state boundaries could not be identified in NRD. Manufacturer provided sampling weights were used to produce national level estimates. The cases of SS and CDI were identified by ICD-9-CM codes using the methodology described above.

 

Outcomes

Our primary outcome of interest was the total and yearly prevalence of CDI in patients with SS from 2007 to 2013. The secondary outcomes were mortality, LOS, and 30-day readmissions in patients with SS with and without CDI.

Statistical Analysis

Weighted data from NIS were used for all analyses. Demographics, hospital characteristics, and outcomes of all patients with SS were obtained. The prevalence of CDI was calculated for each calendar year. The temporal trends of outcomes (LOS and in-hospital mortality) of patients were plotted for patients with SS with and without CDI. A χ² test of trend for proportions was used with the Cochran-Armitage test to calculate statistical significance of changes in prevalence. To test for statistical significance of the temporal trends of LOS, a univariate linear regression was used, with calendar year as a covariate. Independent samples t test, a Mann-Whitney U test, and a χ² test were used to determine statistical significance of parameters between the group with CDI and the group without CDI.

Prolonged LOS was defined either as a LOS > 75th or > 90th percentile of LOS among all patients with SS. To identify if CDI was associated with a prolonged LOS after adjusting for patient and hospital characteristics, a multivariate logistic regression analysis was used. Variables included in the regression model were age, gender, race, Charlson-Deyo Comorbidity Index, hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Data on cases were available for all the above covariates except hospital characteristics, such as teaching status, location, and bed size (these were missing for 0.7% of hospitals).

Stata 13.1.0 (Stata Corp, College Station, TX) and SPSS 23.0 (SPSS Inc., Chicago, IL) were used to perform statistical analyses. A P value of <0.05 was considered statistically significant.

RESULTS

Demographics

A total of 2,031,739 hospitalizations of adults with SS were identified between 2007 and 2013. CDI was present in 166,432 (8.2%) of these patients. Demographic data are displayed in Table 1. CDI was more commonly observed in elderly patients (> 65 years) with SS; 9.3% among the elderly versus 6.6% among individuals < 65 years; P < 0.001. The prevalence of CDI was greater in urban than in rural hospitals (8.4% vs 5.4%; P < 0.001) and greater in teaching than in nonteaching hospitals (8.7% vs 7.7%; P < 0.001). The prevalence of CDI in SS remained stable between 2007 and 2013 (Table 2).

Mortality

In the overall study cohort, the in-hospital mortality for SS was 37%. The in-hospital mortality rate of patients with SS complicated by a CDI was comparable to the mortality rate of patients without a CDI (37.1% vs 37.0%; P = 0.48). The mortality of patients with SS, with or without CDI, progressively decreased from 2007 to 2013 (P value for trend < 0.001 for each group; Figure 1).

Length of Stay

The median LOS for all patients with SS was 9 days. Patients with CDI had a longer median LOS than did those without CDI (13 vs 9 days; P < 0.001). Between 2007 and 2013, the median LOS of CDI group decreased from 14 to 12 days (P < 0.001) while that of non-CDI group decreased from 9 to 8 days (P < 0.001; Figure 2). We also examined LOS among subgroups who were discharged alive and those who died during hospitalization. For patients who were discharged alive, the LOS with and without CDI was 15 days versus 10 days, respectively (P < 0.001). For patients who died during hospitalization, LOS with and without CDI was 10 days versus 6 days, respectively (P < 0.001).

The 75th percentile of LOS of the total SS cohort was 17 days. An LOS > 17 days was observed in 36.9% of SS patients with CDI versus 22.7% without CDI (P < 0.001). After adjusting for patient and provider level variables, the odds of a LOS > 17 days were significantly greater for SS patients with CDI (odds ratio [OR] 2.11; 95% confidence interval [CI], 2.06-2.15; P < 0.001).

The 90th percentile of LOS of the total SS cohort was 29 days. An LOS > 29 days was observed in 17.5% of SS patients with a CDI versus 9.1% without a CDI (P < 0.001). After adjustment for patient and provider level variables, the odds of a LOS > 29 days were significantly greater for SS patients with a CDI (OR 2.25; 95% CI, 2.22-2.28; P < 0.001).

Hospital Readmission

In 2013, patients with SS and CDI had a higher rate of 30-day readmission as compared to patients with SS without CDI (9.8% vs 7.4% respectively; P < 0.001). The multivariate adjusted OR for 30-day readmission for patients with SS and a CDI was 1.26 (95% CI, 1.22-1.31; P < 0.001).

 

Additional Analyses

Lastly, we performed an additional analysis to confirm our hypothesis that a CDI by itself is rarely a cause of SS, and that CDI as the principal diagnosis would constitute an extremely low number of patients with SS in an administrative dataset. In NIS 2013, there were 105,750 cases with CDI as the primary diagnosis. A total of 4470 (4.2%) had a secondary diagnosis of sepsis and only 930 (0.9%) cases had a secondary diagnosis of SS.

DISCUSSION

This is the first study to report on the prevalence and outcome of CDI complicating SS. By using a large nationally representative sample, we found CDI was very prevalent among individuals hospitalized with SS and, at a level in excess of that seen in other populations. Of interest, we did not observe an increase in mortality of SS when complicated by CDI. On the other hand, patients with SS complicated by CDI were more much likely to have a prolonged hospital LOS and a higher risk of 30-day hospital readmission.

The prevalence of CDI exploded between the mid-1990s and mid-2000s, including community, hospital, and intensive care unit (ICU)–related disease.^6,7,17-20 Patients with SS often have multiple risk factors associated with CDI and thus represent a high-risk population for developing CDI.⁷ Our findings are consistent with the suggestion that individuals with SS are at a higher risk of developing CDI. Compared to the rate of CDI in all hospitalized patients, our data suggest an almost 10-fold increase in CDI rate for patients with SS.⁶ Patients with SS and CDI may account for as much as 10% of total CDIs.^6,7 As has been reported for CDI in general, we observed that CDI complicating SS was more common in those > 65 years of age.^4,21 The prevalence of CDI we observed in patients with SS was also higher than has been reported in ICU patients in general (1%), and higher than reported for patients requiring mechanical ventilation (6.6%), including prolonged mechanical ventilation (5.3%); further supporting the conclusion that patients with SS are a particularly high-risk group for acquiring CDI, even compared with other ICU patients.^20,22,23 Similarly, the rate of CDI among SS was 8 times higher than that of recently reported hospital-onset CDI among patients with sepsis in general (incidence 1.08%).²⁴ We have no data regarding why patients with SS have a higher rate of CDI; however, the intensity and duration of antibiotic treatment of these patients may certainly play a role.²⁵ It has recently been reported that CDI in itself can be a precursor leading to intestinal dysbiosis that can increase the risk of subsequent sepsis. Similarly, patients with SS may have higher prevalence of dysbiosis that, in turn, might predispose them to CDI at a higher rate than other individuals.

Following the increase in CDIs in the mid-1990s and the mid-2000s, since 2007 the overall prevalence of CDIs has been stable, albeit at the higher rate. More recently, the Centers for Disease Control and Prevention (CDC) has reported a decrease in hospital onset CDI after 2011.²⁶

The finding that CDI in SS patients was not associated with an increase in mortality is consistent with other reports of CDI in ICU patients in general as well as higher-risk ICU populations such as patients requiring mechanical ventilation, including those on long-term mechanical ventilator support.^{17,18,20,22,23} Why the mortality of ICU patients with CDI is not increased is not completely clear. It has been suggested that this may be related to early recognition and treatment of CDI developing in the ICU.²² Along these lines, it has been previously observed that for patients with CDI on mechanical ventilation, patients who were transferred to the ICU from the ward had worse clinical outcomes compared to patients directly admitted to the ICU, likely due to delayed recognition and treatment in the former.²² Similarly, ICU patients in whom CDI was identified prior to ICU admission had more severe CDI, and mortality that was directly related to CDI was only observed in patients who had CDI identified pre-ICU transfer.¹⁸ The increase in mortality observed in patients with sepsis in general with CDI may reflect similar factors.²⁴ We observed a trend of decreasing mortality in SS patients with or without CDI during 2007 to 2013 consistent to what has been generally reported in SS.^13,14

The increase in LOS observed in SS patients with CDI is also consistent with what has been observed in other ICU populations, as well as in patients with sepsis in general.^17,22-24 Of note, in addition to the increase in median LOS, we found a significant increase in the number of patients with a prolonged LOS associated with having SS with CDI. It is important to note that development of CDI during hospitalization is affected by pre-CDI hospital LOS, so prolonged LOS may not be solely attributable to CDI. The interaction between LOS and CDI remains complex in which higher LOS might be associated with higher incidence of CDI occurrence, and once established, CDI might be associated with changes in LOS for the remaining hospitalization.

Hospitalized patients with CDI have an overall higher resource utilization than those without CDI.²⁷ A recent review has estimated the overall attributable cost of CDI to be $6.3 billion; the attributable cost per case of hospital acquired CDI being 1.5 times the cost of community-acquired CDI.⁵ We did not look at cost directly. However, in the high-CDI risk ICU population requiring prolonged mechanical ventilation, those with CDI had a substantial increase in total costs.²³ Given the substantial increase in LOS associated with CDI complicating SS, there would likely be a significant increase in hospital costs related to providing care for these patients. Further adding to the potential burden of CDI is our finding that CDI and SS was associated with an increase in 30-day hospital readmission rate. This is consistent with a recent report that ICU patients with CDI who are discharged from the hospital have a 25% 30-day hospital readmission rate.²⁸ However, we do not have data either as to the reason for hospital readmission or whether the initial CDI or CDI recurrence played a role. This suggests that, in addition to intervention directed toward preventing CDI, efforts should be directed towards identifying factors that can be modified in CDI patients prior to or after hospital discharge.

This study has several limitations. Using an administrative database (such as NIS) has an inherent limitation of coding errors and reporting bias can lead to misclassification of cohort definition (SS) and outcome (CDI). To minimize bias due to coding errors, we used previously validated ICD-9-CM codes and approach to identify individuals with SS and CDI.^13-15 Although the SS population was identified with ICD-9-CM codes using an administrative database, the in-hospital mortality for our septic population was similar to previously reported mortality of SS, suggesting the population selected was appropriate.¹³ SS due to CDI could not be identified; however, CDI by itself causing SS is rare, as described in recent literature.^29,30 An important potential bias that needs to be acknowledged is the immortal time bias. The occurrence of CDI in itself can be influenced by pre-CDI hospital LOS. Patients who were extremely sick could have died early in their hospital course before they could acquire CDI, which would influence the mortality difference between the group with CDI and group without CDI. Furthermore, we did not have information on either the treatment of CDI or SS or any measures of severity of illness, which could lead to residual confounding despite adjusting for multiple variables. In terms of readmission data, it was necessary to exclude nonresidents of a state for the 30-day readmission analysis, as readmissions could not be tracked across state boundaries by using the NRD. This might have resulted in an underrepresentation of the readmission burden. Lastly, it was not possible to identify mortality after hospital discharge as the NIS provides only in-hospital mortality.

In conclusion, CDI is more prevalent in SS than are other ICU populations or the hospital population in general, and CDI complicating SS is associated with significant increase in LOS and risk of 30-day hospital readmission. How much of the increase in resource utilization and cost are in fact attributable to CDI in this population remains to be studied. Our finding of high prevalence of CDI in the SS population further emphasizes the importance of maintaining and furthering approaches to reduce incidence of hospital acquired CDI. While reducing unnecessary antibiotics is important, a multipronged approach that includes education and infection control interventions has also been shown to reduce the incidence of CDI in the ICU.³¹ Given the economic burden of CDI, implementing these strategies to reduce CDI is warranted. Similarly, the risk of 30-day hospital readmission with CDI highlights the importance of identifying the factors that contribute to hospital readmission prior to initial hospital discharge. Programs to reduce CDI will not only improve outcomes directly attributable to CDI but also decrease the reservoir of CDI. Finally, to the extent that CDI can be reduced in the ICU, the utilization of ICU resources will be more effective.

 

Disclosure

No conflicts of interest or financial disclosures to report. Author Contributions: KC had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis, including and especially any adverse effects. KC, AG, AC, KK, and HC contributed to study design, data analysis, interpretation, and the writing of the manuscript. Guarantor statement: Kshitij Chatterjee takes responsibility for (is the guarantor of) the content of the manuscript, including the data and analysis.

Clostridium difficile infection (CDI) is the most common infectious cause of healthcare-associated diarrhea.¹ Development of a CDI during hospitalization is associated with increases in morbidity, mortality, length of stay (LOS), and cost.^2-5 The prevalence of CDI in hospitalized patients has increased dramatically from the mid-1990s to the mid-2000s to almost 9 cases per 1000 discharges; however, the CDI rate since 2007 appears to have plateaued.^6,7 Antibiotic use has historically been the most important risk factor for acquiring CDI; however, use of acid-suppressing agents, chemotherapy, chronic comorbidities, and healthcare exposure all also increase the risk of CDI.^7-10 The elderly (> 65 years of age) are particularly at risk for developing CDI and having worse clinical outcomes with CDI.^6,7

Patients with septic shock (SS) often have multiple CDI risk factors (in particular, extensive antibiotic exposure) and thus, represent a population at a particularly high risk for acquiring a CDI during hospitalization. However, little data are available on the prevalence of CDI acquired in patients hospitalized with SS. We sought to determine the national-level temporal trends in the prevalence of CDI in patients with SS and the impact of CDI complicating SS on clinical outcomes between 2007 and 2013.

METHODS

Data Source

We used the National Inpatient Sample (NIS) and Nationwide Readmissions Database (NRD) for this study. The NIS is a database developed by the Agency of Healthcare Research and Quality for the Healthcare Cost and Utilization Project (HCUP).¹¹ It is the largest all-payer inpatient database in the United States and has been used by researchers and policy makers to analyze national trends in outcomes and healthcare utilization. The NIS database now approximates a 20% stratified sample of all discharges from all participating US hospitals. Sampling weights are provided by the manufacturer and can be used to produce national-level estimates. Following the redesign of the NIS in 2012, new sampling weights were provided for trend analysis for the years prior to 2012 to account for the new design. Every hospitalization is deidentified and converted into one unique entry that provides information on demographics, hospital characteristics, 1 primary and up to 24 secondary discharge diagnoses, comorbidities, LOS, in-hospital mortality, and procedures performed during stay. The discharge diagnoses are provided in the form of the International Classification of Diseases, 9th Revision-Clinical Modification (ICD-9-CM) codes.

The NRD is a database developed for HCUP that contains about 35 million discharges each year and supports readmission data analyses. In 2013, the NRD contained data from 21 geographically diverse states, accounting for 49.1% of all US hospitalizations. Diagnosis, comorbidities, and outcomes are presented in a similar manner to NIS.

Study Design

This was a retrospective cohort study. Data from the NIS between 2007 and 2013 were used for the analysis. Demographic data obtained included age, gender, race, Charlson-Deyo Comorbidity Index,¹² hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Cases with information missing on key demographic variables (age, gender, and race) were excluded. Only adults (>18 years of age) were included for the analysis.

SS was identified by either (1) ICD-9-CM diagnosis code for SS (785.52) or (2) presence of vasopressor use (00.17) along with ICD-9-CM codes of sepsis, severe sepsis, septicemia, bacteremia, or fungemia. This approach is consistent with what has been utilized in other studies to identify cases of sepsis or SS from administrative databases.^13-15 The appendix provides a complete list of ICD-9-CM codes used in the study. CDI was identified by ICD-9-CM code 008.45 among the secondary diagnosis. This code has been shown to have good accuracy for identifying CDI using administrative data.¹⁶ To minimize the inclusion of cases in which a CDI was present at admission, hospitalizations with a primary diagnosis of CDI were not included as cases of CDI complicating SS.

We used NRD 2013 for estimating the effect of CDI on 30-day readmission after initial hospitalizations with SS. We used the criteria for index admissions and 30-day readmissions as defined by the Centers for Medicare and Medicaid Services. We excluded patients who died during their index admission, patients with index discharges in December due to a lack of sufficient time to capture 30-day readmissions, and patients with missing information on key variables. We also excluded patients who were not a resident of the state of index hospitalization since readmission across state boundaries could not be identified in NRD. Manufacturer provided sampling weights were used to produce national level estimates. The cases of SS and CDI were identified by ICD-9-CM codes using the methodology described above.

 

Outcomes

Our primary outcome of interest was the total and yearly prevalence of CDI in patients with SS from 2007 to 2013. The secondary outcomes were mortality, LOS, and 30-day readmissions in patients with SS with and without CDI.

Statistical Analysis

Weighted data from NIS were used for all analyses. Demographics, hospital characteristics, and outcomes of all patients with SS were obtained. The prevalence of CDI was calculated for each calendar year. The temporal trends of outcomes (LOS and in-hospital mortality) of patients were plotted for patients with SS with and without CDI. A χ² test of trend for proportions was used with the Cochran-Armitage test to calculate statistical significance of changes in prevalence. To test for statistical significance of the temporal trends of LOS, a univariate linear regression was used, with calendar year as a covariate. Independent samples t test, a Mann-Whitney U test, and a χ² test were used to determine statistical significance of parameters between the group with CDI and the group without CDI.

Prolonged LOS was defined either as a LOS > 75th or > 90th percentile of LOS among all patients with SS. To identify if CDI was associated with a prolonged LOS after adjusting for patient and hospital characteristics, a multivariate logistic regression analysis was used. Variables included in the regression model were age, gender, race, Charlson-Deyo Comorbidity Index, hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Data on cases were available for all the above covariates except hospital characteristics, such as teaching status, location, and bed size (these were missing for 0.7% of hospitals).

Stata 13.1.0 (Stata Corp, College Station, TX) and SPSS 23.0 (SPSS Inc., Chicago, IL) were used to perform statistical analyses. A P value of <0.05 was considered statistically significant.

RESULTS

Demographics

A total of 2,031,739 hospitalizations of adults with SS were identified between 2007 and 2013. CDI was present in 166,432 (8.2%) of these patients. Demographic data are displayed in Table 1. CDI was more commonly observed in elderly patients (> 65 years) with SS; 9.3% among the elderly versus 6.6% among individuals < 65 years; P < 0.001. The prevalence of CDI was greater in urban than in rural hospitals (8.4% vs 5.4%; P < 0.001) and greater in teaching than in nonteaching hospitals (8.7% vs 7.7%; P < 0.001). The prevalence of CDI in SS remained stable between 2007 and 2013 (Table 2).

Mortality

In the overall study cohort, the in-hospital mortality for SS was 37%. The in-hospital mortality rate of patients with SS complicated by a CDI was comparable to the mortality rate of patients without a CDI (37.1% vs 37.0%; P = 0.48). The mortality of patients with SS, with or without CDI, progressively decreased from 2007 to 2013 (P value for trend < 0.001 for each group; Figure 1).

Length of Stay

The median LOS for all patients with SS was 9 days. Patients with CDI had a longer median LOS than did those without CDI (13 vs 9 days; P < 0.001). Between 2007 and 2013, the median LOS of CDI group decreased from 14 to 12 days (P < 0.001) while that of non-CDI group decreased from 9 to 8 days (P < 0.001; Figure 2). We also examined LOS among subgroups who were discharged alive and those who died during hospitalization. For patients who were discharged alive, the LOS with and without CDI was 15 days versus 10 days, respectively (P < 0.001). For patients who died during hospitalization, LOS with and without CDI was 10 days versus 6 days, respectively (P < 0.001).

The 75th percentile of LOS of the total SS cohort was 17 days. An LOS > 17 days was observed in 36.9% of SS patients with CDI versus 22.7% without CDI (P < 0.001). After adjusting for patient and provider level variables, the odds of a LOS > 17 days were significantly greater for SS patients with CDI (odds ratio [OR] 2.11; 95% confidence interval [CI], 2.06-2.15; P < 0.001).

The 90th percentile of LOS of the total SS cohort was 29 days. An LOS > 29 days was observed in 17.5% of SS patients with a CDI versus 9.1% without a CDI (P < 0.001). After adjustment for patient and provider level variables, the odds of a LOS > 29 days were significantly greater for SS patients with a CDI (OR 2.25; 95% CI, 2.22-2.28; P < 0.001).

Hospital Readmission

In 2013, patients with SS and CDI had a higher rate of 30-day readmission as compared to patients with SS without CDI (9.8% vs 7.4% respectively; P < 0.001). The multivariate adjusted OR for 30-day readmission for patients with SS and a CDI was 1.26 (95% CI, 1.22-1.31; P < 0.001).

 

Additional Analyses

Lastly, we performed an additional analysis to confirm our hypothesis that a CDI by itself is rarely a cause of SS, and that CDI as the principal diagnosis would constitute an extremely low number of patients with SS in an administrative dataset. In NIS 2013, there were 105,750 cases with CDI as the primary diagnosis. A total of 4470 (4.2%) had a secondary diagnosis of sepsis and only 930 (0.9%) cases had a secondary diagnosis of SS.

DISCUSSION

This is the first study to report on the prevalence and outcome of CDI complicating SS. By using a large nationally representative sample, we found CDI was very prevalent among individuals hospitalized with SS and, at a level in excess of that seen in other populations. Of interest, we did not observe an increase in mortality of SS when complicated by CDI. On the other hand, patients with SS complicated by CDI were more much likely to have a prolonged hospital LOS and a higher risk of 30-day hospital readmission.

The prevalence of CDI exploded between the mid-1990s and mid-2000s, including community, hospital, and intensive care unit (ICU)–related disease.^6,7,17-20 Patients with SS often have multiple risk factors associated with CDI and thus represent a high-risk population for developing CDI.⁷ Our findings are consistent with the suggestion that individuals with SS are at a higher risk of developing CDI. Compared to the rate of CDI in all hospitalized patients, our data suggest an almost 10-fold increase in CDI rate for patients with SS.⁶ Patients with SS and CDI may account for as much as 10% of total CDIs.^6,7 As has been reported for CDI in general, we observed that CDI complicating SS was more common in those > 65 years of age.^4,21 The prevalence of CDI we observed in patients with SS was also higher than has been reported in ICU patients in general (1%), and higher than reported for patients requiring mechanical ventilation (6.6%), including prolonged mechanical ventilation (5.3%); further supporting the conclusion that patients with SS are a particularly high-risk group for acquiring CDI, even compared with other ICU patients.^20,22,23 Similarly, the rate of CDI among SS was 8 times higher than that of recently reported hospital-onset CDI among patients with sepsis in general (incidence 1.08%).²⁴ We have no data regarding why patients with SS have a higher rate of CDI; however, the intensity and duration of antibiotic treatment of these patients may certainly play a role.²⁵ It has recently been reported that CDI in itself can be a precursor leading to intestinal dysbiosis that can increase the risk of subsequent sepsis. Similarly, patients with SS may have higher prevalence of dysbiosis that, in turn, might predispose them to CDI at a higher rate than other individuals.

Following the increase in CDIs in the mid-1990s and the mid-2000s, since 2007 the overall prevalence of CDIs has been stable, albeit at the higher rate. More recently, the Centers for Disease Control and Prevention (CDC) has reported a decrease in hospital onset CDI after 2011.²⁶

The finding that CDI in SS patients was not associated with an increase in mortality is consistent with other reports of CDI in ICU patients in general as well as higher-risk ICU populations such as patients requiring mechanical ventilation, including those on long-term mechanical ventilator support.^{17,18,20,22,23} Why the mortality of ICU patients with CDI is not increased is not completely clear. It has been suggested that this may be related to early recognition and treatment of CDI developing in the ICU.²² Along these lines, it has been previously observed that for patients with CDI on mechanical ventilation, patients who were transferred to the ICU from the ward had worse clinical outcomes compared to patients directly admitted to the ICU, likely due to delayed recognition and treatment in the former.²² Similarly, ICU patients in whom CDI was identified prior to ICU admission had more severe CDI, and mortality that was directly related to CDI was only observed in patients who had CDI identified pre-ICU transfer.¹⁸ The increase in mortality observed in patients with sepsis in general with CDI may reflect similar factors.²⁴ We observed a trend of decreasing mortality in SS patients with or without CDI during 2007 to 2013 consistent to what has been generally reported in SS.^13,14

The increase in LOS observed in SS patients with CDI is also consistent with what has been observed in other ICU populations, as well as in patients with sepsis in general.^17,22-24 Of note, in addition to the increase in median LOS, we found a significant increase in the number of patients with a prolonged LOS associated with having SS with CDI. It is important to note that development of CDI during hospitalization is affected by pre-CDI hospital LOS, so prolonged LOS may not be solely attributable to CDI. The interaction between LOS and CDI remains complex in which higher LOS might be associated with higher incidence of CDI occurrence, and once established, CDI might be associated with changes in LOS for the remaining hospitalization.

Hospitalized patients with CDI have an overall higher resource utilization than those without CDI.²⁷ A recent review has estimated the overall attributable cost of CDI to be $6.3 billion; the attributable cost per case of hospital acquired CDI being 1.5 times the cost of community-acquired CDI.⁵ We did not look at cost directly. However, in the high-CDI risk ICU population requiring prolonged mechanical ventilation, those with CDI had a substantial increase in total costs.²³ Given the substantial increase in LOS associated with CDI complicating SS, there would likely be a significant increase in hospital costs related to providing care for these patients. Further adding to the potential burden of CDI is our finding that CDI and SS was associated with an increase in 30-day hospital readmission rate. This is consistent with a recent report that ICU patients with CDI who are discharged from the hospital have a 25% 30-day hospital readmission rate.²⁸ However, we do not have data either as to the reason for hospital readmission or whether the initial CDI or CDI recurrence played a role. This suggests that, in addition to intervention directed toward preventing CDI, efforts should be directed towards identifying factors that can be modified in CDI patients prior to or after hospital discharge.

This study has several limitations. Using an administrative database (such as NIS) has an inherent limitation of coding errors and reporting bias can lead to misclassification of cohort definition (SS) and outcome (CDI). To minimize bias due to coding errors, we used previously validated ICD-9-CM codes and approach to identify individuals with SS and CDI.^13-15 Although the SS population was identified with ICD-9-CM codes using an administrative database, the in-hospital mortality for our septic population was similar to previously reported mortality of SS, suggesting the population selected was appropriate.¹³ SS due to CDI could not be identified; however, CDI by itself causing SS is rare, as described in recent literature.^29,30 An important potential bias that needs to be acknowledged is the immortal time bias. The occurrence of CDI in itself can be influenced by pre-CDI hospital LOS. Patients who were extremely sick could have died early in their hospital course before they could acquire CDI, which would influence the mortality difference between the group with CDI and group without CDI. Furthermore, we did not have information on either the treatment of CDI or SS or any measures of severity of illness, which could lead to residual confounding despite adjusting for multiple variables. In terms of readmission data, it was necessary to exclude nonresidents of a state for the 30-day readmission analysis, as readmissions could not be tracked across state boundaries by using the NRD. This might have resulted in an underrepresentation of the readmission burden. Lastly, it was not possible to identify mortality after hospital discharge as the NIS provides only in-hospital mortality.

In conclusion, CDI is more prevalent in SS than are other ICU populations or the hospital population in general, and CDI complicating SS is associated with significant increase in LOS and risk of 30-day hospital readmission. How much of the increase in resource utilization and cost are in fact attributable to CDI in this population remains to be studied. Our finding of high prevalence of CDI in the SS population further emphasizes the importance of maintaining and furthering approaches to reduce incidence of hospital acquired CDI. While reducing unnecessary antibiotics is important, a multipronged approach that includes education and infection control interventions has also been shown to reduce the incidence of CDI in the ICU.³¹ Given the economic burden of CDI, implementing these strategies to reduce CDI is warranted. Similarly, the risk of 30-day hospital readmission with CDI highlights the importance of identifying the factors that contribute to hospital readmission prior to initial hospital discharge. Programs to reduce CDI will not only improve outcomes directly attributable to CDI but also decrease the reservoir of CDI. Finally, to the extent that CDI can be reduced in the ICU, the utilization of ICU resources will be more effective.

 

Disclosure

No conflicts of interest or financial disclosures to report. Author Contributions: KC had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis, including and especially any adverse effects. KC, AG, AC, KK, and HC contributed to study design, data analysis, interpretation, and the writing of the manuscript. Guarantor statement: Kshitij Chatterjee takes responsibility for (is the guarantor of) the content of the manuscript, including the data and analysis.

Clostridium difficile infection (CDI) is the most common infectious cause of healthcare-associated diarrhea.¹ Development of a CDI during hospitalization is associated with increases in morbidity, mortality, length of stay (LOS), and cost.^2-5 The prevalence of CDI in hospitalized patients has increased dramatically from the mid-1990s to the mid-2000s to almost 9 cases per 1000 discharges; however, the CDI rate since 2007 appears to have plateaued.^6,7 Antibiotic use has historically been the most important risk factor for acquiring CDI; however, use of acid-suppressing agents, chemotherapy, chronic comorbidities, and healthcare exposure all also increase the risk of CDI.^7-10 The elderly (> 65 years of age) are particularly at risk for developing CDI and having worse clinical outcomes with CDI.^6,7

Patients with septic shock (SS) often have multiple CDI risk factors (in particular, extensive antibiotic exposure) and thus, represent a population at a particularly high risk for acquiring a CDI during hospitalization. However, little data are available on the prevalence of CDI acquired in patients hospitalized with SS. We sought to determine the national-level temporal trends in the prevalence of CDI in patients with SS and the impact of CDI complicating SS on clinical outcomes between 2007 and 2013.

METHODS

Data Source

We used the National Inpatient Sample (NIS) and Nationwide Readmissions Database (NRD) for this study. The NIS is a database developed by the Agency of Healthcare Research and Quality for the Healthcare Cost and Utilization Project (HCUP).¹¹ It is the largest all-payer inpatient database in the United States and has been used by researchers and policy makers to analyze national trends in outcomes and healthcare utilization. The NIS database now approximates a 20% stratified sample of all discharges from all participating US hospitals. Sampling weights are provided by the manufacturer and can be used to produce national-level estimates. Following the redesign of the NIS in 2012, new sampling weights were provided for trend analysis for the years prior to 2012 to account for the new design. Every hospitalization is deidentified and converted into one unique entry that provides information on demographics, hospital characteristics, 1 primary and up to 24 secondary discharge diagnoses, comorbidities, LOS, in-hospital mortality, and procedures performed during stay. The discharge diagnoses are provided in the form of the International Classification of Diseases, 9th Revision-Clinical Modification (ICD-9-CM) codes.

The NRD is a database developed for HCUP that contains about 35 million discharges each year and supports readmission data analyses. In 2013, the NRD contained data from 21 geographically diverse states, accounting for 49.1% of all US hospitalizations. Diagnosis, comorbidities, and outcomes are presented in a similar manner to NIS.

Study Design

This was a retrospective cohort study. Data from the NIS between 2007 and 2013 were used for the analysis. Demographic data obtained included age, gender, race, Charlson-Deyo Comorbidity Index,¹² hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Cases with information missing on key demographic variables (age, gender, and race) were excluded. Only adults (>18 years of age) were included for the analysis.

SS was identified by either (1) ICD-9-CM diagnosis code for SS (785.52) or (2) presence of vasopressor use (00.17) along with ICD-9-CM codes of sepsis, severe sepsis, septicemia, bacteremia, or fungemia. This approach is consistent with what has been utilized in other studies to identify cases of sepsis or SS from administrative databases.^13-15 The appendix provides a complete list of ICD-9-CM codes used in the study. CDI was identified by ICD-9-CM code 008.45 among the secondary diagnosis. This code has been shown to have good accuracy for identifying CDI using administrative data.¹⁶ To minimize the inclusion of cases in which a CDI was present at admission, hospitalizations with a primary diagnosis of CDI were not included as cases of CDI complicating SS.

We used NRD 2013 for estimating the effect of CDI on 30-day readmission after initial hospitalizations with SS. We used the criteria for index admissions and 30-day readmissions as defined by the Centers for Medicare and Medicaid Services. We excluded patients who died during their index admission, patients with index discharges in December due to a lack of sufficient time to capture 30-day readmissions, and patients with missing information on key variables. We also excluded patients who were not a resident of the state of index hospitalization since readmission across state boundaries could not be identified in NRD. Manufacturer provided sampling weights were used to produce national level estimates. The cases of SS and CDI were identified by ICD-9-CM codes using the methodology described above.

 

Outcomes

Our primary outcome of interest was the total and yearly prevalence of CDI in patients with SS from 2007 to 2013. The secondary outcomes were mortality, LOS, and 30-day readmissions in patients with SS with and without CDI.

Statistical Analysis

Weighted data from NIS were used for all analyses. Demographics, hospital characteristics, and outcomes of all patients with SS were obtained. The prevalence of CDI was calculated for each calendar year. The temporal trends of outcomes (LOS and in-hospital mortality) of patients were plotted for patients with SS with and without CDI. A χ² test of trend for proportions was used with the Cochran-Armitage test to calculate statistical significance of changes in prevalence. To test for statistical significance of the temporal trends of LOS, a univariate linear regression was used, with calendar year as a covariate. Independent samples t test, a Mann-Whitney U test, and a χ² test were used to determine statistical significance of parameters between the group with CDI and the group without CDI.

Prolonged LOS was defined either as a LOS > 75th or > 90th percentile of LOS among all patients with SS. To identify if CDI was associated with a prolonged LOS after adjusting for patient and hospital characteristics, a multivariate logistic regression analysis was used. Variables included in the regression model were age, gender, race, Charlson-Deyo Comorbidity Index, hospital characteristics (hospital region, hospital-bed size, urban versus rural location, and teaching status), calendar year, and use of mechanical ventilation. Data on cases were available for all the above covariates except hospital characteristics, such as teaching status, location, and bed size (these were missing for 0.7% of hospitals).

Stata 13.1.0 (Stata Corp, College Station, TX) and SPSS 23.0 (SPSS Inc., Chicago, IL) were used to perform statistical analyses. A P value of <0.05 was considered statistically significant.

RESULTS

Demographics

A total of 2,031,739 hospitalizations of adults with SS were identified between 2007 and 2013. CDI was present in 166,432 (8.2%) of these patients. Demographic data are displayed in Table 1. CDI was more commonly observed in elderly patients (> 65 years) with SS; 9.3% among the elderly versus 6.6% among individuals < 65 years; P < 0.001. The prevalence of CDI was greater in urban than in rural hospitals (8.4% vs 5.4%; P < 0.001) and greater in teaching than in nonteaching hospitals (8.7% vs 7.7%; P < 0.001). The prevalence of CDI in SS remained stable between 2007 and 2013 (Table 2).

Mortality

In the overall study cohort, the in-hospital mortality for SS was 37%. The in-hospital mortality rate of patients with SS complicated by a CDI was comparable to the mortality rate of patients without a CDI (37.1% vs 37.0%; P = 0.48). The mortality of patients with SS, with or without CDI, progressively decreased from 2007 to 2013 (P value for trend < 0.001 for each group; Figure 1).

Length of Stay

The median LOS for all patients with SS was 9 days. Patients with CDI had a longer median LOS than did those without CDI (13 vs 9 days; P < 0.001). Between 2007 and 2013, the median LOS of CDI group decreased from 14 to 12 days (P < 0.001) while that of non-CDI group decreased from 9 to 8 days (P < 0.001; Figure 2). We also examined LOS among subgroups who were discharged alive and those who died during hospitalization. For patients who were discharged alive, the LOS with and without CDI was 15 days versus 10 days, respectively (P < 0.001). For patients who died during hospitalization, LOS with and without CDI was 10 days versus 6 days, respectively (P < 0.001).

The 75th percentile of LOS of the total SS cohort was 17 days. An LOS > 17 days was observed in 36.9% of SS patients with CDI versus 22.7% without CDI (P < 0.001). After adjusting for patient and provider level variables, the odds of a LOS > 17 days were significantly greater for SS patients with CDI (odds ratio [OR] 2.11; 95% confidence interval [CI], 2.06-2.15; P < 0.001).

The 90th percentile of LOS of the total SS cohort was 29 days. An LOS > 29 days was observed in 17.5% of SS patients with a CDI versus 9.1% without a CDI (P < 0.001). After adjustment for patient and provider level variables, the odds of a LOS > 29 days were significantly greater for SS patients with a CDI (OR 2.25; 95% CI, 2.22-2.28; P < 0.001).

Hospital Readmission

In 2013, patients with SS and CDI had a higher rate of 30-day readmission as compared to patients with SS without CDI (9.8% vs 7.4% respectively; P < 0.001). The multivariate adjusted OR for 30-day readmission for patients with SS and a CDI was 1.26 (95% CI, 1.22-1.31; P < 0.001).

 

Additional Analyses

Lastly, we performed an additional analysis to confirm our hypothesis that a CDI by itself is rarely a cause of SS, and that CDI as the principal diagnosis would constitute an extremely low number of patients with SS in an administrative dataset. In NIS 2013, there were 105,750 cases with CDI as the primary diagnosis. A total of 4470 (4.2%) had a secondary diagnosis of sepsis and only 930 (0.9%) cases had a secondary diagnosis of SS.

DISCUSSION

This is the first study to report on the prevalence and outcome of CDI complicating SS. By using a large nationally representative sample, we found CDI was very prevalent among individuals hospitalized with SS and, at a level in excess of that seen in other populations. Of interest, we did not observe an increase in mortality of SS when complicated by CDI. On the other hand, patients with SS complicated by CDI were more much likely to have a prolonged hospital LOS and a higher risk of 30-day hospital readmission.

The prevalence of CDI exploded between the mid-1990s and mid-2000s, including community, hospital, and intensive care unit (ICU)–related disease.^6,7,17-20 Patients with SS often have multiple risk factors associated with CDI and thus represent a high-risk population for developing CDI.⁷ Our findings are consistent with the suggestion that individuals with SS are at a higher risk of developing CDI. Compared to the rate of CDI in all hospitalized patients, our data suggest an almost 10-fold increase in CDI rate for patients with SS.⁶ Patients with SS and CDI may account for as much as 10% of total CDIs.^6,7 As has been reported for CDI in general, we observed that CDI complicating SS was more common in those > 65 years of age.^4,21 The prevalence of CDI we observed in patients with SS was also higher than has been reported in ICU patients in general (1%), and higher than reported for patients requiring mechanical ventilation (6.6%), including prolonged mechanical ventilation (5.3%); further supporting the conclusion that patients with SS are a particularly high-risk group for acquiring CDI, even compared with other ICU patients.^20,22,23 Similarly, the rate of CDI among SS was 8 times higher than that of recently reported hospital-onset CDI among patients with sepsis in general (incidence 1.08%).²⁴ We have no data regarding why patients with SS have a higher rate of CDI; however, the intensity and duration of antibiotic treatment of these patients may certainly play a role.²⁵ It has recently been reported that CDI in itself can be a precursor leading to intestinal dysbiosis that can increase the risk of subsequent sepsis. Similarly, patients with SS may have higher prevalence of dysbiosis that, in turn, might predispose them to CDI at a higher rate than other individuals.

Following the increase in CDIs in the mid-1990s and the mid-2000s, since 2007 the overall prevalence of CDIs has been stable, albeit at the higher rate. More recently, the Centers for Disease Control and Prevention (CDC) has reported a decrease in hospital onset CDI after 2011.²⁶

The finding that CDI in SS patients was not associated with an increase in mortality is consistent with other reports of CDI in ICU patients in general as well as higher-risk ICU populations such as patients requiring mechanical ventilation, including those on long-term mechanical ventilator support.^{17,18,20,22,23} Why the mortality of ICU patients with CDI is not increased is not completely clear. It has been suggested that this may be related to early recognition and treatment of CDI developing in the ICU.²² Along these lines, it has been previously observed that for patients with CDI on mechanical ventilation, patients who were transferred to the ICU from the ward had worse clinical outcomes compared to patients directly admitted to the ICU, likely due to delayed recognition and treatment in the former.²² Similarly, ICU patients in whom CDI was identified prior to ICU admission had more severe CDI, and mortality that was directly related to CDI was only observed in patients who had CDI identified pre-ICU transfer.¹⁸ The increase in mortality observed in patients with sepsis in general with CDI may reflect similar factors.²⁴ We observed a trend of decreasing mortality in SS patients with or without CDI during 2007 to 2013 consistent to what has been generally reported in SS.^13,14

The increase in LOS observed in SS patients with CDI is also consistent with what has been observed in other ICU populations, as well as in patients with sepsis in general.^17,22-24 Of note, in addition to the increase in median LOS, we found a significant increase in the number of patients with a prolonged LOS associated with having SS with CDI. It is important to note that development of CDI during hospitalization is affected by pre-CDI hospital LOS, so prolonged LOS may not be solely attributable to CDI. The interaction between LOS and CDI remains complex in which higher LOS might be associated with higher incidence of CDI occurrence, and once established, CDI might be associated with changes in LOS for the remaining hospitalization.

Hospitalized patients with CDI have an overall higher resource utilization than those without CDI.²⁷ A recent review has estimated the overall attributable cost of CDI to be $6.3 billion; the attributable cost per case of hospital acquired CDI being 1.5 times the cost of community-acquired CDI.⁵ We did not look at cost directly. However, in the high-CDI risk ICU population requiring prolonged mechanical ventilation, those with CDI had a substantial increase in total costs.²³ Given the substantial increase in LOS associated with CDI complicating SS, there would likely be a significant increase in hospital costs related to providing care for these patients. Further adding to the potential burden of CDI is our finding that CDI and SS was associated with an increase in 30-day hospital readmission rate. This is consistent with a recent report that ICU patients with CDI who are discharged from the hospital have a 25% 30-day hospital readmission rate.²⁸ However, we do not have data either as to the reason for hospital readmission or whether the initial CDI or CDI recurrence played a role. This suggests that, in addition to intervention directed toward preventing CDI, efforts should be directed towards identifying factors that can be modified in CDI patients prior to or after hospital discharge.

This study has several limitations. Using an administrative database (such as NIS) has an inherent limitation of coding errors and reporting bias can lead to misclassification of cohort definition (SS) and outcome (CDI). To minimize bias due to coding errors, we used previously validated ICD-9-CM codes and approach to identify individuals with SS and CDI.^13-15 Although the SS population was identified with ICD-9-CM codes using an administrative database, the in-hospital mortality for our septic population was similar to previously reported mortality of SS, suggesting the population selected was appropriate.¹³ SS due to CDI could not be identified; however, CDI by itself causing SS is rare, as described in recent literature.^29,30 An important potential bias that needs to be acknowledged is the immortal time bias. The occurrence of CDI in itself can be influenced by pre-CDI hospital LOS. Patients who were extremely sick could have died early in their hospital course before they could acquire CDI, which would influence the mortality difference between the group with CDI and group without CDI. Furthermore, we did not have information on either the treatment of CDI or SS or any measures of severity of illness, which could lead to residual confounding despite adjusting for multiple variables. In terms of readmission data, it was necessary to exclude nonresidents of a state for the 30-day readmission analysis, as readmissions could not be tracked across state boundaries by using the NRD. This might have resulted in an underrepresentation of the readmission burden. Lastly, it was not possible to identify mortality after hospital discharge as the NIS provides only in-hospital mortality.

