Over 900,000 incident and recurrent venous thromboembolism (VTE) events occur in the United States each year, resulting in nearly 300,000 fatalities.[1] VTE, including deep vein thrombosis (DVT) and pulmonary embolism (PE), is among the most common causes of death in the United States, with more people dying annually from VTE than motor vehicle accidents and breast cancer.[2]

Accordingly, healthcare policy makers and regulators have placed greater emphasis on VTE prevention, including use of VTE prophylaxis measures in the Centers for Medicare and Medicaid Services (CMS) value‐based purchasing (pay for performance) program and the Joint Commission's adoption of a national hospital patient safety goal related to anticoagulation therapy.[3, 4] Beginning in 2008, VTE events following hip and knee procedures were included as 1 of 10 hospital‐acquired conditions for which CMS would not pay for associated additional costs of care.[5]

A typical 300‐bed hospital can expect roughly 150 cases of hospital‐acquired VTE annually.[6] Up to 75% of these cases will occur on the medicine service, where nearly every patient has 1 or more VTE risk factor.[7] Although effective preventive modalities exist, prophylaxis rates among medical patients have been noted to be <50%.[8, 9] While quality improvement interventions have been shown to be effective in improving compliance with VTE prophylaxis, there are few studies describing effectiveness of these interventions in electronic health record (EHR) environments.[10] As EHR implementation accelerates, it will be essential to