In conclusion, CDI is more prevalent in SS than are other ICU populations or the hospital population in general, and CDI complicating SS is associated with significant increase in LOS and risk of 30-day hospital readmission. How much of the increase in resource utilization and cost are in fact attributable to CDI in this population remains to be studied. Our finding of high prevalence of CDI in the SS population further emphasizes the importance of maintaining and furthering approaches to reduce incidence of hospital acquired CDI. While reducing unnecessary antibiotics is important, a multipronged approach that includes education and infection control interventions has also been shown to reduce the incidence of CDI in the ICU.³¹ Given the economic burden of CDI, implementing these strategies to reduce CDI is warranted. Similarly, the risk of 30-day hospital readmission with CDI highlights the importance of identifying the factors that contribute to hospital readmission prior to initial hospital discharge. Programs to reduce CDI will not only improve outcomes directly attributable to CDI but also decrease the reservoir of CDI. Finally, to the extent that CDI can be reduced in the ICU, the utilization of ICU resources will be more effective.

 

Disclosure

No conflicts of interest or financial disclosures to report. Author Contributions: KC had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis, including and especially any adverse effects. KC, AG, AC, KK, and HC contributed to study design, data analysis, interpretation, and the writing of the manuscript. Guarantor statement: Kshitij Chatterjee takes responsibility for (is the guarantor of) the content of the manuscript, including the data and analysis.

Medication reconciliation at hospital discharge is a critical component of the posthospital transition of care.¹ Effective reconciliation involves a clear process for documenting a current medication list, identifying and resolving discrepancies, and then documenting decisions and instructions around which medications should be continued, modified, or stopped.² Existing studies^3-5 suggest that medication discrepancies are common during hospital discharge transitions of care and lead to preventable adverse drug events, patient disability, and increased healthcare utilization following hospital discharge, including physician office visits, emergency department (ED) visits, and hospitalizations.^6-8

While the majority of studies of medication discrepancies have been conducted in general medical patients, few have examined how gaps in discharge medication reconciliation might affect surgical patients.^9,10 Two prior studies^9,10 suggest that medication discrepancies may occur more frequently for surgical patients, compared with medical patients, particularly discrepancies in reordering home medications postoperatively, raising patient safety concerns for more than 50 million patients hospitalized for surgery each year.¹¹ In particular, little is known about the appropriate discharge reconciliation of chronic cardiovascular medications, such as beta-blockers, renin-angiotensin system inhibitors, and statins in surgical patients, despite perioperative practice guidelines recommending continuation or rapid reinitiation of these medications after noncardiac surgery.¹² Problems with chronic cardiovascular medications have been implicated as major contributors to ED visits and hospitalizations for adverse drug events,^13,14 further highlighting the importance of safe and appropriate management of these medications.

To better understand the current state and impact of postoperative discharge medication reconciliation of chronic cardiovascular medications in surgical patients, we examined (1) the appropriate discharge reconciliation of 4 cardiovascular medication classes, and (2) the associations between the appropriate discharge reconciliation of these medication classes and postdischarge acute hospital and ambulatory visits in patients hospitalized for elective noncardiac surgery at an academic medical center.

METHODS

Study Design and Patient Selection

We performed a retrospective analysis of data collected as part of a cohort study of hospitalized surgical patients admitted between January 2007 and December 2011. The original study was designed to assess the impact of a social marketing intervention on guideline-appropriate perioperative beta-blocker use in surgical patients. The study was conducted at 1 academic medical center that had 2 campuses with full noncardiac operative facilities and a 600-bed total capacity. Both sites had preoperative clinics, and patients were recruited by review of preoperative clinic records. Institutional review boards responsible for all sites approved the study.

For this analysis, we included adults (age >18 years) undergoing elective noncardiac surgery, who were expected to remain hospitalized for at least 1 day and were taking antiplatelet agents (aspirin, aspirin-dipyridamole, or clopidogrel), beta-blockers, renin-angiotensin system inhibitors (angiotensin-converting-enzyme inhibitors or angiotensin-receptor blockers), or statin lipid-lowering agents.

Data Collection

Data Sources. We collected data from a structured review of medical records as well as from discharge abstract information obtained from administrative data systems. Data regarding patient demographics (age, sex, and race/ethnicity), medical history, preoperative cardiovascular medications, surgical procedure and service, and attending surgeon were obtained from a medical record review of comprehensive preoperative clinic evaluations. Data regarding complications during hospitalization were obtained from medical record review and administrative data (Supplement for International Classification of Diseases, Ninth Revision codes).¹⁵ Research assistants abstracting data were trained by using a comprehensive reference manual providing specific criteria for classifying chart abstraction data. Research assistants also were directly observed during initial chart abstractions and underwent random chart validation audits by a senior investigator to ensure consistency. Any discrepancies in coding were resolved by consensus among senior investigators.

Definition of Key Predictor: Appropriate Reconciliation. We abstracted discharge medication lists from the electronic medical record. We defined the appropriate reconciliation of cardiovascular medications at discharge as documentation in discharge instructions, medication reconciliation tools, or discharge summaries that a preadmission cardiovascular medication was being continued at discharge, or, if the medication was not continued, documentation of a new contraindication to the medication or complication precluding its use during hospitalization. Medication continuity was considered appropriate if it was continued at discharge irrespective of changes in dosage. By using this measure for individual medications, we also assessed appropriate reconciliation as an “all-or-none” complete versus incomplete measure (appropriate reconciliation of all preoperative cardiovascular medication classes the patient was taking).¹⁶

Definition of Outcomes. Our coprimary outcomes were acute hospital visits (ED visits or hospitalizations) and unplanned ambulatory visits (primary care or surgical) at 30 days after surgery. Postoperative ambulatory visits that were not planned prior to surgery were defined as unplanned. Outcomes were ascertained by patient reports during follow-up telephone questionnaires administered by trained research staff and verified by medical record review.

Definition of Covariates. Using these data, we calculated a Revised Cardiac Risk Index (RCRI) score,¹⁷ which estimates the risk of perioperative cardiac complications in patients undergoing surgery. Through chart abstraction data supplemented by diagnosis codes from administrative data, we also constructed variables indicating occurrences of postoperative complications anytime during hospitalization that might pose contraindications to continuation of the 4 cardiovascular medication classes studied. For example, if a chart indicated that the patient had an acute rise in creatinine (elevation of baseline creatinine by 50% or absolute rise of 1 mg/dL in patients with baseline creatinine greater than 3 mg/dL) during hospitalization and a preoperative renin-angiotensin system inhibitor was not prescribed at discharge, we would have considered discontinuation appropriate. Other complications we abstracted were hypotension (systolic blood pressure less than 90 mmHg) for beta-blockers and renin-angiotensin system inhibitors, bradycardia (heart rate less than 50 bpm) for beta-blockers, acute kidney injury (defined above) and hyperkalemia for renin-angiotensin system inhibitors, and bleeding (any site) for antiplatelet agents.

 

Statistical Analysis

We used χ² and Kruskal-Wallis tests to compare baseline patient characteristics. To assess associations between appropriate medication reconciliation and patient outcomes, we used multilevel mixed-effects logistic regression to account for the clustering of patients by the attending surgeon. We adjusted for baseline patient demographics, surgical service, the number of baseline cardiovascular medications, and individual RCRI criteria. We constructed separate models for all-or-none appropriate reconciliation and for each individual medication class.

As a sensitivity analysis, we constructed similar models by using a simplified definition of appropriate reconciliation based entirely on medication continuity (continued or not continued at discharge) without taking potential contraindications during hospitalization into account. For complete versus incomplete reconciliation, we also constructed models with an interaction term between the number of baseline cardiovascular medications and appropriate medication reconciliation to test the hypothesis that inappropriate reconciliation would be more likely with an increasing number of preoperative cardiovascular medications. Because this interaction term was not statistically significant, we did not include it in the final models for ease of reporting and interpretability. We performed all statistical analyses using Stata 14 (StataCorp, LLC, College Station, Texas), and used 2-sided statistical tests and a P value of less than .05 to define statistical significance.

RESULTS

Patient Characteristics

A total of 849 patients were enrolled, of which 752 (88.6%) were taking at least 1 of the specified cardiovascular medications in the preoperative period. Their mean age was 61.5; 50.9% were male, 72.6% were non-Hispanic white, and 89.4% had RCRI scores of 0 or 1 (Table 1). The majority (63.8%) were undergoing general surgery, orthopedic surgery, or neurosurgery procedures. In the preoperative period, 327 (43.5%) patients were taking antiplatelet agents, 624 (83.0%) were taking beta-blockers, 361 (48.0%) were taking renin-angiotensin system inhibitors, and 406 (54.0%) were taking statins (Table 2). Among patients taking antiplatelet agents, 271 (82.9%) were taking aspirin alone, 21 (6.4%) were taking clopidogrel alone, and 35 (10.7%) were taking dual antiplatelet therapy with aspirin and clopidogrel. Nearly three-quarters of the patients (551, 73.3%) were taking medications from 2 or more classes, and the proportion of patients with inappropriate reconciliation increased with the number of preoperative cardiovascular medications.

Patients with and without appropriate reconciliation of all preoperative cardiovascular medications were similar in age, sex, and race/ethnicity (Table 1). Patients with inappropriate reconciliation of at least 1 medication were more likely to be on the urology and renal/liver transplant surgical services, have higher RCRI scores, and be taking antiplatelet agents, statins, renin-angiotensin system inhibitors, and 3 or more cardiovascular medications in the preoperative period.

Appropriate Medication Reconciliation

Four hundred thirty-six patients (58.0%) had their baseline cardiovascular medications appropriately reconciled. Among all patients with appropriately reconciled medications, 1 (0.2%) had beta-blockers discontinued due to a documented episode of hypotension; 17 (3.9%) had renin-angiotensin system inhibitors discontinued due to episodes of acute kidney injury, hypotension, or hyperkalemia; and 1 (0.2%) had antiplatelet agents discontinued due to bleeding. For individual medications, appropriate reconciliation between the preoperative and discharge periods occurred for 156 of the 327 patients on antiplatelet agents (47.7%), 507 of the 624 patients on beta-blockers (81.3%), 259 of the 361 patients on renin-angiotensin system inhibitors (71.8%), and 302 of the 406 patients on statins (74.4%; Table 2).

Associations Between Medication Reconciliation and Outcomes

Thirty-day outcome data on acute hospital visits were available for 679 (90.3%) patients. Of these, 146 (21.5%) were seen in the ED or were hospitalized, and 111 (16.3%) were seen for an unplanned primary care or surgical outpatient visit at 30 days after surgery. Patients with incomplete outcome data were more likely to have complete medication reconciliation compared with those with complete outcome data (71.2% vs 56.6%, P = 0.02). As shown in Table 3, the proportion of patients with 30-day acute hospital visits was nonstatistically significantly lower in patients with complete medication reconciliation (20.8% vs 22.4%, P = 0.63) and the appropriate reconciliation of beta-blockers (21.9% vs 23.6%, P = 0.71) and renin-angiotensin system inhibitors (19.6% vs 20.0%, P = 0.93), and nonsignificantly higher with the appropriate reconciliation of antiplatelet agents (23.9% vs 19.9%, P = 0.40). Acute hospital visits were statistically significantly lower with the appropriate reconciliation of statins (17.9% vs 31.9%, P = 0.004).

In hierarchical multivariable models, complete appropriate medication reconciliation was not associated with acute hospital visits (adjusted odds ratio [AOR], 0.94; 95% confidence interval [CI], 0.63-1.41). For individual medications, appropriate reconciliation of statins was associated with lower odds of unplanned hospital visits (AOR, 0.47; 95% CI, 0.26-0.85), but there were no statistically significant associations between appropriate reconciliation of antiplatelet agents, beta-blockers, or renin-angiotensin system inhibitors and hospital visits (Table 3). Similarly, the proportion of patients with 30-day unplanned ambulatory visits was not statistically different among patients with complete reconciliation or appropriate reconciliation of individual medications (Table 4). Adjusted analyses were consistent with the unadjusted point estimates and demonstrated no statistically significant associations.

 

Sensitivity Analysis

Overall, 430 (57.2%) patients had complete cardiovascular medication continuity without considering potential contraindications during hospitalization. Associations between medication continuity and acute hospital and ambulatory visits were similar to the primary analyses.

DISCUSSION

In this study of 752 patients hospitalized for elective noncardiac surgery, we found significant gaps in the appropriate reconciliation of commonly prescribed cardiovascular medications, with inappropriate discontinuation ranging from 18.8% to 52.3% for individual medications. Unplanned postdischarge healthcare utilization was high, with acute hospital visits documented in 21.5% of patients and unplanned ambulatory visits in 16.3% at 30 days after surgery. However, medication reconciliation gaps were not consistently associated with ED visits, hospitalizations, or unplanned ambulatory visits.

Our finding of large gaps in postoperative medication reconciliation is consistent with existing studies of medication reconciliation in surgical patients.^9,10,18 One study found medication discrepancies in 40.2% of postoperative patients receiving usual care and discrepancies judged to have the potential to cause harm (such as the omission of beta-blockers) in 29.9%.⁹ Consistent with our findings, this study also found that most postoperative medication discrepancies were omissions in reordering home medications, though at a rate somewhat higher than those seen in medical patients at discharge.⁵ While hospitalization by itself increases the risk of unintentional discontinuation of chronic medications,³ our results, along with existing literature, suggest that the risk for omission of chronic medications is unacceptably high.

We also found significant variation in reconciliation among cardiovascular medications, with appropriate reconciliation occurring least frequently for antiplatelet agents and most frequently for beta-blockers. The low rates of appropriate reconciliation for antiplatelet agents may be attributable to deliberate withholding of antiplatelet therapy in the postoperative period based on clinical assessments of surgical bleeding risk in the absence of active bleeding. Perioperative management of antiplatelet agents for noncardiac surgery remains an unclear and controversial topic, which may also contribute to the variation noted.¹⁹ Conversely, beta-blockers demonstrated high rates of preoperative use (over 80% of patients) and appropriate reconciliation. Both findings are likely attributable in part to the timing of the study, which began prior to the publication of the Perioperative Ischemic Evaluation trial, which more definitively demonstrated the potential harms of perioperative beta-blocker therapy.²⁰

Despite a high proportion of patients with discontinuous medications at discharge, we found no associations between the appropriate reconciliation of beta-blockers, renin-angiotensin system inhibitors, and antiplatelet agents and acute hospital or ambulatory visits in the first 30 days after discharge. One explanation for this discrepancy is that, although we focused on cardiovascular medications commonly implicated in acute hospital visits, the vast majority of patients in our study had low perioperative cardiovascular risk as assessed by the RCRI. Previous studies have demonstrated that the benefit of perioperative beta-blocker therapy is predominantly in patients with moderate to high perioperative cardiovascular risk.^21,22 It is possible that the detrimental effects of the discontinuation of chronic cardiovascular medications are more prominent in populations at a higher risk of perioperative cardiovascular complications or that complications will occur later than 30 days after discharge. Similarly, while the benefits of continuation of renin-angiotensin system inhibitors are less clear,²³ few patients in our cohort had a history of congestive heart failure (6.3%) or coronary artery disease (13.0%), 2 conditions in which the impact of perioperative discontinuation of renin-angiotensin inhibitor or beta-blocker therapy would likely be more pronounced.^24,25 An additional explanation for the lack of associations is that, while multiple studies have demonstrated that medication errors are common, the proportion of errors with the potential for harm is much lower, and the proportion that causes actual harm is lower still.^5,26,27 Thus, while we likely captured high-severity medication errors leading to acute hospital or unplanned ambulatory visits, we would not have captured medication errors with lower severity clinical consequences that did not result in medical encounters.

We did find an association between the continuation of statin therapy and reduced ED visits and hospitalizations. This finding is supported by previous studies of patients undergoing noncardiac surgery, including 1 demonstrating an association between immediate postoperative statin therapy and reduced in-hospital mortality²⁸ and another study demonstrating an association between postoperative statin therapy and reductions in a composite endpoint of 30-day mortality, atrial fibrillation, and nonfatal myocardial infarction.²⁹ Alternatively, this finding could reflect the effects of unaddressed confounding by factors contributing to statin discontinuation and poor health outcomes leading to acute hospital visits, such as acute elevations in liver enzymes.

Our study has important implications for patients undergoing elective noncardiac surgery and the healthcare providers caring for them. First, inappropriate omissions of chronic cardiovascular medications at discharge are common; clinicians should increase their general awareness and focus on appropriately reconciling these medications, for even if our results do not connect medication discontinuity to readmissions or unexpected clinical encounters, their impact on patients’ understanding of their medications remains a potential concern. Second, the overall high rates of unplanned postdischarge healthcare utilization in this study highlight the need for close postdischarge monitoring of patients undergoing elective surgical procedures and for further research to identify preventable etiologies of postdischarge healthcare utilization in this population. Third, further study is needed to identify specific patient populations and medication classes, in which appropriate reconciliation is associated with patient outcomes that may benefit from more intensive discharge medication reconciliation interventions.

Our study has limitations. First, the majority of patients in this single-center study were at low risk of perioperative cardiovascular events, and our results may not be generalizable to higher-risk patients undergoing elective surgery. Second, discharge reconciliation was based on documentation of medication reconciliation and not on patient-reported medication adherence. In addition, the ability to judge the accuracy of discharge medication reconciliation is in part dependent on the accuracy of the admission medication reconciliation. Thus, although we used preoperative medication regimens documented during preadmission visits to comprehensive preoperative clinics for comparison, discrepancies in these preoperative regimens could have affected our analysis of appropriate discharge reconciliation. Third, inadequate documentation of clinical reasons for discontinuing medications may have led to residual confounding by indication in our observational study. Finally, the outcomes available to us may have been relatively insensitive to other adverse effects of medication discontinuity, such as patient symptoms (eg, angina severity), patient awareness of medications, or work placed on primary care physicians needing to “clean up” erroneous medication lists.

In conclusion, gaps in appropriate discharge reconciliation of chronic cardiovascular medications were common but not consistently associated with postdischarge acute hospital or unplanned ambulatory visits in a relatively low-risk cohort of patients undergoing elective surgery. While appropriate medication reconciliation should always be a priority, further study is needed to identify medication reconciliation approaches associated with postdischarge healthcare utilization and other patient outcomes.

 

Disclosure

Dr. Lee reports receiving grant support from the Health Resources and Services Administration (T32HP19025). Dr. Vittinghoff reports receiving grant support from the Agency for Healthcare Research and Quality. Dr. Auerbach and Dr. Fleischmann report receiving grant support from the National Institutes of Health. Dr. Auerbach also reports receiving honorarium as Editor-in-Chief of the Journal of Hospital Medicine. Dr. Corbett reports receiving grant and travel support from Simon Fraser University. The remaining authors have no disclosures to report.

Medication reconciliation at hospital discharge is a critical component of the posthospital transition of care.¹ Effective reconciliation involves a clear process for documenting a current medication list, identifying and resolving discrepancies, and then documenting decisions and instructions around which medications should be continued, modified, or stopped.² Existing studies^3-5 suggest that medication discrepancies are common during hospital discharge transitions of care and lead to preventable adverse drug events, patient disability, and increased healthcare utilization following hospital discharge, including physician office visits, emergency department (ED) visits, and hospitalizations.^6-8

While the majority of studies of medication discrepancies have been conducted in general medical patients, few have examined how gaps in discharge medication reconciliation might affect surgical patients.^9,10 Two prior studies^9,10 suggest that medication discrepancies may occur more frequently for surgical patients, compared with medical patients, particularly discrepancies in reordering home medications postoperatively, raising patient safety concerns for more than 50 million patients hospitalized for surgery each year.¹¹ In particular, little is known about the appropriate discharge reconciliation of chronic cardiovascular medications, such as beta-blockers, renin-angiotensin system inhibitors, and statins in surgical patients, despite perioperative practice guidelines recommending continuation or rapid reinitiation of these medications after noncardiac surgery.¹² Problems with chronic cardiovascular medications have been implicated as major contributors to ED visits and hospitalizations for adverse drug events,^13,14 further highlighting the importance of safe and appropriate management of these medications.

To better understand the current state and impact of postoperative discharge medication reconciliation of chronic cardiovascular medications in surgical patients, we examined (1) the appropriate discharge reconciliation of 4 cardiovascular medication classes, and (2) the associations between the appropriate discharge reconciliation of these medication classes and postdischarge acute hospital and ambulatory visits in patients hospitalized for elective noncardiac surgery at an academic medical center.

METHODS

Study Design and Patient Selection

We performed a retrospective analysis of data collected as part of a cohort study of hospitalized surgical patients admitted between January 2007 and December 2011. The original study was designed to assess the impact of a social marketing intervention on guideline-appropriate perioperative beta-blocker use in surgical patients. The study was conducted at 1 academic medical center that had 2 campuses with full noncardiac operative facilities and a 600-bed total capacity. Both sites had preoperative clinics, and patients were recruited by review of preoperative clinic records. Institutional review boards responsible for all sites approved the study.

For this analysis, we included adults (age >18 years) undergoing elective noncardiac surgery, who were expected to remain hospitalized for at least 1 day and were taking antiplatelet agents (aspirin, aspirin-dipyridamole, or clopidogrel), beta-blockers, renin-angiotensin system inhibitors (angiotensin-converting-enzyme inhibitors or angiotensin-receptor blockers), or statin lipid-lowering agents.

Data Collection

Data Sources. We collected data from a structured review of medical records as well as from discharge abstract information obtained from administrative data systems. Data regarding patient demographics (age, sex, and race/ethnicity), medical history, preoperative cardiovascular medications, surgical procedure and service, and attending surgeon were obtained from a medical record review of comprehensive preoperative clinic evaluations. Data regarding complications during hospitalization were obtained from medical record review and administrative data (Supplement for International Classification of Diseases, Ninth Revision codes).¹⁵ Research assistants abstracting data were trained by using a comprehensive reference manual providing specific criteria for classifying chart abstraction data. Research assistants also were directly observed during initial chart abstractions and underwent random chart validation audits by a senior investigator to ensure consistency. Any discrepancies in coding were resolved by consensus among senior investigators.

Definition of Key Predictor: Appropriate Reconciliation. We abstracted discharge medication lists from the electronic medical record. We defined the appropriate reconciliation of cardiovascular medications at discharge as documentation in discharge instructions, medication reconciliation tools, or discharge summaries that a preadmission cardiovascular medication was being continued at discharge, or, if the medication was not continued, documentation of a new contraindication to the medication or complication precluding its use during hospitalization. Medication continuity was considered appropriate if it was continued at discharge irrespective of changes in dosage. By using this measure for individual medications, we also assessed appropriate reconciliation as an “all-or-none” complete versus incomplete measure (appropriate reconciliation of all preoperative cardiovascular medication classes the patient was taking).¹⁶

Definition of Outcomes. Our coprimary outcomes were acute hospital visits (ED visits or hospitalizations) and unplanned ambulatory visits (primary care or surgical) at 30 days after surgery. Postoperative ambulatory visits that were not planned prior to surgery were defined as unplanned. Outcomes were ascertained by patient reports during follow-up telephone questionnaires administered by trained research staff and verified by medical record review.

Definition of Covariates. Using these data, we calculated a Revised Cardiac Risk Index (RCRI) score,¹⁷ which estimates the risk of perioperative cardiac complications in patients undergoing surgery. Through chart abstraction data supplemented by diagnosis codes from administrative data, we also constructed variables indicating occurrences of postoperative complications anytime during hospitalization that might pose contraindications to continuation of the 4 cardiovascular medication classes studied. For example, if a chart indicated that the patient had an acute rise in creatinine (elevation of baseline creatinine by 50% or absolute rise of 1 mg/dL in patients with baseline creatinine greater than 3 mg/dL) during hospitalization and a preoperative renin-angiotensin system inhibitor was not prescribed at discharge, we would have considered discontinuation appropriate. Other complications we abstracted were hypotension (systolic blood pressure less than 90 mmHg) for beta-blockers and renin-angiotensin system inhibitors, bradycardia (heart rate less than 50 bpm) for beta-blockers, acute kidney injury (defined above) and hyperkalemia for renin-angiotensin system inhibitors, and bleeding (any site) for antiplatelet agents.

 

Statistical Analysis

We used χ² and Kruskal-Wallis tests to compare baseline patient characteristics. To assess associations between appropriate medication reconciliation and patient outcomes, we used multilevel mixed-effects logistic regression to account for the clustering of patients by the attending surgeon. We adjusted for baseline patient demographics, surgical service, the number of baseline cardiovascular medications, and individual RCRI criteria. We constructed separate models for all-or-none appropriate reconciliation and for each individual medication class.

As a sensitivity analysis, we constructed similar models by using a simplified definition of appropriate reconciliation based entirely on medication continuity (continued or not continued at discharge) without taking potential contraindications during hospitalization into account. For complete versus incomplete reconciliation, we also constructed models with an interaction term between the number of baseline cardiovascular medications and appropriate medication reconciliation to test the hypothesis that inappropriate reconciliation would be more likely with an increasing number of preoperative cardiovascular medications. Because this interaction term was not statistically significant, we did not include it in the final models for ease of reporting and interpretability. We performed all statistical analyses using Stata 14 (StataCorp, LLC, College Station, Texas), and used 2-sided statistical tests and a P value of less than .05 to define statistical significance.

RESULTS

Patient Characteristics

A total of 849 patients were enrolled, of which 752 (88.6%) were taking at least 1 of the specified cardiovascular medications in the preoperative period. Their mean age was 61.5; 50.9% were male, 72.6% were non-Hispanic white, and 89.4% had RCRI scores of 0 or 1 (Table 1). The majority (63.8%) were undergoing general surgery, orthopedic surgery, or neurosurgery procedures. In the preoperative period, 327 (43.5%) patients were taking antiplatelet agents, 624 (83.0%) were taking beta-blockers, 361 (48.0%) were taking renin-angiotensin system inhibitors, and 406 (54.0%) were taking statins (Table 2). Among patients taking antiplatelet agents, 271 (82.9%) were taking aspirin alone, 21 (6.4%) were taking clopidogrel alone, and 35 (10.7%) were taking dual antiplatelet therapy with aspirin and clopidogrel. Nearly three-quarters of the patients (551, 73.3%) were taking medications from 2 or more classes, and the proportion of patients with inappropriate reconciliation increased with the number of preoperative cardiovascular medications.

Patients with and without appropriate reconciliation of all preoperative cardiovascular medications were similar in age, sex, and race/ethnicity (Table 1). Patients with inappropriate reconciliation of at least 1 medication were more likely to be on the urology and renal/liver transplant surgical services, have higher RCRI scores, and be taking antiplatelet agents, statins, renin-angiotensin system inhibitors, and 3 or more cardiovascular medications in the preoperative period.

Appropriate Medication Reconciliation

Four hundred thirty-six patients (58.0%) had their baseline cardiovascular medications appropriately reconciled. Among all patients with appropriately reconciled medications, 1 (0.2%) had beta-blockers discontinued due to a documented episode of hypotension; 17 (3.9%) had renin-angiotensin system inhibitors discontinued due to episodes of acute kidney injury, hypotension, or hyperkalemia; and 1 (0.2%) had antiplatelet agents discontinued due to bleeding. For individual medications, appropriate reconciliation between the preoperative and discharge periods occurred for 156 of the 327 patients on antiplatelet agents (47.7%), 507 of the 624 patients on beta-blockers (81.3%), 259 of the 361 patients on renin-angiotensin system inhibitors (71.8%), and 302 of the 406 patients on statins (74.4%; Table 2).

Associations Between Medication Reconciliation and Outcomes

Thirty-day outcome data on acute hospital visits were available for 679 (90.3%) patients. Of these, 146 (21.5%) were seen in the ED or were hospitalized, and 111 (16.3%) were seen for an unplanned primary care or surgical outpatient visit at 30 days after surgery. Patients with incomplete outcome data were more likely to have complete medication reconciliation compared with those with complete outcome data (71.2% vs 56.6%, P = 0.02). As shown in Table 3, the proportion of patients with 30-day acute hospital visits was nonstatistically significantly lower in patients with complete medication reconciliation (20.8% vs 22.4%, P = 0.63) and the appropriate reconciliation of beta-blockers (21.9% vs 23.6%, P = 0.71) and renin-angiotensin system inhibitors (19.6% vs 20.0%, P = 0.93), and nonsignificantly higher with the appropriate reconciliation of antiplatelet agents (23.9% vs 19.9%, P = 0.40). Acute hospital visits were statistically significantly lower with the appropriate reconciliation of statins (17.9% vs 31.9%, P = 0.004).

In hierarchical multivariable models, complete appropriate medication reconciliation was not associated with acute hospital visits (adjusted odds ratio [AOR], 0.94; 95% confidence interval [CI], 0.63-1.41). For individual medications, appropriate reconciliation of statins was associated with lower odds of unplanned hospital visits (AOR, 0.47; 95% CI, 0.26-0.85), but there were no statistically significant associations between appropriate reconciliation of antiplatelet agents, beta-blockers, or renin-angiotensin system inhibitors and hospital visits (Table 3). Similarly, the proportion of patients with 30-day unplanned ambulatory visits was not statistically different among patients with complete reconciliation or appropriate reconciliation of individual medications (Table 4). Adjusted analyses were consistent with the unadjusted point estimates and demonstrated no statistically significant associations.

 

Sensitivity Analysis

Overall, 430 (57.2%) patients had complete cardiovascular medication continuity without considering potential contraindications during hospitalization. Associations between medication continuity and acute hospital and ambulatory visits were similar to the primary analyses.

DISCUSSION

In this study of 752 patients hospitalized for elective noncardiac surgery, we found significant gaps in the appropriate reconciliation of commonly prescribed cardiovascular medications, with inappropriate discontinuation ranging from 18.8% to 52.3% for individual medications. Unplanned postdischarge healthcare utilization was high, with acute hospital visits documented in 21.5% of patients and unplanned ambulatory visits in 16.3% at 30 days after surgery. However, medication reconciliation gaps were not consistently associated with ED visits, hospitalizations, or unplanned ambulatory visits.

Our finding of large gaps in postoperative medication reconciliation is consistent with existing studies of medication reconciliation in surgical patients.^9,10,18 One study found medication discrepancies in 40.2% of postoperative patients receiving usual care and discrepancies judged to have the potential to cause harm (such as the omission of beta-blockers) in 29.9%.⁹ Consistent with our findings, this study also found that most postoperative medication discrepancies were omissions in reordering home medications, though at a rate somewhat higher than those seen in medical patients at discharge.⁵ While hospitalization by itself increases the risk of unintentional discontinuation of chronic medications,³ our results, along with existing literature, suggest that the risk for omission of chronic medications is unacceptably high.

We also found significant variation in reconciliation among cardiovascular medications, with appropriate reconciliation occurring least frequently for antiplatelet agents and most frequently for beta-blockers. The low rates of appropriate reconciliation for antiplatelet agents may be attributable to deliberate withholding of antiplatelet therapy in the postoperative period based on clinical assessments of surgical bleeding risk in the absence of active bleeding. Perioperative management of antiplatelet agents for noncardiac surgery remains an unclear and controversial topic, which may also contribute to the variation noted.¹⁹ Conversely, beta-blockers demonstrated high rates of preoperative use (over 80% of patients) and appropriate reconciliation. Both findings are likely attributable in part to the timing of the study, which began prior to the publication of the Perioperative Ischemic Evaluation trial, which more definitively demonstrated the potential harms of perioperative beta-blocker therapy.²⁰

Despite a high proportion of patients with discontinuous medications at discharge, we found no associations between the appropriate reconciliation of beta-blockers, renin-angiotensin system inhibitors, and antiplatelet agents and acute hospital or ambulatory visits in the first 30 days after discharge. One explanation for this discrepancy is that, although we focused on cardiovascular medications commonly implicated in acute hospital visits, the vast majority of patients in our study had low perioperative cardiovascular risk as assessed by the RCRI. Previous studies have demonstrated that the benefit of perioperative beta-blocker therapy is predominantly in patients with moderate to high perioperative cardiovascular risk.^21,22 It is possible that the detrimental effects of the discontinuation of chronic cardiovascular medications are more prominent in populations at a higher risk of perioperative cardiovascular complications or that complications will occur later than 30 days after discharge. Similarly, while the benefits of continuation of renin-angiotensin system inhibitors are less clear,²³ few patients in our cohort had a history of congestive heart failure (6.3%) or coronary artery disease (13.0%), 2 conditions in which the impact of perioperative discontinuation of renin-angiotensin inhibitor or beta-blocker therapy would likely be more pronounced.^24,25 An additional explanation for the lack of associations is that, while multiple studies have demonstrated that medication errors are common, the proportion of errors with the potential for harm is much lower, and the proportion that causes actual harm is lower still.^5,26,27 Thus, while we likely captured high-severity medication errors leading to acute hospital or unplanned ambulatory visits, we would not have captured medication errors with lower severity clinical consequences that did not result in medical encounters.

We did find an association between the continuation of statin therapy and reduced ED visits and hospitalizations. This finding is supported by previous studies of patients undergoing noncardiac surgery, including 1 demonstrating an association between immediate postoperative statin therapy and reduced in-hospital mortality²⁸ and another study demonstrating an association between postoperative statin therapy and reductions in a composite endpoint of 30-day mortality, atrial fibrillation, and nonfatal myocardial infarction.²⁹ Alternatively, this finding could reflect the effects of unaddressed confounding by factors contributing to statin discontinuation and poor health outcomes leading to acute hospital visits, such as acute elevations in liver enzymes.

Our study has important implications for patients undergoing elective noncardiac surgery and the healthcare providers caring for them. First, inappropriate omissions of chronic cardiovascular medications at discharge are common; clinicians should increase their general awareness and focus on appropriately reconciling these medications, for even if our results do not connect medication discontinuity to readmissions or unexpected clinical encounters, their impact on patients’ understanding of their medications remains a potential concern. Second, the overall high rates of unplanned postdischarge healthcare utilization in this study highlight the need for close postdischarge monitoring of patients undergoing elective surgical procedures and for further research to identify preventable etiologies of postdischarge healthcare utilization in this population. Third, further study is needed to identify specific patient populations and medication classes, in which appropriate reconciliation is associated with patient outcomes that may benefit from more intensive discharge medication reconciliation interventions.

Our study has limitations. First, the majority of patients in this single-center study were at low risk of perioperative cardiovascular events, and our results may not be generalizable to higher-risk patients undergoing elective surgery. Second, discharge reconciliation was based on documentation of medication reconciliation and not on patient-reported medication adherence. In addition, the ability to judge the accuracy of discharge medication reconciliation is in part dependent on the accuracy of the admission medication reconciliation. Thus, although we used preoperative medication regimens documented during preadmission visits to comprehensive preoperative clinics for comparison, discrepancies in these preoperative regimens could have affected our analysis of appropriate discharge reconciliation. Third, inadequate documentation of clinical reasons for discontinuing medications may have led to residual confounding by indication in our observational study. Finally, the outcomes available to us may have been relatively insensitive to other adverse effects of medication discontinuity, such as patient symptoms (eg, angina severity), patient awareness of medications, or work placed on primary care physicians needing to “clean up” erroneous medication lists.

In conclusion, gaps in appropriate discharge reconciliation of chronic cardiovascular medications were common but not consistently associated with postdischarge acute hospital or unplanned ambulatory visits in a relatively low-risk cohort of patients undergoing elective surgery. While appropriate medication reconciliation should always be a priority, further study is needed to identify medication reconciliation approaches associated with postdischarge healthcare utilization and other patient outcomes.

 

Disclosure

Dr. Lee reports receiving grant support from the Health Resources and Services Administration (T32HP19025). Dr. Vittinghoff reports receiving grant support from the Agency for Healthcare Research and Quality. Dr. Auerbach and Dr. Fleischmann report receiving grant support from the National Institutes of Health. Dr. Auerbach also reports receiving honorarium as Editor-in-Chief of the Journal of Hospital Medicine. Dr. Corbett reports receiving grant and travel support from Simon Fraser University. The remaining authors have no disclosures to report.

Medication reconciliation at hospital discharge is a critical component of the posthospital transition of care.¹ Effective reconciliation involves a clear process for documenting a current medication list, identifying and resolving discrepancies, and then documenting decisions and instructions around which medications should be continued, modified, or stopped.² Existing studies^3-5 suggest that medication discrepancies are common during hospital discharge transitions of care and lead to preventable adverse drug events, patient disability, and increased healthcare utilization following hospital discharge, including physician office visits, emergency department (ED) visits, and hospitalizations.^6-8

While the majority of studies of medication discrepancies have been conducted in general medical patients, few have examined how gaps in discharge medication reconciliation might affect surgical patients.^9,10 Two prior studies^9,10 suggest that medication discrepancies may occur more frequently for surgical patients, compared with medical patients, particularly discrepancies in reordering home medications postoperatively, raising patient safety concerns for more than 50 million patients hospitalized for surgery each year.¹¹ In particular, little is known about the appropriate discharge reconciliation of chronic cardiovascular medications, such as beta-blockers, renin-angiotensin system inhibitors, and statins in surgical patients, despite perioperative practice guidelines recommending continuation or rapid reinitiation of these medications after noncardiac surgery.¹² Problems with chronic cardiovascular medications have been implicated as major contributors to ED visits and hospitalizations for adverse drug events,^13,14 further highlighting the importance of safe and appropriate management of these medications.

To better understand the current state and impact of postoperative discharge medication reconciliation of chronic cardiovascular medications in surgical patients, we examined (1) the appropriate discharge reconciliation of 4 cardiovascular medication classes, and (2) the associations between the appropriate discharge reconciliation of these medication classes and postdischarge acute hospital and ambulatory visits in patients hospitalized for elective noncardiac surgery at an academic medical center.

METHODS

Study Design and Patient Selection

We performed a retrospective analysis of data collected as part of a cohort study of hospitalized surgical patients admitted between January 2007 and December 2011. The original study was designed to assess the impact of a social marketing intervention on guideline-appropriate perioperative beta-blocker use in surgical patients. The study was conducted at 1 academic medical center that had 2 campuses with full noncardiac operative facilities and a 600-bed total capacity. Both sites had preoperative clinics, and patients were recruited by review of preoperative clinic records. Institutional review boards responsible for all sites approved the study.

For this analysis, we included adults (age >18 years) undergoing elective noncardiac surgery, who were expected to remain hospitalized for at least 1 day and were taking antiplatelet agents (aspirin, aspirin-dipyridamole, or clopidogrel), beta-blockers, renin-angiotensin system inhibitors (angiotensin-converting-enzyme inhibitors or angiotensin-receptor blockers), or statin lipid-lowering agents.

Data Collection

Data Sources. We collected data from a structured review of medical records as well as from discharge abstract information obtained from administrative data systems. Data regarding patient demographics (age, sex, and race/ethnicity), medical history, preoperative cardiovascular medications, surgical procedure and service, and attending surgeon were obtained from a medical record review of comprehensive preoperative clinic evaluations. Data regarding complications during hospitalization were obtained from medical record review and administrative data (Supplement for International Classification of Diseases, Ninth Revision codes).¹⁵ Research assistants abstracting data were trained by using a comprehensive reference manual providing specific criteria for classifying chart abstraction data. Research assistants also were directly observed during initial chart abstractions and underwent random chart validation audits by a senior investigator to ensure consistency. Any discrepancies in coding were resolved by consensus among senior investigators.

Definition of Key Predictor: Appropriate Reconciliation. We abstracted discharge medication lists from the electronic medical record. We defined the appropriate reconciliation of cardiovascular medications at discharge as documentation in discharge instructions, medication reconciliation tools, or discharge summaries that a preadmission cardiovascular medication was being continued at discharge, or, if the medication was not continued, documentation of a new contraindication to the medication or complication precluding its use during hospitalization. Medication continuity was considered appropriate if it was continued at discharge irrespective of changes in dosage. By using this measure for individual medications, we also assessed appropriate reconciliation as an “all-or-none” complete versus incomplete measure (appropriate reconciliation of all preoperative cardiovascular medication classes the patient was taking).¹⁶

Definition of Outcomes. Our coprimary outcomes were acute hospital visits (ED visits or hospitalizations) and unplanned ambulatory visits (primary care or surgical) at 30 days after surgery. Postoperative ambulatory visits that were not planned prior to surgery were defined as unplanned. Outcomes were ascertained by patient reports during follow-up telephone questionnaires administered by trained research staff and verified by medical record review.

Definition of Covariates. Using these data, we calculated a Revised Cardiac Risk Index (RCRI) score,¹⁷ which estimates the risk of perioperative cardiac complications in patients undergoing surgery. Through chart abstraction data supplemented by diagnosis codes from administrative data, we also constructed variables indicating occurrences of postoperative complications anytime during hospitalization that might pose contraindications to continuation of the 4 cardiovascular medication classes studied. For example, if a chart indicated that the patient had an acute rise in creatinine (elevation of baseline creatinine by 50% or absolute rise of 1 mg/dL in patients with baseline creatinine greater than 3 mg/dL) during hospitalization and a preoperative renin-angiotensin system inhibitor was not prescribed at discharge, we would have considered discontinuation appropriate. Other complications we abstracted were hypotension (systolic blood pressure less than 90 mmHg) for beta-blockers and renin-angiotensin system inhibitors, bradycardia (heart rate less than 50 bpm) for beta-blockers, acute kidney injury (defined above) and hyperkalemia for renin-angiotensin system inhibitors, and bleeding (any site) for antiplatelet agents.

 

Statistical Analysis

We used χ² and Kruskal-Wallis tests to compare baseline patient characteristics. To assess associations between appropriate medication reconciliation and patient outcomes, we used multilevel mixed-effects logistic regression to account for the clustering of patients by the attending surgeon. We adjusted for baseline patient demographics, surgical service, the number of baseline cardiovascular medications, and individual RCRI criteria. We constructed separate models for all-or-none appropriate reconciliation and for each individual medication class.

As a sensitivity analysis, we constructed similar models by using a simplified definition of appropriate reconciliation based entirely on medication continuity (continued or not continued at discharge) without taking potential contraindications during hospitalization into account. For complete versus incomplete reconciliation, we also constructed models with an interaction term between the number of baseline cardiovascular medications and appropriate medication reconciliation to test the hypothesis that inappropriate reconciliation would be more likely with an increasing number of preoperative cardiovascular medications. Because this interaction term was not statistically significant, we did not include it in the final models for ease of reporting and interpretability. We performed all statistical analyses using Stata 14 (StataCorp, LLC, College Station, Texas), and used 2-sided statistical tests and a P value of less than .05 to define statistical significance.

RESULTS

Patient Characteristics

A total of 849 patients were enrolled, of which 752 (88.6%) were taking at least 1 of the specified cardiovascular medications in the preoperative period. Their mean age was 61.5; 50.9% were male, 72.6% were non-Hispanic white, and 89.4% had RCRI scores of 0 or 1 (Table 1). The majority (63.8%) were undergoing general surgery, orthopedic surgery, or neurosurgery procedures. In the preoperative period, 327 (43.5%) patients were taking antiplatelet agents, 624 (83.0%) were taking beta-blockers, 361 (48.0%) were taking renin-angiotensin system inhibitors, and 406 (54.0%) were taking statins (Table 2). Among patients taking antiplatelet agents, 271 (82.9%) were taking aspirin alone, 21 (6.4%) were taking clopidogrel alone, and 35 (10.7%) were taking dual antiplatelet therapy with aspirin and clopidogrel. Nearly three-quarters of the patients (551, 73.3%) were taking medications from 2 or more classes, and the proportion of patients with inappropriate reconciliation increased with the number of preoperative cardiovascular medications.

Patients with and without appropriate reconciliation of all preoperative cardiovascular medications were similar in age, sex, and race/ethnicity (Table 1). Patients with inappropriate reconciliation of at least 1 medication were more likely to be on the urology and renal/liver transplant surgical services, have higher RCRI scores, and be taking antiplatelet agents, statins, renin-angiotensin system inhibitors, and 3 or more cardiovascular medications in the preoperative period.

Appropriate Medication Reconciliation

Four hundred thirty-six patients (58.0%) had their baseline cardiovascular medications appropriately reconciled. Among all patients with appropriately reconciled medications, 1 (0.2%) had beta-blockers discontinued due to a documented episode of hypotension; 17 (3.9%) had renin-angiotensin system inhibitors discontinued due to episodes of acute kidney injury, hypotension, or hyperkalemia; and 1 (0.2%) had antiplatelet agents discontinued due to bleeding. For individual medications, appropriate reconciliation between the preoperative and discharge periods occurred for 156 of the 327 patients on antiplatelet agents (47.7%), 507 of the 624 patients on beta-blockers (81.3%), 259 of the 361 patients on renin-angiotensin system inhibitors (71.8%), and 302 of the 406 patients on statins (74.4%; Table 2).

Associations Between Medication Reconciliation and Outcomes

Thirty-day outcome data on acute hospital visits were available for 679 (90.3%) patients. Of these, 146 (21.5%) were seen in the ED or were hospitalized, and 111 (16.3%) were seen for an unplanned primary care or surgical outpatient visit at 30 days after surgery. Patients with incomplete outcome data were more likely to have complete medication reconciliation compared with those with complete outcome data (71.2% vs 56.6%, P = 0.02). As shown in Table 3, the proportion of patients with 30-day acute hospital visits was nonstatistically significantly lower in patients with complete medication reconciliation (20.8% vs 22.4%, P = 0.63) and the appropriate reconciliation of beta-blockers (21.9% vs 23.6%, P = 0.71) and renin-angiotensin system inhibitors (19.6% vs 20.0%, P = 0.93), and nonsignificantly higher with the appropriate reconciliation of antiplatelet agents (23.9% vs 19.9%, P = 0.40). Acute hospital visits were statistically significantly lower with the appropriate reconciliation of statins (17.9% vs 31.9%, P = 0.004).

In hierarchical multivariable models, complete appropriate medication reconciliation was not associated with acute hospital visits (adjusted odds ratio [AOR], 0.94; 95% confidence interval [CI], 0.63-1.41). For individual medications, appropriate reconciliation of statins was associated with lower odds of unplanned hospital visits (AOR, 0.47; 95% CI, 0.26-0.85), but there were no statistically significant associations between appropriate reconciliation of antiplatelet agents, beta-blockers, or renin-angiotensin system inhibitors and hospital visits (Table 3). Similarly, the proportion of patients with 30-day unplanned ambulatory visits was not statistically different among patients with complete reconciliation or appropriate reconciliation of individual medications (Table 4). Adjusted analyses were consistent with the unadjusted point estimates and demonstrated no statistically significant associations.

 

Sensitivity Analysis

Overall, 430 (57.2%) patients had complete cardiovascular medication continuity without considering potential contraindications during hospitalization. Associations between medication continuity and acute hospital and ambulatory visits were similar to the primary analyses.

DISCUSSION

In this study of 752 patients hospitalized for elective noncardiac surgery, we found significant gaps in the appropriate reconciliation of commonly prescribed cardiovascular medications, with inappropriate discontinuation ranging from 18.8% to 52.3% for individual medications. Unplanned postdischarge healthcare utilization was high, with acute hospital visits documented in 21.5% of patients and unplanned ambulatory visits in 16.3% at 30 days after surgery. However, medication reconciliation gaps were not consistently associated with ED visits, hospitalizations, or unplanned ambulatory visits.

Our finding of large gaps in postoperative medication reconciliation is consistent with existing studies of medication reconciliation in surgical patients.^9,10,18 One study found medication discrepancies in 40.2% of postoperative patients receiving usual care and discrepancies judged to have the potential to cause harm (such as the omission of beta-blockers) in 29.9%.⁹ Consistent with our findings, this study also found that most postoperative medication discrepancies were omissions in reordering home medications, though at a rate somewhat higher than those seen in medical patients at discharge.⁵ While hospitalization by itself increases the risk of unintentional discontinuation of chronic medications,³ our results, along with existing literature, suggest that the risk for omission of chronic medications is unacceptably high.

We also found significant variation in reconciliation among cardiovascular medications, with appropriate reconciliation occurring least frequently for antiplatelet agents and most frequently for beta-blockers. The low rates of appropriate reconciliation for antiplatelet agents may be attributable to deliberate withholding of antiplatelet therapy in the postoperative period based on clinical assessments of surgical bleeding risk in the absence of active bleeding. Perioperative management of antiplatelet agents for noncardiac surgery remains an unclear and controversial topic, which may also contribute to the variation noted.¹⁹ Conversely, beta-blockers demonstrated high rates of preoperative use (over 80% of patients) and appropriate reconciliation. Both findings are likely attributable in part to the timing of the study, which began prior to the publication of the Perioperative Ischemic Evaluation trial, which more definitively demonstrated the potential harms of perioperative beta-blocker therapy.²⁰

Despite a high proportion of patients with discontinuous medications at discharge, we found no associations between the appropriate reconciliation of beta-blockers, renin-angiotensin system inhibitors, and antiplatelet agents and acute hospital or ambulatory visits in the first 30 days after discharge. One explanation for this discrepancy is that, although we focused on cardiovascular medications commonly implicated in acute hospital visits, the vast majority of patients in our study had low perioperative cardiovascular risk as assessed by the RCRI. Previous studies have demonstrated that the benefit of perioperative beta-blocker therapy is predominantly in patients with moderate to high perioperative cardiovascular risk.^21,22 It is possible that the detrimental effects of the discontinuation of chronic cardiovascular medications are more prominent in populations at a higher risk of perioperative cardiovascular complications or that complications will occur later than 30 days after discharge. Similarly, while the benefits of continuation of renin-angiotensin system inhibitors are less clear,²³ few patients in our cohort had a history of congestive heart failure (6.3%) or coronary artery disease (13.0%), 2 conditions in which the impact of perioperative discontinuation of renin-angiotensin inhibitor or beta-blocker therapy would likely be more pronounced.^24,25 An additional explanation for the lack of associations is that, while multiple studies have demonstrated that medication errors are common, the proportion of errors with the potential for harm is much lower, and the proportion that causes actual harm is lower still.^5,26,27 Thus, while we likely captured high-severity medication errors leading to acute hospital or unplanned ambulatory visits, we would not have captured medication errors with lower severity clinical consequences that did not result in medical encounters.

We did find an association between the continuation of statin therapy and reduced ED visits and hospitalizations. This finding is supported by previous studies of patients undergoing noncardiac surgery, including 1 demonstrating an association between immediate postoperative statin therapy and reduced in-hospital mortality²⁸ and another study demonstrating an association between postoperative statin therapy and reductions in a composite endpoint of 30-day mortality, atrial fibrillation, and nonfatal myocardial infarction.²⁹ Alternatively, this finding could reflect the effects of unaddressed confounding by factors contributing to statin discontinuation and poor health outcomes leading to acute hospital visits, such as acute elevations in liver enzymes.

Our study has important implications for patients undergoing elective noncardiac surgery and the healthcare providers caring for them. First, inappropriate omissions of chronic cardiovascular medications at discharge are common; clinicians should increase their general awareness and focus on appropriately reconciling these medications, for even if our results do not connect medication discontinuity to readmissions or unexpected clinical encounters, their impact on patients’ understanding of their medications remains a potential concern. Second, the overall high rates of unplanned postdischarge healthcare utilization in this study highlight the need for close postdischarge monitoring of patients undergoing elective surgical procedures and for further research to identify preventable etiologies of postdischarge healthcare utilization in this population. Third, further study is needed to identify specific patient populations and medication classes, in which appropriate reconciliation is associated with patient outcomes that may benefit from more intensive discharge medication reconciliation interventions.

Our study has limitations. First, the majority of patients in this single-center study were at low risk of perioperative cardiovascular events, and our results may not be generalizable to higher-risk patients undergoing elective surgery. Second, discharge reconciliation was based on documentation of medication reconciliation and not on patient-reported medication adherence. In addition, the ability to judge the accuracy of discharge medication reconciliation is in part dependent on the accuracy of the admission medication reconciliation. Thus, although we used preoperative medication regimens documented during preadmission visits to comprehensive preoperative clinics for comparison, discrepancies in these preoperative regimens could have affected our analysis of appropriate discharge reconciliation. Third, inadequate documentation of clinical reasons for discontinuing medications may have led to residual confounding by indication in our observational study. Finally, the outcomes available to us may have been relatively insensitive to other adverse effects of medication discontinuity, such as patient symptoms (eg, angina severity), patient awareness of medications, or work placed on primary care physicians needing to “clean up” erroneous medication lists.

In conclusion, gaps in appropriate discharge reconciliation of chronic cardiovascular medications were common but not consistently associated with postdischarge acute hospital or unplanned ambulatory visits in a relatively low-risk cohort of patients undergoing elective surgery. While appropriate medication reconciliation should always be a priority, further study is needed to identify medication reconciliation approaches associated with postdischarge healthcare utilization and other patient outcomes.

 

Disclosure

Dr. Lee reports receiving grant support from the Health Resources and Services Administration (T32HP19025). Dr. Vittinghoff reports receiving grant support from the Agency for Healthcare Research and Quality. Dr. Auerbach and Dr. Fleischmann report receiving grant support from the National Institutes of Health. Dr. Auerbach also reports receiving honorarium as Editor-in-Chief of the Journal of Hospital Medicine. Dr. Corbett reports receiving grant and travel support from Simon Fraser University. The remaining authors have no disclosures to report.

Influenza-associated morbidity poses a significant hospital burden.¹ A study from the United States estimated that seasonal influenza is responsible for 3.1 million hospitalization days per year.²

Assessment of hospital burden during influenza seasons presents a challenge due to several possible factors, such as inaccurate recording of diagnosis³ and incomplete age group data. Although great emphasis has historically been placed on older age groups, a study from England and Wales showed that the number of hospitalizations and deaths resulting from influenza was significantly higher in children as compared with adults.⁴ Moreover, excess visits to emergency departments in New York City because of fever and respiratory morbidity during influenza seasons were found mostly among school-age children, whereas in adults, the surplus was small to nonexistent.⁵

Studies examining influenza-related hospitalizations evaluated numbers and rates of hospitalization.^6-11 However, information regarding length of hospitalizations, hospitalizations during the influenza season that were not influenza related, or comparisons between influenza seasons and summer seasons is scarce. These determinants are of great importance for hospital preparedness towards influenza seasons. The aim of the current study was to estimate excess hospitalizations and length of hospitalization during influenza seasons, as compared with the summer, in different age groups and selected diagnoses in Israel.

METHODS

Data Sources

Hospitalization data of internal medicine and pediatric departments in 28 acute care hospitals in Israel between 2005 and 2013 were obtained from the National Hospital Discharges Database managed by the Health Information Division (HID) in the Israel Ministry of Health (MOH). The information included number of discharges (including in-hospital deaths), number of hospitalization days, and the mean length of stay (LOS) per discharge for all diagnoses and for primary or secondary diagnoses of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Bed occupancy rates for internal medicine and pediatric departments were based on the National Patient Flow Database managed by the HID.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Classification

Hospitalizations data were analyzed for all ages, for specific age groups (the first year of life [0], ages 1-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84, and 85 years and older), for all diagnoses, and for primary or secondary discharge diagnosis of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Duration of Influenza Season

The beginning and the end of the influenza season were determined by the National Influenza surveillance program, which includes on average 22 community sentinel clinics, throughout Israel, each influenza season. These clinics send nose-throat samples from a convenience sample of patients with influenza-like illness (ILI), from week 40 of each year until the end of the influenza season in the subsequent year. These samples are analyzed for the presence of influenza virus by real-time reverse transcription polymerase chain reaction (RT-PCR) at the Central Virology Laboratory of Israel. Based on influenza virus detection in nose-throat samples from patients with ILI attending the community sentinel clinics, we determined the first and last month of each influenza season. The first month in which positive influenza samples were identified in sequence was defined as the first month of the season. The month in which the sequence of positive influenza samples stopped was defined as the last month of the season.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Analysis

Rates. Rates of monthly hospitalizations and monthly hospitalization days were calculated per 100,000 residents for all ages and for the specific age groups. Estimated average population sizes in different years for all ages and for specific age groups were obtained from the Central Bureau of Statistics (http://www.cbs.gov.il/reader/shnaton/templ_shnaton.html?num_tab=st02_01&CYear=2014). Monthly LOS was not converted to rates.

 

Hospitalizations. Mean monthly rate of hospitalizations during influenza and summer seasons was calculated by dividing the sum of hospital discharge rates during influenza/summer seasons of the entire evaluation period (2005/2006 to 2012/2013) by the number of influenza/summer activity months of that period.

Hospitalization Days. The measure “hospitalization days” refers to the hospitalization days of all patients who were discharged during influenza seasons. Mean monthly rate of hospitalization days during the influenza season and summer season was calculated using the procedure described for monthly mean rate of hospitalizations.

Length of Stay. The measure “length of stay” refers to the number of days that individual patients stayed in the hospital during an admission in the evaluated seasons.

Mean monthly LOS during the influenza and summer seasons for all patients (in both internal medicine and pediatric departments) and by age group was calculated by dividing the sum of monthly LOS during influenza seasons/summer season of the entire evaluation period (2005/2006 to 2012/2013 except for the 2009/2010 season) by the number of influenza/summer activity months of that period.

LOS for each specific month of the evaluation period for a single patient was calculated by dividing the number of hospitalization days of all patients that were discharged that month (stratified by age group) by the number of discharges in the same month.

Bed Occupancy. Bed occupancy rates for internal medicine and pediatric departments of the seasons evaluated were computed as a weighted rate based on the hospitalization days and licensed inpatient beds for the period of each influenza and summer season. The calculation took into account the number of days of each month and was based on the monthly reporting of hospital inpatient days in these departments and on the number of inpatient beds according to standard license documents issued by the MOH for each hospital.

Difference Between Influenza and Summer Seasons. Differences in mean monthly rates of hospitalizations, mean monthly rate of hospitalization days, and LOS during influenza seasons and the preceding summer were calculated as absolute numbers per month and as a percentage. The difference between bed occupancy during the influenza seasons and the preceding summers was expressed in percentage. Differences were computed for all diagnoses and for ICD9 480-487 and 390-519.

Statistical Analysis

Mean and standard deviation for monthly hospitalization rates, rates of monthly hospitalization days, and for LOS were calculated for all the influenza and summer seasons that were evaluated. Differences and statistical significance for these parameters were evaluated using a two-tailed Wilcoxon-Mann-Whitney test adjusted for ties, with 95% confidence interval for mean locations. The null hypothesis of the Wilcoxon test used was that the mean ranks of the influenza and summer season observations were equal.

Mean of bed occupancy percentage was calculated for influenza and summer seasons, with the difference and statistical significance being evaluated using a χ² test. P value of < 0.05 was considered statistically significant. SAS Version 9.1 and R program version 3.3.1 software were used for analysis.

RESULTS

Influenza Seasons

The length of influenza seasons varied, with the shortest season lasting 3 months (2006-2007) and the longest season lasting six months (2010-2011 and 2011-2012; Table 1). Of the 14 first and last months of the 7 influenza seasons, 9 had influenza activity throughout the month, 2 had 3 weeks of influenza activity, and 3 had 2 weeks of influenza activity (Table 1).

Hospitalizations

A total of 452,209 hospital discharges occurred in pediatric and internal medicine departments during the influenza seasons that were evaluated. The mean monthly rate of hospitalizations (as defined in Methods) for all patients was 4.8% higher during influenza seasons as compared with the preceding summer seasons (panel A in Figure; Supplementary Table 1). Analysis by age groups revealed a statistically significant increase during influenza seasons in the younger and older age groups (panel A in Figure; Supplementary Table 1). Specifically, the increase for all diagnoses was 32.8%, 16.7%, and 14.1% for infants <1 year, children aged 1-4 years, and adults ≥85 years, respectively (panel A in Figure; Supplementary Table 1).

The mean monthly rate of hospitalizations for all ages due to the diagnosis of respiratory/cardiovascular diseases and influenza/pneumonia was 18.6% and 60.8% higher, respectively, during influenza seasons compared with the preceding summers (panels B and C in Figure). These differences were statistically significant (panels B and C in Figure; Supplementary Table 1).

The increase in mean monthly hospitalization rates for patients with a diagnosis of respiratory/cardiovascular diseases and pneumonia/influenza was highest among infants <1 year and children aged 1-4 years (panels B and C in Figure; Supplementary Table 1). Increases were also observed among other age groups. However, they were more modest and reached statistical significance for respiratory/cardiovascular diseases in the age groups of ≤34 years and ≥75 years (panel B in Figure; Supplementary Table 1). The increases in mean monthly hospitalization rates for pneumonia/influenza were statistically significant in all age groups and were greater than 40% among adults ≥55 years (panel C in Figure; Supplementary Table 1).

Statistically significant decreases in mean monthly hospitalization rates during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 1). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or pneumonia/influenza (panels B and C in Figure; Supplementary Table 1).

 

Hospitalization Days

The mean monthly rate of hospitalization days per 100,000 residents showed a similar trend to that of the hospitalization rates (panels A, B, and C in Figure; Supplementary Table 2), with the most prominent increases observed among infants and children <5 years and adults ≥65 years.

The mean monthly rate of hospitalization days per 100,000 during influenza seasons for all ages due to all diagnoses was 8% higher (P < 0.001) as compared with the summer seasons (panel A in Figure; Supplementary Table 2). Statistically significant increases were also found among patients diagnosed with respiratory/cardiovascular diseases and for influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

A significant increase was also observed among infants and children <5 years and adults ≥65 years with all diagnoses (panel A in Figure; Supplementary Table 2). The increase in the monthly mean rate of hospitalization days was statistically significant for respiratory/cardiovascular diseases in most age groups, except the 35-64 age groups (panel B in Figure; Supplementary Table 2). A statistically significant increase in the monthly mean rate of hospitalization days for influenza/pneumonia was seen in all age

groups except for the 15-24 age group (panel C in Figure; Supplementary Table 2).

Children <5 years of age showed the largest increases during the influenza season as compared with the summer, with an up to 155.9% increase in the mean monthly rate of hospitalization days due to influenza/pneumonia (panel C in Figure; Supplementary Table 2), and an up to 206.6% increase for respiratory/cardiovascular diseases in infants <1 year of age (panel B in Figure; Supplementary Table 2). In adults, the largest increases were observed among those ≥75 years; the rates for influenza/pneumonia increased by about 40% (panel C in Figure; Supplementary Table 2), and the rates for respiratory/cardiovascular diseases increased by 14.8%-20.7% as compared with the summer months (panel B in Figure; Supplementary Table 2).

Statistically significant decreases in monthly mean rate of hospitalization days during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 2). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

Hospital Length of Stay

The longest mean monthly LOS due to all diagnoses (for both influenza and summer seasons) was observed in adults ≥65 years of age (Table 2). The longest mean monthly LOS due to influenza/pneumonia (for both influenza and summer seasons) was observed in adults ≥55 years or older, and for the diagnosis of respiratory/cardiovascular diseases, infants <1 year and adults ≥55 years had the longest LOS.

The differences between influenza and summer seasons in mean monthly LOS were mostly small or not observed in any of the diagnostic categories examined. The mean monthly LOS due to a diagnosis of influenza/pneumonia was shorter during the influenza seasons than summer seasons in most age groups. These differences were statistically significant in children <5 years and adults ≥45 years (Table 2).

The mean LOS due to respiratory/cardiovascular diseases was significantly shorter during influenza seasons than summer seasons in children under 5.

Bed Occupancy

Mean bed occupancy was significantly higher during influenza seasons compared with the preceding summer seasons, both in internal medicine and pediatric departments (Table 3). The differences were higher in pediatric departments as compared with internal medicine departments for most years evaluated.

DISCUSSION

Our study demonstrates trends of excess hospitalizations during influenza as compared with summer seasons and identifies patient groups that contribute mostly to changes in hospital burden between these seasons.

Overall, the present study demonstrates differences between influenza and summer seasons for all measures tested: hospitalizations, hospitalization days, LOS, and bed occupancy. These differences were due primarily to excess number of hospitalizations and hospitalization days, rather than to longer LOS.

Our results concerning hospitalizations for all diagnoses are consistent with a United States report showing about 5% more hospitalizations following emergency department visits during winter compared with summer.¹²

The increase in hospitalizations and total hospitalization days in older age groups reflects the probability of severe diseases in a population with multiple comorbidities, and is consistent with a 90% influenza-related mortality due to respiratory and cardiovascular diseases reported in patients 65 and older.¹³ The increase in hospitalization and total hospitalization days in the age groups <5 years during influenza seasons are consistent with studies showing that the risk of children to contract influenza is higher than that of adults surrounding them. In this regard, outbreak investigations during the 2009 influenza pandemic showed that influenza attack rates in children were higher than those of adults.¹⁴

Nationwide studies from Singapore and Taiwan also showed more hospitalizations related to influenza in young children and older adults.^15,16

The increase in hospitalization days for all patients should be interpreted while taking into account the mean monthly LOS per patient (Table 2). In most age groups, a small decrease in the mean LOS for individual patients with the diagnosis of influenza/pneumonia was observed (Table 2). This decrease may suggest a need to shorten hospitalization slightly in order to accommodate new patients. Similarly, the decrease in hospitalization rates from all diagnoses during influenza seasons in the 5-54 years age groups (Figure) may stem, at least in part, from the shortage of available hospital beds due to patient overload. Additional study is required to further explore these decreases and their possible effects on morbidity and mortality.

Influenza vaccine guidelines in Israel following the 2009 influenza pandemic recommend influenza vaccination for all individuals age 6 months and older. However, influenza vaccination in Israel has remained low. Specifically, vaccination rates among children below the age of 5 years have been approximately 21%, as compared with 60%-65% in adults 65 years and over.¹⁷ Given the low rate of vaccination in children, we believe that there would be minimal or no difference in hospitalization of children under the age of 5 years, between the pre- and postpandemic years. Israel has started a school-based influenza vaccination program for the 2016-2017 influenza season in an effort to increase childhood influenza vaccination. It would be important to see if the expansion and continuation of the program would have an effect on influenza season hospitalizations.

Our study has several advantages. To the best of our knowledge, it is the first study examining differences in hospital burden between influenza and summer seasons on a national level. As such, it constitutes one of the largest studies on the subject. In addition, our study relies on original data, rather than estimates. Analysis of specific months of each year in which influenza virus circulates provides a targeted analysis of influenza seasons, rather than the entire winter season. The comparison with summer months is of great importance for preparatory plans by health systems, as it takes into account the degree of variation between the seasons. The analysis of 6 influenza seasons in our study intended to take into account season-to-season disease variability. Such variability among influenza seasons has been described previously due to changes in the virus itself, the population immune status, and the weather.¹⁸

We used several diagnosis categories to evaluate different aspects of hospital burden. Although the category of “all diagnoses” provided a broad assessment of hospital burden, influenza/pneumonia or pulmonary/cardiovascular disease constituted a more specific measure of influenza-associated burden.

Evaluating LOS added to the accuracy of hospital burden estimates, and our age-group analysis highlighted the specific age groups responsible for changes in hospital burden. Thus, the use of several measures to assess influenza season morbidity provides a comprehensive picture of the hospitalization dynamics between influenza and summer seasons. In this regard, the trends observed in our study for hospitalizations and total hospitalization days correspond to those observed in bed occupancy, especially for hospitalization rates due to all causes.

Our study has several limitations. We did not rely on laboratory diagnosis of influenza to determine burden. Because obtaining specimens for viral detection is usually based on individual clinical judgement, and patients hospitalized with influenza-related complications can often test negative for the virus due to time elapsed from disease onset, relying on a laboratory-based analysis may lead to underestimation of hospital burden. On the other hand, it is possible that patients with morbidity not specifically related to influenza were included in our analysis. Respiratory syncytial virus (RSV), for example, can also cause respiratory illness during the fall and winter.¹⁹ However, in Israel, RSV epidemic usually occurs before the influenza epidemic.^17,20 Thus, it is expected that only a small percentage of hospital admissions due to RSV would occur during the influenza season. Another limitation of our study relates to the small number of months in the beginning and end of influenza seasons in which influenza activity was recorded only during part of the month. Thus, hospital burden may have been underestimated during these “incomplete” months. Future studies using time series analysis methods will contribute to a more accurate estimation of such differences, as well as account for variability in influenza activity.

Our results clearly highlight the issues that challenge hospitals in Israel, and possibly other countries, during influenza seasons, such as the most affected age groups and the shortening of hospital stay. Thus, our findings are most relevant for hospital preparedness towards influenza seasons, particularly in terms of the need for additional hospital beds and personnel.

 

Acknowledgment

We would like to thank Anneke ifrah for English language editing

Disclosure

All authors report no conflict of interest relevant to this article. No financial support was provided relevant to this article.

Influenza-associated morbidity poses a significant hospital burden.¹ A study from the United States estimated that seasonal influenza is responsible for 3.1 million hospitalization days per year.²

Assessment of hospital burden during influenza seasons presents a challenge due to several possible factors, such as inaccurate recording of diagnosis³ and incomplete age group data. Although great emphasis has historically been placed on older age groups, a study from England and Wales showed that the number of hospitalizations and deaths resulting from influenza was significantly higher in children as compared with adults.⁴ Moreover, excess visits to emergency departments in New York City because of fever and respiratory morbidity during influenza seasons were found mostly among school-age children, whereas in adults, the surplus was small to nonexistent.⁵

Studies examining influenza-related hospitalizations evaluated numbers and rates of hospitalization.^6-11 However, information regarding length of hospitalizations, hospitalizations during the influenza season that were not influenza related, or comparisons between influenza seasons and summer seasons is scarce. These determinants are of great importance for hospital preparedness towards influenza seasons. The aim of the current study was to estimate excess hospitalizations and length of hospitalization during influenza seasons, as compared with the summer, in different age groups and selected diagnoses in Israel.

METHODS

Data Sources

Hospitalization data of internal medicine and pediatric departments in 28 acute care hospitals in Israel between 2005 and 2013 were obtained from the National Hospital Discharges Database managed by the Health Information Division (HID) in the Israel Ministry of Health (MOH). The information included number of discharges (including in-hospital deaths), number of hospitalization days, and the mean length of stay (LOS) per discharge for all diagnoses and for primary or secondary diagnoses of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Bed occupancy rates for internal medicine and pediatric departments were based on the National Patient Flow Database managed by the HID.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Classification

Hospitalizations data were analyzed for all ages, for specific age groups (the first year of life [0], ages 1-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84, and 85 years and older), for all diagnoses, and for primary or secondary discharge diagnosis of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Duration of Influenza Season

The beginning and the end of the influenza season were determined by the National Influenza surveillance program, which includes on average 22 community sentinel clinics, throughout Israel, each influenza season. These clinics send nose-throat samples from a convenience sample of patients with influenza-like illness (ILI), from week 40 of each year until the end of the influenza season in the subsequent year. These samples are analyzed for the presence of influenza virus by real-time reverse transcription polymerase chain reaction (RT-PCR) at the Central Virology Laboratory of Israel. Based on influenza virus detection in nose-throat samples from patients with ILI attending the community sentinel clinics, we determined the first and last month of each influenza season. The first month in which positive influenza samples were identified in sequence was defined as the first month of the season. The month in which the sequence of positive influenza samples stopped was defined as the last month of the season.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Analysis

Rates. Rates of monthly hospitalizations and monthly hospitalization days were calculated per 100,000 residents for all ages and for the specific age groups. Estimated average population sizes in different years for all ages and for specific age groups were obtained from the Central Bureau of Statistics (http://www.cbs.gov.il/reader/shnaton/templ_shnaton.html?num_tab=st02_01&CYear=2014). Monthly LOS was not converted to rates.

 

Hospitalizations. Mean monthly rate of hospitalizations during influenza and summer seasons was calculated by dividing the sum of hospital discharge rates during influenza/summer seasons of the entire evaluation period (2005/2006 to 2012/2013) by the number of influenza/summer activity months of that period.

Hospitalization Days. The measure “hospitalization days” refers to the hospitalization days of all patients who were discharged during influenza seasons. Mean monthly rate of hospitalization days during the influenza season and summer season was calculated using the procedure described for monthly mean rate of hospitalizations.

Length of Stay. The measure “length of stay” refers to the number of days that individual patients stayed in the hospital during an admission in the evaluated seasons.

Mean monthly LOS during the influenza and summer seasons for all patients (in both internal medicine and pediatric departments) and by age group was calculated by dividing the sum of monthly LOS during influenza seasons/summer season of the entire evaluation period (2005/2006 to 2012/2013 except for the 2009/2010 season) by the number of influenza/summer activity months of that period.

LOS for each specific month of the evaluation period for a single patient was calculated by dividing the number of hospitalization days of all patients that were discharged that month (stratified by age group) by the number of discharges in the same month.

Bed Occupancy. Bed occupancy rates for internal medicine and pediatric departments of the seasons evaluated were computed as a weighted rate based on the hospitalization days and licensed inpatient beds for the period of each influenza and summer season. The calculation took into account the number of days of each month and was based on the monthly reporting of hospital inpatient days in these departments and on the number of inpatient beds according to standard license documents issued by the MOH for each hospital.

Difference Between Influenza and Summer Seasons. Differences in mean monthly rates of hospitalizations, mean monthly rate of hospitalization days, and LOS during influenza seasons and the preceding summer were calculated as absolute numbers per month and as a percentage. The difference between bed occupancy during the influenza seasons and the preceding summers was expressed in percentage. Differences were computed for all diagnoses and for ICD9 480-487 and 390-519.

Statistical Analysis

Mean and standard deviation for monthly hospitalization rates, rates of monthly hospitalization days, and for LOS were calculated for all the influenza and summer seasons that were evaluated. Differences and statistical significance for these parameters were evaluated using a two-tailed Wilcoxon-Mann-Whitney test adjusted for ties, with 95% confidence interval for mean locations. The null hypothesis of the Wilcoxon test used was that the mean ranks of the influenza and summer season observations were equal.

Mean of bed occupancy percentage was calculated for influenza and summer seasons, with the difference and statistical significance being evaluated using a χ² test. P value of < 0.05 was considered statistically significant. SAS Version 9.1 and R program version 3.3.1 software were used for analysis.

RESULTS

Influenza Seasons

The length of influenza seasons varied, with the shortest season lasting 3 months (2006-2007) and the longest season lasting six months (2010-2011 and 2011-2012; Table 1). Of the 14 first and last months of the 7 influenza seasons, 9 had influenza activity throughout the month, 2 had 3 weeks of influenza activity, and 3 had 2 weeks of influenza activity (Table 1).

Hospitalizations

A total of 452,209 hospital discharges occurred in pediatric and internal medicine departments during the influenza seasons that were evaluated. The mean monthly rate of hospitalizations (as defined in Methods) for all patients was 4.8% higher during influenza seasons as compared with the preceding summer seasons (panel A in Figure; Supplementary Table 1). Analysis by age groups revealed a statistically significant increase during influenza seasons in the younger and older age groups (panel A in Figure; Supplementary Table 1). Specifically, the increase for all diagnoses was 32.8%, 16.7%, and 14.1% for infants <1 year, children aged 1-4 years, and adults ≥85 years, respectively (panel A in Figure; Supplementary Table 1).

The mean monthly rate of hospitalizations for all ages due to the diagnosis of respiratory/cardiovascular diseases and influenza/pneumonia was 18.6% and 60.8% higher, respectively, during influenza seasons compared with the preceding summers (panels B and C in Figure). These differences were statistically significant (panels B and C in Figure; Supplementary Table 1).

The increase in mean monthly hospitalization rates for patients with a diagnosis of respiratory/cardiovascular diseases and pneumonia/influenza was highest among infants <1 year and children aged 1-4 years (panels B and C in Figure; Supplementary Table 1). Increases were also observed among other age groups. However, they were more modest and reached statistical significance for respiratory/cardiovascular diseases in the age groups of ≤34 years and ≥75 years (panel B in Figure; Supplementary Table 1). The increases in mean monthly hospitalization rates for pneumonia/influenza were statistically significant in all age groups and were greater than 40% among adults ≥55 years (panel C in Figure; Supplementary Table 1).

Statistically significant decreases in mean monthly hospitalization rates during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 1). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or pneumonia/influenza (panels B and C in Figure; Supplementary Table 1).

 

Hospitalization Days

The mean monthly rate of hospitalization days per 100,000 residents showed a similar trend to that of the hospitalization rates (panels A, B, and C in Figure; Supplementary Table 2), with the most prominent increases observed among infants and children <5 years and adults ≥65 years.

The mean monthly rate of hospitalization days per 100,000 during influenza seasons for all ages due to all diagnoses was 8% higher (P < 0.001) as compared with the summer seasons (panel A in Figure; Supplementary Table 2). Statistically significant increases were also found among patients diagnosed with respiratory/cardiovascular diseases and for influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

A significant increase was also observed among infants and children <5 years and adults ≥65 years with all diagnoses (panel A in Figure; Supplementary Table 2). The increase in the monthly mean rate of hospitalization days was statistically significant for respiratory/cardiovascular diseases in most age groups, except the 35-64 age groups (panel B in Figure; Supplementary Table 2). A statistically significant increase in the monthly mean rate of hospitalization days for influenza/pneumonia was seen in all age

groups except for the 15-24 age group (panel C in Figure; Supplementary Table 2).

Children <5 years of age showed the largest increases during the influenza season as compared with the summer, with an up to 155.9% increase in the mean monthly rate of hospitalization days due to influenza/pneumonia (panel C in Figure; Supplementary Table 2), and an up to 206.6% increase for respiratory/cardiovascular diseases in infants <1 year of age (panel B in Figure; Supplementary Table 2). In adults, the largest increases were observed among those ≥75 years; the rates for influenza/pneumonia increased by about 40% (panel C in Figure; Supplementary Table 2), and the rates for respiratory/cardiovascular diseases increased by 14.8%-20.7% as compared with the summer months (panel B in Figure; Supplementary Table 2).

Statistically significant decreases in monthly mean rate of hospitalization days during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 2). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

Hospital Length of Stay

The longest mean monthly LOS due to all diagnoses (for both influenza and summer seasons) was observed in adults ≥65 years of age (Table 2). The longest mean monthly LOS due to influenza/pneumonia (for both influenza and summer seasons) was observed in adults ≥55 years or older, and for the diagnosis of respiratory/cardiovascular diseases, infants <1 year and adults ≥55 years had the longest LOS.

The differences between influenza and summer seasons in mean monthly LOS were mostly small or not observed in any of the diagnostic categories examined. The mean monthly LOS due to a diagnosis of influenza/pneumonia was shorter during the influenza seasons than summer seasons in most age groups. These differences were statistically significant in children <5 years and adults ≥45 years (Table 2).

The mean LOS due to respiratory/cardiovascular diseases was significantly shorter during influenza seasons than summer seasons in children under 5.

Bed Occupancy

Mean bed occupancy was significantly higher during influenza seasons compared with the preceding summer seasons, both in internal medicine and pediatric departments (Table 3). The differences were higher in pediatric departments as compared with internal medicine departments for most years evaluated.

DISCUSSION

Our study demonstrates trends of excess hospitalizations during influenza as compared with summer seasons and identifies patient groups that contribute mostly to changes in hospital burden between these seasons.

Overall, the present study demonstrates differences between influenza and summer seasons for all measures tested: hospitalizations, hospitalization days, LOS, and bed occupancy. These differences were due primarily to excess number of hospitalizations and hospitalization days, rather than to longer LOS.

Our results concerning hospitalizations for all diagnoses are consistent with a United States report showing about 5% more hospitalizations following emergency department visits during winter compared with summer.¹²

The increase in hospitalizations and total hospitalization days in older age groups reflects the probability of severe diseases in a population with multiple comorbidities, and is consistent with a 90% influenza-related mortality due to respiratory and cardiovascular diseases reported in patients 65 and older.¹³ The increase in hospitalization and total hospitalization days in the age groups <5 years during influenza seasons are consistent with studies showing that the risk of children to contract influenza is higher than that of adults surrounding them. In this regard, outbreak investigations during the 2009 influenza pandemic showed that influenza attack rates in children were higher than those of adults.¹⁴

Nationwide studies from Singapore and Taiwan also showed more hospitalizations related to influenza in young children and older adults.^15,16

The increase in hospitalization days for all patients should be interpreted while taking into account the mean monthly LOS per patient (Table 2). In most age groups, a small decrease in the mean LOS for individual patients with the diagnosis of influenza/pneumonia was observed (Table 2). This decrease may suggest a need to shorten hospitalization slightly in order to accommodate new patients. Similarly, the decrease in hospitalization rates from all diagnoses during influenza seasons in the 5-54 years age groups (Figure) may stem, at least in part, from the shortage of available hospital beds due to patient overload. Additional study is required to further explore these decreases and their possible effects on morbidity and mortality.

Influenza vaccine guidelines in Israel following the 2009 influenza pandemic recommend influenza vaccination for all individuals age 6 months and older. However, influenza vaccination in Israel has remained low. Specifically, vaccination rates among children below the age of 5 years have been approximately 21%, as compared with 60%-65% in adults 65 years and over.¹⁷ Given the low rate of vaccination in children, we believe that there would be minimal or no difference in hospitalization of children under the age of 5 years, between the pre- and postpandemic years. Israel has started a school-based influenza vaccination program for the 2016-2017 influenza season in an effort to increase childhood influenza vaccination. It would be important to see if the expansion and continuation of the program would have an effect on influenza season hospitalizations.

Our study has several advantages. To the best of our knowledge, it is the first study examining differences in hospital burden between influenza and summer seasons on a national level. As such, it constitutes one of the largest studies on the subject. In addition, our study relies on original data, rather than estimates. Analysis of specific months of each year in which influenza virus circulates provides a targeted analysis of influenza seasons, rather than the entire winter season. The comparison with summer months is of great importance for preparatory plans by health systems, as it takes into account the degree of variation between the seasons. The analysis of 6 influenza seasons in our study intended to take into account season-to-season disease variability. Such variability among influenza seasons has been described previously due to changes in the virus itself, the population immune status, and the weather.¹⁸

We used several diagnosis categories to evaluate different aspects of hospital burden. Although the category of “all diagnoses” provided a broad assessment of hospital burden, influenza/pneumonia or pulmonary/cardiovascular disease constituted a more specific measure of influenza-associated burden.

Evaluating LOS added to the accuracy of hospital burden estimates, and our age-group analysis highlighted the specific age groups responsible for changes in hospital burden. Thus, the use of several measures to assess influenza season morbidity provides a comprehensive picture of the hospitalization dynamics between influenza and summer seasons. In this regard, the trends observed in our study for hospitalizations and total hospitalization days correspond to those observed in bed occupancy, especially for hospitalization rates due to all causes.

Our study has several limitations. We did not rely on laboratory diagnosis of influenza to determine burden. Because obtaining specimens for viral detection is usually based on individual clinical judgement, and patients hospitalized with influenza-related complications can often test negative for the virus due to time elapsed from disease onset, relying on a laboratory-based analysis may lead to underestimation of hospital burden. On the other hand, it is possible that patients with morbidity not specifically related to influenza were included in our analysis. Respiratory syncytial virus (RSV), for example, can also cause respiratory illness during the fall and winter.¹⁹ However, in Israel, RSV epidemic usually occurs before the influenza epidemic.^17,20 Thus, it is expected that only a small percentage of hospital admissions due to RSV would occur during the influenza season. Another limitation of our study relates to the small number of months in the beginning and end of influenza seasons in which influenza activity was recorded only during part of the month. Thus, hospital burden may have been underestimated during these “incomplete” months. Future studies using time series analysis methods will contribute to a more accurate estimation of such differences, as well as account for variability in influenza activity.

Our results clearly highlight the issues that challenge hospitals in Israel, and possibly other countries, during influenza seasons, such as the most affected age groups and the shortening of hospital stay. Thus, our findings are most relevant for hospital preparedness towards influenza seasons, particularly in terms of the need for additional hospital beds and personnel.

 

Acknowledgment

We would like to thank Anneke ifrah for English language editing

Disclosure

All authors report no conflict of interest relevant to this article. No financial support was provided relevant to this article.

Influenza-associated morbidity poses a significant hospital burden.¹ A study from the United States estimated that seasonal influenza is responsible for 3.1 million hospitalization days per year.²

Assessment of hospital burden during influenza seasons presents a challenge due to several possible factors, such as inaccurate recording of diagnosis³ and incomplete age group data. Although great emphasis has historically been placed on older age groups, a study from England and Wales showed that the number of hospitalizations and deaths resulting from influenza was significantly higher in children as compared with adults.⁴ Moreover, excess visits to emergency departments in New York City because of fever and respiratory morbidity during influenza seasons were found mostly among school-age children, whereas in adults, the surplus was small to nonexistent.⁵

Studies examining influenza-related hospitalizations evaluated numbers and rates of hospitalization.^6-11 However, information regarding length of hospitalizations, hospitalizations during the influenza season that were not influenza related, or comparisons between influenza seasons and summer seasons is scarce. These determinants are of great importance for hospital preparedness towards influenza seasons. The aim of the current study was to estimate excess hospitalizations and length of hospitalization during influenza seasons, as compared with the summer, in different age groups and selected diagnoses in Israel.

METHODS

Data Sources

Hospitalization data of internal medicine and pediatric departments in 28 acute care hospitals in Israel between 2005 and 2013 were obtained from the National Hospital Discharges Database managed by the Health Information Division (HID) in the Israel Ministry of Health (MOH). The information included number of discharges (including in-hospital deaths), number of hospitalization days, and the mean length of stay (LOS) per discharge for all diagnoses and for primary or secondary diagnoses of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Bed occupancy rates for internal medicine and pediatric departments were based on the National Patient Flow Database managed by the HID.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Classification

Hospitalizations data were analyzed for all ages, for specific age groups (the first year of life [0], ages 1-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84, and 85 years and older), for all diagnoses, and for primary or secondary discharge diagnosis of respiratory/cardiovascular disease (ICD9 390-519) and influenza/pneumonia (ICD9 480-487).

Duration of Influenza Season

The beginning and the end of the influenza season were determined by the National Influenza surveillance program, which includes on average 22 community sentinel clinics, throughout Israel, each influenza season. These clinics send nose-throat samples from a convenience sample of patients with influenza-like illness (ILI), from week 40 of each year until the end of the influenza season in the subsequent year. These samples are analyzed for the presence of influenza virus by real-time reverse transcription polymerase chain reaction (RT-PCR) at the Central Virology Laboratory of Israel. Based on influenza virus detection in nose-throat samples from patients with ILI attending the community sentinel clinics, we determined the first and last month of each influenza season. The first month in which positive influenza samples were identified in sequence was defined as the first month of the season. The month in which the sequence of positive influenza samples stopped was defined as the last month of the season.

The 2009-2010 pandemic influenza season was excluded from analysis due to different morbidity patterns and timing (April 2009 until August 2010) as compared with seasonal influenza.

Data Analysis

Rates. Rates of monthly hospitalizations and monthly hospitalization days were calculated per 100,000 residents for all ages and for the specific age groups. Estimated average population sizes in different years for all ages and for specific age groups were obtained from the Central Bureau of Statistics (http://www.cbs.gov.il/reader/shnaton/templ_shnaton.html?num_tab=st02_01&CYear=2014). Monthly LOS was not converted to rates.

 

Hospitalizations. Mean monthly rate of hospitalizations during influenza and summer seasons was calculated by dividing the sum of hospital discharge rates during influenza/summer seasons of the entire evaluation period (2005/2006 to 2012/2013) by the number of influenza/summer activity months of that period.

Hospitalization Days. The measure “hospitalization days” refers to the hospitalization days of all patients who were discharged during influenza seasons. Mean monthly rate of hospitalization days during the influenza season and summer season was calculated using the procedure described for monthly mean rate of hospitalizations.

Length of Stay. The measure “length of stay” refers to the number of days that individual patients stayed in the hospital during an admission in the evaluated seasons.

Mean monthly LOS during the influenza and summer seasons for all patients (in both internal medicine and pediatric departments) and by age group was calculated by dividing the sum of monthly LOS during influenza seasons/summer season of the entire evaluation period (2005/2006 to 2012/2013 except for the 2009/2010 season) by the number of influenza/summer activity months of that period.

LOS for each specific month of the evaluation period for a single patient was calculated by dividing the number of hospitalization days of all patients that were discharged that month (stratified by age group) by the number of discharges in the same month.

Bed Occupancy. Bed occupancy rates for internal medicine and pediatric departments of the seasons evaluated were computed as a weighted rate based on the hospitalization days and licensed inpatient beds for the period of each influenza and summer season. The calculation took into account the number of days of each month and was based on the monthly reporting of hospital inpatient days in these departments and on the number of inpatient beds according to standard license documents issued by the MOH for each hospital.

Difference Between Influenza and Summer Seasons. Differences in mean monthly rates of hospitalizations, mean monthly rate of hospitalization days, and LOS during influenza seasons and the preceding summer were calculated as absolute numbers per month and as a percentage. The difference between bed occupancy during the influenza seasons and the preceding summers was expressed in percentage. Differences were computed for all diagnoses and for ICD9 480-487 and 390-519.

Statistical Analysis

Mean and standard deviation for monthly hospitalization rates, rates of monthly hospitalization days, and for LOS were calculated for all the influenza and summer seasons that were evaluated. Differences and statistical significance for these parameters were evaluated using a two-tailed Wilcoxon-Mann-Whitney test adjusted for ties, with 95% confidence interval for mean locations. The null hypothesis of the Wilcoxon test used was that the mean ranks of the influenza and summer season observations were equal.

Mean of bed occupancy percentage was calculated for influenza and summer seasons, with the difference and statistical significance being evaluated using a χ² test. P value of < 0.05 was considered statistically significant. SAS Version 9.1 and R program version 3.3.1 software were used for analysis.

RESULTS

Influenza Seasons

The length of influenza seasons varied, with the shortest season lasting 3 months (2006-2007) and the longest season lasting six months (2010-2011 and 2011-2012; Table 1). Of the 14 first and last months of the 7 influenza seasons, 9 had influenza activity throughout the month, 2 had 3 weeks of influenza activity, and 3 had 2 weeks of influenza activity (Table 1).

Hospitalizations

A total of 452,209 hospital discharges occurred in pediatric and internal medicine departments during the influenza seasons that were evaluated. The mean monthly rate of hospitalizations (as defined in Methods) for all patients was 4.8% higher during influenza seasons as compared with the preceding summer seasons (panel A in Figure; Supplementary Table 1). Analysis by age groups revealed a statistically significant increase during influenza seasons in the younger and older age groups (panel A in Figure; Supplementary Table 1). Specifically, the increase for all diagnoses was 32.8%, 16.7%, and 14.1% for infants <1 year, children aged 1-4 years, and adults ≥85 years, respectively (panel A in Figure; Supplementary Table 1).

The mean monthly rate of hospitalizations for all ages due to the diagnosis of respiratory/cardiovascular diseases and influenza/pneumonia was 18.6% and 60.8% higher, respectively, during influenza seasons compared with the preceding summers (panels B and C in Figure). These differences were statistically significant (panels B and C in Figure; Supplementary Table 1).

The increase in mean monthly hospitalization rates for patients with a diagnosis of respiratory/cardiovascular diseases and pneumonia/influenza was highest among infants <1 year and children aged 1-4 years (panels B and C in Figure; Supplementary Table 1). Increases were also observed among other age groups. However, they were more modest and reached statistical significance for respiratory/cardiovascular diseases in the age groups of ≤34 years and ≥75 years (panel B in Figure; Supplementary Table 1). The increases in mean monthly hospitalization rates for pneumonia/influenza were statistically significant in all age groups and were greater than 40% among adults ≥55 years (panel C in Figure; Supplementary Table 1).

Statistically significant decreases in mean monthly hospitalization rates during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 1). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or pneumonia/influenza (panels B and C in Figure; Supplementary Table 1).

 

Hospitalization Days

The mean monthly rate of hospitalization days per 100,000 residents showed a similar trend to that of the hospitalization rates (panels A, B, and C in Figure; Supplementary Table 2), with the most prominent increases observed among infants and children <5 years and adults ≥65 years.

The mean monthly rate of hospitalization days per 100,000 during influenza seasons for all ages due to all diagnoses was 8% higher (P < 0.001) as compared with the summer seasons (panel A in Figure; Supplementary Table 2). Statistically significant increases were also found among patients diagnosed with respiratory/cardiovascular diseases and for influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

A significant increase was also observed among infants and children <5 years and adults ≥65 years with all diagnoses (panel A in Figure; Supplementary Table 2). The increase in the monthly mean rate of hospitalization days was statistically significant for respiratory/cardiovascular diseases in most age groups, except the 35-64 age groups (panel B in Figure; Supplementary Table 2). A statistically significant increase in the monthly mean rate of hospitalization days for influenza/pneumonia was seen in all age

groups except for the 15-24 age group (panel C in Figure; Supplementary Table 2).

Children <5 years of age showed the largest increases during the influenza season as compared with the summer, with an up to 155.9% increase in the mean monthly rate of hospitalization days due to influenza/pneumonia (panel C in Figure; Supplementary Table 2), and an up to 206.6% increase for respiratory/cardiovascular diseases in infants <1 year of age (panel B in Figure; Supplementary Table 2). In adults, the largest increases were observed among those ≥75 years; the rates for influenza/pneumonia increased by about 40% (panel C in Figure; Supplementary Table 2), and the rates for respiratory/cardiovascular diseases increased by 14.8%-20.7% as compared with the summer months (panel B in Figure; Supplementary Table 2).

Statistically significant decreases in monthly mean rate of hospitalization days during influenza seasons were observed for all diagnoses in the 5-54 age groups (panel A in Figure; Supplementary Table 2). Decreases were not seen for the diagnoses of respiratory/cardiovascular diseases or influenza/pneumonia (panels B and C in Figure; Supplementary Table 2).

Hospital Length of Stay

The longest mean monthly LOS due to all diagnoses (for both influenza and summer seasons) was observed in adults ≥65 years of age (Table 2). The longest mean monthly LOS due to influenza/pneumonia (for both influenza and summer seasons) was observed in adults ≥55 years or older, and for the diagnosis of respiratory/cardiovascular diseases, infants <1 year and adults ≥55 years had the longest LOS.

The differences between influenza and summer seasons in mean monthly LOS were mostly small or not observed in any of the diagnostic categories examined. The mean monthly LOS due to a diagnosis of influenza/pneumonia was shorter during the influenza seasons than summer seasons in most age groups. These differences were statistically significant in children <5 years and adults ≥45 years (Table 2).

The mean LOS due to respiratory/cardiovascular diseases was significantly shorter during influenza seasons than summer seasons in children under 5.

Bed Occupancy

Mean bed occupancy was significantly higher during influenza seasons compared with the preceding summer seasons, both in internal medicine and pediatric departments (Table 3). The differences were higher in pediatric departments as compared with internal medicine departments for most years evaluated.

DISCUSSION

Our study demonstrates trends of excess hospitalizations during influenza as compared with summer seasons and identifies patient groups that contribute mostly to changes in hospital burden between these seasons.

Overall, the present study demonstrates differences between influenza and summer seasons for all measures tested: hospitalizations, hospitalization days, LOS, and bed occupancy. These differences were due primarily to excess number of hospitalizations and hospitalization days, rather than to longer LOS.

Our results concerning hospitalizations for all diagnoses are consistent with a United States report showing about 5% more hospitalizations following emergency department visits during winter compared with summer.¹²

The increase in hospitalizations and total hospitalization days in older age groups reflects the probability of severe diseases in a population with multiple comorbidities, and is consistent with a 90% influenza-related mortality due to respiratory and cardiovascular diseases reported in patients 65 and older.¹³ The increase in hospitalization and total hospitalization days in the age groups <5 years during influenza seasons are consistent with studies showing that the risk of children to contract influenza is higher than that of adults surrounding them. In this regard, outbreak investigations during the 2009 influenza pandemic showed that influenza attack rates in children were higher than those of adults.¹⁴

Nationwide studies from Singapore and Taiwan also showed more hospitalizations related to influenza in young children and older adults.^15,16

The increase in hospitalization days for all patients should be interpreted while taking into account the mean monthly LOS per patient (Table 2). In most age groups, a small decrease in the mean LOS for individual patients with the diagnosis of influenza/pneumonia was observed (Table 2). This decrease may suggest a need to shorten hospitalization slightly in order to accommodate new patients. Similarly, the decrease in hospitalization rates from all diagnoses during influenza seasons in the 5-54 years age groups (Figure) may stem, at least in part, from the shortage of available hospital beds due to patient overload. Additional study is required to further explore these decreases and their possible effects on morbidity and mortality.

Influenza vaccine guidelines in Israel following the 2009 influenza pandemic recommend influenza vaccination for all individuals age 6 months and older. However, influenza vaccination in Israel has remained low. Specifically, vaccination rates among children below the age of 5 years have been approximately 21%, as compared with 60%-65% in adults 65 years and over.¹⁷ Given the low rate of vaccination in children, we believe that there would be minimal or no difference in hospitalization of children under the age of 5 years, between the pre- and postpandemic years. Israel has started a school-based influenza vaccination program for the 2016-2017 influenza season in an effort to increase childhood influenza vaccination. It would be important to see if the expansion and continuation of the program would have an effect on influenza season hospitalizations.

Our study has several advantages. To the best of our knowledge, it is the first study examining differences in hospital burden between influenza and summer seasons on a national level. As such, it constitutes one of the largest studies on the subject. In addition, our study relies on original data, rather than estimates. Analysis of specific months of each year in which influenza virus circulates provides a targeted analysis of influenza seasons, rather than the entire winter season. The comparison with summer months is of great importance for preparatory plans by health systems, as it takes into account the degree of variation between the seasons. The analysis of 6 influenza seasons in our study intended to take into account season-to-season disease variability. Such variability among influenza seasons has been described previously due to changes in the virus itself, the population immune status, and the weather.¹⁸

We used several diagnosis categories to evaluate different aspects of hospital burden. Although the category of “all diagnoses” provided a broad assessment of hospital burden, influenza/pneumonia or pulmonary/cardiovascular disease constituted a more specific measure of influenza-associated burden.

Evaluating LOS added to the accuracy of hospital burden estimates, and our age-group analysis highlighted the specific age groups responsible for changes in hospital burden. Thus, the use of several measures to assess influenza season morbidity provides a comprehensive picture of the hospitalization dynamics between influenza and summer seasons. In this regard, the trends observed in our study for hospitalizations and total hospitalization days correspond to those observed in bed occupancy, especially for hospitalization rates due to all causes.

Our study has several limitations. We did not rely on laboratory diagnosis of influenza to determine burden. Because obtaining specimens for viral detection is usually based on individual clinical judgement, and patients hospitalized with influenza-related complications can often test negative for the virus due to time elapsed from disease onset, relying on a laboratory-based analysis may lead to underestimation of hospital burden. On the other hand, it is possible that patients with morbidity not specifically related to influenza were included in our analysis. Respiratory syncytial virus (RSV), for example, can also cause respiratory illness during the fall and winter.¹⁹ However, in Israel, RSV epidemic usually occurs before the influenza epidemic.^17,20 Thus, it is expected that only a small percentage of hospital admissions due to RSV would occur during the influenza season. Another limitation of our study relates to the small number of months in the beginning and end of influenza seasons in which influenza activity was recorded only during part of the month. Thus, hospital burden may have been underestimated during these “incomplete” months. Future studies using time series analysis methods will contribute to a more accurate estimation of such differences, as well as account for variability in influenza activity.

Our results clearly highlight the issues that challenge hospitals in Israel, and possibly other countries, during influenza seasons, such as the most affected age groups and the shortening of hospital stay. Thus, our findings are most relevant for hospital preparedness towards influenza seasons, particularly in terms of the need for additional hospital beds and personnel.

 

Acknowledgment

We would like to thank Anneke ifrah for English language editing

Disclosure

All authors report no conflict of interest relevant to this article. No financial support was provided relevant to this article.

Thrombophilia is a prothrombotic state, either acquired or inherited, leading to a thrombotic predisposition.¹ The most common heritable thrombophilias include factor V Leiden (FVL) and prothrombin G20210A. The most common acquired thrombophilia is the presence of phospholipid antibodies.¹ Thrombotic risk varies with thrombophilia type. For example, deficiencies of antithrombin, protein C and protein S, and the presence of phospholipid antibodies, confer higher risk than FVL and prothrombin G20210A.^2-5 Other thrombophilias (eg, methylenetetrahydrofolate reductase mutation, increased factor VIII activity) are relatively uncommon and/or their impact on thrombosis risk appears to be either minimal or unknown.^1-6 There is little clinical evidence that testing for thrombophilia impacts subsequent thrombosis prevention.^5,7,8 Multiple clinical guidelines and medical societies recommend against the routine and indiscriminate use of thrombophilia testing.^8-13 In general, thrombophilia testing should be considered only if the result would lead to changes in anticoagulant initiation, intensity, and/or duration, or might inform interventions to prevent thrombosis in asymptomatic family members.^8-13 However, thrombophilia testing rarely changes the acute management of a thrombotic event and may have harmful effects on patients and their family members because positive results may unnecessarily increase anxiety and negative results may provide false reassurance.^6,14-18 The cost-effectiveness of thrombophilia testing is unknown. Economic models have sought to quantify cost-effectiveness, but conclusions from these studies are limited.⁷

The utility of thrombophilia testing in emergency department (ED) and inpatient settings is further limited because patients are often treated and discharged before thrombophilia test results are available. Additionally, in these settings, multiple factors increase the risk of false-positive or false-negative results (eg, acute thrombosis, acute illness, pregnancy, and anticoagulant therapy).^19,20 The purpose of this study was to systematically assess thrombophilia testing patterns in the ED and hospitalized patients at an academic medical center and to quantify the proportion of tests associated with minimal clinical utility. We hypothesize that the majority of thrombophilia tests completed in the inpatient setting are associated with minimal clinical utility.

METHODS

Setting and Patients

This study was conducted at University of Utah Health Care (UUHC) University Hospital, a 488-bed academic medical center with a level I trauma center, primary stroke center, and 50-bed ED. Laboratory services for UUHC, including thrombophilia testing, are provided by a national reference laboratory, Associated Regional and University Pathologists Laboratories. This study included patients ≥18 years of age who received thrombophilia testing (Supplementary Table 1) during an ED visit or inpatient admission at University Hospital between July 1, 2014 and December 31, 2014. There were no exclusion criteria. An institutional electronic data repository was used to identify patients matching inclusion criteria. All study activities were reviewed and approved by the UUHC Institutional Review Board with a waiver of informed consent.

Outcomes

An electronic database query was used to identify patients, collect patient demographic information, and collect test characteristics. Each patient’s electronic medical record was manually reviewed to collect all other outcomes. Indication for thrombophilia testing was identified by manual review of provider notes. Thrombophilia tests occurring in situations associated with minimal clinical utility were defined as tests meeting at least one of the following criteria: patient discharged before test results were available for review; test type not recommended by published guidelines or by UUHC Thrombosis Service physicians for thrombophilia testing (Supplementary Table 2); test performed in situations associated with decreased accuracy; test was a duplicate test as a result of different thrombophilia panels containing identical tests; and test followed a provoked venous thromboembolism (VTE). Testing in situations associated with decreased accuracy are summarized in Supplementary Table 3 and included at least one of the following at the time of the test: anticoagulant therapy, acute thrombosis, pregnant or <8 weeks postpartum, and receiving estrogen-containing medications. Only test types known to be affected by the respective situation were included. Testing following a provoked VTE was defined as testing prompted by an acute thrombosis and performed within 3 months following major surgery (defined administratively as any surgery performed in an operating room), during pregnancy, <8 weeks postpartum, or while on estrogen-containing medications. Thrombophilia testing during anticoagulant therapy was defined as testing within 4 half-lives of anticoagulant administration based on medication administration records. Anticoagulant therapy changes were identified by comparing prior-to-admission and discharge medication lists.

 

Data Analysis

Patient and laboratory characteristics were summarized using descriptive statistics, including mean and standard deviation (SD) for continuous variables and proportions for categorical variables. Data analysis was performed using Excel (Version 2013, Microsoft Corporation. Redmond, Washington).

RESULTS

During the 6-month study period, 163 patients received at least 1 thrombophilia test during an ED visit or inpatient admission. Patient characteristics are summarized in Table 1. Tested patients were most commonly inpatients (96%) and female (71%). A total of 1451 thrombophilia tests were performed with a mean (± SD) of 8.9 ± 6.0 tests per patient. Testing characteristics are summarized in Table 2. Of the 39 different test types performed, the most commonly ordered were cardiolipin IgG and IgM antibodies (9% each), lupus anticoagulant (9%), and β₂-glycoprotein 1 IgG and IgM antibodies (8% each). When combined with testing for phosphatidyl antibodies, antiphospholipid tests accounted for 70% of all tests. Overall, 134 (9%) test results were positive. The mean time for results to become available was 2.2 ± 2.5 days. The frequency of test types with corresponding positivity rates and mean time for results to become available are summarized in Supplementary Table 4.

The indications for thrombophilia testing are summarized in Table 3. Ischemic stroke was the most common indication for testing (50% of tests; 35% of patients), followed by VTE (21% of tests; 21% of patients), and pregnancy-related conditions (eg, preeclampsia, intrauterine fetal demise; 15% of tests; 25% of patients). Overall, 911 tests (63%) occurred in situations associated with minimal clinical utility, with 126 patients (77%) receiving at least one of these tests (Table 4).

Anticoagulant therapy was changed in 43 patients (26%) in the following ways: initiated in 35 patients (21%), transitioned to a different anticoagulant in 6 patients (4%), and discontinued in 2 patients (1%). Of the 35 patients initiating anticoagulant therapy, 29 had documented thrombosis (24 had VTE, 4 had cerebral venous sinus thrombosis [CVST], and 1 had basilar artery thrombosis). Overall, 2 instances were identified in which initiation of anticoagulant therapy at discharge was in response to thrombophilia test results. In the first instance, warfarin without a parenteral anticoagulant bridge was initiated for a 54-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG antibodies, lupus anticoagulant, and protein S deficiency. In the second instance, warfarin with an enoxaparin bridge was initiated for a 26-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG and IgM antibodies, cardiolipin IgG antibodies, lupus anticoagulant, protein C deficiency, and antithrombin deficiency. Of the 163 patients receiving thrombophilia testing, only 2 patients (1%) had clear documentation of being offered genetic consultation.

DISCUSSION

In this retrospective analysis, 1451 thrombophilia tests were performed in 163 patients over 6 months. Tested patients were relatively young, which is likely explained by the number of patients tested for pregnancy-related conditions and the fact that a stroke or VTE in younger patients more frequently prompted providers to suspect thrombophilia. Nearly three-fourths of patients were female, which is likely due to testing for pregnancy-related conditions and possibly diagnostic suspicion bias given the comparative predilection of antiphospholipid syndrome for women. The patient characteristics in our study are consistent with other studies evaluating thrombophilia testing.^21,22

Thrombophilia testing was most frequently prompted by stroke, VTE, and pregnancy-related conditions. Only 26% of patients had acute thrombosis identified during the admission, primarily because of the high proportion of tests for cryptogenic strokes and pregnancy-related conditions. Thrombophilia testing is recommended in patients who have had a stroke when the stroke is considered to be cryptogenic after a standard stroke evaluation.²³ Thrombophilia testing in pregnancy-related conditions is controversial but is often considered in situations such as stillbirths with severe placental pathology and/or significant growth restriction, or in mothers with a personal or family history of thrombosis.²⁴ The proportion of testing for pregnancy-related conditions may be greater than at other institutions because UUHC Maternal Fetal Medicine is a referral center for women with conditions associated with hypercoagulability. Anticoagulant therapy was initiated in 21% of patients, but specifically in response to thrombophilia testing in only 2 instances; in most cases, anticoagulant therapy was initiated regardless of thrombophilia test results.

The results of this study confirm our hypothesis because the majority of thrombophilia tests occurred in situations associated with minimal clinical utility. Testing in these situations was not isolated to specific patients or medical services because 77% of tested patients received at least 1 test associated with minimal clinical utility. Our study took a conservative approach in defining scenarios associated with minimal clinical utility because other situations can also affect testing accuracy (eg, hepatic disease, nephrotic syndrome) but were not included in our analysis of this outcome.

The results of this study highlight opportunities to improve thrombophilia testing practices at our institution and may be generalizable to institutions with similar testing patterns. Because multiple medical services order thrombophilia tests, strategies to improve testing practices are still being determined. The results of this study can serve as a baseline for comparison after strategies are implemented. The most common situation associated with minimal clinical utility was the use of test types not generally recommended by guidelines or UUHC Thrombosis Service physicians for thrombophilia testing (eg, β₂-glycoprotein 1 IgA antibodies, phosphatidyl antibodies). We intend to require a hematology or thrombosis specialty consult prior to ordering these tests. This intervention alone could potentially decrease unnecessary testing by a third. Another consideration is to require a specialty consult prior to any inpatient thrombophilia testing. This strategy has been found to decrease inappropriate testing at other institutions.²¹ We also intend to streamline available thrombophilia testing panels because a poorly designed panel could lead to ordering of multiple tests associated with minimal clinical utility. At least 12 different thrombophilia panels are currently available in our computerized physician order entry system (see Supplementary Table 5). We hypothesize that current panel designs contribute to providers inadvertently ordering unintended or duplicate tests and that reducing the number of available panels and clearly delineating what tests are contained in each panel is likely to reduce unnecessary testing. Other strategies being considered include using electronic clinical decision support tools, implementing strict ordering criteria for all inpatient testing, and establishing a thrombosis stewardship program.

Our study was unique in at least 2 ways. First, previous studies describing thrombophilia testing have described testing patterns for patients with specific indications (eg, VTE), whereas our study described all thrombophilia tests regardless of indication. This allows for testing pattern comparisons across indications and medical services, increasing the generalizability of our results. Second, this study quantifies tests occurring in situations associated with a practical definition of minimal clinical utility.

Our study has several limitations: (1) Many variables were reliant on provider notes and other documentation, which allows for potential misclassification of variables. (2) It was not always possible to determine the ultimate utility of each test in clinical management decisions, and our study did not investigate the impact of thrombophilia testing on duration of anticoagulant therapy. Additionally, select situations could benefit from testing regardless if anticoagulant therapy is altered (eg, informing contraceptive choices). (3) Testing performed following a provoked acute thrombosis was defined as testing within 3 months following administratively defined major surgery. This definition could have included some minor procedures that do not substantially increase VTE risk, resulting in underestimated clinical utility. (4) The UUHC University Hospital serves as a referral hospital for a large geographical area, and investigators did not have access to outpatient records for a large proportion of discharged patients. As a result, frequency of repeat testing could not be assessed, possibly resulting in overestimated clinical utility. (5) In categorizing indications for testing, testing for CVST was subcategorized under testing for ischemic stroke based on presenting symptoms rather than on underlying pathophysiology. The rationale for this categorization is that patients with CVST were often tested based on presenting symptoms. Additionally, tests for CVST were ordered by the neurology service, which also ordered tests for all other ischemic stroke indications. (6) The purpose of our study was to investigate the subset of the hospital’s patient population that received thrombophilia testing, and patients were identified by tests received and not by diagnosis codes. As a result, we are unable to provide the proportion of total patients treated at the hospital for specific conditions who were tested (eg, the proportion of stroke patients that received thrombophilia testing). (7) Current practice guidelines do not recommend testing for phosphatidyl antibodies, even when traditional antiphospholipid testing is negative.^25-27 Although expert panels continue to explore associations between phosphatidyl antibodies and pregnancy morbidity and thrombotic events, the low level of evidence is insufficient to guide clinical management.²⁸ Therefore, we categorized all phosphatidyl testing as associated with minimal clinical utility.

 

CONCLUSIONS

In a large academic medical center, the majority of tests occurred in situations associated with minimal clinical utility. Strategies to improve thrombophilia testing practices are needed in order to minimize potentially inappropriate testing, provide more cost-effective care, and promote value-driven outcomes.

Disclosure

S.W. received financial support for this submitted work via a Bristol-Myers-Squibb grant. G.F. received financial support from Portola Pharmaceuticals for consulting and lectures that were not related to this submitted work.

Thrombophilia is a prothrombotic state, either acquired or inherited, leading to a thrombotic predisposition.¹ The most common heritable thrombophilias include factor V Leiden (FVL) and prothrombin G20210A. The most common acquired thrombophilia is the presence of phospholipid antibodies.¹ Thrombotic risk varies with thrombophilia type. For example, deficiencies of antithrombin, protein C and protein S, and the presence of phospholipid antibodies, confer higher risk than FVL and prothrombin G20210A.^2-5 Other thrombophilias (eg, methylenetetrahydrofolate reductase mutation, increased factor VIII activity) are relatively uncommon and/or their impact on thrombosis risk appears to be either minimal or unknown.^1-6 There is little clinical evidence that testing for thrombophilia impacts subsequent thrombosis prevention.^5,7,8 Multiple clinical guidelines and medical societies recommend against the routine and indiscriminate use of thrombophilia testing.^8-13 In general, thrombophilia testing should be considered only if the result would lead to changes in anticoagulant initiation, intensity, and/or duration, or might inform interventions to prevent thrombosis in asymptomatic family members.^8-13 However, thrombophilia testing rarely changes the acute management of a thrombotic event and may have harmful effects on patients and their family members because positive results may unnecessarily increase anxiety and negative results may provide false reassurance.^6,14-18 The cost-effectiveness of thrombophilia testing is unknown. Economic models have sought to quantify cost-effectiveness, but conclusions from these studies are limited.⁷

The utility of thrombophilia testing in emergency department (ED) and inpatient settings is further limited because patients are often treated and discharged before thrombophilia test results are available. Additionally, in these settings, multiple factors increase the risk of false-positive or false-negative results (eg, acute thrombosis, acute illness, pregnancy, and anticoagulant therapy).^19,20 The purpose of this study was to systematically assess thrombophilia testing patterns in the ED and hospitalized patients at an academic medical center and to quantify the proportion of tests associated with minimal clinical utility. We hypothesize that the majority of thrombophilia tests completed in the inpatient setting are associated with minimal clinical utility.

METHODS

Setting and Patients

This study was conducted at University of Utah Health Care (UUHC) University Hospital, a 488-bed academic medical center with a level I trauma center, primary stroke center, and 50-bed ED. Laboratory services for UUHC, including thrombophilia testing, are provided by a national reference laboratory, Associated Regional and University Pathologists Laboratories. This study included patients ≥18 years of age who received thrombophilia testing (Supplementary Table 1) during an ED visit or inpatient admission at University Hospital between July 1, 2014 and December 31, 2014. There were no exclusion criteria. An institutional electronic data repository was used to identify patients matching inclusion criteria. All study activities were reviewed and approved by the UUHC Institutional Review Board with a waiver of informed consent.

Outcomes

An electronic database query was used to identify patients, collect patient demographic information, and collect test characteristics. Each patient’s electronic medical record was manually reviewed to collect all other outcomes. Indication for thrombophilia testing was identified by manual review of provider notes. Thrombophilia tests occurring in situations associated with minimal clinical utility were defined as tests meeting at least one of the following criteria: patient discharged before test results were available for review; test type not recommended by published guidelines or by UUHC Thrombosis Service physicians for thrombophilia testing (Supplementary Table 2); test performed in situations associated with decreased accuracy; test was a duplicate test as a result of different thrombophilia panels containing identical tests; and test followed a provoked venous thromboembolism (VTE). Testing in situations associated with decreased accuracy are summarized in Supplementary Table 3 and included at least one of the following at the time of the test: anticoagulant therapy, acute thrombosis, pregnant or <8 weeks postpartum, and receiving estrogen-containing medications. Only test types known to be affected by the respective situation were included. Testing following a provoked VTE was defined as testing prompted by an acute thrombosis and performed within 3 months following major surgery (defined administratively as any surgery performed in an operating room), during pregnancy, <8 weeks postpartum, or while on estrogen-containing medications. Thrombophilia testing during anticoagulant therapy was defined as testing within 4 half-lives of anticoagulant administration based on medication administration records. Anticoagulant therapy changes were identified by comparing prior-to-admission and discharge medication lists.

 

Data Analysis

Patient and laboratory characteristics were summarized using descriptive statistics, including mean and standard deviation (SD) for continuous variables and proportions for categorical variables. Data analysis was performed using Excel (Version 2013, Microsoft Corporation. Redmond, Washington).

RESULTS

During the 6-month study period, 163 patients received at least 1 thrombophilia test during an ED visit or inpatient admission. Patient characteristics are summarized in Table 1. Tested patients were most commonly inpatients (96%) and female (71%). A total of 1451 thrombophilia tests were performed with a mean (± SD) of 8.9 ± 6.0 tests per patient. Testing characteristics are summarized in Table 2. Of the 39 different test types performed, the most commonly ordered were cardiolipin IgG and IgM antibodies (9% each), lupus anticoagulant (9%), and β₂-glycoprotein 1 IgG and IgM antibodies (8% each). When combined with testing for phosphatidyl antibodies, antiphospholipid tests accounted for 70% of all tests. Overall, 134 (9%) test results were positive. The mean time for results to become available was 2.2 ± 2.5 days. The frequency of test types with corresponding positivity rates and mean time for results to become available are summarized in Supplementary Table 4.

The indications for thrombophilia testing are summarized in Table 3. Ischemic stroke was the most common indication for testing (50% of tests; 35% of patients), followed by VTE (21% of tests; 21% of patients), and pregnancy-related conditions (eg, preeclampsia, intrauterine fetal demise; 15% of tests; 25% of patients). Overall, 911 tests (63%) occurred in situations associated with minimal clinical utility, with 126 patients (77%) receiving at least one of these tests (Table 4).

Anticoagulant therapy was changed in 43 patients (26%) in the following ways: initiated in 35 patients (21%), transitioned to a different anticoagulant in 6 patients (4%), and discontinued in 2 patients (1%). Of the 35 patients initiating anticoagulant therapy, 29 had documented thrombosis (24 had VTE, 4 had cerebral venous sinus thrombosis [CVST], and 1 had basilar artery thrombosis). Overall, 2 instances were identified in which initiation of anticoagulant therapy at discharge was in response to thrombophilia test results. In the first instance, warfarin without a parenteral anticoagulant bridge was initiated for a 54-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG antibodies, lupus anticoagulant, and protein S deficiency. In the second instance, warfarin with an enoxaparin bridge was initiated for a 26-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG and IgM antibodies, cardiolipin IgG antibodies, lupus anticoagulant, protein C deficiency, and antithrombin deficiency. Of the 163 patients receiving thrombophilia testing, only 2 patients (1%) had clear documentation of being offered genetic consultation.

DISCUSSION

In this retrospective analysis, 1451 thrombophilia tests were performed in 163 patients over 6 months. Tested patients were relatively young, which is likely explained by the number of patients tested for pregnancy-related conditions and the fact that a stroke or VTE in younger patients more frequently prompted providers to suspect thrombophilia. Nearly three-fourths of patients were female, which is likely due to testing for pregnancy-related conditions and possibly diagnostic suspicion bias given the comparative predilection of antiphospholipid syndrome for women. The patient characteristics in our study are consistent with other studies evaluating thrombophilia testing.^21,22

Thrombophilia testing was most frequently prompted by stroke, VTE, and pregnancy-related conditions. Only 26% of patients had acute thrombosis identified during the admission, primarily because of the high proportion of tests for cryptogenic strokes and pregnancy-related conditions. Thrombophilia testing is recommended in patients who have had a stroke when the stroke is considered to be cryptogenic after a standard stroke evaluation.²³ Thrombophilia testing in pregnancy-related conditions is controversial but is often considered in situations such as stillbirths with severe placental pathology and/or significant growth restriction, or in mothers with a personal or family history of thrombosis.²⁴ The proportion of testing for pregnancy-related conditions may be greater than at other institutions because UUHC Maternal Fetal Medicine is a referral center for women with conditions associated with hypercoagulability. Anticoagulant therapy was initiated in 21% of patients, but specifically in response to thrombophilia testing in only 2 instances; in most cases, anticoagulant therapy was initiated regardless of thrombophilia test results.

The results of this study confirm our hypothesis because the majority of thrombophilia tests occurred in situations associated with minimal clinical utility. Testing in these situations was not isolated to specific patients or medical services because 77% of tested patients received at least 1 test associated with minimal clinical utility. Our study took a conservative approach in defining scenarios associated with minimal clinical utility because other situations can also affect testing accuracy (eg, hepatic disease, nephrotic syndrome) but were not included in our analysis of this outcome.

The results of this study highlight opportunities to improve thrombophilia testing practices at our institution and may be generalizable to institutions with similar testing patterns. Because multiple medical services order thrombophilia tests, strategies to improve testing practices are still being determined. The results of this study can serve as a baseline for comparison after strategies are implemented. The most common situation associated with minimal clinical utility was the use of test types not generally recommended by guidelines or UUHC Thrombosis Service physicians for thrombophilia testing (eg, β₂-glycoprotein 1 IgA antibodies, phosphatidyl antibodies). We intend to require a hematology or thrombosis specialty consult prior to ordering these tests. This intervention alone could potentially decrease unnecessary testing by a third. Another consideration is to require a specialty consult prior to any inpatient thrombophilia testing. This strategy has been found to decrease inappropriate testing at other institutions.²¹ We also intend to streamline available thrombophilia testing panels because a poorly designed panel could lead to ordering of multiple tests associated with minimal clinical utility. At least 12 different thrombophilia panels are currently available in our computerized physician order entry system (see Supplementary Table 5). We hypothesize that current panel designs contribute to providers inadvertently ordering unintended or duplicate tests and that reducing the number of available panels and clearly delineating what tests are contained in each panel is likely to reduce unnecessary testing. Other strategies being considered include using electronic clinical decision support tools, implementing strict ordering criteria for all inpatient testing, and establishing a thrombosis stewardship program.

Our study was unique in at least 2 ways. First, previous studies describing thrombophilia testing have described testing patterns for patients with specific indications (eg, VTE), whereas our study described all thrombophilia tests regardless of indication. This allows for testing pattern comparisons across indications and medical services, increasing the generalizability of our results. Second, this study quantifies tests occurring in situations associated with a practical definition of minimal clinical utility.

Our study has several limitations: (1) Many variables were reliant on provider notes and other documentation, which allows for potential misclassification of variables. (2) It was not always possible to determine the ultimate utility of each test in clinical management decisions, and our study did not investigate the impact of thrombophilia testing on duration of anticoagulant therapy. Additionally, select situations could benefit from testing regardless if anticoagulant therapy is altered (eg, informing contraceptive choices). (3) Testing performed following a provoked acute thrombosis was defined as testing within 3 months following administratively defined major surgery. This definition could have included some minor procedures that do not substantially increase VTE risk, resulting in underestimated clinical utility. (4) The UUHC University Hospital serves as a referral hospital for a large geographical area, and investigators did not have access to outpatient records for a large proportion of discharged patients. As a result, frequency of repeat testing could not be assessed, possibly resulting in overestimated clinical utility. (5) In categorizing indications for testing, testing for CVST was subcategorized under testing for ischemic stroke based on presenting symptoms rather than on underlying pathophysiology. The rationale for this categorization is that patients with CVST were often tested based on presenting symptoms. Additionally, tests for CVST were ordered by the neurology service, which also ordered tests for all other ischemic stroke indications. (6) The purpose of our study was to investigate the subset of the hospital’s patient population that received thrombophilia testing, and patients were identified by tests received and not by diagnosis codes. As a result, we are unable to provide the proportion of total patients treated at the hospital for specific conditions who were tested (eg, the proportion of stroke patients that received thrombophilia testing). (7) Current practice guidelines do not recommend testing for phosphatidyl antibodies, even when traditional antiphospholipid testing is negative.^25-27 Although expert panels continue to explore associations between phosphatidyl antibodies and pregnancy morbidity and thrombotic events, the low level of evidence is insufficient to guide clinical management.²⁸ Therefore, we categorized all phosphatidyl testing as associated with minimal clinical utility.

 

CONCLUSIONS

In a large academic medical center, the majority of tests occurred in situations associated with minimal clinical utility. Strategies to improve thrombophilia testing practices are needed in order to minimize potentially inappropriate testing, provide more cost-effective care, and promote value-driven outcomes.

Disclosure

S.W. received financial support for this submitted work via a Bristol-Myers-Squibb grant. G.F. received financial support from Portola Pharmaceuticals for consulting and lectures that were not related to this submitted work.

Thrombophilia is a prothrombotic state, either acquired or inherited, leading to a thrombotic predisposition.¹ The most common heritable thrombophilias include factor V Leiden (FVL) and prothrombin G20210A. The most common acquired thrombophilia is the presence of phospholipid antibodies.¹ Thrombotic risk varies with thrombophilia type. For example, deficiencies of antithrombin, protein C and protein S, and the presence of phospholipid antibodies, confer higher risk than FVL and prothrombin G20210A.^2-5 Other thrombophilias (eg, methylenetetrahydrofolate reductase mutation, increased factor VIII activity) are relatively uncommon and/or their impact on thrombosis risk appears to be either minimal or unknown.^1-6 There is little clinical evidence that testing for thrombophilia impacts subsequent thrombosis prevention.^5,7,8 Multiple clinical guidelines and medical societies recommend against the routine and indiscriminate use of thrombophilia testing.^8-13 In general, thrombophilia testing should be considered only if the result would lead to changes in anticoagulant initiation, intensity, and/or duration, or might inform interventions to prevent thrombosis in asymptomatic family members.^8-13 However, thrombophilia testing rarely changes the acute management of a thrombotic event and may have harmful effects on patients and their family members because positive results may unnecessarily increase anxiety and negative results may provide false reassurance.^6,14-18 The cost-effectiveness of thrombophilia testing is unknown. Economic models have sought to quantify cost-effectiveness, but conclusions from these studies are limited.⁷

The utility of thrombophilia testing in emergency department (ED) and inpatient settings is further limited because patients are often treated and discharged before thrombophilia test results are available. Additionally, in these settings, multiple factors increase the risk of false-positive or false-negative results (eg, acute thrombosis, acute illness, pregnancy, and anticoagulant therapy).^19,20 The purpose of this study was to systematically assess thrombophilia testing patterns in the ED and hospitalized patients at an academic medical center and to quantify the proportion of tests associated with minimal clinical utility. We hypothesize that the majority of thrombophilia tests completed in the inpatient setting are associated with minimal clinical utility.

METHODS

Setting and Patients

This study was conducted at University of Utah Health Care (UUHC) University Hospital, a 488-bed academic medical center with a level I trauma center, primary stroke center, and 50-bed ED. Laboratory services for UUHC, including thrombophilia testing, are provided by a national reference laboratory, Associated Regional and University Pathologists Laboratories. This study included patients ≥18 years of age who received thrombophilia testing (Supplementary Table 1) during an ED visit or inpatient admission at University Hospital between July 1, 2014 and December 31, 2014. There were no exclusion criteria. An institutional electronic data repository was used to identify patients matching inclusion criteria. All study activities were reviewed and approved by the UUHC Institutional Review Board with a waiver of informed consent.

Outcomes

An electronic database query was used to identify patients, collect patient demographic information, and collect test characteristics. Each patient’s electronic medical record was manually reviewed to collect all other outcomes. Indication for thrombophilia testing was identified by manual review of provider notes. Thrombophilia tests occurring in situations associated with minimal clinical utility were defined as tests meeting at least one of the following criteria: patient discharged before test results were available for review; test type not recommended by published guidelines or by UUHC Thrombosis Service physicians for thrombophilia testing (Supplementary Table 2); test performed in situations associated with decreased accuracy; test was a duplicate test as a result of different thrombophilia panels containing identical tests; and test followed a provoked venous thromboembolism (VTE). Testing in situations associated with decreased accuracy are summarized in Supplementary Table 3 and included at least one of the following at the time of the test: anticoagulant therapy, acute thrombosis, pregnant or <8 weeks postpartum, and receiving estrogen-containing medications. Only test types known to be affected by the respective situation were included. Testing following a provoked VTE was defined as testing prompted by an acute thrombosis and performed within 3 months following major surgery (defined administratively as any surgery performed in an operating room), during pregnancy, <8 weeks postpartum, or while on estrogen-containing medications. Thrombophilia testing during anticoagulant therapy was defined as testing within 4 half-lives of anticoagulant administration based on medication administration records. Anticoagulant therapy changes were identified by comparing prior-to-admission and discharge medication lists.

 

Data Analysis

Patient and laboratory characteristics were summarized using descriptive statistics, including mean and standard deviation (SD) for continuous variables and proportions for categorical variables. Data analysis was performed using Excel (Version 2013, Microsoft Corporation. Redmond, Washington).

RESULTS

During the 6-month study period, 163 patients received at least 1 thrombophilia test during an ED visit or inpatient admission. Patient characteristics are summarized in Table 1. Tested patients were most commonly inpatients (96%) and female (71%). A total of 1451 thrombophilia tests were performed with a mean (± SD) of 8.9 ± 6.0 tests per patient. Testing characteristics are summarized in Table 2. Of the 39 different test types performed, the most commonly ordered were cardiolipin IgG and IgM antibodies (9% each), lupus anticoagulant (9%), and β₂-glycoprotein 1 IgG and IgM antibodies (8% each). When combined with testing for phosphatidyl antibodies, antiphospholipid tests accounted for 70% of all tests. Overall, 134 (9%) test results were positive. The mean time for results to become available was 2.2 ± 2.5 days. The frequency of test types with corresponding positivity rates and mean time for results to become available are summarized in Supplementary Table 4.

The indications for thrombophilia testing are summarized in Table 3. Ischemic stroke was the most common indication for testing (50% of tests; 35% of patients), followed by VTE (21% of tests; 21% of patients), and pregnancy-related conditions (eg, preeclampsia, intrauterine fetal demise; 15% of tests; 25% of patients). Overall, 911 tests (63%) occurred in situations associated with minimal clinical utility, with 126 patients (77%) receiving at least one of these tests (Table 4).

Anticoagulant therapy was changed in 43 patients (26%) in the following ways: initiated in 35 patients (21%), transitioned to a different anticoagulant in 6 patients (4%), and discontinued in 2 patients (1%). Of the 35 patients initiating anticoagulant therapy, 29 had documented thrombosis (24 had VTE, 4 had cerebral venous sinus thrombosis [CVST], and 1 had basilar artery thrombosis). Overall, 2 instances were identified in which initiation of anticoagulant therapy at discharge was in response to thrombophilia test results. In the first instance, warfarin without a parenteral anticoagulant bridge was initiated for a 54-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG antibodies, lupus anticoagulant, and protein S deficiency. In the second instance, warfarin with an enoxaparin bridge was initiated for a 26-year-old patient with a cryptogenic stroke who tested positive for β₂-glycoprotein 1 IgG and IgM antibodies, cardiolipin IgG antibodies, lupus anticoagulant, protein C deficiency, and antithrombin deficiency. Of the 163 patients receiving thrombophilia testing, only 2 patients (1%) had clear documentation of being offered genetic consultation.

DISCUSSION

In this retrospective analysis, 1451 thrombophilia tests were performed in 163 patients over 6 months. Tested patients were relatively young, which is likely explained by the number of patients tested for pregnancy-related conditions and the fact that a stroke or VTE in younger patients more frequently prompted providers to suspect thrombophilia. Nearly three-fourths of patients were female, which is likely due to testing for pregnancy-related conditions and possibly diagnostic suspicion bias given the comparative predilection of antiphospholipid syndrome for women. The patient characteristics in our study are consistent with other studies evaluating thrombophilia testing.^21,22

Thrombophilia testing was most frequently prompted by stroke, VTE, and pregnancy-related conditions. Only 26% of patients had acute thrombosis identified during the admission, primarily because of the high proportion of tests for cryptogenic strokes and pregnancy-related conditions. Thrombophilia testing is recommended in patients who have had a stroke when the stroke is considered to be cryptogenic after a standard stroke evaluation.²³ Thrombophilia testing in pregnancy-related conditions is controversial but is often considered in situations such as stillbirths with severe placental pathology and/or significant growth restriction, or in mothers with a personal or family history of thrombosis.²⁴ The proportion of testing for pregnancy-related conditions may be greater than at other institutions because UUHC Maternal Fetal Medicine is a referral center for women with conditions associated with hypercoagulability. Anticoagulant therapy was initiated in 21% of patients, but specifically in response to thrombophilia testing in only 2 instances; in most cases, anticoagulant therapy was initiated regardless of thrombophilia test results.

The results of this study confirm our hypothesis because the majority of thrombophilia tests occurred in situations associated with minimal clinical utility. Testing in these situations was not isolated to specific patients or medical services because 77% of tested patients received at least 1 test associated with minimal clinical utility. Our study took a conservative approach in defining scenarios associated with minimal clinical utility because other situations can also affect testing accuracy (eg, hepatic disease, nephrotic syndrome) but were not included in our analysis of this outcome.

The results of this study highlight opportunities to improve thrombophilia testing practices at our institution and may be generalizable to institutions with similar testing patterns. Because multiple medical services order thrombophilia tests, strategies to improve testing practices are still being determined. The results of this study can serve as a baseline for comparison after strategies are implemented. The most common situation associated with minimal clinical utility was the use of test types not generally recommended by guidelines or UUHC Thrombosis Service physicians for thrombophilia testing (eg, β₂-glycoprotein 1 IgA antibodies, phosphatidyl antibodies). We intend to require a hematology or thrombosis specialty consult prior to ordering these tests. This intervention alone could potentially decrease unnecessary testing by a third. Another consideration is to require a specialty consult prior to any inpatient thrombophilia testing. This strategy has been found to decrease inappropriate testing at other institutions.²¹ We also intend to streamline available thrombophilia testing panels because a poorly designed panel could lead to ordering of multiple tests associated with minimal clinical utility. At least 12 different thrombophilia panels are currently available in our computerized physician order entry system (see Supplementary Table 5). We hypothesize that current panel designs contribute to providers inadvertently ordering unintended or duplicate tests and that reducing the number of available panels and clearly delineating what tests are contained in each panel is likely to reduce unnecessary testing. Other strategies being considered include using electronic clinical decision support tools, implementing strict ordering criteria for all inpatient testing, and establishing a thrombosis stewardship program.

Our study was unique in at least 2 ways. First, previous studies describing thrombophilia testing have described testing patterns for patients with specific indications (eg, VTE), whereas our study described all thrombophilia tests regardless of indication. This allows for testing pattern comparisons across indications and medical services, increasing the generalizability of our results. Second, this study quantifies tests occurring in situations associated with a practical definition of minimal clinical utility.

Our study has several limitations: (1) Many variables were reliant on provider notes and other documentation, which allows for potential misclassification of variables. (2) It was not always possible to determine the ultimate utility of each test in clinical management decisions, and our study did not investigate the impact of thrombophilia testing on duration of anticoagulant therapy. Additionally, select situations could benefit from testing regardless if anticoagulant therapy is altered (eg, informing contraceptive choices). (3) Testing performed following a provoked acute thrombosis was defined as testing within 3 months following administratively defined major surgery. This definition could have included some minor procedures that do not substantially increase VTE risk, resulting in underestimated clinical utility. (4) The UUHC University Hospital serves as a referral hospital for a large geographical area, and investigators did not have access to outpatient records for a large proportion of discharged patients. As a result, frequency of repeat testing could not be assessed, possibly resulting in overestimated clinical utility. (5) In categorizing indications for testing, testing for CVST was subcategorized under testing for ischemic stroke based on presenting symptoms rather than on underlying pathophysiology. The rationale for this categorization is that patients with CVST were often tested based on presenting symptoms. Additionally, tests for CVST were ordered by the neurology service, which also ordered tests for all other ischemic stroke indications. (6) The purpose of our study was to investigate the subset of the hospital’s patient population that received thrombophilia testing, and patients were identified by tests received and not by diagnosis codes. As a result, we are unable to provide the proportion of total patients treated at the hospital for specific conditions who were tested (eg, the proportion of stroke patients that received thrombophilia testing). (7) Current practice guidelines do not recommend testing for phosphatidyl antibodies, even when traditional antiphospholipid testing is negative.^25-27 Although expert panels continue to explore associations between phosphatidyl antibodies and pregnancy morbidity and thrombotic events, the low level of evidence is insufficient to guide clinical management.²⁸ Therefore, we categorized all phosphatidyl testing as associated with minimal clinical utility.

 

CONCLUSIONS

In a large academic medical center, the majority of tests occurred in situations associated with minimal clinical utility. Strategies to improve thrombophilia testing practices are needed in order to minimize potentially inappropriate testing, provide more cost-effective care, and promote value-driven outcomes.

Disclosure

S.W. received financial support for this submitted work via a Bristol-Myers-Squibb grant. G.F. received financial support from Portola Pharmaceuticals for consulting and lectures that were not related to this submitted work.

Discussions about cardiopulmonary resuscitation (CPR) can be difficult due to their association with end of life. The Patient Self Determination Act (H.R.4449 — 101st Congress [1989-1990]) and institutional standards mandate collaboration between care providers and patients regarding goals of care in emergency situations such as cardiopulmonary arrest. The default option is to provide CPR, which may involve chest compressions, intubation, and/or defibrillation. Yet numerous studies show that a significant number of patients have no code preference documented in their medical chart, and even fewer report a conversation with their care provider about their wishes regarding CPR.^1-3 CPR is an invasive and potentially painful procedure with a higher chance of failure than success⁴, and yet many patients report that their provider did not discuss with them the risks and benefits of resuscitation.^5,6 Further highlighting the importance of individual discussions about CPR preferences is the reality that factors such as age and disease burden further skew the likelihood of survival after cardiopulmonary arrest.⁷

Complicating the lack of appropriate provider and patient discussion of the risks and benefits of resuscitation are significant misunderstandings about CPR in the lay population. Patients routinely overestimate the likelihood of survival following CPR.^8,9 This may be partially due to the portrayal of CPR in the lay media as highly efficacious.¹⁰ Other factors known to prevent effective provider-and-patient discussions about CPR preferences are providers’ discomfort with the subject¹¹ and perceived time constraints.¹²

Informational videos have been developed to assist patients with decision making about CPR and have been shown to impact patients’ choices in the setting of life-limiting diseases such as advanced cancer,^13-14 serious illness with a prognosis of less than 1 year,¹⁵ and dementia.¹⁶ While discussion of code status is vitally important in end-of-life planning for seriously ill individuals, delayed discussion of CPR preferences is associated with a significant increase in the number of invasive procedures performed at the end of life, increased length of stay in the hospital, and increased medical cost.¹⁷ Despite clear evidence that earlier discussion of resuscitation options are valuable, no studies have examined the impact of a video about code status options in the general patient population.

Here we present our findings of a randomized trial in patients hospitalized on the general medicine wards who were 65 years of age or older, regardless of illness severity or diagnosis. The video tool was a supplement for, rather than a replacement of, standard provider and patient communication about code preferences, and we compared patients who watched the video against controls who had standard discussions with their providers. Our video detailed the process of chest compressions and intubation during CPR and explained the differences between the code statuses: full code, do not resuscitate (DNR), and do not resuscitate/do not intubate (DNR/DNI). We found a significant difference between the 2 groups, with significantly more individuals in the video group choosing DNR/DNI. These findings suggest that video support tools may be a useful supplement to traditional provider discussions about code preferences in the general patient population.

METHODS

We enrolled patients from the general medicine wards at the Minneapolis VA Hospital from September 28, 2015 to October 23, 2015. Eligibility criteria included age 65 years or older, ability to provide informed consent, and ability to communicate in English. Study recruitment and data collection were performed by a study coordinator who was a house staff physician and had no role in the care of the participants. The medical charts of all general medicine patients were reviewed to determine if they met the age criteria. The physician of record for potential participants was contacted to assess if the patient was able to provide informed consent and communicate in English. Eligible patients were approached and informed consent was obtained from those who chose to participate in the study. After obtaining informed consent, patients were randomized using a random number generator to the intervention or usual-care arm of the study.

Those who were assigned to the intervention arm watched a 6-minute long video explaining the code-preference choices of full code, DNR, or DNR/DNI. Full code was described as possibly including CPR, intubation, and/or defibrillation depending on the clinical situation. Do not resuscitate was described as meaning no CPR or defibrillation but possible intubation in the case of respiratory failure. Do not resuscitate/do not intubate was explained as meaning no CPR, no defibrillation, and no intubation but rather permitting “natural death” to occur. The video showed a mock code with chest compressions, defibrillation, and intubation on a mannequin as well as palliative care specialists who discussed potential complications and survival rates of inhospital resuscitation.

The video was created at the University of Minnesota with the departments of palliative care and internal medicine (www.mmcgmeservices.org/codestat.html). After viewing the video, participants in the intervention arm filled out a questionnaire designed to assess their knowledge and beliefs about CPR and trust in their medical care providers. They were asked to circle their code preference. The participants’ medical teams were made aware of the code preferences and were counseled to discuss code preferences further if it was different from their previously documented code preference.

Participants in the control arm were assigned to usual care. At the institution where this study occurred, a discussion about code preferences between the patient and their medical team is considered the standard of care. After informed consent was obtained, participants filled out the same questionnaire as the participants in the intervention arm. They were asked to circle their code status preference. If they chose to ask questions about resuscitation, these were answered, but the study coordinator did not volunteer information about resuscitation or intervene in the medical care of the participants in any way.

Data collection included demographic and medical information from the participants’ charts for race, sex, age, and primary diagnosis for hospitalization (Table). We also collected data on the presence or absence of end-stage kidney disease; progressive pulmonary diseases including chronic obstructive pulmonary disease and interstitial lung disease; cirrhosis of the liver; chronic heart failure; or active cancer (defined as currently undergoing treatment or metastatic). In both study arms, the questionnaire included questions to assess patient trust in their medical team, beliefs about resuscitation, and desire to continue life-prolonging interventions in the absence of likely recovery to the point of being discharged from the hospital. Possible responses occurred on a continuum from “agree,” “agree somewhat,” “neither agree nor disagree,” “disagree somewhat,” and “disagree.”

All participants’ demographic characteristics and outcomes were described using proportions for categorical variables and means ± standard deviation for continuous variables. The primary outcome was participants’ stated code preference (full code, DNR, or DNR/DNI). Secondary outcomes included comparison of trust in medical providers, resuscitation beliefs, and desire for life-prolonging interventions as obtained from the questionnaire.

We analyzed code preferences between the intervention and control groups using Fisher exact test. We used analysis of variance (ANOVA) to compare questionnaire responses between the 2 groups. All reported P values are 2-sided with P < 0.05 considered significant. The project originally targeted a sample size of 194 participants for 80% power to detect a 20% difference in the code preference choices between intervention and control groups. Given the short time frame available to enroll participants, the target sample size was not reached. Propitiously, the effect size was greater than originally expected.

 

RESULTS

Study Participants

A total of 273 potentially eligible patients were approached to participate and 119 (44%) enrolled. (Figure 1). Of the 154 patients that were deemed eligible after initial screening, 42 patients were unable to give consent due to the severity of their illness or because of their mental status. Another 112 patients declined participation in the study, citing reasons such as disinterest in the consent paperwork, desire to spend time with visitors, and unease with the subject matter. Patients who declined participation did not differ significantly by age, sex, or race from those enrolled in the study.

Among the 119 participants, 60 were randomized to the control arm, and 59 were randomized to the intervention arm. Participants in the 2 arms did not differ significantly in age, sex, or race (P > 0.05), although all 4 women in the study were randomized to the intervention arm. Eighty-seven percent of the study population identified as white with the remainder including black, Asian, Pacific Islander, Native American, or declining to answer. The mean age was 75.8 years in the control arm vs. 75.2 years in the intervention arm.

Primary diagnoses in the study group ranged widely from relatively minor skin infections to acute pancreatitis. The control arm and the intervention arm did not differ significantly in the incidence of heart failure, pulmonary disease, renal dialysis, cirrhosis, stroke, or active cancer (P > 0.05). Patients were considered as having a stroke if they had suffered a stroke during their hospital admission or if they had long-term sequelae of prior stroke. Patients were considered as having active cancer if they were currently undergoing treatment or had metastases. Participants were considered as having multiple morbidities if they possessed 2 or more of the listed conditions. Between the control arm and the intervention arm, there was no significant difference in the number of participants with multiple morbidities (27% in the control group and 24% in the video group).

Code Status Preference

There was a significant difference in the code status preferences of the intervention arm and the control arm (P < 0.00001; Figure 2). In the control arm, 71% of participants chose full code, 12% chose DNR, and 17% chose DNR/DNI. In the intervention arm, only 37% chose full code, 7% chose DNR, and 56% chose DNR/DNI.

Secondary outcomes

Participants in the control and intervention arms were asked about their trust in their medical team (Question 1, Figure 3). There was no significant difference, but a trend towards less trust in the intervention group (P = 0.083) was seen with 93% of the control arm and 76% of the intervention arm agreeing with the statement “My doctors and healthcare team want what is best for me.”

Question 2, “If I choose to avoid resuscitation efforts, I will not receive care,” was designed to assess participants’ knowledge and perception about the care they would receive if they chose DNR/DNI as their code status. No significant difference was seen between the control and the interventions arms, with 28% of the control group agreeing with the statement, compared to 22% of the video group.

For question 3, participants were asked to respond to the statement “I would like to live as long as possible, even if I never leave the hospital.” No significant differences were seen between the control and the intervention arms, with 22% of both groups agreeing with the statement.

When we examined participant responses by the code status chosen, a significantly higher percentage of participants who chose full code agreed with the statement in question 3 (P = 0.0133). Of participants who chose full code, 27% agreed with the statement, compared to 18% of participants who chose DNR and 12% of participants who chose DNR/DNI. There was no significant difference (P > 0.05) between participant code status choice and either Question 1 or 2.

DISCUSSION

This study examined the effect of watching a video about CPR and intubation on the code status preferences of hospitalized patients. Participants who viewed a video about CPR and intubation were more likely to choose to forgo these treatments. Participants who chose CPR and intubation were more likely to agree that they would want to live as long as possible even if that time were spent in a medical setting.

To our knowledge, this is the first study to examine the role of a video decision support tool about code choices in the general hospital population, regardless of prognosis. Previous work has trialed the use of video support tools in hospitalized patients with a prognosis of less than 1 year,¹⁵ patients admitted to the ICU,¹⁸ and outpatients with cancer¹⁸ and those with dementia.¹⁶ Unlike previous studies, our study included a variety of illness severity.

Discussions about resuscitation are important for all adults admitted to the hospital because of the unpredictable nature of illness and the importance of providing high-quality care at the end of life.  A recent study indicates that in-hospital cardiopulmonary arrest occurs in almost 1 per 1000 hospital days.¹⁹ These discussions are particularly salient for patients 65 years and older because of the higher incidence of death in this group. Inpatient admission is often a result of a change in health status, making it an important time for patients to reassess their resuscitation preferences based on their physical state and known comorbidities.

Video tools supplement the traditional code status discussion in several key ways. They provide a visual simulation of the procedures that occur during a typical resuscitation. These tools can help patients understand what CPR and intubation entail and transmit information that might be missed in verbal discussions. Visual media is now a common way for patients to obtain medical information^20-22 and may be particularly helpful to patients who have low health literacy.²³Video tools also help ensure that patients receive all the facts about resuscitation irrespective of how busy their provider may be or how comfortable the provider is with the topic. Lastly, video tools can reinforce information that is shared in the initial code status discussion. Given the significant differences in code status preference between our control and video arms, it is clear that the video tool has a significant impact on patient choices.

While we feel that our study clearly indicates the utility of video tools in code status discussion in hospitalized patients, there are some limitations. The current study enrolled participants who were predominantly white and male. All participants were recruited from the Minneapolis Veterans Affairs Health Care System, Minnesota. The relatively homogenous study population may impact the study’s generalizability. Another potential limitation of our study was the large number of eligible participants who declined to participate (41%), with many citing that they did not want to sign the consent paperwork. Additionally, the study coordinator was not blinded to the randomization of the participants, which could result in ascertainment bias. Also of concern was a trend, albeit nonsignificant, towards less trust in the healthcare team in the video group. Because the study was not designed to assess trust in the healthcare team both before and after the intervention, it is unclear if this difference was a result of the video.

Another area of potential concern is that visual images can be edited to sway viewers’ opinions based on the way content is presented. In our video, we included input from palliative care and internal medicine specialists. Cardiopulmonary resuscitation and intubation were performed on a CPR mannequin. The risks and benefits of CPR and intubation were discussed, as were the implications of choosing DNR or DNR/DNI code statuses.

The questionnaire that we used to assess participants’ knowledge and beliefs about resuscitation showed no differences between the control and the intervention arms of the study. We were surprised that a significant number of participants in the intervention group agreed with the statement, “If I choose to avoid resuscitation efforts, I will not receive care.” Our video specifically addressed the common belief that choosing DNR/DNI or DNR code statuses means that a patient will not continue to receive medical care. It is possible that participants were confused by the way the question was worded or that they understood the question to apply only to care received after a cardiopulmonary arrest had occurred.

This study and several others^14-16 show that the use of video tools impacts participants’ code status preferences. There is clinical and humanistic importance in helping patients make informed decisions regarding whether or not they would want CPR and/or intubation if their heart were to stop or if they were to stop breathing. The data suggest that video tools are an efficient way to improve patient care and should be made widely available.

 

Disclosures: The authors report no conflicts of interest.

Discussions about cardiopulmonary resuscitation (CPR) can be difficult due to their association with end of life. The Patient Self Determination Act (H.R.4449 — 101st Congress [1989-1990]) and institutional standards mandate collaboration between care providers and patients regarding goals of care in emergency situations such as cardiopulmonary arrest. The default option is to provide CPR, which may involve chest compressions, intubation, and/or defibrillation. Yet numerous studies show that a significant number of patients have no code preference documented in their medical chart, and even fewer report a conversation with their care provider about their wishes regarding CPR.^1-3 CPR is an invasive and potentially painful procedure with a higher chance of failure than success⁴, and yet many patients report that their provider did not discuss with them the risks and benefits of resuscitation.^5,6 Further highlighting the importance of individual discussions about CPR preferences is the reality that factors such as age and disease burden further skew the likelihood of survival after cardiopulmonary arrest.⁷

Complicating the lack of appropriate provider and patient discussion of the risks and benefits of resuscitation are significant misunderstandings about CPR in the lay population. Patients routinely overestimate the likelihood of survival following CPR.^8,9 This may be partially due to the portrayal of CPR in the lay media as highly efficacious.¹⁰ Other factors known to prevent effective provider-and-patient discussions about CPR preferences are providers’ discomfort with the subject¹¹ and perceived time constraints.¹²

Informational videos have been developed to assist patients with decision making about CPR and have been shown to impact patients’ choices in the setting of life-limiting diseases such as advanced cancer,^13-14 serious illness with a prognosis of less than 1 year,¹⁵ and dementia.¹⁶ While discussion of code status is vitally important in end-of-life planning for seriously ill individuals, delayed discussion of CPR preferences is associated with a significant increase in the number of invasive procedures performed at the end of life, increased length of stay in the hospital, and increased medical cost.¹⁷ Despite clear evidence that earlier discussion of resuscitation options are valuable, no studies have examined the impact of a video about code status options in the general patient population.

Here we present our findings of a randomized trial in patients hospitalized on the general medicine wards who were 65 years of age or older, regardless of illness severity or diagnosis. The video tool was a supplement for, rather than a replacement of, standard provider and patient communication about code preferences, and we compared patients who watched the video against controls who had standard discussions with their providers. Our video detailed the process of chest compressions and intubation during CPR and explained the differences between the code statuses: full code, do not resuscitate (DNR), and do not resuscitate/do not intubate (DNR/DNI). We found a significant difference between the 2 groups, with significantly more individuals in the video group choosing DNR/DNI. These findings suggest that video support tools may be a useful supplement to traditional provider discussions about code preferences in the general patient population.

METHODS

We enrolled patients from the general medicine wards at the Minneapolis VA Hospital from September 28, 2015 to October 23, 2015. Eligibility criteria included age 65 years or older, ability to provide informed consent, and ability to communicate in English. Study recruitment and data collection were performed by a study coordinator who was a house staff physician and had no role in the care of the participants. The medical charts of all general medicine patients were reviewed to determine if they met the age criteria. The physician of record for potential participants was contacted to assess if the patient was able to provide informed consent and communicate in English. Eligible patients were approached and informed consent was obtained from those who chose to participate in the study. After obtaining informed consent, patients were randomized using a random number generator to the intervention or usual-care arm of the study.

Those who were assigned to the intervention arm watched a 6-minute long video explaining the code-preference choices of full code, DNR, or DNR/DNI. Full code was described as possibly including CPR, intubation, and/or defibrillation depending on the clinical situation. Do not resuscitate was described as meaning no CPR or defibrillation but possible intubation in the case of respiratory failure. Do not resuscitate/do not intubate was explained as meaning no CPR, no defibrillation, and no intubation but rather permitting “natural death” to occur. The video showed a mock code with chest compressions, defibrillation, and intubation on a mannequin as well as palliative care specialists who discussed potential complications and survival rates of inhospital resuscitation.

The video was created at the University of Minnesota with the departments of palliative care and internal medicine (www.mmcgmeservices.org/codestat.html). After viewing the video, participants in the intervention arm filled out a questionnaire designed to assess their knowledge and beliefs about CPR and trust in their medical care providers. They were asked to circle their code preference. The participants’ medical teams were made aware of the code preferences and were counseled to discuss code preferences further if it was different from their previously documented code preference.

Participants in the control arm were assigned to usual care. At the institution where this study occurred, a discussion about code preferences between the patient and their medical team is considered the standard of care. After informed consent was obtained, participants filled out the same questionnaire as the participants in the intervention arm. They were asked to circle their code status preference. If they chose to ask questions about resuscitation, these were answered, but the study coordinator did not volunteer information about resuscitation or intervene in the medical care of the participants in any way.

Data collection included demographic and medical information from the participants’ charts for race, sex, age, and primary diagnosis for hospitalization (Table). We also collected data on the presence or absence of end-stage kidney disease; progressive pulmonary diseases including chronic obstructive pulmonary disease and interstitial lung disease; cirrhosis of the liver; chronic heart failure; or active cancer (defined as currently undergoing treatment or metastatic). In both study arms, the questionnaire included questions to assess patient trust in their medical team, beliefs about resuscitation, and desire to continue life-prolonging interventions in the absence of likely recovery to the point of being discharged from the hospital. Possible responses occurred on a continuum from “agree,” “agree somewhat,” “neither agree nor disagree,” “disagree somewhat,” and “disagree.”

All participants’ demographic characteristics and outcomes were described using proportions for categorical variables and means ± standard deviation for continuous variables. The primary outcome was participants’ stated code preference (full code, DNR, or DNR/DNI). Secondary outcomes included comparison of trust in medical providers, resuscitation beliefs, and desire for life-prolonging interventions as obtained from the questionnaire.

We analyzed code preferences between the intervention and control groups using Fisher exact test. We used analysis of variance (ANOVA) to compare questionnaire responses between the 2 groups. All reported P values are 2-sided with P < 0.05 considered significant. The project originally targeted a sample size of 194 participants for 80% power to detect a 20% difference in the code preference choices between intervention and control groups. Given the short time frame available to enroll participants, the target sample size was not reached. Propitiously, the effect size was greater than originally expected.

 

RESULTS

Study Participants

A total of 273 potentially eligible patients were approached to participate and 119 (44%) enrolled. (Figure 1). Of the 154 patients that were deemed eligible after initial screening, 42 patients were unable to give consent due to the severity of their illness or because of their mental status. Another 112 patients declined participation in the study, citing reasons such as disinterest in the consent paperwork, desire to spend time with visitors, and unease with the subject matter. Patients who declined participation did not differ significantly by age, sex, or race from those enrolled in the study.

Among the 119 participants, 60 were randomized to the control arm, and 59 were randomized to the intervention arm. Participants in the 2 arms did not differ significantly in age, sex, or race (P > 0.05), although all 4 women in the study were randomized to the intervention arm. Eighty-seven percent of the study population identified as white with the remainder including black, Asian, Pacific Islander, Native American, or declining to answer. The mean age was 75.8 years in the control arm vs. 75.2 years in the intervention arm.

Primary diagnoses in the study group ranged widely from relatively minor skin infections to acute pancreatitis. The control arm and the intervention arm did not differ significantly in the incidence of heart failure, pulmonary disease, renal dialysis, cirrhosis, stroke, or active cancer (P > 0.05). Patients were considered as having a stroke if they had suffered a stroke during their hospital admission or if they had long-term sequelae of prior stroke. Patients were considered as having active cancer if they were currently undergoing treatment or had metastases. Participants were considered as having multiple morbidities if they possessed 2 or more of the listed conditions. Between the control arm and the intervention arm, there was no significant difference in the number of participants with multiple morbidities (27% in the control group and 24% in the video group).

Code Status Preference

There was a significant difference in the code status preferences of the intervention arm and the control arm (P < 0.00001; Figure 2). In the control arm, 71% of participants chose full code, 12% chose DNR, and 17% chose DNR/DNI. In the intervention arm, only 37% chose full code, 7% chose DNR, and 56% chose DNR/DNI.

Secondary outcomes

Participants in the control and intervention arms were asked about their trust in their medical team (Question 1, Figure 3). There was no significant difference, but a trend towards less trust in the intervention group (P = 0.083) was seen with 93% of the control arm and 76% of the intervention arm agreeing with the statement “My doctors and healthcare team want what is best for me.”

Question 2, “If I choose to avoid resuscitation efforts, I will not receive care,” was designed to assess participants’ knowledge and perception about the care they would receive if they chose DNR/DNI as their code status. No significant difference was seen between the control and the interventions arms, with 28% of the control group agreeing with the statement, compared to 22% of the video group.

For question 3, participants were asked to respond to the statement “I would like to live as long as possible, even if I never leave the hospital.” No significant differences were seen between the control and the intervention arms, with 22% of both groups agreeing with the statement.

When we examined participant responses by the code status chosen, a significantly higher percentage of participants who chose full code agreed with the statement in question 3 (P = 0.0133). Of participants who chose full code, 27% agreed with the statement, compared to 18% of participants who chose DNR and 12% of participants who chose DNR/DNI. There was no significant difference (P > 0.05) between participant code status choice and either Question 1 or 2.

DISCUSSION

This study examined the effect of watching a video about CPR and intubation on the code status preferences of hospitalized patients. Participants who viewed a video about CPR and intubation were more likely to choose to forgo these treatments. Participants who chose CPR and intubation were more likely to agree that they would want to live as long as possible even if that time were spent in a medical setting.

To our knowledge, this is the first study to examine the role of a video decision support tool about code choices in the general hospital population, regardless of prognosis. Previous work has trialed the use of video support tools in hospitalized patients with a prognosis of less than 1 year,¹⁵ patients admitted to the ICU,¹⁸ and outpatients with cancer¹⁸ and those with dementia.¹⁶ Unlike previous studies, our study included a variety of illness severity.

Discussions about resuscitation are important for all adults admitted to the hospital because of the unpredictable nature of illness and the importance of providing high-quality care at the end of life.  A recent study indicates that in-hospital cardiopulmonary arrest occurs in almost 1 per 1000 hospital days.¹⁹ These discussions are particularly salient for patients 65 years and older because of the higher incidence of death in this group. Inpatient admission is often a result of a change in health status, making it an important time for patients to reassess their resuscitation preferences based on their physical state and known comorbidities.

Video tools supplement the traditional code status discussion in several key ways. They provide a visual simulation of the procedures that occur during a typical resuscitation. These tools can help patients understand what CPR and intubation entail and transmit information that might be missed in verbal discussions. Visual media is now a common way for patients to obtain medical information^20-22 and may be particularly helpful to patients who have low health literacy.²³Video tools also help ensure that patients receive all the facts about resuscitation irrespective of how busy their provider may be or how comfortable the provider is with the topic. Lastly, video tools can reinforce information that is shared in the initial code status discussion. Given the significant differences in code status preference between our control and video arms, it is clear that the video tool has a significant impact on patient choices.

While we feel that our study clearly indicates the utility of video tools in code status discussion in hospitalized patients, there are some limitations. The current study enrolled participants who were predominantly white and male. All participants were recruited from the Minneapolis Veterans Affairs Health Care System, Minnesota. The relatively homogenous study population may impact the study’s generalizability. Another potential limitation of our study was the large number of eligible participants who declined to participate (41%), with many citing that they did not want to sign the consent paperwork. Additionally, the study coordinator was not blinded to the randomization of the participants, which could result in ascertainment bias. Also of concern was a trend, albeit nonsignificant, towards less trust in the healthcare team in the video group. Because the study was not designed to assess trust in the healthcare team both before and after the intervention, it is unclear if this difference was a result of the video.

Another area of potential concern is that visual images can be edited to sway viewers’ opinions based on the way content is presented. In our video, we included input from palliative care and internal medicine specialists. Cardiopulmonary resuscitation and intubation were performed on a CPR mannequin. The risks and benefits of CPR and intubation were discussed, as were the implications of choosing DNR or DNR/DNI code statuses.

The questionnaire that we used to assess participants’ knowledge and beliefs about resuscitation showed no differences between the control and the intervention arms of the study. We were surprised that a significant number of participants in the intervention group agreed with the statement, “If I choose to avoid resuscitation efforts, I will not receive care.” Our video specifically addressed the common belief that choosing DNR/DNI or DNR code statuses means that a patient will not continue to receive medical care. It is possible that participants were confused by the way the question was worded or that they understood the question to apply only to care received after a cardiopulmonary arrest had occurred.

This study and several others^14-16 show that the use of video tools impacts participants’ code status preferences. There is clinical and humanistic importance in helping patients make informed decisions regarding whether or not they would want CPR and/or intubation if their heart were to stop or if they were to stop breathing. The data suggest that video tools are an efficient way to improve patient care and should be made widely available.

 

Disclosures: The authors report no conflicts of interest.

Discussions about cardiopulmonary resuscitation (CPR) can be difficult due to their association with end of life. The Patient Self Determination Act (H.R.4449 — 101st Congress [1989-1990]) and institutional standards mandate collaboration between care providers and patients regarding goals of care in emergency situations such as cardiopulmonary arrest. The default option is to provide CPR, which may involve chest compressions, intubation, and/or defibrillation. Yet numerous studies show that a significant number of patients have no code preference documented in their medical chart, and even fewer report a conversation with their care provider about their wishes regarding CPR.^1-3 CPR is an invasive and potentially painful procedure with a higher chance of failure than success⁴, and yet many patients report that their provider did not discuss with them the risks and benefits of resuscitation.^5,6 Further highlighting the importance of individual discussions about CPR preferences is the reality that factors such as age and disease burden further skew the likelihood of survival after cardiopulmonary arrest.⁷

Complicating the lack of appropriate provider and patient discussion of the risks and benefits of resuscitation are significant misunderstandings about CPR in the lay population. Patients routinely overestimate the likelihood of survival following CPR.^8,9 This may be partially due to the portrayal of CPR in the lay media as highly efficacious.¹⁰ Other factors known to prevent effective provider-and-patient discussions about CPR preferences are providers’ discomfort with the subject¹¹ and perceived time constraints.¹²

Informational videos have been developed to assist patients with decision making about CPR and have been shown to impact patients’ choices in the setting of life-limiting diseases such as advanced cancer,^13-14 serious illness with a prognosis of less than 1 year,¹⁵ and dementia.¹⁶ While discussion of code status is vitally important in end-of-life planning for seriously ill individuals, delayed discussion of CPR preferences is associated with a significant increase in the number of invasive procedures performed at the end of life, increased length of stay in the hospital, and increased medical cost.¹⁷ Despite clear evidence that earlier discussion of resuscitation options are valuable, no studies have examined the impact of a video about code status options in the general patient population.

Here we present our findings of a randomized trial in patients hospitalized on the general medicine wards who were 65 years of age or older, regardless of illness severity or diagnosis. The video tool was a supplement for, rather than a replacement of, standard provider and patient communication about code preferences, and we compared patients who watched the video against controls who had standard discussions with their providers. Our video detailed the process of chest compressions and intubation during CPR and explained the differences between the code statuses: full code, do not resuscitate (DNR), and do not resuscitate/do not intubate (DNR/DNI). We found a significant difference between the 2 groups, with significantly more individuals in the video group choosing DNR/DNI. These findings suggest that video support tools may be a useful supplement to traditional provider discussions about code preferences in the general patient population.

METHODS

We enrolled patients from the general medicine wards at the Minneapolis VA Hospital from September 28, 2015 to October 23, 2015. Eligibility criteria included age 65 years or older, ability to provide informed consent, and ability to communicate in English. Study recruitment and data collection were performed by a study coordinator who was a house staff physician and had no role in the care of the participants. The medical charts of all general medicine patients were reviewed to determine if they met the age criteria. The physician of record for potential participants was contacted to assess if the patient was able to provide informed consent and communicate in English. Eligible patients were approached and informed consent was obtained from those who chose to participate in the study. After obtaining informed consent, patients were randomized using a random number generator to the intervention or usual-care arm of the study.

Those who were assigned to the intervention arm watched a 6-minute long video explaining the code-preference choices of full code, DNR, or DNR/DNI. Full code was described as possibly including CPR, intubation, and/or defibrillation depending on the clinical situation. Do not resuscitate was described as meaning no CPR or defibrillation but possible intubation in the case of respiratory failure. Do not resuscitate/do not intubate was explained as meaning no CPR, no defibrillation, and no intubation but rather permitting “natural death” to occur. The video showed a mock code with chest compressions, defibrillation, and intubation on a mannequin as well as palliative care specialists who discussed potential complications and survival rates of inhospital resuscitation.

The video was created at the University of Minnesota with the departments of palliative care and internal medicine (www.mmcgmeservices.org/codestat.html). After viewing the video, participants in the intervention arm filled out a questionnaire designed to assess their knowledge and beliefs about CPR and trust in their medical care providers. They were asked to circle their code preference. The participants’ medical teams were made aware of the code preferences and were counseled to discuss code preferences further if it was different from their previously documented code preference.

Participants in the control arm were assigned to usual care. At the institution where this study occurred, a discussion about code preferences between the patient and their medical team is considered the standard of care. After informed consent was obtained, participants filled out the same questionnaire as the participants in the intervention arm. They were asked to circle their code status preference. If they chose to ask questions about resuscitation, these were answered, but the study coordinator did not volunteer information about resuscitation or intervene in the medical care of the participants in any way.

Data collection included demographic and medical information from the participants’ charts for race, sex, age, and primary diagnosis for hospitalization (Table). We also collected data on the presence or absence of end-stage kidney disease; progressive pulmonary diseases including chronic obstructive pulmonary disease and interstitial lung disease; cirrhosis of the liver; chronic heart failure; or active cancer (defined as currently undergoing treatment or metastatic). In both study arms, the questionnaire included questions to assess patient trust in their medical team, beliefs about resuscitation, and desire to continue life-prolonging interventions in the absence of likely recovery to the point of being discharged from the hospital. Possible responses occurred on a continuum from “agree,” “agree somewhat,” “neither agree nor disagree,” “disagree somewhat,” and “disagree.”

All participants’ demographic characteristics and outcomes were described using proportions for categorical variables and means ± standard deviation for continuous variables. The primary outcome was participants’ stated code preference (full code, DNR, or DNR/DNI). Secondary outcomes included comparison of trust in medical providers, resuscitation beliefs, and desire for life-prolonging interventions as obtained from the questionnaire.

We analyzed code preferences between the intervention and control groups using Fisher exact test. We used analysis of variance (ANOVA) to compare questionnaire responses between the 2 groups. All reported P values are 2-sided with P < 0.05 considered significant. The project originally targeted a sample size of 194 participants for 80% power to detect a 20% difference in the code preference choices between intervention and control groups. Given the short time frame available to enroll participants, the target sample size was not reached. Propitiously, the effect size was greater than originally expected.

 

RESULTS

Study Participants

A total of 273 potentially eligible patients were approached to participate and 119 (44%) enrolled. (Figure 1). Of the 154 patients that were deemed eligible after initial screening, 42 patients were unable to give consent due to the severity of their illness or because of their mental status. Another 112 patients declined participation in the study, citing reasons such as disinterest in the consent paperwork, desire to spend time with visitors, and unease with the subject matter. Patients who declined participation did not differ significantly by age, sex, or race from those enrolled in the study.

Among the 119 participants, 60 were randomized to the control arm, and 59 were randomized to the intervention arm. Participants in the 2 arms did not differ significantly in age, sex, or race (P > 0.05), although all 4 women in the study were randomized to the intervention arm. Eighty-seven percent of the study population identified as white with the remainder including black, Asian, Pacific Islander, Native American, or declining to answer. The mean age was 75.8 years in the control arm vs. 75.2 years in the intervention arm.

Primary diagnoses in the study group ranged widely from relatively minor skin infections to acute pancreatitis. The control arm and the intervention arm did not differ significantly in the incidence of heart failure, pulmonary disease, renal dialysis, cirrhosis, stroke, or active cancer (P > 0.05). Patients were considered as having a stroke if they had suffered a stroke during their hospital admission or if they had long-term sequelae of prior stroke. Patients were considered as having active cancer if they were currently undergoing treatment or had metastases. Participants were considered as having multiple morbidities if they possessed 2 or more of the listed conditions. Between the control arm and the intervention arm, there was no significant difference in the number of participants with multiple morbidities (27% in the control group and 24% in the video group).

Code Status Preference

There was a significant difference in the code status preferences of the intervention arm and the control arm (P < 0.00001; Figure 2). In the control arm, 71% of participants chose full code, 12% chose DNR, and 17% chose DNR/DNI. In the intervention arm, only 37% chose full code, 7% chose DNR, and 56% chose DNR/DNI.

Secondary outcomes

Participants in the control and intervention arms were asked about their trust in their medical team (Question 1, Figure 3). There was no significant difference, but a trend towards less trust in the intervention group (P = 0.083) was seen with 93% of the control arm and 76% of the intervention arm agreeing with the statement “My doctors and healthcare team want what is best for me.”

Question 2, “If I choose to avoid resuscitation efforts, I will not receive care,” was designed to assess participants’ knowledge and perception about the care they would receive if they chose DNR/DNI as their code status. No significant difference was seen between the control and the interventions arms, with 28% of the control group agreeing with the statement, compared to 22% of the video group.

For question 3, participants were asked to respond to the statement “I would like to live as long as possible, even if I never leave the hospital.” No significant differences were seen between the control and the intervention arms, with 22% of both groups agreeing with the statement.

When we examined participant responses by the code status chosen, a significantly higher percentage of participants who chose full code agreed with the statement in question 3 (P = 0.0133). Of participants who chose full code, 27% agreed with the statement, compared to 18% of participants who chose DNR and 12% of participants who chose DNR/DNI. There was no significant difference (P > 0.05) between participant code status choice and either Question 1 or 2.

DISCUSSION

This study examined the effect of watching a video about CPR and intubation on the code status preferences of hospitalized patients. Participants who viewed a video about CPR and intubation were more likely to choose to forgo these treatments. Participants who chose CPR and intubation were more likely to agree that they would want to live as long as possible even if that time were spent in a medical setting.

To our knowledge, this is the first study to examine the role of a video decision support tool about code choices in the general hospital population, regardless of prognosis. Previous work has trialed the use of video support tools in hospitalized patients with a prognosis of less than 1 year,¹⁵ patients admitted to the ICU,¹⁸ and outpatients with cancer¹⁸ and those with dementia.¹⁶ Unlike previous studies, our study included a variety of illness severity.

Discussions about resuscitation are important for all adults admitted to the hospital because of the unpredictable nature of illness and the importance of providing high-quality care at the end of life.  A recent study indicates that in-hospital cardiopulmonary arrest occurs in almost 1 per 1000 hospital days.¹⁹ These discussions are particularly salient for patients 65 years and older because of the higher incidence of death in this group. Inpatient admission is often a result of a change in health status, making it an important time for patients to reassess their resuscitation preferences based on their physical state and known comorbidities.

Video tools supplement the traditional code status discussion in several key ways. They provide a visual simulation of the procedures that occur during a typical resuscitation. These tools can help patients understand what CPR and intubation entail and transmit information that might be missed in verbal discussions. Visual media is now a common way for patients to obtain medical information^20-22 and may be particularly helpful to patients who have low health literacy.²³Video tools also help ensure that patients receive all the facts about resuscitation irrespective of how busy their provider may be or how comfortable the provider is with the topic. Lastly, video tools can reinforce information that is shared in the initial code status discussion. Given the significant differences in code status preference between our control and video arms, it is clear that the video tool has a significant impact on patient choices.

While we feel that our study clearly indicates the utility of video tools in code status discussion in hospitalized patients, there are some limitations. The current study enrolled participants who were predominantly white and male. All participants were recruited from the Minneapolis Veterans Affairs Health Care System, Minnesota. The relatively homogenous study population may impact the study’s generalizability. Another potential limitation of our study was the large number of eligible participants who declined to participate (41%), with many citing that they did not want to sign the consent paperwork. Additionally, the study coordinator was not blinded to the randomization of the participants, which could result in ascertainment bias. Also of concern was a trend, albeit nonsignificant, towards less trust in the healthcare team in the video group. Because the study was not designed to assess trust in the healthcare team both before and after the intervention, it is unclear if this difference was a result of the video.

Another area of potential concern is that visual images can be edited to sway viewers’ opinions based on the way content is presented. In our video, we included input from palliative care and internal medicine specialists. Cardiopulmonary resuscitation and intubation were performed on a CPR mannequin. The risks and benefits of CPR and intubation were discussed, as were the implications of choosing DNR or DNR/DNI code statuses.

The questionnaire that we used to assess participants’ knowledge and beliefs about resuscitation showed no differences between the control and the intervention arms of the study. We were surprised that a significant number of participants in the intervention group agreed with the statement, “If I choose to avoid resuscitation efforts, I will not receive care.” Our video specifically addressed the common belief that choosing DNR/DNI or DNR code statuses means that a patient will not continue to receive medical care. It is possible that participants were confused by the way the question was worded or that they understood the question to apply only to care received after a cardiopulmonary arrest had occurred.

This study and several others^14-16 show that the use of video tools impacts participants’ code status preferences. There is clinical and humanistic importance in helping patients make informed decisions regarding whether or not they would want CPR and/or intubation if their heart were to stop or if they were to stop breathing. The data suggest that video tools are an efficient way to improve patient care and should be made widely available.

 

Disclosures: The authors report no conflicts of interest.

The past 3 decades have seen an evolution in the understanding, diagnosis, and treatment of groin pain, both chronic and acute, in athletes and non-athletes alike. Groin pain and groin injury are common. Most cases are transient, with patients returning to their activities within weeks or months. There has also been increasing awareness of a definitive population of patients who do not get better, or who improve and plateau before reaching preinjury level of performance.^1-3 Several authors have brought more attention to the injury, introducing vocabulary, theories, diagnostic testing, and diagnoses, which now constitute a knowledge base.^1,3-5

As stated in almost every article on groin pain and diagnosis, lack of cohesive agreement and vocabulary, and consistent protocols and procedures, has abounded, making general understanding and agreement in this area inconsistent.^1,6-8In this article, members of a tertiary-care group specializing in chronic groin pain, athletic pubalgia (sports hernia), and inguinal herniorrhaphy outline their clinical examination, diagnostic algorithm, imaging protocol, treatment strategy, and outcomes for a population of patients referred by physicians and allied health professionals for a suspected diagnosis of athletic pubalgia.

Background

The pubic symphysis acts as a stabilizing central anchor with elaborate involvement of the anterior structures, including the rectus abdominis, adductor longus, and inguinal ligaments.^3,7,9 Literature from Europe, Australia, and the United States has described groin pain, mostly in professional athletes, involving these pubic structures and attachments. Several publications have been addressing chronic groin pain, and each has its own diagnostic algorithm, imaging protocol, and treatment strategy.^3,6,9-18

Terminology specific to groin pain in athletes is not new, and has a varied history dating to the early 20th century. Terms such as sportsman hernia¹⁹ and subsequently sports hernia²⁰, have recently been embraced by the lay population. In 1999, Gibbon²¹ described shearing of the common adductor–rectus abdominis anatomical and functional unit and referenced a 1902 anatomical text that describes vertical ligamentous fibers contiguous with rectus sheath and adductor muscles, both attaching to the pubis. Injury to this region is the basis of pubalgia, a term originally used in 1984 by Brunet to describe a pain syndrome at the pubis.²²

Many authors have proposed replacing sports hernia with athletic pubalgia.^{1,3,6,7,10,14,18,23} These terms refer to a group of musculoskeletal processes that occur in and around the pubic symphysis and that share similar mechanisms of injury and common clinical manifestations. The condition was originally described in high-performance athletes, and at one point the term sports hernia was reserved for this patient population.⁵ According to many authors, presence of an inguinal hernia excludes the diagnosis.^1,2,5Magnetic resonance imaging (MRI) has helped to advance and define our understanding of the injury.¹⁰ As the history of the literature suggests, earlier concepts of chronic pain focused either on the medial aspect of the inguinal canal and its structures or on the pubic attachments. Many specialists in the area have concluded that the chronic groin pain injury can and often does embody both elements.^3,9 Correlation with MRI findings, injury seen during surgical procedures, and cadaveric studies have directed our understanding to a structure, the pre-pubic aponeurotic complex (P-PAC), or rectus aponeurotic plate.^12,24,25 Anatomically, the P-PAC, which has several fascial components, attaches posteriorly to the pubic bone and, to a degree, the pubic symphyseal cartilaginous disc. Major contributions to the P-PAC are fibers from the rectus abdominis tendon, the medial aspect of the transversalis and internal oblique muscles (the conjoint tendon, according to some), the inguinal ligament, and the adductor longus tendon.²⁶When communicating with referring physicians, we use the term athletic pubalgia to indicate a specific injury. The athletic pubalgia injury can be defined as serial microtearing,¹ or complete tearing, of the posterior attachment of the P-PAC off the anterior pubis.^3,10 Complete tearing or displacement can occur unilaterally or across the midline to the other side. As athletic pubalgia is a specific anatomical injury rather than a broad category of findings, an additional pathologic diagnosis, such as inguinal hernia, does not exclude the diagnosis of athletic pubalgia. Unfortunately, the terms sports hernia and sportsman hernia, commonly used in the media and in professional communities, have largely confused the broader understanding of nuances and of the differences between the specific injuries and MRI findings.¹⁸

 

Our Experience

In our practice, we see groin pain patients referred by internists, physiatrists, physical therapists, trainers, general surgeons, urologists, gynecologists, and orthopedic surgeons. In many cases, patients have been through several consultations and work-ups, as their pain syndrome does not fall under a specific category. Patients without inguinal hernia, hip injury, urologic, or gynecologic issues typically are referred to a physiatrist or a physical therapist. Often, there are marginal improvements with physical therapy, but in some cases the injury never completely resolves, and the patient continues to have pain with activity or return to sports.

Most of our patients are nonprofessional athletes, men and women who range widely in age and participate casually or regularly in sporting events. Most lack the rigorous training, conditioning, and close supervision that professional athletes receive. Many other patients are nonprofessional but elite athletes who train 7 days a week for marathons, ultramarathons, triathlons, obstacle course races (“mudders”), and similar events.

Work-Up

A single algorithm is used for all patients initially referred to the surgeon’s office for pelvic or groin pain. The initial interview directs attention to injury onset and mechanism, duration of rest or physical therapy after surgery, pain quality and pain levels, and antagonistic movements and positions. Examination starts with assessment for inguinal, femoral, and umbilical hernias. Resisted sit-up, leg-raise, adduction, and hip assessment tests are performed. The P-PAC is examined with a maneuver similar to the one used for inguinal hernia, as it allows for better assessment of the transversalis fascia (over the direct space) to determine if the inguinal canal floor is attenuated and bulges forward with the Valsalva maneuver. Then, the lateral aspect of the rectus muscle is assessed for pain, usually with the head raised to contract the muscle, to determine tenderness along the lateral border. The rectus edge is traced down to the pubis at its attachment, the superolateral border of the P-PAC. Examination proceeds medially, over the rectus attachment, toward the pubic symphysis, continuing the assessment for tenderness. Laterally, the conjoint tendon and inguinal ligament medial attachments are assessed at the level of the pubic tubercle, which represents the lateral border of the P-PAC. Finally, the examination continues to the inferior border with assessment of the adductor longus attachment, which is best performed with the leg in an adducted position. In the acute or semiacute setting (pain within 1 year of injury onset), tenderness is often elicited. With long-standing injuries, pain is often not elicited, but the patient experiences pain along this axis during activity or afterward.

Patients with positive history and physical examination findings proceed through an MRI protocol designed to detect pathology of the pubic symphysis, hips, and inguinal canals (Figures 1A-1D).

Imaging includes use of axial single-shot fast spin-echo with and without the Valsalva maneuver. Use of this maneuver increases sensitivity in detecting abdominal wall deficiencies, such as inguinal hernias (Figures 2A, 2B).

We developed this pubalgia protocol (Table) to specifically assess each of the potential areas of concern.

Treatment

Patients who report sustaining an acute groin injury within the previous 6 months are treated nonoperatively. A combination of rest, nonsteroidal anti-inflammatory drugs, and physical therapy is generally recommended.^2,10 In cases of failed nonoperative management, patients are evaluated for surgery. No single operation is recommended for all patients.^1,6,14,27,28 (Larson²⁶ recently reviewed results from several trials involving a variety of surgical repairs and found return-to-sports rates ranging from 80% to 100%.) Findings from the physical examination and from the properly protocolled MRI examination are used in planning surgery to correct any pathology that could be contributing to symptoms or destabilization of the structures attaching to the pubis. Disruption of the P-PAC from the pubis would be repaired, for example. Additional injuries, such as partial or complete detachment of the conjoint tendon or inguinal ligament, may be repaired as well. If the transversalis fascia is attenuated and bulging forward, the inguinal floor is closed. Adductor longus tendon pathology is addressed, most commonly with partial tendinolysis. Often, concomitant inguinal hernias are found, and these may be repaired in open fashion while other maneuvers are being performed, or laparoscopically.

Materials and Methods

After receiving study approval from our Institutional Review Board, we retrospectively searched for all MRIs performed by our radiology department between March 1, 2011 and March 31, 2013 on patients referred for an indication of groin pain, sports hernia, or athletic pubalgia. Patients were excluded if they were younger than 18 years any time during their care. Some patients previously or subsequently underwent computed tomography or ultrasonography. MRIs were reviewed and positive findings were compiled in a database. Charts were reviewed to identify which patients in the dataset underwent surgery, after MRI, to address their presenting chief complaint. Surgery date and procedure(s) performed were recorded. Patients were interviewed by telephone as part of the in-office postoperative follow-up.

 

Results

One hundred nineteen MRIs were performed on 117 patients (97 men, 83%). Mean age was 39.8 years. Seventy-nine patients (68%) had an MRI finding of athletic pubalgia, 67 (57%) had an acetabular labral tear in one or both hip joints, and 41 (35%) had a true inguinal hernia. Concomitant findings were common: 47 cases of athletic pubalgia and labral tear(s), 28 cases of athletic pubalgia and inguinal hernia, and 15 cases of all 3 (athletic pubalgia, labral tear, inguinal hernia).

Use of breath-hold axial single-shot fast spin-echo sequences with and without the Valsalva maneuver increased sensitivity in detecting pathologies—inguinal hernia and Gilmore groin in particular. On 24 of the 119 MRIs, the Valsalva maneuver either revealed the finding or made it significantly more apparent.

Of all patients referred for MRI for chronic groin pain, 48 (41%) subsequently underwent surgery. In 29 surgeries, the rectus abdominis, adductor longus, and/or pre-pubic aponeurotic plate were repaired; in 13 cases, herniorrhaphy was performed as well; in 2 cases, masses involving the spermatic cord were removed.

The most common surgery (30 cases) was herniorrhaphy, which was performed as a single procedure, multiple procedures, or in combination with procedures not related to a true hernia. Eighteen patients underwent surgery only for hernia repair.

Of the 79 patients with MRI-positive athletic pubalgia, 39 subsequently underwent surgery, and 31 (79%) of these were followed up by telephone. Mean duration of rest after surgery was 6.2 weeks. Twelve patients (39%) had physical therapy after surgery, some as early as 4 weeks, and some have continued their therapy since surgery. Of the 31 patients who were followed up after surgery, 23 (74%) resumed previous activity levels. Return to previous activity level took these patients a mean of 17.9 weeks. When asked if outcomes satisfied their expectations, 28 patients (90%) said yes, and 3 said no.

Forty patients with MRI-positive athletic pubalgia were nonoperatively treated, and 28 (70%) of these patients were followed up. In this group, mean duration of rest after surgery was 6.9 weeks. Thirteen patients (46%) participated in physical therapy, for a mean duration of 10.8 weeks. Of the patients followed up, 19 (68%) returned to previous activity level. Twenty-one patients (75%) were satisfied with their outcome.

Discussion

Diagnosis and treatment of chronic groin pain have had a long, confusing, and frustrating history for both patients and the medical professionals who provide them with care.^3,6,7,10 Historically, the problem has been, in part, the lack of diagnostic capabilities. Currently, however, pubalgia MRI protocol allows the exact pathology to be demonstrated.³ As already noted, concomitant hip pathology or inguinal hernia is not unusual⁸; any structural abnormality in the area is a potential destabilizer of the structures attached to the pubis.¹⁸ Solving only one of these issues may offer only partial resolution of symptoms and thereby reduce the rate of successful treatment of groin pain.

Diagnostic algorithms are being developed. In addition, nonoperative treatments are being tried for some of the issues. Physicians are giving diagnostic and therapeutic steroid injections in the pubic cleft, along the rectus abdominis/adductor longus complex, or posterior to the P-PAC. Platelet-rich plasma injection therapy has had limited success.²⁹This article provides a snapshot of what a tertiary-care group of physicians specializing in chronic groin pain sees in an unfiltered setting. We think this is instructive for several reasons.

First, many patients in our population have visited a multitude of specialists without receiving a diagnosis or being referred appropriately. Simply, many specialists do not know the next step in treating groin pain and thus do not make the appropriate referral. Until recently, the literature has not been helpful. It has poorly described the constellation of injuries comprising chronic groin pain. More significantly, groin injuries have been presented as ambiguous injuries lacking effective treatment. Over the past decade, however, abundant literature on the correlation of these injuries with specific MRI findings has made the case otherwise.

Second, a specific MRI pubalgia protocol is needed. Inability to make a correct diagnosis, because of improper MRI, continues to add to the confusion surrounding the injury and undoubtedly prolongs the general medical community’s thinking that diagnosis and treatment of chronic groin pain are elusive. Our data support this point in many ways. Although all patients in this study were seen by a medical professional before coming to our office, none had received a diagnosis of occult hernia or attenuated transversalis fascia; nevertheless, we identified inguinal hernia, Gilmore groin, or both in 44% of these patients. These findings are not surprising, as MRI was the crucial link in diagnosis. In addition, the point made by other groin pain specialists—that a hernia precludes a pubalgia diagnosis^1,2,5—is not supported by our data. Inguinal hernia can and does exist in conjunction with pubalgia. More than half the patients in our study had a combined diagnosis. We contend that, much as hip labral pathology occurs concomitantly with pubalgia,²³ inguinal hernia may be a predisposing factor as well. A defect in the direct or indirect space can destabilize the area and place additional strain on the pubic attachments.

In our experience, the dynamic Valsalva sequence improves detection of true hernias and anterior abdominal wall deficiencies and should be included in each protocol for the evaluation of acute or chronic groin pain.

Shear forces and injury at the pubis can occur outside professional athletics. Our patient population is nonprofessional athletes, teenagers to retirees, and all can develop athletic pubalgia. Ninety percent of surveyed patients who received a diagnosis and were treated surgically were satisfied with their outcomes.

Am J Orthop. 2017;46(4):E251-E256. Copyright Frontline Medical Communications Inc. 2017. All rights reserved.

The past 3 decades have seen an evolution in the understanding, diagnosis, and treatment of groin pain, both chronic and acute, in athletes and non-athletes alike. Groin pain and groin injury are common. Most cases are transient, with patients returning to their activities within weeks or months. There has also been increasing awareness of a definitive population of patients who do not get better, or who improve and plateau before reaching preinjury level of performance.^1-3 Several authors have brought more attention to the injury, introducing vocabulary, theories, diagnostic testing, and diagnoses, which now constitute a knowledge base.^1,3-5

As stated in almost every article on groin pain and diagnosis, lack of cohesive agreement and vocabulary, and consistent protocols and procedures, has abounded, making general understanding and agreement in this area inconsistent.^1,6-8In this article, members of a tertiary-care group specializing in chronic groin pain, athletic pubalgia (sports hernia), and inguinal herniorrhaphy outline their clinical examination, diagnostic algorithm, imaging protocol, treatment strategy, and outcomes for a population of patients referred by physicians and allied health professionals for a suspected diagnosis of athletic pubalgia.

Background

The pubic symphysis acts as a stabilizing central anchor with elaborate involvement of the anterior structures, including the rectus abdominis, adductor longus, and inguinal ligaments.^3,7,9 Literature from Europe, Australia, and the United States has described groin pain, mostly in professional athletes, involving these pubic structures and attachments. Several publications have been addressing chronic groin pain, and each has its own diagnostic algorithm, imaging protocol, and treatment strategy.^3,6,9-18

Terminology specific to groin pain in athletes is not new, and has a varied history dating to the early 20th century. Terms such as sportsman hernia¹⁹ and subsequently sports hernia²⁰, have recently been embraced by the lay population. In 1999, Gibbon²¹ described shearing of the common adductor–rectus abdominis anatomical and functional unit and referenced a 1902 anatomical text that describes vertical ligamentous fibers contiguous with rectus sheath and adductor muscles, both attaching to the pubis. Injury to this region is the basis of pubalgia, a term originally used in 1984 by Brunet to describe a pain syndrome at the pubis.²²

Many authors have proposed replacing sports hernia with athletic pubalgia.^{1,3,6,7,10,14,18,23} These terms refer to a group of musculoskeletal processes that occur in and around the pubic symphysis and that share similar mechanisms of injury and common clinical manifestations. The condition was originally described in high-performance athletes, and at one point the term sports hernia was reserved for this patient population.⁵ According to many authors, presence of an inguinal hernia excludes the diagnosis.^1,2,5Magnetic resonance imaging (MRI) has helped to advance and define our understanding of the injury.¹⁰ As the history of the literature suggests, earlier concepts of chronic pain focused either on the medial aspect of the inguinal canal and its structures or on the pubic attachments. Many specialists in the area have concluded that the chronic groin pain injury can and often does embody both elements.^3,9 Correlation with MRI findings, injury seen during surgical procedures, and cadaveric studies have directed our understanding to a structure, the pre-pubic aponeurotic complex (P-PAC), or rectus aponeurotic plate.^12,24,25 Anatomically, the P-PAC, which has several fascial components, attaches posteriorly to the pubic bone and, to a degree, the pubic symphyseal cartilaginous disc. Major contributions to the P-PAC are fibers from the rectus abdominis tendon, the medial aspect of the transversalis and internal oblique muscles (the conjoint tendon, according to some), the inguinal ligament, and the adductor longus tendon.²⁶When communicating with referring physicians, we use the term athletic pubalgia to indicate a specific injury. The athletic pubalgia injury can be defined as serial microtearing,¹ or complete tearing, of the posterior attachment of the P-PAC off the anterior pubis.^3,10 Complete tearing or displacement can occur unilaterally or across the midline to the other side. As athletic pubalgia is a specific anatomical injury rather than a broad category of findings, an additional pathologic diagnosis, such as inguinal hernia, does not exclude the diagnosis of athletic pubalgia. Unfortunately, the terms sports hernia and sportsman hernia, commonly used in the media and in professional communities, have largely confused the broader understanding of nuances and of the differences between the specific injuries and MRI findings.¹⁸

 

Our Experience

In our practice, we see groin pain patients referred by internists, physiatrists, physical therapists, trainers, general surgeons, urologists, gynecologists, and orthopedic surgeons. In many cases, patients have been through several consultations and work-ups, as their pain syndrome does not fall under a specific category. Patients without inguinal hernia, hip injury, urologic, or gynecologic issues typically are referred to a physiatrist or a physical therapist. Often, there are marginal improvements with physical therapy, but in some cases the injury never completely resolves, and the patient continues to have pain with activity or return to sports.

Most of our patients are nonprofessional athletes, men and women who range widely in age and participate casually or regularly in sporting events. Most lack the rigorous training, conditioning, and close supervision that professional athletes receive. Many other patients are nonprofessional but elite athletes who train 7 days a week for marathons, ultramarathons, triathlons, obstacle course races (“mudders”), and similar events.

Work-Up

A single algorithm is used for all patients initially referred to the surgeon’s office for pelvic or groin pain. The initial interview directs attention to injury onset and mechanism, duration of rest or physical therapy after surgery, pain quality and pain levels, and antagonistic movements and positions. Examination starts with assessment for inguinal, femoral, and umbilical hernias. Resisted sit-up, leg-raise, adduction, and hip assessment tests are performed. The P-PAC is examined with a maneuver similar to the one used for inguinal hernia, as it allows for better assessment of the transversalis fascia (over the direct space) to determine if the inguinal canal floor is attenuated and bulges forward with the Valsalva maneuver. Then, the lateral aspect of the rectus muscle is assessed for pain, usually with the head raised to contract the muscle, to determine tenderness along the lateral border. The rectus edge is traced down to the pubis at its attachment, the superolateral border of the P-PAC. Examination proceeds medially, over the rectus attachment, toward the pubic symphysis, continuing the assessment for tenderness. Laterally, the conjoint tendon and inguinal ligament medial attachments are assessed at the level of the pubic tubercle, which represents the lateral border of the P-PAC. Finally, the examination continues to the inferior border with assessment of the adductor longus attachment, which is best performed with the leg in an adducted position. In the acute or semiacute setting (pain within 1 year of injury onset), tenderness is often elicited. With long-standing injuries, pain is often not elicited, but the patient experiences pain along this axis during activity or afterward.

Patients with positive history and physical examination findings proceed through an MRI protocol designed to detect pathology of the pubic symphysis, hips, and inguinal canals (Figures 1A-1D).

Imaging includes use of axial single-shot fast spin-echo with and without the Valsalva maneuver. Use of this maneuver increases sensitivity in detecting abdominal wall deficiencies, such as inguinal hernias (Figures 2A, 2B).

We developed this pubalgia protocol (Table) to specifically assess each of the potential areas of concern.

Treatment

Patients who report sustaining an acute groin injury within the previous 6 months are treated nonoperatively. A combination of rest, nonsteroidal anti-inflammatory drugs, and physical therapy is generally recommended.^2,10 In cases of failed nonoperative management, patients are evaluated for surgery. No single operation is recommended for all patients.^1,6,14,27,28 (Larson²⁶ recently reviewed results from several trials involving a variety of surgical repairs and found return-to-sports rates ranging from 80% to 100%.) Findings from the physical examination and from the properly protocolled MRI examination are used in planning surgery to correct any pathology that could be contributing to symptoms or destabilization of the structures attaching to the pubis. Disruption of the P-PAC from the pubis would be repaired, for example. Additional injuries, such as partial or complete detachment of the conjoint tendon or inguinal ligament, may be repaired as well. If the transversalis fascia is attenuated and bulging forward, the inguinal floor is closed. Adductor longus tendon pathology is addressed, most commonly with partial tendinolysis. Often, concomitant inguinal hernias are found, and these may be repaired in open fashion while other maneuvers are being performed, or laparoscopically.

Materials and Methods

After receiving study approval from our Institutional Review Board, we retrospectively searched for all MRIs performed by our radiology department between March 1, 2011 and March 31, 2013 on patients referred for an indication of groin pain, sports hernia, or athletic pubalgia. Patients were excluded if they were younger than 18 years any time during their care. Some patients previously or subsequently underwent computed tomography or ultrasonography. MRIs were reviewed and positive findings were compiled in a database. Charts were reviewed to identify which patients in the dataset underwent surgery, after MRI, to address their presenting chief complaint. Surgery date and procedure(s) performed were recorded. Patients were interviewed by telephone as part of the in-office postoperative follow-up.

 

Results

One hundred nineteen MRIs were performed on 117 patients (97 men, 83%). Mean age was 39.8 years. Seventy-nine patients (68%) had an MRI finding of athletic pubalgia, 67 (57%) had an acetabular labral tear in one or both hip joints, and 41 (35%) had a true inguinal hernia. Concomitant findings were common: 47 cases of athletic pubalgia and labral tear(s), 28 cases of athletic pubalgia and inguinal hernia, and 15 cases of all 3 (athletic pubalgia, labral tear, inguinal hernia).

Use of breath-hold axial single-shot fast spin-echo sequences with and without the Valsalva maneuver increased sensitivity in detecting pathologies—inguinal hernia and Gilmore groin in particular. On 24 of the 119 MRIs, the Valsalva maneuver either revealed the finding or made it significantly more apparent.

Of all patients referred for MRI for chronic groin pain, 48 (41%) subsequently underwent surgery. In 29 surgeries, the rectus abdominis, adductor longus, and/or pre-pubic aponeurotic plate were repaired; in 13 cases, herniorrhaphy was performed as well; in 2 cases, masses involving the spermatic cord were removed.

The most common surgery (30 cases) was herniorrhaphy, which was performed as a single procedure, multiple procedures, or in combination with procedures not related to a true hernia. Eighteen patients underwent surgery only for hernia repair.

Of the 79 patients with MRI-positive athletic pubalgia, 39 subsequently underwent surgery, and 31 (79%) of these were followed up by telephone. Mean duration of rest after surgery was 6.2 weeks. Twelve patients (39%) had physical therapy after surgery, some as early as 4 weeks, and some have continued their therapy since surgery. Of the 31 patients who were followed up after surgery, 23 (74%) resumed previous activity levels. Return to previous activity level took these patients a mean of 17.9 weeks. When asked if outcomes satisfied their expectations, 28 patients (90%) said yes, and 3 said no.

Forty patients with MRI-positive athletic pubalgia were nonoperatively treated, and 28 (70%) of these patients were followed up. In this group, mean duration of rest after surgery was 6.9 weeks. Thirteen patients (46%) participated in physical therapy, for a mean duration of 10.8 weeks. Of the patients followed up, 19 (68%) returned to previous activity level. Twenty-one patients (75%) were satisfied with their outcome.

Discussion

Diagnosis and treatment of chronic groin pain have had a long, confusing, and frustrating history for both patients and the medical professionals who provide them with care.^3,6,7,10 Historically, the problem has been, in part, the lack of diagnostic capabilities. Currently, however, pubalgia MRI protocol allows the exact pathology to be demonstrated.³ As already noted, concomitant hip pathology or inguinal hernia is not unusual⁸; any structural abnormality in the area is a potential destabilizer of the structures attached to the pubis.¹⁸ Solving only one of these issues may offer only partial resolution of symptoms and thereby reduce the rate of successful treatment of groin pain.

Diagnostic algorithms are being developed. In addition, nonoperative treatments are being tried for some of the issues. Physicians are giving diagnostic and therapeutic steroid injections in the pubic cleft, along the rectus abdominis/adductor longus complex, or posterior to the P-PAC. Platelet-rich plasma injection therapy has had limited success.²⁹This article provides a snapshot of what a tertiary-care group of physicians specializing in chronic groin pain sees in an unfiltered setting. We think this is instructive for several reasons.

First, many patients in our population have visited a multitude of specialists without receiving a diagnosis or being referred appropriately. Simply, many specialists do not know the next step in treating groin pain and thus do not make the appropriate referral. Until recently, the literature has not been helpful. It has poorly described the constellation of injuries comprising chronic groin pain. More significantly, groin injuries have been presented as ambiguous injuries lacking effective treatment. Over the past decade, however, abundant literature on the correlation of these injuries with specific MRI findings has made the case otherwise.

Second, a specific MRI pubalgia protocol is needed. Inability to make a correct diagnosis, because of improper MRI, continues to add to the confusion surrounding the injury and undoubtedly prolongs the general medical community’s thinking that diagnosis and treatment of chronic groin pain are elusive. Our data support this point in many ways. Although all patients in this study were seen by a medical professional before coming to our office, none had received a diagnosis of occult hernia or attenuated transversalis fascia; nevertheless, we identified inguinal hernia, Gilmore groin, or both in 44% of these patients. These findings are not surprising, as MRI was the crucial link in diagnosis. In addition, the point made by other groin pain specialists—that a hernia precludes a pubalgia diagnosis^1,2,5—is not supported by our data. Inguinal hernia can and does exist in conjunction with pubalgia. More than half the patients in our study had a combined diagnosis. We contend that, much as hip labral pathology occurs concomitantly with pubalgia,²³ inguinal hernia may be a predisposing factor as well. A defect in the direct or indirect space can destabilize the area and place additional strain on the pubic attachments.

In our experience, the dynamic Valsalva sequence improves detection of true hernias and anterior abdominal wall deficiencies and should be included in each protocol for the evaluation of acute or chronic groin pain.

Shear forces and injury at the pubis can occur outside professional athletics. Our patient population is nonprofessional athletes, teenagers to retirees, and all can develop athletic pubalgia. Ninety percent of surveyed patients who received a diagnosis and were treated surgically were satisfied with their outcomes.

Am J Orthop. 2017;46(4):E251-E256. Copyright Frontline Medical Communications Inc. 2017. All rights reserved.

The past 3 decades have seen an evolution in the understanding, diagnosis, and treatment of groin pain, both chronic and acute, in athletes and non-athletes alike. Groin pain and groin injury are common. Most cases are transient, with patients returning to their activities within weeks or months. There has also been increasing awareness of a definitive population of patients who do not get better, or who improve and plateau before reaching preinjury level of performance.^1-3 Several authors have brought more attention to the injury, introducing vocabulary, theories, diagnostic testing, and diagnoses, which now constitute a knowledge base.^1,3-5

As stated in almost every article on groin pain and diagnosis, lack of cohesive agreement and vocabulary, and consistent protocols and procedures, has abounded, making general understanding and agreement in this area inconsistent.^1,6-8In this article, members of a tertiary-care group specializing in chronic groin pain, athletic pubalgia (sports hernia), and inguinal herniorrhaphy outline their clinical examination, diagnostic algorithm, imaging protocol, treatment strategy, and outcomes for a population of patients referred by physicians and allied health professionals for a suspected diagnosis of athletic pubalgia.

Background

The pubic symphysis acts as a stabilizing central anchor with elaborate involvement of the anterior structures, including the rectus abdominis, adductor longus, and inguinal ligaments.^3,7,9 Literature from Europe, Australia, and the United States has described groin pain, mostly in professional athletes, involving these pubic structures and attachments. Several publications have been addressing chronic groin pain, and each has its own diagnostic algorithm, imaging protocol, and treatment strategy.^3,6,9-18

Terminology specific to groin pain in athletes is not new, and has a varied history dating to the early 20th century. Terms such as sportsman hernia¹⁹ and subsequently sports hernia²⁰, have recently been embraced by the lay population. In 1999, Gibbon²¹ described shearing of the common adductor–rectus abdominis anatomical and functional unit and referenced a 1902 anatomical text that describes vertical ligamentous fibers contiguous with rectus sheath and adductor muscles, both attaching to the pubis. Injury to this region is the basis of pubalgia, a term originally used in 1984 by Brunet to describe a pain syndrome at the pubis.²²

Many authors have proposed replacing sports hernia with athletic pubalgia.^{1,3,6,7,10,14,18,23} These terms refer to a group of musculoskeletal processes that occur in and around the pubic symphysis and that share similar mechanisms of injury and common clinical manifestations. The condition was originally described in high-performance athletes, and at one point the term sports hernia was reserved for this patient population.⁵ According to many authors, presence of an inguinal hernia excludes the diagnosis.^1,2,5Magnetic resonance imaging (MRI) has helped to advance and define our understanding of the injury.¹⁰ As the history of the literature suggests, earlier concepts of chronic pain focused either on the medial aspect of the inguinal canal and its structures or on the pubic attachments. Many specialists in the area have concluded that the chronic groin pain injury can and often does embody both elements.^3,9 Correlation with MRI findings, injury seen during surgical procedures, and cadaveric studies have directed our understanding to a structure, the pre-pubic aponeurotic complex (P-PAC), or rectus aponeurotic plate.^12,24,25 Anatomically, the P-PAC, which has several fascial components, attaches posteriorly to the pubic bone and, to a degree, the pubic symphyseal cartilaginous disc. Major contributions to the P-PAC are fibers from the rectus abdominis tendon, the medial aspect of the transversalis and internal oblique muscles (the conjoint tendon, according to some), the inguinal ligament, and the adductor longus tendon.²⁶When communicating with referring physicians, we use the term athletic pubalgia to indicate a specific injury. The athletic pubalgia injury can be defined as serial microtearing,¹ or complete tearing, of the posterior attachment of the P-PAC off the anterior pubis.^3,10 Complete tearing or displacement can occur unilaterally or across the midline to the other side. As athletic pubalgia is a specific anatomical injury rather than a broad category of findings, an additional pathologic diagnosis, such as inguinal hernia, does not exclude the diagnosis of athletic pubalgia. Unfortunately, the terms sports hernia and sportsman hernia, commonly used in the media and in professional communities, have largely confused the broader understanding of nuances and of the differences between the specific injuries and MRI findings.¹⁸

 

Our Experience

In our practice, we see groin pain patients referred by internists, physiatrists, physical therapists, trainers, general surgeons, urologists, gynecologists, and orthopedic surgeons. In many cases, patients have been through several consultations and work-ups, as their pain syndrome does not fall under a specific category. Patients without inguinal hernia, hip injury, urologic, or gynecologic issues typically are referred to a physiatrist or a physical therapist. Often, there are marginal improvements with physical therapy, but in some cases the injury never completely resolves, and the patient continues to have pain with activity or return to sports.

Most of our patients are nonprofessional athletes, men and women who range widely in age and participate casually or regularly in sporting events. Most lack the rigorous training, conditioning, and close supervision that professional athletes receive. Many other patients are nonprofessional but elite athletes who train 7 days a week for marathons, ultramarathons, triathlons, obstacle course races (“mudders”), and similar events.

Work-Up

A single algorithm is used for all patients initially referred to the surgeon’s office for pelvic or groin pain. The initial interview directs attention to injury onset and mechanism, duration of rest or physical therapy after surgery, pain quality and pain levels, and antagonistic movements and positions. Examination starts with assessment for inguinal, femoral, and umbilical hernias. Resisted sit-up, leg-raise, adduction, and hip assessment tests are performed. The P-PAC is examined with a maneuver similar to the one used for inguinal hernia, as it allows for better assessment of the transversalis fascia (over the direct space) to determine if the inguinal canal floor is attenuated and bulges forward with the Valsalva maneuver. Then, the lateral aspect of the rectus muscle is assessed for pain, usually with the head raised to contract the muscle, to determine tenderness along the lateral border. The rectus edge is traced down to the pubis at its attachment, the superolateral border of the P-PAC. Examination proceeds medially, over the rectus attachment, toward the pubic symphysis, continuing the assessment for tenderness. Laterally, the conjoint tendon and inguinal ligament medial attachments are assessed at the level of the pubic tubercle, which represents the lateral border of the P-PAC. Finally, the examination continues to the inferior border with assessment of the adductor longus attachment, which is best performed with the leg in an adducted position. In the acute or semiacute setting (pain within 1 year of injury onset), tenderness is often elicited. With long-standing injuries, pain is often not elicited, but the patient experiences pain along this axis during activity or afterward.

Patients with positive history and physical examination findings proceed through an MRI protocol designed to detect pathology of the pubic symphysis, hips, and inguinal canals (Figures 1A-1D).

Imaging includes use of axial single-shot fast spin-echo with and without the Valsalva maneuver. Use of this maneuver increases sensitivity in detecting abdominal wall deficiencies, such as inguinal hernias (Figures 2A, 2B).

We developed this pubalgia protocol (Table) to specifically assess each of the potential areas of concern.

Treatment

Patients who report sustaining an acute groin injury within the previous 6 months are treated nonoperatively. A combination of rest, nonsteroidal anti-inflammatory drugs, and physical therapy is generally recommended.^2,10 In cases of failed nonoperative management, patients are evaluated for surgery. No single operation is recommended for all patients.^1,6,14,27,28 (Larson²⁶ recently reviewed results from several trials involving a variety of surgical repairs and found return-to-sports rates ranging from 80% to 100%.) Findings from the physical examination and from the properly protocolled MRI examination are used in planning surgery to correct any pathology that could be contributing to symptoms or destabilization of the structures attaching to the pubis. Disruption of the P-PAC from the pubis would be repaired, for example. Additional injuries, such as partial or complete detachment of the conjoint tendon or inguinal ligament, may be repaired as well. If the transversalis fascia is attenuated and bulging forward, the inguinal floor is closed. Adductor longus tendon pathology is addressed, most commonly with partial tendinolysis. Often, concomitant inguinal hernias are found, and these may be repaired in open fashion while other maneuvers are being performed, or laparoscopically.

Materials and Methods

After receiving study approval from our Institutional Review Board, we retrospectively searched for all MRIs performed by our radiology department between March 1, 2011 and March 31, 2013 on patients referred for an indication of groin pain, sports hernia, or athletic pubalgia. Patients were excluded if they were younger than 18 years any time during their care. Some patients previously or subsequently underwent computed tomography or ultrasonography. MRIs were reviewed and positive findings were compiled in a database. Charts were reviewed to identify which patients in the dataset underwent surgery, after MRI, to address their presenting chief complaint. Surgery date and procedure(s) performed were recorded. Patients were interviewed by telephone as part of the in-office postoperative follow-up.

 

Results

One hundred nineteen MRIs were performed on 117 patients (97 men, 83%). Mean age was 39.8 years. Seventy-nine patients (68%) had an MRI finding of athletic pubalgia, 67 (57%) had an acetabular labral tear in one or both hip joints, and 41 (35%) had a true inguinal hernia. Concomitant findings were common: 47 cases of athletic pubalgia and labral tear(s), 28 cases of athletic pubalgia and inguinal hernia, and 15 cases of all 3 (athletic pubalgia, labral tear, inguinal hernia).

Use of breath-hold axial single-shot fast spin-echo sequences with and without the Valsalva maneuver increased sensitivity in detecting pathologies—inguinal hernia and Gilmore groin in particular. On 24 of the 119 MRIs, the Valsalva maneuver either revealed the finding or made it significantly more apparent.

Of all patients referred for MRI for chronic groin pain, 48 (41%) subsequently underwent surgery. In 29 surgeries, the rectus abdominis, adductor longus, and/or pre-pubic aponeurotic plate were repaired; in 13 cases, herniorrhaphy was performed as well; in 2 cases, masses involving the spermatic cord were removed.

The most common surgery (30 cases) was herniorrhaphy, which was performed as a single procedure, multiple procedures, or in combination with procedures not related to a true hernia. Eighteen patients underwent surgery only for hernia repair.

Of the 79 patients with MRI-positive athletic pubalgia, 39 subsequently underwent surgery, and 31 (79%) of these were followed up by telephone. Mean duration of rest after surgery was 6.2 weeks. Twelve patients (39%) had physical therapy after surgery, some as early as 4 weeks, and some have continued their therapy since surgery. Of the 31 patients who were followed up after surgery, 23 (74%) resumed previous activity levels. Return to previous activity level took these patients a mean of 17.9 weeks. When asked if outcomes satisfied their expectations, 28 patients (90%) said yes, and 3 said no.

Forty patients with MRI-positive athletic pubalgia were nonoperatively treated, and 28 (70%) of these patients were followed up. In this group, mean duration of rest after surgery was 6.9 weeks. Thirteen patients (46%) participated in physical therapy, for a mean duration of 10.8 weeks. Of the patients followed up, 19 (68%) returned to previous activity level. Twenty-one patients (75%) were satisfied with their outcome.

Discussion

Diagnosis and treatment of chronic groin pain have had a long, confusing, and frustrating history for both patients and the medical professionals who provide them with care.^3,6,7,10 Historically, the problem has been, in part, the lack of diagnostic capabilities. Currently, however, pubalgia MRI protocol allows the exact pathology to be demonstrated.³ As already noted, concomitant hip pathology or inguinal hernia is not unusual⁸; any structural abnormality in the area is a potential destabilizer of the structures attached to the pubis.¹⁸ Solving only one of these issues may offer only partial resolution of symptoms and thereby reduce the rate of successful treatment of groin pain.

Diagnostic algorithms are being developed. In addition, nonoperative treatments are being tried for some of the issues. Physicians are giving diagnostic and therapeutic steroid injections in the pubic cleft, along the rectus abdominis/adductor longus complex, or posterior to the P-PAC. Platelet-rich plasma injection therapy has had limited success.²⁹This article provides a snapshot of what a tertiary-care group of physicians specializing in chronic groin pain sees in an unfiltered setting. We think this is instructive for several reasons.

First, many patients in our population have visited a multitude of specialists without receiving a diagnosis or being referred appropriately. Simply, many specialists do not know the next step in treating groin pain and thus do not make the appropriate referral. Until recently, the literature has not been helpful. It has poorly described the constellation of injuries comprising chronic groin pain. More significantly, groin injuries have been presented as ambiguous injuries lacking effective treatment. Over the past decade, however, abundant literature on the correlation of these injuries with specific MRI findings has made the case otherwise.

Second, a specific MRI pubalgia protocol is needed. Inability to make a correct diagnosis, because of improper MRI, continues to add to the confusion surrounding the injury and undoubtedly prolongs the general medical community’s thinking that diagnosis and treatment of chronic groin pain are elusive. Our data support this point in many ways. Although all patients in this study were seen by a medical professional before coming to our office, none had received a diagnosis of occult hernia or attenuated transversalis fascia; nevertheless, we identified inguinal hernia, Gilmore groin, or both in 44% of these patients. These findings are not surprising, as MRI was the crucial link in diagnosis. In addition, the point made by other groin pain specialists—that a hernia precludes a pubalgia diagnosis^1,2,5—is not supported by our data. Inguinal hernia can and does exist in conjunction with pubalgia. More than half the patients in our study had a combined diagnosis. We contend that, much as hip labral pathology occurs concomitantly with pubalgia,²³ inguinal hernia may be a predisposing factor as well. A defect in the direct or indirect space can destabilize the area and place additional strain on the pubic attachments.

In our experience, the dynamic Valsalva sequence improves detection of true hernias and anterior abdominal wall deficiencies and should be included in each protocol for the evaluation of acute or chronic groin pain.

Shear forces and injury at the pubis can occur outside professional athletics. Our patient population is nonprofessional athletes, teenagers to retirees, and all can develop athletic pubalgia. Ninety percent of surveyed patients who received a diagnosis and were treated surgically were satisfied with their outcomes.

Am J Orthop. 2017;46(4):E251-E256. Copyright Frontline Medical Communications Inc. 2017. All rights reserved.

Hospitals typically allocate beds based on historical patient volumes. If funding decreases, hospitals will usually try to maximize resource utilization by allocating beds to attain occupancies close to 100% for significant periods of time. This will invariably cause days in which hospital occupancy exceeds capacity, at which time critical entry points (such as the emergency department and operating room) will become blocked. This creates significant concerns over the patient quality of care.

Hospital administrators have very few options when hospital occupancy exceeds 100%. They could postpone admissions for “planned” cases, bring in additional staff to increase capacity, or instigate additional methods to increase hospital discharges such as expanding care resources in the community. All options are costly, bothersome, or cannot be actioned immediately. The need for these options could be minimized by enabling hospital administrators to make more informed decisions regarding hospital bed management by knowing the likely number of discharges in the next 24 hours.

Predicting the number of people who will be discharged in the next day can be approached in several ways. One approach would be to calculate each patient’s expected length of stay and then use the variation around that estimate to calculate each day’s discharge probability. Several studies have attempted to model hospital length of stay using a broad assortment of methodologies, but a mechanism to accurately predict this outcome has been elusive^1,2 (with Verburg et al.³ concluding in their study’s abstract that “…it is difficult to predict length of stay…”). A second approach would be to use survival analysis methods to generate each patient’s hazard of discharge over time, which could be directly converted to an expected daily risk of discharge. However, this approach is complicated by the concurrent need to include time-dependent covariates and consider the competing risk of death in hospital, which can complicate survival modeling.^4,5 A third approach would be the implementation of a longitudinal analysis using marginal models to predict the daily probability of discharge,⁶ but this method quickly overwhelms computer resources when large datasets are present.

In this study, we decided to use nonparametric models to predict the daily number of hospital discharges. We first identified patient groups with distinct discharge patterns. We then calculated the conditional daily discharge probability of patients in each of these groups. Finally, these conditional daily discharge probabilities were then summed for each hospital day to generate the expected number of discharges in the next 24 hours. This paper details the methods we used to create our model and the accuracy of its predictions.

 

METHODS

Study Setting and Databases Used for Analysis

The study took place at The Ottawa Hospital, a 1000-bed teaching hospital with 3 campuses that is the primary referral center in our region. The study was approved by our local research ethics board.

The Patient Registry Database records the date and time of admission for each patient (defined as the moment that a patient’s admission request is registered in the patient registration) and discharge (defined as the time when the patient’s discharge from hospital was entered into the patient registration) for hospital encounters. Emergency department encounters were also identified in the Patient Registry Database along with admission service, patient age and sex, and patient location throughout the admission. The Laboratory Database records all laboratory studies and results on all patients at the hospital.

Study Cohort

We used the Patient Registry Database to identify all people aged 1 year or more who were admitted to the hospital between January 1, 2013, and December 31, 2015. This time frame was selected to (i) ensure that data were complete; and (ii) complete calendar years of data were available for both derivation (patient-days in 2013-2014) and validation (2015) cohorts. Patients who were observed in the emergency room without admission to hospital were not included.

Study Outcome

The study outcome was the number of patients discharged from the hospital each day. For the analysis, the reference point for each day was 1 second past midnight; therefore, values for time-dependent covariates up to and including midnight were used to predict the number of discharges in the next 24 hours.

Study Covariates

Baseline (ie, time-independent) covariates included patient age and sex, admission service, hospital campus, whether or not the patient was admitted from the emergency department (all determined from the Patient Registry Database), and the Laboratory-based Acute Physiological Score (LAPS). The latter, which was calculated with the Laboratory Database using results for 14 tests (arterial pH, PaCO2, PaO2, anion gap, hematocrit, total white blood cell count, serum albumin, total bilirubin, creatinine, urea nitrogen, glucose, sodium, bicarbonate, and troponin I) measured in the 24-hour time frame preceding hospitalization, was derived by Escobar and colleagues⁷ to measure severity of illness and was subsequently validated in our hospital.⁸ The independent association of each laboratory perturbation with risk of death in hospital is reflected by the number of points assigned to each lab value with the total LAPS being the sum of these values. Time-dependent covariates included weekday in hospital and whether or not patients were in the intensive care unit.

Analysis

We used 3 stages to create a model to predict the daily expected number of discharges: we identified discharge risk strata containing patients having similar discharge patterns using data from patients in the derivation cohort (first stage); then, we generated the preliminary probability of discharge by determining the daily discharge probability in each discharge risk strata (second stage); finally, we modified the probability from the second stage based on the weekday and admission service and summed these probabilities to create the expected number of discharges on a particular date (third stage).

The first stage identified discharge risk strata based on the covariates listed above. This was determined by using a survival tree approach⁹ with proportional hazard regression models to generate the “splits.” These models were offered all covariates listed in the Study Covariates section. Admission service was clustered within 4 departments (obstetrics/gynecology, psychiatry, surgery, and medicine) and day of week was “binarized” into weekday/weekend-holiday (because the use of categorical variables with large numbers of groups can “stunt” regression trees due to small numbers of patients—and, therefore, statistical power—in each subgroup). The proportional hazards model identified the covariate having the strongest association with time to discharge (based on the Wald X² value divided by the degrees of freedom). This variable was then used to split the cohort into subgroups (with continuous covariates being categorized into quartiles). The proportional hazards model was then repeated in each subgroup (with the previous splitting variable[s] excluded from the model). This process continued until no variable was associated with time to discharge with a P value less than .0001. This survival-tree was then used to cluster all patients into distinct discharge risk strata.

In the second stage, we generated the preliminary probability of discharge for a specific date. This was calculated by assigning all patients in hospital to their discharge risk strata (Appendix). We then measured the probability of discharge on each hospitalization day in all discharge risk strata using data from the previous 180 days (we only used the prior 180 days of data to account for temporal changes in hospital discharge patterns). For example, consider a 75-year-old patient on her third hospital day under obstetrics/gynecology on December 19, 2015 (a Saturday). This patient would be assigned to risk stratum #133 (Appendix A). We then measured the probability of discharge of all patients in this discharge risk stratum hospitalized in the previous 6 months (ie, between June 22, 2015, and December 18, 2015) on each hospital day. For risk stratum #133, the probability of discharge on hospital day 3 was 0.1111; therefore, our sample patient’s preliminary expected discharge probability was 0.1111.

To attain stable daily discharge probability estimates, a minimum of 50 patients per discharge risk stratum-hospitalization day combination was required. If there were less than 50 patients for a particular hospitalization day in a particular discharge risk stratum, we grouped hospitalization days in that risk stratum together until the minimum of 50 patients was collected.

The third (and final) stage accounted for the lack of granularity when we created the discharge risk strata in the first stage. As we mentioned above, admission service was clustered into 4 departments and the day of week was clustered into weekend/weekday. However, important variations in discharge probabilities could still exist within departments and between particular days of the week.¹⁰ Therefore, we created a correction factor to adjust the preliminary expected number of discharges based on the admission division and day of week. This correction factor used data from the 180 days prior to the analysis date within which the expected daily number of discharges was calculated (using the methods above). The correction factor was the relative difference between the observed and expected number of discharges within each division-day of week grouping.

For example, to calculate the correction factor for our sample patient presented above (75-year-old patient on hospital day 3 under gynecology on Saturday, December 19, 2015), we measured the observed number of discharges from gynecology on Saturdays between June 22, 2015, and December 18, 2015, (n = 206) and the expected number of discharges (n = 195.255) resulting in a correction factor of (observed-expected)/expected = (195.255-206)/195.206 = 0.05503. Therefore, the final expected discharge probability for our sample patient was 0.1111+0.1111*0.05503=0.1172. The expected number of discharges on a particular date was the preliminary expected number of discharges on that date (generated in the second stage) multiplied by the correction factor for the corresponding division-day or week group.

 

RESULTS

There were 192,859 admissions involving patients more than 1 year of age that spent at least part of their hospitalization between January 1, 2013, and December 31, 2015 (Table). Patients were middle-aged and slightly female predominant, with about half being admitted from the emergency department. Approximately 80% of admissions were to surgical or medical services. More than 95% of admissions ended with a discharge from the hospital with the remainder ending in a death. Almost 30% of hospitalization days occurred on weekends or holidays. Hospitalizations in the derivation (2013-2014) and validation (2015) group were essentially the same, except there was a slight drop in hospital length of stay (from a median of 4 days to 3 days) between the 2 periods.

Patient and hospital covariates importantly influenced the daily conditional probability of discharge (Figure 1). Patients admitted to the obstetrics/gynecology department were notably more likely to be discharged from hospital with no influence from the day of week. In contrast, the probability of discharge decreased notably on the weekends in the other departments. Patients on the ward were much more likely to be discharged than those in the intensive care unit, with increasing age associated with a decreased discharge likelihood in the former but not the latter patients. Finally, discharge probabilities varied only slightly between campuses at our hospital with discharge risk decreasing as severity of illness (as measured by LAPS) increased.

The TEND model contained 142 discharge risk strata (Appendix A). Weekend-holiday status had the strongest association with discharge probability (ie, it was the first splitting variable). The most complex discharge risk strata contained 6 covariates. The daily conditional probability of discharge during the first 2 weeks of hospitalization varied extensively between discharge risk strata (Figure 2). Overall, the conditional discharge probability increased from the first to the second day, remained relatively stable for several days, and then slowly decreased over time. However, this pattern and day-to-day variability differed extensively between risk strata.

The observed daily number of discharges in the validation cohort varied extensively (median 139; interquartile range [IQR] 95-160; range 39-214). The TEND model accurately predicted the daily number of discharges with the expected daily number being strongly associated with the observed number (adjusted R² = 89.2%; P < 0.0001; Figure 3). Calibration decreased but remained significant when we limited the analyses by hospital campus (General: R² = 46.3%; P < 0.0001; Civic: R² = 47.9%; P < 0.0001; Heart Institute: R² = 18.1%; P < 0.0001). The expected number of daily discharges was an unbiased estimator of the observed number of discharges (its parameter estimate in a linear regression model with the observed number of discharges as the outcome variable was 1.0005; 95% confidence interval, 0.9647-1.0363). The absolute difference in the observed and expected daily number of discharges was small (median 1.6; IQR −6.8 to 9.4; range −37 to 63.4) as was the relative difference (median 1.4%; IQR −5.5% to 7.1%; range −40.9% to 43.4%). The expected number of discharges was within 20% of the observed number of discharges in 95.1% of days in 2015.

DISCUSSION

Knowing how many patients will soon be discharged from the hospital should greatly facilitate hospital planning. This study showed that the TEND model used simple patient and hospitalization covariates to accurately predict the number of patients who will be discharged from hospital in the next day.

We believe that this study has several notable findings. First, we think that using a nonparametric approach to predicting the daily number of discharges importantly increased accuracy. This approach allowed us to generate expected likelihoods based on actual discharge probabilities at our hospital in the most recent 6 months of hospitalization-days within patients having discharge patterns that were very similar to the patient in question (ie, discharge risk strata, Appendix A). This ensured that trends in hospitalization habits were accounted for without the need of a period variable in our model. In addition, the lack of parameters in the model will make it easier to transplant it to other hospitals. Second, we think that the accuracy of the predictions were remarkable given the relative “crudeness” of our predictors. By using relatively simple factors, the TEND model was able to output accurate predictions for the number of daily discharges (Figure 3).

This study joins several others that have attempted to accomplish the difficult task of predicting the number of hospital discharges by using digitized data. Barnes et al.¹¹ created a model using regression random forest methods in a single medical service within a hospital to predict the daily number of discharges with impressive accuracy (mean daily number of discharges observed 8.29, expected 8.51). Interestingly, the model in this study was more accurate at predicting discharge likelihood than physicians. Levin et al.¹² derived a model using discrete time logistic regression to predict the likelihood of discharge from a pediatric intensive care unit, finding that physician orders (captured via electronic order entry) could be categorized and used to significantly increase the accuracy of discharge likelihood. This study demonstrates the potential opportunities within health-related data from hospital data warehouses to improve prediction. We believe that continued work in this field will result in the increased use of digital data to help hospital administrators manage patient beds more efficiently and effectively than currently used resource intensive manual methods.^13,14

Several issues should be kept in mind when interpreting our findings. First, our analysis is limited to a single institution in Canada. It will be important to determine if the TEND model methodology generalizes to other hospitals in different jurisdictions. Such an external validation, especially in multiple hospitals, will be important to show that the TEND model methodology works in other facilities. Hospitals could implement the TEND model if they are able to record daily values for each of the variables required to assign patients to a discharge risk stratum (Appendix A) and calculate within each the daily probability of discharge. Hospitals could derive their own discharge risk strata to account for covariates, which we did not include in our study but could be influential, such as insurance status. These discharge risk estimates could also be incorporated into the electronic medical record or hospital dashboards (as long as the data required to generate the estimates are available). These interventions would permit the expected number of hospital discharges (and even the patient-level probability of discharge) to be calculated on a daily basis. Second, 2 potential biases could have influenced the identification of our discharge risk strata (Appendix A). In this process, we used survival tree methods to separate patient-days into clusters having progressively more homogenous discharge patterns. Each split was determined by using a proportional hazards model that ignored the competing risks of death in hospital. In addition, the model expressed age and LAPS as continuous variables, whereas these covariates had to be categorized to create our risk strata groupings. The strength of a covariate’s association with an outcome will decrease when a continuous variable is categorized.¹⁵ Both of these issues might have biased our final risk strata categorization (Appendix A). Third, we limited our model to include simple covariates whose values could be determined relatively easily within most hospital administrative data systems. While this increases the generalizability to other hospital information systems, we believe that the introduction of other covariates to the model—such as daily vital signs, laboratory results, medications, or time from operations—could increase prediction accuracy. Finally, it is uncertain whether or not knowing the predicted number of discharges will improve the efficiency of bed management within the hospital. It seems logical that an accurate prediction of the number of beds that will be made available in the next day should improve decisions regarding the number of patients who could be admitted electively to the hospital. It remains to be seen, however, whether this truly happens.

In summary, we found that the TEND model used a handful of patient and hospitalization factors to accurately predict the expected number of discharges from hospital in the next day. Further work is required to implement this model into our institution’s data warehouse and then determine whether this prediction will improve the efficiency of bed management at our hospital.

 

Disclosure: CvW is supported by a University of Ottawa Department of Medicine Clinician Scientist Chair. The authors have no conflicts of interest

 

Hospitals typically allocate beds based on historical patient volumes. If funding decreases, hospitals will usually try to maximize resource utilization by allocating beds to attain occupancies close to 100% for significant periods of time. This will invariably cause days in which hospital occupancy exceeds capacity, at which time critical entry points (such as the emergency department and operating room) will become blocked. This creates significant concerns over the patient quality of care.

Hospital administrators have very few options when hospital occupancy exceeds 100%. They could postpone admissions for “planned” cases, bring in additional staff to increase capacity, or instigate additional methods to increase hospital discharges such as expanding care resources in the community. All options are costly, bothersome, or cannot be actioned immediately. The need for these options could be minimized by enabling hospital administrators to make more informed decisions regarding hospital bed management by knowing the likely number of discharges in the next 24 hours.

Predicting the number of people who will be discharged in the next day can be approached in several ways. One approach would be to calculate each patient’s expected length of stay and then use the variation around that estimate to calculate each day’s discharge probability. Several studies have attempted to model hospital length of stay using a broad assortment of methodologies, but a mechanism to accurately predict this outcome has been elusive^1,2 (with Verburg et al.³ concluding in their study’s abstract that “…it is difficult to predict length of stay…”). A second approach would be to use survival analysis methods to generate each patient’s hazard of discharge over time, which could be directly converted to an expected daily risk of discharge. However, this approach is complicated by the concurrent need to include time-dependent covariates and consider the competing risk of death in hospital, which can complicate survival modeling.^4,5 A third approach would be the implementation of a longitudinal analysis using marginal models to predict the daily probability of discharge,⁶ but this method quickly overwhelms computer resources when large datasets are present.

In this study, we decided to use nonparametric models to predict the daily number of hospital discharges. We first identified patient groups with distinct discharge patterns. We then calculated the conditional daily discharge probability of patients in each of these groups. Finally, these conditional daily discharge probabilities were then summed for each hospital day to generate the expected number of discharges in the next 24 hours. This paper details the methods we used to create our model and the accuracy of its predictions.

 

METHODS

Study Setting and Databases Used for Analysis

The study took place at The Ottawa Hospital, a 1000-bed teaching hospital with 3 campuses that is the primary referral center in our region. The study was approved by our local research ethics board.

The Patient Registry Database records the date and time of admission for each patient (defined as the moment that a patient’s admission request is registered in the patient registration) and discharge (defined as the time when the patient’s discharge from hospital was entered into the patient registration) for hospital encounters. Emergency department encounters were also identified in the Patient Registry Database along with admission service, patient age and sex, and patient location throughout the admission. The Laboratory Database records all laboratory studies and results on all patients at the hospital.

Study Cohort

We used the Patient Registry Database to identify all people aged 1 year or more who were admitted to the hospital between January 1, 2013, and December 31, 2015. This time frame was selected to (i) ensure that data were complete; and (ii) complete calendar years of data were available for both derivation (patient-days in 2013-2014) and validation (2015) cohorts. Patients who were observed in the emergency room without admission to hospital were not included.

Study Outcome

The study outcome was the number of patients discharged from the hospital each day. For the analysis, the reference point for each day was 1 second past midnight; therefore, values for time-dependent covariates up to and including midnight were used to predict the number of discharges in the next 24 hours.

Study Covariates

Baseline (ie, time-independent) covariates included patient age and sex, admission service, hospital campus, whether or not the patient was admitted from the emergency department (all determined from the Patient Registry Database), and the Laboratory-based Acute Physiological Score (LAPS). The latter, which was calculated with the Laboratory Database using results for 14 tests (arterial pH, PaCO2, PaO2, anion gap, hematocrit, total white blood cell count, serum albumin, total bilirubin, creatinine, urea nitrogen, glucose, sodium, bicarbonate, and troponin I) measured in the 24-hour time frame preceding hospitalization, was derived by Escobar and colleagues⁷ to measure severity of illness and was subsequently validated in our hospital.⁸ The independent association of each laboratory perturbation with risk of death in hospital is reflected by the number of points assigned to each lab value with the total LAPS being the sum of these values. Time-dependent covariates included weekday in hospital and whether or not patients were in the intensive care unit.

Analysis

We used 3 stages to create a model to predict the daily expected number of discharges: we identified discharge risk strata containing patients having similar discharge patterns using data from patients in the derivation cohort (first stage); then, we generated the preliminary probability of discharge by determining the daily discharge probability in each discharge risk strata (second stage); finally, we modified the probability from the second stage based on the weekday and admission service and summed these probabilities to create the expected number of discharges on a particular date (third stage).

The first stage identified discharge risk strata based on the covariates listed above. This was determined by using a survival tree approach⁹ with proportional hazard regression models to generate the “splits.” These models were offered all covariates listed in the Study Covariates section. Admission service was clustered within 4 departments (obstetrics/gynecology, psychiatry, surgery, and medicine) and day of week was “binarized” into weekday/weekend-holiday (because the use of categorical variables with large numbers of groups can “stunt” regression trees due to small numbers of patients—and, therefore, statistical power—in each subgroup). The proportional hazards model identified the covariate having the strongest association with time to discharge (based on the Wald X² value divided by the degrees of freedom). This variable was then used to split the cohort into subgroups (with continuous covariates being categorized into quartiles). The proportional hazards model was then repeated in each subgroup (with the previous splitting variable[s] excluded from the model). This process continued until no variable was associated with time to discharge with a P value less than .0001. This survival-tree was then used to cluster all patients into distinct discharge risk strata.

In the second stage, we generated the preliminary probability of discharge for a specific date. This was calculated by assigning all patients in hospital to their discharge risk strata (Appendix). We then measured the probability of discharge on each hospitalization day in all discharge risk strata using data from the previous 180 days (we only used the prior 180 days of data to account for temporal changes in hospital discharge patterns). For example, consider a 75-year-old patient on her third hospital day under obstetrics/gynecology on December 19, 2015 (a Saturday). This patient would be assigned to risk stratum #133 (Appendix A). We then measured the probability of discharge of all patients in this discharge risk stratum hospitalized in the previous 6 months (ie, between June 22, 2015, and December 18, 2015) on each hospital day. For risk stratum #133, the probability of discharge on hospital day 3 was 0.1111; therefore, our sample patient’s preliminary expected discharge probability was 0.1111.

To attain stable daily discharge probability estimates, a minimum of 50 patients per discharge risk stratum-hospitalization day combination was required. If there were less than 50 patients for a particular hospitalization day in a particular discharge risk stratum, we grouped hospitalization days in that risk stratum together until the minimum of 50 patients was collected.

The third (and final) stage accounted for the lack of granularity when we created the discharge risk strata in the first stage. As we mentioned above, admission service was clustered into 4 departments and the day of week was clustered into weekend/weekday. However, important variations in discharge probabilities could still exist within departments and between particular days of the week.¹⁰ Therefore, we created a correction factor to adjust the preliminary expected number of discharges based on the admission division and day of week. This correction factor used data from the 180 days prior to the analysis date within which the expected daily number of discharges was calculated (using the methods above). The correction factor was the relative difference between the observed and expected number of discharges within each division-day of week grouping.

For example, to calculate the correction factor for our sample patient presented above (75-year-old patient on hospital day 3 under gynecology on Saturday, December 19, 2015), we measured the observed number of discharges from gynecology on Saturdays between June 22, 2015, and December 18, 2015, (n = 206) and the expected number of discharges (n = 195.255) resulting in a correction factor of (observed-expected)/expected = (195.255-206)/195.206 = 0.05503. Therefore, the final expected discharge probability for our sample patient was 0.1111+0.1111*0.05503=0.1172. The expected number of discharges on a particular date was the preliminary expected number of discharges on that date (generated in the second stage) multiplied by the correction factor for the corresponding division-day or week group.

 

RESULTS

There were 192,859 admissions involving patients more than 1 year of age that spent at least part of their hospitalization between January 1, 2013, and December 31, 2015 (Table). Patients were middle-aged and slightly female predominant, with about half being admitted from the emergency department. Approximately 80% of admissions were to surgical or medical services. More than 95% of admissions ended with a discharge from the hospital with the remainder ending in a death. Almost 30% of hospitalization days occurred on weekends or holidays. Hospitalizations in the derivation (2013-2014) and validation (2015) group were essentially the same, except there was a slight drop in hospital length of stay (from a median of 4 days to 3 days) between the 2 periods.

Patient and hospital covariates importantly influenced the daily conditional probability of discharge (Figure 1). Patients admitted to the obstetrics/gynecology department were notably more likely to be discharged from hospital with no influence from the day of week. In contrast, the probability of discharge decreased notably on the weekends in the other departments. Patients on the ward were much more likely to be discharged than those in the intensive care unit, with increasing age associated with a decreased discharge likelihood in the former but not the latter patients. Finally, discharge probabilities varied only slightly between campuses at our hospital with discharge risk decreasing as severity of illness (as measured by LAPS) increased.

The TEND model contained 142 discharge risk strata (Appendix A). Weekend-holiday status had the strongest association with discharge probability (ie, it was the first splitting variable). The most complex discharge risk strata contained 6 covariates. The daily conditional probability of discharge during the first 2 weeks of hospitalization varied extensively between discharge risk strata (Figure 2). Overall, the conditional discharge probability increased from the first to the second day, remained relatively stable for several days, and then slowly decreased over time. However, this pattern and day-to-day variability differed extensively between risk strata.

The observed daily number of discharges in the validation cohort varied extensively (median 139; interquartile range [IQR] 95-160; range 39-214). The TEND model accurately predicted the daily number of discharges with the expected daily number being strongly associated with the observed number (adjusted R² = 89.2%; P < 0.0001; Figure 3). Calibration decreased but remained significant when we limited the analyses by hospital campus (General: R² = 46.3%; P < 0.0001; Civic: R² = 47.9%; P < 0.0001; Heart Institute: R² = 18.1%; P < 0.0001). The expected number of daily discharges was an unbiased estimator of the observed number of discharges (its parameter estimate in a linear regression model with the observed number of discharges as the outcome variable was 1.0005; 95% confidence interval, 0.9647-1.0363). The absolute difference in the observed and expected daily number of discharges was small (median 1.6; IQR −6.8 to 9.4; range −37 to 63.4) as was the relative difference (median 1.4%; IQR −5.5% to 7.1%; range −40.9% to 43.4%). The expected number of discharges was within 20% of the observed number of discharges in 95.1% of days in 2015.

DISCUSSION

Knowing how many patients will soon be discharged from the hospital should greatly facilitate hospital planning. This study showed that the TEND model used simple patient and hospitalization covariates to accurately predict the number of patients who will be discharged from hospital in the next day.

We believe that this study has several notable findings. First, we think that using a nonparametric approach to predicting the daily number of discharges importantly increased accuracy. This approach allowed us to generate expected likelihoods based on actual discharge probabilities at our hospital in the most recent 6 months of hospitalization-days within patients having discharge patterns that were very similar to the patient in question (ie, discharge risk strata, Appendix A). This ensured that trends in hospitalization habits were accounted for without the need of a period variable in our model. In addition, the lack of parameters in the model will make it easier to transplant it to other hospitals. Second, we think that the accuracy of the predictions were remarkable given the relative “crudeness” of our predictors. By using relatively simple factors, the TEND model was able to output accurate predictions for the number of daily discharges (Figure 3).

This study joins several others that have attempted to accomplish the difficult task of predicting the number of hospital discharges by using digitized data. Barnes et al.¹¹ created a model using regression random forest methods in a single medical service within a hospital to predict the daily number of discharges with impressive accuracy (mean daily number of discharges observed 8.29, expected 8.51). Interestingly, the model in this study was more accurate at predicting discharge likelihood than physicians. Levin et al.¹² derived a model using discrete time logistic regression to predict the likelihood of discharge from a pediatric intensive care unit, finding that physician orders (captured via electronic order entry) could be categorized and used to significantly increase the accuracy of discharge likelihood. This study demonstrates the potential opportunities within health-related data from hospital data warehouses to improve prediction. We believe that continued work in this field will result in the increased use of digital data to help hospital administrators manage patient beds more efficiently and effectively than currently used resource intensive manual methods.^13,14

Several issues should be kept in mind when interpreting our findings. First, our analysis is limited to a single institution in Canada. It will be important to determine if the TEND model methodology generalizes to other hospitals in different jurisdictions. Such an external validation, especially in multiple hospitals, will be important to show that the TEND model methodology works in other facilities. Hospitals could implement the TEND model if they are able to record daily values for each of the variables required to assign patients to a discharge risk stratum (Appendix A) and calculate within each the daily probability of discharge. Hospitals could derive their own discharge risk strata to account for covariates, which we did not include in our study but could be influential, such as insurance status. These discharge risk estimates could also be incorporated into the electronic medical record or hospital dashboards (as long as the data required to generate the estimates are available). These interventions would permit the expected number of hospital discharges (and even the patient-level probability of discharge) to be calculated on a daily basis. Second, 2 potential biases could have influenced the identification of our discharge risk strata (Appendix A). In this process, we used survival tree methods to separate patient-days into clusters having progressively more homogenous discharge patterns. Each split was determined by using a proportional hazards model that ignored the competing risks of death in hospital. In addition, the model expressed age and LAPS as continuous variables, whereas these covariates had to be categorized to create our risk strata groupings. The strength of a covariate’s association with an outcome will decrease when a continuous variable is categorized.¹⁵ Both of these issues might have biased our final risk strata categorization (Appendix A). Third, we limited our model to include simple covariates whose values could be determined relatively easily within most hospital administrative data systems. While this increases the generalizability to other hospital information systems, we believe that the introduction of other covariates to the model—such as daily vital signs, laboratory results, medications, or time from operations—could increase prediction accuracy. Finally, it is uncertain whether or not knowing the predicted number of discharges will improve the efficiency of bed management within the hospital. It seems logical that an accurate prediction of the number of beds that will be made available in the next day should improve decisions regarding the number of patients who could be admitted electively to the hospital. It remains to be seen, however, whether this truly happens.

In summary, we found that the TEND model used a handful of patient and hospitalization factors to accurately predict the expected number of discharges from hospital in the next day. Further work is required to implement this model into our institution’s data warehouse and then determine whether this prediction will improve the efficiency of bed management at our hospital.

 

Disclosure: CvW is supported by a University of Ottawa Department of Medicine Clinician Scientist Chair. The authors have no conflicts of interest

 

Hospitals typically allocate beds based on historical patient volumes. If funding decreases, hospitals will usually try to maximize resource utilization by allocating beds to attain occupancies close to 100% for significant periods of time. This will invariably cause days in which hospital occupancy exceeds capacity, at which time critical entry points (such as the emergency department and operating room) will become blocked. This creates significant concerns over the patient quality of care.

Hospital administrators have very few options when hospital occupancy exceeds 100%. They could postpone admissions for “planned” cases, bring in additional staff to increase capacity, or instigate additional methods to increase hospital discharges such as expanding care resources in the community. All options are costly, bothersome, or cannot be actioned immediately. The need for these options could be minimized by enabling hospital administrators to make more informed decisions regarding hospital bed management by knowing the likely number of discharges in the next 24 hours.

Predicting the number of people who will be discharged in the next day can be approached in several ways. One approach would be to calculate each patient’s expected length of stay and then use the variation around that estimate to calculate each day’s discharge probability. Several studies have attempted to model hospital length of stay using a broad assortment of methodologies, but a mechanism to accurately predict this outcome has been elusive^1,2 (with Verburg et al.³ concluding in their study’s abstract that “…it is difficult to predict length of stay…”). A second approach would be to use survival analysis methods to generate each patient’s hazard of discharge over time, which could be directly converted to an expected daily risk of discharge. However, this approach is complicated by the concurrent need to include time-dependent covariates and consider the competing risk of death in hospital, which can complicate survival modeling.^4,5 A third approach would be the implementation of a longitudinal analysis using marginal models to predict the daily probability of discharge,⁶ but this method quickly overwhelms computer resources when large datasets are present.

In this study, we decided to use nonparametric models to predict the daily number of hospital discharges. We first identified patient groups with distinct discharge patterns. We then calculated the conditional daily discharge probability of patients in each of these groups. Finally, these conditional daily discharge probabilities were then summed for each hospital day to generate the expected number of discharges in the next 24 hours. This paper details the methods we used to create our model and the accuracy of its predictions.

 

METHODS

Study Setting and Databases Used for Analysis

The study took place at The Ottawa Hospital, a 1000-bed teaching hospital with 3 campuses that is the primary referral center in our region. The study was approved by our local research ethics board.

The Patient Registry Database records the date and time of admission for each patient (defined as the moment that a patient’s admission request is registered in the patient registration) and discharge (defined as the time when the patient’s discharge from hospital was entered into the patient registration) for hospital encounters. Emergency department encounters were also identified in the Patient Registry Database along with admission service, patient age and sex, and patient location throughout the admission. The Laboratory Database records all laboratory studies and results on all patients at the hospital.

Study Cohort

We used the Patient Registry Database to identify all people aged 1 year or more who were admitted to the hospital between January 1, 2013, and December 31, 2015. This time frame was selected to (i) ensure that data were complete; and (ii) complete calendar years of data were available for both derivation (patient-days in 2013-2014) and validation (2015) cohorts. Patients who were observed in the emergency room without admission to hospital were not included.

Study Outcome

The study outcome was the number of patients discharged from the hospital each day. For the analysis, the reference point for each day was 1 second past midnight; therefore, values for time-dependent covariates up to and including midnight were used to predict the number of discharges in the next 24 hours.

Study Covariates

Baseline (ie, time-independent) covariates included patient age and sex, admission service, hospital campus, whether or not the patient was admitted from the emergency department (all determined from the Patient Registry Database), and the Laboratory-based Acute Physiological Score (LAPS). The latter, which was calculated with the Laboratory Database using results for 14 tests (arterial pH, PaCO2, PaO2, anion gap, hematocrit, total white blood cell count, serum albumin, total bilirubin, creatinine, urea nitrogen, glucose, sodium, bicarbonate, and troponin I) measured in the 24-hour time frame preceding hospitalization, was derived by Escobar and colleagues⁷ to measure severity of illness and was subsequently validated in our hospital.⁸ The independent association of each laboratory perturbation with risk of death in hospital is reflected by the number of points assigned to each lab value with the total LAPS being the sum of these values. Time-dependent covariates included weekday in hospital and whether or not patients were in the intensive care unit.

Analysis

We used 3 stages to create a model to predict the daily expected number of discharges: we identified discharge risk strata containing patients having similar discharge patterns using data from patients in the derivation cohort (first stage); then, we generated the preliminary probability of discharge by determining the daily discharge probability in each discharge risk strata (second stage); finally, we modified the probability from the second stage based on the weekday and admission service and summed these probabilities to create the expected number of discharges on a particular date (third stage).

The first stage identified discharge risk strata based on the covariates listed above. This was determined by using a survival tree approach⁹ with proportional hazard regression models to generate the “splits.” These models were offered all covariates listed in the Study Covariates section. Admission service was clustered within 4 departments (obstetrics/gynecology, psychiatry, surgery, and medicine) and day of week was “binarized” into weekday/weekend-holiday (because the use of categorical variables with large numbers of groups can “stunt” regression trees due to small numbers of patients—and, therefore, statistical power—in each subgroup). The proportional hazards model identified the covariate having the strongest association with time to discharge (based on the Wald X² value divided by the degrees of freedom). This variable was then used to split the cohort into subgroups (with continuous covariates being categorized into quartiles). The proportional hazards model was then repeated in each subgroup (with the previous splitting variable[s] excluded from the model). This process continued until no variable was associated with time to discharge with a P value less than .0001. This survival-tree was then used to cluster all patients into distinct discharge risk strata.

In the second stage, we generated the preliminary probability of discharge for a specific date. This was calculated by assigning all patients in hospital to their discharge risk strata (Appendix). We then measured the probability of discharge on each hospitalization day in all discharge risk strata using data from the previous 180 days (we only used the prior 180 days of data to account for temporal changes in hospital discharge patterns). For example, consider a 75-year-old patient on her third hospital day under obstetrics/gynecology on December 19, 2015 (a Saturday). This patient would be assigned to risk stratum #133 (Appendix A). We then measured the probability of discharge of all patients in this discharge risk stratum hospitalized in the previous 6 months (ie, between June 22, 2015, and December 18, 2015) on each hospital day. For risk stratum #133, the probability of discharge on hospital day 3 was 0.1111; therefore, our sample patient’s preliminary expected discharge probability was 0.1111.

To attain stable daily discharge probability estimates, a minimum of 50 patients per discharge risk stratum-hospitalization day combination was required. If there were less than 50 patients for a particular hospitalization day in a particular discharge risk stratum, we grouped hospitalization days in that risk stratum together until the minimum of 50 patients was collected.

The third (and final) stage accounted for the lack of granularity when we created the discharge risk strata in the first stage. As we mentioned above, admission service was clustered into 4 departments and the day of week was clustered into weekend/weekday. However, important variations in discharge probabilities could still exist within departments and between particular days of the week.¹⁰ Therefore, we created a correction factor to adjust the preliminary expected number of discharges based on the admission division and day of week. This correction factor used data from the 180 days prior to the analysis date within which the expected daily number of discharges was calculated (using the methods above). The correction factor was the relative difference between the observed and expected number of discharges within each division-day of week grouping.

For example, to calculate the correction factor for our sample patient presented above (75-year-old patient on hospital day 3 under gynecology on Saturday, December 19, 2015), we measured the observed number of discharges from gynecology on Saturdays between June 22, 2015, and December 18, 2015, (n = 206) and the expected number of discharges (n = 195.255) resulting in a correction factor of (observed-expected)/expected = (195.255-206)/195.206 = 0.05503. Therefore, the final expected discharge probability for our sample patient was 0.1111+0.1111*0.05503=0.1172. The expected number of discharges on a particular date was the preliminary expected number of discharges on that date (generated in the second stage) multiplied by the correction factor for the corresponding division-day or week group.

 

RESULTS

There were 192,859 admissions involving patients more than 1 year of age that spent at least part of their hospitalization between January 1, 2013, and December 31, 2015 (Table). Patients were middle-aged and slightly female predominant, with about half being admitted from the emergency department. Approximately 80% of admissions were to surgical or medical services. More than 95% of admissions ended with a discharge from the hospital with the remainder ending in a death. Almost 30% of hospitalization days occurred on weekends or holidays. Hospitalizations in the derivation (2013-2014) and validation (2015) group were essentially the same, except there was a slight drop in hospital length of stay (from a median of 4 days to 3 days) between the 2 periods.

Patient and hospital covariates importantly influenced the daily conditional probability of discharge (Figure 1). Patients admitted to the obstetrics/gynecology department were notably more likely to be discharged from hospital with no influence from the day of week. In contrast, the probability of discharge decreased notably on the weekends in the other departments. Patients on the ward were much more likely to be discharged than those in the intensive care unit, with increasing age associated with a decreased discharge likelihood in the former but not the latter patients. Finally, discharge probabilities varied only slightly between campuses at our hospital with discharge risk decreasing as severity of illness (as measured by LAPS) increased.

The TEND model contained 142 discharge risk strata (Appendix A). Weekend-holiday status had the strongest association with discharge probability (ie, it was the first splitting variable). The most complex discharge risk strata contained 6 covariates. The daily conditional probability of discharge during the first 2 weeks of hospitalization varied extensively between discharge risk strata (Figure 2). Overall, the conditional discharge probability increased from the first to the second day, remained relatively stable for several days, and then slowly decreased over time. However, this pattern and day-to-day variability differed extensively between risk strata.

The observed daily number of discharges in the validation cohort varied extensively (median 139; interquartile range [IQR] 95-160; range 39-214). The TEND model accurately predicted the daily number of discharges with the expected daily number being strongly associated with the observed number (adjusted R² = 89.2%; P < 0.0001; Figure 3). Calibration decreased but remained significant when we limited the analyses by hospital campus (General: R² = 46.3%; P < 0.0001; Civic: R² = 47.9%; P < 0.0001; Heart Institute: R² = 18.1%; P < 0.0001). The expected number of daily discharges was an unbiased estimator of the observed number of discharges (its parameter estimate in a linear regression model with the observed number of discharges as the outcome variable was 1.0005; 95% confidence interval, 0.9647-1.0363). The absolute difference in the observed and expected daily number of discharges was small (median 1.6; IQR −6.8 to 9.4; range −37 to 63.4) as was the relative difference (median 1.4%; IQR −5.5% to 7.1%; range −40.9% to 43.4%). The expected number of discharges was within 20% of the observed number of discharges in 95.1% of days in 2015.

DISCUSSION

Knowing how many patients will soon be discharged from the hospital should greatly facilitate hospital planning. This study showed that the TEND model used simple patient and hospitalization covariates to accurately predict the number of patients who will be discharged from hospital in the next day.

We believe that this study has several notable findings. First, we think that using a nonparametric approach to predicting the daily number of discharges importantly increased accuracy. This approach allowed us to generate expected likelihoods based on actual discharge probabilities at our hospital in the most recent 6 months of hospitalization-days within patients having discharge patterns that were very similar to the patient in question (ie, discharge risk strata, Appendix A). This ensured that trends in hospitalization habits were accounted for without the need of a period variable in our model. In addition, the lack of parameters in the model will make it easier to transplant it to other hospitals. Second, we think that the accuracy of the predictions were remarkable given the relative “crudeness” of our predictors. By using relatively simple factors, the TEND model was able to output accurate predictions for the number of daily discharges (Figure 3).

This study joins several others that have attempted to accomplish the difficult task of predicting the number of hospital discharges by using digitized data. Barnes et al.¹¹ created a model using regression random forest methods in a single medical service within a hospital to predict the daily number of discharges with impressive accuracy (mean daily number of discharges observed 8.29, expected 8.51). Interestingly, the model in this study was more accurate at predicting discharge likelihood than physicians. Levin et al.¹² derived a model using discrete time logistic regression to predict the likelihood of discharge from a pediatric intensive care unit, finding that physician orders (captured via electronic order entry) could be categorized and used to significantly increase the accuracy of discharge likelihood. This study demonstrates the potential opportunities within health-related data from hospital data warehouses to improve prediction. We believe that continued work in this field will result in the increased use of digital data to help hospital administrators manage patient beds more efficiently and effectively than currently used resource intensive manual methods.^13,14

Several issues should be kept in mind when interpreting our findings. First, our analysis is limited to a single institution in Canada. It will be important to determine if the TEND model methodology generalizes to other hospitals in different jurisdictions. Such an external validation, especially in multiple hospitals, will be important to show that the TEND model methodology works in other facilities. Hospitals could implement the TEND model if they are able to record daily values for each of the variables required to assign patients to a discharge risk stratum (Appendix A) and calculate within each the daily probability of discharge. Hospitals could derive their own discharge risk strata to account for covariates, which we did not include in our study but could be influential, such as insurance status. These discharge risk estimates could also be incorporated into the electronic medical record or hospital dashboards (as long as the data required to generate the estimates are available). These interventions would permit the expected number of hospital discharges (and even the patient-level probability of discharge) to be calculated on a daily basis. Second, 2 potential biases could have influenced the identification of our discharge risk strata (Appendix A). In this process, we used survival tree methods to separate patient-days into clusters having progressively more homogenous discharge patterns. Each split was determined by using a proportional hazards model that ignored the competing risks of death in hospital. In addition, the model expressed age and LAPS as continuous variables, whereas these covariates had to be categorized to create our risk strata groupings. The strength of a covariate’s association with an outcome will decrease when a continuous variable is categorized.¹⁵ Both of these issues might have biased our final risk strata categorization (Appendix A). Third, we limited our model to include simple covariates whose values could be determined relatively easily within most hospital administrative data systems. While this increases the generalizability to other hospital information systems, we believe that the introduction of other covariates to the model—such as daily vital signs, laboratory results, medications, or time from operations—could increase prediction accuracy. Finally, it is uncertain whether or not knowing the predicted number of discharges will improve the efficiency of bed management within the hospital. It seems logical that an accurate prediction of the number of beds that will be made available in the next day should improve decisions regarding the number of patients who could be admitted electively to the hospital. It remains to be seen, however, whether this truly happens.

In summary, we found that the TEND model used a handful of patient and hospitalization factors to accurately predict the expected number of discharges from hospital in the next day. Further work is required to implement this model into our institution’s data warehouse and then determine whether this prediction will improve the efficiency of bed management at our hospital.

 

Disclosure: CvW is supported by a University of Ottawa Department of Medicine Clinician Scientist Chair. The authors have no conflicts of interest

 

Secondary to rising healthcare costs in the United States, broad efforts are underway to identify and reduce waste in the health system.1,2 A recent systematic review exhibited that many physicians inaccurately estimate the cost of medications.3 Raising awareness of medication costs among prescribers is one potential way to promote high-value care.

Some evidence suggests that cost transparency may help prescribers understand how medication orders drive costs. In a previous study carried out at the Johns Hopkins Hospital, fee data were displayed to providers for diagnostic laboratory tests.4 An 8.6% decrease (95% confidence interval [CI], –8.99% to –8.19%) in test ordering was observed when costs were displayed vs a 5.6% increase (95% CI, 4.90% to 6.39%) in ordering when costs were not displayed during a 6-month intervention period (P < 0.001). Conversely, a similar study that investigated the impact of cost transparency on inpatient imaging utilization did not demonstrate a significant influence of cost display.5 This suggests that cost transparency may work in some areas of care but not in others. A systematic review that investigated price-display interventions for imaging, laboratory studies, and medications reported 10 studies that demonstrated a statistically significant decrease in expenditures without an effect on patient safety.6

Informing prescribers of institution-specific medication costs within and between drug classes may enable the selection of less expensive, therapeutically equivalent drugs. Prior studies investigating the effect of medication cost display were conducted in a variety of patient care settings, including ambulatory clinics,7 urgent care centers,8 and operating rooms,9^,10 with some yielding positive results in terms of ordering and cost11^,12 and others having no impact.13^,14 Currently, there is little evidence specifically addressing the effect of cost display for medications in the inpatient setting.

As part of an institutional initiative to control pharmaceutical expenditures, informational messaging for several high-cost drugs was initiated at our tertiary care hospital in April 2015. The goal of our study was to assess the effect of these medication cost messages on ordering practices. We hypothesized that the display of inpatient pharmaceutical costs at the time of order entry would result in a reduction in ordering.

METHODS

Setting, Intervention, and Participants

As part of an effort to educate prescribers about the high cost of medications, 9 intravenous (IV) medications were selected by the Johns Hopkins Hospital Pharmacy and Therapeutics Committee as targets for drug cost messaging. The intention of the committee was to implement a rapid, low-cost, proof-of-concept, quality-improvement project that was not designed as prospective research. Representatives from the pharmacy and clinicians from relevant clinical areas participated in preimplementation discussions to help identify medications that were subjectively felt to be overused at our institution and potentially modifiable through provider education. The criteria for selecting drug targets included a variety of factors, such as medications infrequently ordered but representing a significant cost per dose (eg, eculizumab and ribavirin), frequently ordered medications with less expensive substitutes (eg, linezolid and voriconazole), and high-cost medications without direct therapeutic alternatives (eg, calcitonin). From April 10, 2015, to October 5, 2015, the computerized Provider Order Entry System (cPOE), Sunrise Clinical Manager (Allscripts Corporation, Chicago, IL), displayed the cost for targeted medications. Seven of the medication alerts also included a reasonable therapeutic alternative and its cost. There were no restrictions placed on ordering; prescribers were able to choose the high-cost medications at their discretion.

Despite the fact that this initiative was not designed as a research project, we felt it was important to formally evaluate the impact of the drug cost messaging effort to inform future quality-improvement interventions. Each medication was compared to its preintervention baseline utilization dating back to January 1, 2013. For the 7 medications with alternatives offered, we also analyzed use of the suggested alternative during these time periods.

Data Sources and Measurement

Our study utilized data obtained from the pharmacy order verification system and the cPOE database. Data were collected over a period of 143 weeks from January 1, 2013, to October 5, 2015, to allow for a baseline period (January 1, 2013, to April 9, 2015) and an intervention period (April 10, 2015, to October 5, 2015). Data elements extracted included drug characteristics (dosage form, route, cost, strength, name, and quantity), patient characteristics (race, gender, and age), clinical setting (facility location, inpatient or outpatient), and billing information (provider name, doses dispensed from pharmacy, order number, revenue or procedure code, record number, date of service, and unique billing number) for each admission. Using these elements, we generated the following 8 variables to use in our analyses: week, month, period identifier, drug name, dosage form, weekly orders, weekly patient days, and number of weekly orders per 10,000 patient days. Average wholesale price (AWP), referred to as medication cost in this manuscript, was used to report all drug costs in all associated cost calculations. While the actual cost of acquisition and price charged to the patient may vary based on several factors, including manufacturer and payer, we chose to use AWP as a generalizable estimate of the cost of acquisition of the drug for the hospital.

 

Variables

“Week” and “month” were defined as the week and month of our study, respectively. The “period identifier” was a binary variable that identified the time period before and after the intervention. “Weekly orders” was defined as the total number of new orders placed per week for each specified drug included in our study. For example, if a patient received 2 discrete, new orders for a medication in a given week, 2 orders would be counted toward the “weekly orders” variable. “Patient days,” defined as the total number of patients treated at our facility, was summated for each week of our study to yield “weekly patient days.” To derive the “number of weekly orders per 10,000 patient days,” we divided weekly orders by weekly patient days and multiplied the resultant figure by 10,000.

Statistical Analysis

Segmented regression, a form of interrupted time series analysis, is a quasi-experimental design that was used to determine the immediate and sustained effects of the drug cost messages on the rate of medication ordering.^15-17 The model enabled the use of comparison groups (alternative medications, as described above) to enhance internal validity.

In time series data, outcomes may not be independent over time. Autocorrelation of the error terms can arise when outcomes are more similar at time points closer together than outcomes at time points further apart. Failure to account for autocorrelation of the error terms may lead to underestimated standard errors. The presence of autocorrelation, assessed by calculating the Durbin-Watson statistic, was significant among our data. To adjust for this, we employed a Prais-Winsten estimation to adjust the error term (ε_t) calculated in our models.

Two segmented linear regression models were used to estimate trends in ordering before and after the intervention. The presence or absence of a comparator drug determined which model was to be used. When only single medications were under study, as in the case of eculizumab and calcitonin, our regression model was as follows:

Y_t = (β₀) + (β₁)(Time_t) + (β₂)(Intervention_t) + (β₃)(Post-Intervention Time_t) + (ε_t)

In our single-drug model, Y_t denoted the number of orders per 10,000 patient days at week “t”; Time_t was a continuous variable that indicated the number of weeks prior to or after the study intervention (April 10, 2015) and ranged from –116 to 27 weeks. Post-Intervention Time_t was a continuous variable that denoted the number of weeks since the start of the intervention and is coded as zero for all time periods prior to the intervention. β₀ was the estimated baseline number of orders per 10,000 patient days at the beginning of the study. β₁ is the trend of orders per 10,000 patient days per week during the preintervention period; β₂ represents an estimate of the change in the number of orders per 10,000 patient days immediately after the intervention; β₃ denotes the difference between preintervention and postintervention slopes; and ε_t is the “error term,” which represents autocorrelation and random variability of the data.

As mentioned previously, alternative dosage forms of 7 medications included in our study were utilized as comparison groups. In these instances (when multiple drugs were included in our analyses), the following regression model was applied:

Y _t = ( β ₀ ) + ( β ₁ )(Time _t ) + ( β ₂ )(Intervention _t ) + ( β ₃ )(Post-Intervention Time _t ) + ( β ₄ )(Cohort) + ( β ₅ )(Cohort)(Time _t ) + ( β ₆ )(Cohort)(Intervention _t ) + ( β ₇ )(Cohort)(Post-Intervention Time _t ) + ( ε _t )

 

Here, 3 coefficients were added (β₄-β₇) to describe an additional cohort of orders. Cohort, a binary indicator variable, held a value of either 0 or 1 when the model was used to describe the treatment or comparison group, respectively. The coefficients β₄-β₇ described the treatment group, and β₀-β₃ described the comparison group. β₄ was the difference in the number of baseline orders per 10,000 patient days between treatment and comparison groups; Β₅ represented the difference between the estimated ordering trends of treatment and comparison groups; and Β₆ indicated the difference in immediate changes in the number of orders per 10,000 patient days in the 2 groups following the intervention.

The number of orders per week was recorded for each medicine, which enabled a large number of data points to be included in our analyses. This allowed for more accurate and stable estimates to be made in our regression model. A total of 143 data points were collected for each study group, 116 before and 27 following each intervention.

All analyses were conducted by using STATA version 13.1 (StataCorp LP, College Station, TX).

RESULTS

Initial results pertaining to 9 IV medications were examined (Table). Following the implementation of cost messaging, no significant changes were observed in order frequency or trend for IV formulations of eculizumab, calcitonin, levetiracetam, linezolid, mycophenolate, ribavirin, voriconazole, and levothyroxine (Figures 1 and 2). However, a significant decrease in the number of oral ribavirin orders (Figure 2), the control group for the IV form, was observed (–16.3 orders per 10,000 patient days; P = .004; 95% CI, –27.2 to –5.31).

From March 26, 2015, to January 21, 2016, the use of IV pantoprazole was restricted at our facility due to a national shortage, requiring clinical pharmacy review and approval. Additionally, on September 15, 2014, oncology order sets that allowed antibiotic prescribing without an infectious disease provider’s approval replaced oral voriconazole with oral posaconazole. Consequently, both IV pantoprazole and oral voriconazole exhibited a sharp, sustained decrease in ordering following these policy changes (Figure 3). Although an apparent difference in ordering followed the policy changes, no significant change was noted in IV pantoprazole and oral voriconazole following our intervention.

DISCUSSION

Our results suggest that the passive strategy of displaying cost alone was not effective in altering prescriber ordering patterns for the selected medications. This may be due to a lack of awareness regarding direct financial impact on the patient, importance of costs in medical decision-making, or a perceived lack of alternatives or suitability of recommended alternatives. These results may prove valuable to hospital and pharmacy leadership as they develop strategies to curb medication expense.

Changes observed in IV pantoprazole ordering are instructive. Due to a national shortage, the IV form of this medication underwent a restriction, which required approval by the pharmacy prior to dispensing. This restriction was instituted independently of our study and led to a 73% decrease from usage rates prior to policy implementation (Figure 3). Ordering was restricted according to defined criteria for IV use. The restriction did not apply to oral pantoprazole, and no significant change in ordering of the oral formulation was noted during the evaluated period (Figure 3).

Oral voriconazole also exhibited a marked reduction in ordering of 46% following a change in hospital formulary that occurred independent of our study. The change in formulary involved a replacement of oral voriconazole with oral posaconazole in several oncology order sets, allowing antifungal administration without authorization by an infectious disease clinician. A dramatic reduction in ordering resulted from the removal of oral voriconazole from these order sets. Figure 3 represents the effect of the policy change on ordering. The frequency of orders for IV voriconazole, which was the target of our study intervention, did not change during this period (Figure 3).

The dramatic effect of policy changes, as observed with pantoprazole and voriconazole, suggests that a more active strategy may have a greater impact on prescriber behavior when it comes to medication ordering in the inpatient setting. It also highlights several potential sources of confounding that may introduce bias to cost-transparency studies.

This study has multiple limitations. First, as with all observational study designs, causation cannot be drawn with certainty from our results. While we were able to compare medications to their preintervention baselines, the data could have been impacted by longitudinal or seasonal trends in medication ordering, which may have been impacted by seasonal variability in disease prevalence, changes in resistance patterns, and annual cycling of house staff in an academic medical center. While there appear to be potential seasonal patterns regarding prescribing patterns for some of the medications included in this analysis, we also believe the linear regressions capture the overall trends in prescribing adequately. Nonstationarity, or trends in the mean and variance of the outcome that are not related to the intervention, may introduce bias in the interpretation of our findings. However, we believe the parameters included in our models, namely the immediate change in the intercept following the intervention and the change in the trend of the rate of prescribing over time from pre- to postintervention, provide substantial protections from faulty interpretation. Our models are limited to the extent that these parameters do not account for nonstationarity. Additionally, we did not collect data on dosing frequency or duration of treatment, which would have been dependent on factors that are not readily quantified, such as indication, clinical rationale, or patient response. Thus, we were not able to evaluate the impact of the intervention on these factors.

Although intended to enhance internal validity, comparison groups were also subject to external influence. For example, we observed a significant, short-lived rise in oral ribavirin (a control medication) ordering during the preintervention baseline period that appeared to be independent of our intervention and may speak to the unaccounted-for longitudinal variability detailed above.

Finally, the clinical indication and setting may be important. Previous studies performed at the same hospital with price displays showed a reduction in laboratory ordering but no change in imaging.18,19 One might speculate that ordering fewer laboratory tests is viewed by providers as eliminating waste rather than choosing a less expensive option to accomplish the same diagnostic task at hand. Therapeutics may be more similar to radiology tests, because patients presumably need the treatment and often do not have the option of simply not ordering without a concerted effort to reevaluate the treatment plan. Additionally, in a tertiary care teaching center such as ours, a junior clinician, oftentimes at the behest of a more senior colleague, enters most orders. In an environment in which the ordering prescriber has more autonomy or when the order is driven by a junior practitioner rather than an attending (such as daily laboratories), results may be different. Additionally, institutions that incentivize prescribers directly to practice cost-conscious care may experience different results from similar interventions.

We conclude that, in the case of medication cost messaging, a strategy of displaying cost information alone was insufficient to affect prescriber ordering behavior. Coupling cost transparency with educational interventions and active stewardship to impact clinical practice is worthy of further study.

 

Disclosures: The authors state that there were no external sponsors for this work. The Johns Hopkins Hospital and University “funded” this work by paying the salaries of the authors. The author team maintained independence and made all decisions regarding the study design, data collection, data analysis, interpretation of results, writing of the research report, and decision to submit it for publication. Dr. Shermock had full access to all the study data and takes responsibility for the integrity of the data and accuracy of the data analysis.

Secondary to rising healthcare costs in the United States, broad efforts are underway to identify and reduce waste in the health system.1,2 A recent systematic review exhibited that many physicians inaccurately estimate the cost of medications.3 Raising awareness of medication costs among prescribers is one potential way to promote high-value care.

Some evidence suggests that cost transparency may help prescribers understand how medication orders drive costs. In a previous study carried out at the Johns Hopkins Hospital, fee data were displayed to providers for diagnostic laboratory tests.4 An 8.6% decrease (95% confidence interval [CI], –8.99% to –8.19%) in test ordering was observed when costs were displayed vs a 5.6% increase (95% CI, 4.90% to 6.39%) in ordering when costs were not displayed during a 6-month intervention period (P < 0.001). Conversely, a similar study that investigated the impact of cost transparency on inpatient imaging utilization did not demonstrate a significant influence of cost display.5 This suggests that cost transparency may work in some areas of care but not in others. A systematic review that investigated price-display interventions for imaging, laboratory studies, and medications reported 10 studies that demonstrated a statistically significant decrease in expenditures without an effect on patient safety.6

Informing prescribers of institution-specific medication costs within and between drug classes may enable the selection of less expensive, therapeutically equivalent drugs. Prior studies investigating the effect of medication cost display were conducted in a variety of patient care settings, including ambulatory clinics,7 urgent care centers,8 and operating rooms,9^,10 with some yielding positive results in terms of ordering and cost11^,12 and others having no impact.13^,14 Currently, there is little evidence specifically addressing the effect of cost display for medications in the inpatient setting.

As part of an institutional initiative to control pharmaceutical expenditures, informational messaging for several high-cost drugs was initiated at our tertiary care hospital in April 2015. The goal of our study was to assess the effect of these medication cost messages on ordering practices. We hypothesized that the display of inpatient pharmaceutical costs at the time of order entry would result in a reduction in ordering.

METHODS

Setting, Intervention, and Participants

As part of an effort to educate prescribers about the high cost of medications, 9 intravenous (IV) medications were selected by the Johns Hopkins Hospital Pharmacy and Therapeutics Committee as targets for drug cost messaging. The intention of the committee was to implement a rapid, low-cost, proof-of-concept, quality-improvement project that was not designed as prospective research. Representatives from the pharmacy and clinicians from relevant clinical areas participated in preimplementation discussions to help identify medications that were subjectively felt to be overused at our institution and potentially modifiable through provider education. The criteria for selecting drug targets included a variety of factors, such as medications infrequently ordered but representing a significant cost per dose (eg, eculizumab and ribavirin), frequently ordered medications with less expensive substitutes (eg, linezolid and voriconazole), and high-cost medications without direct therapeutic alternatives (eg, calcitonin). From April 10, 2015, to October 5, 2015, the computerized Provider Order Entry System (cPOE), Sunrise Clinical Manager (Allscripts Corporation, Chicago, IL), displayed the cost for targeted medications. Seven of the medication alerts also included a reasonable therapeutic alternative and its cost. There were no restrictions placed on ordering; prescribers were able to choose the high-cost medications at their discretion.

Despite the fact that this initiative was not designed as a research project, we felt it was important to formally evaluate the impact of the drug cost messaging effort to inform future quality-improvement interventions. Each medication was compared to its preintervention baseline utilization dating back to January 1, 2013. For the 7 medications with alternatives offered, we also analyzed use of the suggested alternative during these time periods.

Data Sources and Measurement

Our study utilized data obtained from the pharmacy order verification system and the cPOE database. Data were collected over a period of 143 weeks from January 1, 2013, to October 5, 2015, to allow for a baseline period (January 1, 2013, to April 9, 2015) and an intervention period (April 10, 2015, to October 5, 2015). Data elements extracted included drug characteristics (dosage form, route, cost, strength, name, and quantity), patient characteristics (race, gender, and age), clinical setting (facility location, inpatient or outpatient), and billing information (provider name, doses dispensed from pharmacy, order number, revenue or procedure code, record number, date of service, and unique billing number) for each admission. Using these elements, we generated the following 8 variables to use in our analyses: week, month, period identifier, drug name, dosage form, weekly orders, weekly patient days, and number of weekly orders per 10,000 patient days. Average wholesale price (AWP), referred to as medication cost in this manuscript, was used to report all drug costs in all associated cost calculations. While the actual cost of acquisition and price charged to the patient may vary based on several factors, including manufacturer and payer, we chose to use AWP as a generalizable estimate of the cost of acquisition of the drug for the hospital.

 

Variables

“Week” and “month” were defined as the week and month of our study, respectively. The “period identifier” was a binary variable that identified the time period before and after the intervention. “Weekly orders” was defined as the total number of new orders placed per week for each specified drug included in our study. For example, if a patient received 2 discrete, new orders for a medication in a given week, 2 orders would be counted toward the “weekly orders” variable. “Patient days,” defined as the total number of patients treated at our facility, was summated for each week of our study to yield “weekly patient days.” To derive the “number of weekly orders per 10,000 patient days,” we divided weekly orders by weekly patient days and multiplied the resultant figure by 10,000.

Statistical Analysis

Segmented regression, a form of interrupted time series analysis, is a quasi-experimental design that was used to determine the immediate and sustained effects of the drug cost messages on the rate of medication ordering.^15-17 The model enabled the use of comparison groups (alternative medications, as described above) to enhance internal validity.

In time series data, outcomes may not be independent over time. Autocorrelation of the error terms can arise when outcomes are more similar at time points closer together than outcomes at time points further apart. Failure to account for autocorrelation of the error terms may lead to underestimated standard errors. The presence of autocorrelation, assessed by calculating the Durbin-Watson statistic, was significant among our data. To adjust for this, we employed a Prais-Winsten estimation to adjust the error term (ε_t) calculated in our models.

Two segmented linear regression models were used to estimate trends in ordering before and after the intervention. The presence or absence of a comparator drug determined which model was to be used. When only single medications were under study, as in the case of eculizumab and calcitonin, our regression model was as follows:

Y_t = (β₀) + (β₁)(Time_t) + (β₂)(Intervention_t) + (β₃)(Post-Intervention Time_t) + (ε_t)

In our single-drug model, Y_t denoted the number of orders per 10,000 patient days at week “t”; Time_t was a continuous variable that indicated the number of weeks prior to or after the study intervention (April 10, 2015) and ranged from –116 to 27 weeks. Post-Intervention Time_t was a continuous variable that denoted the number of weeks since the start of the intervention and is coded as zero for all time periods prior to the intervention. β₀ was the estimated baseline number of orders per 10,000 patient days at the beginning of the study. β₁ is the trend of orders per 10,000 patient days per week during the preintervention period; β₂ represents an estimate of the change in the number of orders per 10,000 patient days immediately after the intervention; β₃ denotes the difference between preintervention and postintervention slopes; and ε_t is the “error term,” which represents autocorrelation and random variability of the data.

As mentioned previously, alternative dosage forms of 7 medications included in our study were utilized as comparison groups. In these instances (when multiple drugs were included in our analyses), the following regression model was applied:

Y _t = ( β ₀ ) + ( β ₁ )(Time _t ) + ( β ₂ )(Intervention _t ) + ( β ₃ )(Post-Intervention Time _t ) + ( β ₄ )(Cohort) + ( β ₅ )(Cohort)(Time _t ) + ( β ₆ )(Cohort)(Intervention _t ) + ( β ₇ )(Cohort)(Post-Intervention Time _t ) + ( ε _t )

 

Here, 3 coefficients were added (β₄-β₇) to describe an additional cohort of orders. Cohort, a binary indicator variable, held a value of either 0 or 1 when the model was used to describe the treatment or comparison group, respectively. The coefficients β₄-β₇ described the treatment group, and β₀-β₃ described the comparison group. β₄ was the difference in the number of baseline orders per 10,000 patient days between treatment and comparison groups; Β₅ represented the difference between the estimated ordering trends of treatment and comparison groups; and Β₆ indicated the difference in immediate changes in the number of orders per 10,000 patient days in the 2 groups following the intervention.

The number of orders per week was recorded for each medicine, which enabled a large number of data points to be included in our analyses. This allowed for more accurate and stable estimates to be made in our regression model. A total of 143 data points were collected for each study group, 116 before and 27 following each intervention.

All analyses were conducted by using STATA version 13.1 (StataCorp LP, College Station, TX).

RESULTS

Initial results pertaining to 9 IV medications were examined (Table). Following the implementation of cost messaging, no significant changes were observed in order frequency or trend for IV formulations of eculizumab, calcitonin, levetiracetam, linezolid, mycophenolate, ribavirin, voriconazole, and levothyroxine (Figures 1 and 2). However, a significant decrease in the number of oral ribavirin orders (Figure 2), the control group for the IV form, was observed (–16.3 orders per 10,000 patient days; P = .004; 95% CI, –27.2 to –5.31).

From March 26, 2015, to January 21, 2016, the use of IV pantoprazole was restricted at our facility due to a national shortage, requiring clinical pharmacy review and approval. Additionally, on September 15, 2014, oncology order sets that allowed antibiotic prescribing without an infectious disease provider’s approval replaced oral voriconazole with oral posaconazole. Consequently, both IV pantoprazole and oral voriconazole exhibited a sharp, sustained decrease in ordering following these policy changes (Figure 3). Although an apparent difference in ordering followed the policy changes, no significant change was noted in IV pantoprazole and oral voriconazole following our intervention.

DISCUSSION

Our results suggest that the passive strategy of displaying cost alone was not effective in altering prescriber ordering patterns for the selected medications. This may be due to a lack of awareness regarding direct financial impact on the patient, importance of costs in medical decision-making, or a perceived lack of alternatives or suitability of recommended alternatives. These results may prove valuable to hospital and pharmacy leadership as they develop strategies to curb medication expense.

Changes observed in IV pantoprazole ordering are instructive. Due to a national shortage, the IV form of this medication underwent a restriction, which required approval by the pharmacy prior to dispensing. This restriction was instituted independently of our study and led to a 73% decrease from usage rates prior to policy implementation (Figure 3). Ordering was restricted according to defined criteria for IV use. The restriction did not apply to oral pantoprazole, and no significant change in ordering of the oral formulation was noted during the evaluated period (Figure 3).

Oral voriconazole also exhibited a marked reduction in ordering of 46% following a change in hospital formulary that occurred independent of our study. The change in formulary involved a replacement of oral voriconazole with oral posaconazole in several oncology order sets, allowing antifungal administration without authorization by an infectious disease clinician. A dramatic reduction in ordering resulted from the removal of oral voriconazole from these order sets. Figure 3 represents the effect of the policy change on ordering. The frequency of orders for IV voriconazole, which was the target of our study intervention, did not change during this period (Figure 3).

The dramatic effect of policy changes, as observed with pantoprazole and voriconazole, suggests that a more active strategy may have a greater impact on prescriber behavior when it comes to medication ordering in the inpatient setting. It also highlights several potential sources of confounding that may introduce bias to cost-transparency studies.

This study has multiple limitations. First, as with all observational study designs, causation cannot be drawn with certainty from our results. While we were able to compare medications to their preintervention baselines, the data could have been impacted by longitudinal or seasonal trends in medication ordering, which may have been impacted by seasonal variability in disease prevalence, changes in resistance patterns, and annual cycling of house staff in an academic medical center. While there appear to be potential seasonal patterns regarding prescribing patterns for some of the medications included in this analysis, we also believe the linear regressions capture the overall trends in prescribing adequately. Nonstationarity, or trends in the mean and variance of the outcome that are not related to the intervention, may introduce bias in the interpretation of our findings. However, we believe the parameters included in our models, namely the immediate change in the intercept following the intervention and the change in the trend of the rate of prescribing over time from pre- to postintervention, provide substantial protections from faulty interpretation. Our models are limited to the extent that these parameters do not account for nonstationarity. Additionally, we did not collect data on dosing frequency or duration of treatment, which would have been dependent on factors that are not readily quantified, such as indication, clinical rationale, or patient response. Thus, we were not able to evaluate the impact of the intervention on these factors.

Although intended to enhance internal validity, comparison groups were also subject to external influence. For example, we observed a significant, short-lived rise in oral ribavirin (a control medication) ordering during the preintervention baseline period that appeared to be independent of our intervention and may speak to the unaccounted-for longitudinal variability detailed above.

Finally, the clinical indication and setting may be important. Previous studies performed at the same hospital with price displays showed a reduction in laboratory ordering but no change in imaging.18,19 One might speculate that ordering fewer laboratory tests is viewed by providers as eliminating waste rather than choosing a less expensive option to accomplish the same diagnostic task at hand. Therapeutics may be more similar to radiology tests, because patients presumably need the treatment and often do not have the option of simply not ordering without a concerted effort to reevaluate the treatment plan. Additionally, in a tertiary care teaching center such as ours, a junior clinician, oftentimes at the behest of a more senior colleague, enters most orders. In an environment in which the ordering prescriber has more autonomy or when the order is driven by a junior practitioner rather than an attending (such as daily laboratories), results may be different. Additionally, institutions that incentivize prescribers directly to practice cost-conscious care may experience different results from similar interventions.

We conclude that, in the case of medication cost messaging, a strategy of displaying cost information alone was insufficient to affect prescriber ordering behavior. Coupling cost transparency with educational interventions and active stewardship to impact clinical practice is worthy of further study.

 

Disclosures: The authors state that there were no external sponsors for this work. The Johns Hopkins Hospital and University “funded” this work by paying the salaries of the authors. The author team maintained independence and made all decisions regarding the study design, data collection, data analysis, interpretation of results, writing of the research report, and decision to submit it for publication. Dr. Shermock had full access to all the study data and takes responsibility for the integrity of the data and accuracy of the data analysis.

Secondary to rising healthcare costs in the United States, broad efforts are underway to identify and reduce waste in the health system.1,2 A recent systematic review exhibited that many physicians inaccurately estimate the cost of medications.3 Raising awareness of medication costs among prescribers is one potential way to promote high-value care.

Some evidence suggests that cost transparency may help prescribers understand how medication orders drive costs. In a previous study carried out at the Johns Hopkins Hospital, fee data were displayed to providers for diagnostic laboratory tests.4 An 8.6% decrease (95% confidence interval [CI], –8.99% to –8.19%) in test ordering was observed when costs were displayed vs a 5.6% increase (95% CI, 4.90% to 6.39%) in ordering when costs were not displayed during a 6-month intervention period (P < 0.001). Conversely, a similar study that investigated the impact of cost transparency on inpatient imaging utilization did not demonstrate a significant influence of cost display.5 This suggests that cost transparency may work in some areas of care but not in others. A systematic review that investigated price-display interventions for imaging, laboratory studies, and medications reported 10 studies that demonstrated a statistically significant decrease in expenditures without an effect on patient safety.6

Informing prescribers of institution-specific medication costs within and between drug classes may enable the selection of less expensive, therapeutically equivalent drugs. Prior studies investigating the effect of medication cost display were conducted in a variety of patient care settings, including ambulatory clinics,7 urgent care centers,8 and operating rooms,9^,10 with some yielding positive results in terms of ordering and cost11^,12 and others having no impact.13^,14 Currently, there is little evidence specifically addressing the effect of cost display for medications in the inpatient setting.

As part of an institutional initiative to control pharmaceutical expenditures, informational messaging for several high-cost drugs was initiated at our tertiary care hospital in April 2015. The goal of our study was to assess the effect of these medication cost messages on ordering practices. We hypothesized that the display of inpatient pharmaceutical costs at the time of order entry would result in a reduction in ordering.

METHODS

Setting, Intervention, and Participants

As part of an effort to educate prescribers about the high cost of medications, 9 intravenous (IV) medications were selected by the Johns Hopkins Hospital Pharmacy and Therapeutics Committee as targets for drug cost messaging. The intention of the committee was to implement a rapid, low-cost, proof-of-concept, quality-improvement project that was not designed as prospective research. Representatives from the pharmacy and clinicians from relevant clinical areas participated in preimplementation discussions to help identify medications that were subjectively felt to be overused at our institution and potentially modifiable through provider education. The criteria for selecting drug targets included a variety of factors, such as medications infrequently ordered but representing a significant cost per dose (eg, eculizumab and ribavirin), frequently ordered medications with less expensive substitutes (eg, linezolid and voriconazole), and high-cost medications without direct therapeutic alternatives (eg, calcitonin). From April 10, 2015, to October 5, 2015, the computerized Provider Order Entry System (cPOE), Sunrise Clinical Manager (Allscripts Corporation, Chicago, IL), displayed the cost for targeted medications. Seven of the medication alerts also included a reasonable therapeutic alternative and its cost. There were no restrictions placed on ordering; prescribers were able to choose the high-cost medications at their discretion.

Despite the fact that this initiative was not designed as a research project, we felt it was important to formally evaluate the impact of the drug cost messaging effort to inform future quality-improvement interventions. Each medication was compared to its preintervention baseline utilization dating back to January 1, 2013. For the 7 medications with alternatives offered, we also analyzed use of the suggested alternative during these time periods.

Data Sources and Measurement

Our study utilized data obtained from the pharmacy order verification system and the cPOE database. Data were collected over a period of 143 weeks from January 1, 2013, to October 5, 2015, to allow for a baseline period (January 1, 2013, to April 9, 2015) and an intervention period (April 10, 2015, to October 5, 2015). Data elements extracted included drug characteristics (dosage form, route, cost, strength, name, and quantity), patient characteristics (race, gender, and age), clinical setting (facility location, inpatient or outpatient), and billing information (provider name, doses dispensed from pharmacy, order number, revenue or procedure code, record number, date of service, and unique billing number) for each admission. Using these elements, we generated the following 8 variables to use in our analyses: week, month, period identifier, drug name, dosage form, weekly orders, weekly patient days, and number of weekly orders per 10,000 patient days. Average wholesale price (AWP), referred to as medication cost in this manuscript, was used to report all drug costs in all associated cost calculations. While the actual cost of acquisition and price charged to the patient may vary based on several factors, including manufacturer and payer, we chose to use AWP as a generalizable estimate of the cost of acquisition of the drug for the hospital.

 

Variables

“Week” and “month” were defined as the week and month of our study, respectively. The “period identifier” was a binary variable that identified the time period before and after the intervention. “Weekly orders” was defined as the total number of new orders placed per week for each specified drug included in our study. For example, if a patient received 2 discrete, new orders for a medication in a given week, 2 orders would be counted toward the “weekly orders” variable. “Patient days,” defined as the total number of patients treated at our facility, was summated for each week of our study to yield “weekly patient days.” To derive the “number of weekly orders per 10,000 patient days,” we divided weekly orders by weekly patient days and multiplied the resultant figure by 10,000.

Statistical Analysis

Segmented regression, a form of interrupted time series analysis, is a quasi-experimental design that was used to determine the immediate and sustained effects of the drug cost messages on the rate of medication ordering.^15-17 The model enabled the use of comparison groups (alternative medications, as described above) to enhance internal validity.

In time series data, outcomes may not be independent over time. Autocorrelation of the error terms can arise when outcomes are more similar at time points closer together than outcomes at time points further apart. Failure to account for autocorrelation of the error terms may lead to underestimated standard errors. The presence of autocorrelation, assessed by calculating the Durbin-Watson statistic, was significant among our data. To adjust for this, we employed a Prais-Winsten estimation to adjust the error term (ε_t) calculated in our models.

Two segmented linear regression models were used to estimate trends in ordering before and after the intervention. The presence or absence of a comparator drug determined which model was to be used. When only single medications were under study, as in the case of eculizumab and calcitonin, our regression model was as follows:

Y_t = (β₀) + (β₁)(Time_t) + (β₂)(Intervention_t) + (β₃)(Post-Intervention Time_t) + (ε_t)

In our single-drug model, Y_t denoted the number of orders per 10,000 patient days at week “t”; Time_t was a continuous variable that indicated the number of weeks prior to or after the study intervention (April 10, 2015) and ranged from –116 to 27 weeks. Post-Intervention Time_t was a continuous variable that denoted the number of weeks since the start of the intervention and is coded as zero for all time periods prior to the intervention. β₀ was the estimated baseline number of orders per 10,000 patient days at the beginning of the study. β₁ is the trend of orders per 10,000 patient days per week during the preintervention period; β₂ represents an estimate of the change in the number of orders per 10,000 patient days immediately after the intervention; β₃ denotes the difference between preintervention and postintervention slopes; and ε_t is the “error term,” which represents autocorrelation and random variability of the data.

As mentioned previously, alternative dosage forms of 7 medications included in our study were utilized as comparison groups. In these instances (when multiple drugs were included in our analyses), the following regression model was applied:

Y _t = ( β ₀ ) + ( β ₁ )(Time _t ) + ( β ₂ )(Intervention _t ) + ( β ₃ )(Post-Intervention Time _t ) + ( β ₄ )(Cohort) + ( β ₅ )(Cohort)(Time _t ) + ( β ₆ )(Cohort)(Intervention _t ) + ( β ₇ )(Cohort)(Post-Intervention Time _t ) + ( ε _t )

 

Here, 3 coefficients were added (β₄-β₇) to describe an additional cohort of orders. Cohort, a binary indicator variable, held a value of either 0 or 1 when the model was used to describe the treatment or comparison group, respectively. The coefficients β₄-β₇ described the treatment group, and β₀-β₃ described the comparison group. β₄ was the difference in the number of baseline orders per 10,000 patient days between treatment and comparison groups; Β₅ represented the difference between the estimated ordering trends of treatment and comparison groups; and Β₆ indicated the difference in immediate changes in the number of orders per 10,000 patient days in the 2 groups following the intervention.

The number of orders per week was recorded for each medicine, which enabled a large number of data points to be included in our analyses. This allowed for more accurate and stable estimates to be made in our regression model. A total of 143 data points were collected for each study group, 116 before and 27 following each intervention.

All analyses were conducted by using STATA version 13.1 (StataCorp LP, College Station, TX).

RESULTS

Initial results pertaining to 9 IV medications were examined (Table). Following the implementation of cost messaging, no significant changes were observed in order frequency or trend for IV formulations of eculizumab, calcitonin, levetiracetam, linezolid, mycophenolate, ribavirin, voriconazole, and levothyroxine (Figures 1 and 2). However, a significant decrease in the number of oral ribavirin orders (Figure 2), the control group for the IV form, was observed (–16.3 orders per 10,000 patient days; P = .004; 95% CI, –27.2 to –5.31).

From March 26, 2015, to January 21, 2016, the use of IV pantoprazole was restricted at our facility due to a national shortage, requiring clinical pharmacy review and approval. Additionally, on September 15, 2014, oncology order sets that allowed antibiotic prescribing without an infectious disease provider’s approval replaced oral voriconazole with oral posaconazole. Consequently, both IV pantoprazole and oral voriconazole exhibited a sharp, sustained decrease in ordering following these policy changes (Figure 3). Although an apparent difference in ordering followed the policy changes, no significant change was noted in IV pantoprazole and oral voriconazole following our intervention.

DISCUSSION

Our results suggest that the passive strategy of displaying cost alone was not effective in altering prescriber ordering patterns for the selected medications. This may be due to a lack of awareness regarding direct financial impact on the patient, importance of costs in medical decision-making, or a perceived lack of alternatives or suitability of recommended alternatives. These results may prove valuable to hospital and pharmacy leadership as they develop strategies to curb medication expense.

Changes observed in IV pantoprazole ordering are instructive. Due to a national shortage, the IV form of this medication underwent a restriction, which required approval by the pharmacy prior to dispensing. This restriction was instituted independently of our study and led to a 73% decrease from usage rates prior to policy implementation (Figure 3). Ordering was restricted according to defined criteria for IV use. The restriction did not apply to oral pantoprazole, and no significant change in ordering of the oral formulation was noted during the evaluated period (Figure 3).

Oral voriconazole also exhibited a marked reduction in ordering of 46% following a change in hospital formulary that occurred independent of our study. The change in formulary involved a replacement of oral voriconazole with oral posaconazole in several oncology order sets, allowing antifungal administration without authorization by an infectious disease clinician. A dramatic reduction in ordering resulted from the removal of oral voriconazole from these order sets. Figure 3 represents the effect of the policy change on ordering. The frequency of orders for IV voriconazole, which was the target of our study intervention, did not change during this period (Figure 3).

The dramatic effect of policy changes, as observed with pantoprazole and voriconazole, suggests that a more active strategy may have a greater impact on prescriber behavior when it comes to medication ordering in the inpatient setting. It also highlights several potential sources of confounding that may introduce bias to cost-transparency studies.

This study has multiple limitations. First, as with all observational study designs, causation cannot be drawn with certainty from our results. While we were able to compare medications to their preintervention baselines, the data could have been impacted by longitudinal or seasonal trends in medication ordering, which may have been impacted by seasonal variability in disease prevalence, changes in resistance patterns, and annual cycling of house staff in an academic medical center. While there appear to be potential seasonal patterns regarding prescribing patterns for some of the medications included in this analysis, we also believe the linear regressions capture the overall trends in prescribing adequately. Nonstationarity, or trends in the mean and variance of the outcome that are not related to the intervention, may introduce bias in the interpretation of our findings. However, we believe the parameters included in our models, namely the immediate change in the intercept following the intervention and the change in the trend of the rate of prescribing over time from pre- to postintervention, provide substantial protections from faulty interpretation. Our models are limited to the extent that these parameters do not account for nonstationarity. Additionally, we did not collect data on dosing frequency or duration of treatment, which would have been dependent on factors that are not readily quantified, such as indication, clinical rationale, or patient response. Thus, we were not able to evaluate the impact of the intervention on these factors.

Although intended to enhance internal validity, comparison groups were also subject to external influence. For example, we observed a significant, short-lived rise in oral ribavirin (a control medication) ordering during the preintervention baseline period that appeared to be independent of our intervention and may speak to the unaccounted-for longitudinal variability detailed above.

Finally, the clinical indication and setting may be important. Previous studies performed at the same hospital with price displays showed a reduction in laboratory ordering but no change in imaging.18,19 One might speculate that ordering fewer laboratory tests is viewed by providers as eliminating waste rather than choosing a less expensive option to accomplish the same diagnostic task at hand. Therapeutics may be more similar to radiology tests, because patients presumably need the treatment and often do not have the option of simply not ordering without a concerted effort to reevaluate the treatment plan. Additionally, in a tertiary care teaching center such as ours, a junior clinician, oftentimes at the behest of a more senior colleague, enters most orders. In an environment in which the ordering prescriber has more autonomy or when the order is driven by a junior practitioner rather than an attending (such as daily laboratories), results may be different. Additionally, institutions that incentivize prescribers directly to practice cost-conscious care may experience different results from similar interventions.

We conclude that, in the case of medication cost messaging, a strategy of displaying cost information alone was insufficient to affect prescriber ordering behavior. Coupling cost transparency with educational interventions and active stewardship to impact clinical practice is worthy of further study.

 

Disclosures: The authors state that there were no external sponsors for this work. The Johns Hopkins Hospital and University “funded” this work by paying the salaries of the authors. The author team maintained independence and made all decisions regarding the study design, data collection, data analysis, interpretation of results, writing of the research report, and decision to submit it for publication. Dr. Shermock had full access to all the study data and takes responsibility for the integrity of the data and accuracy of the data analysis.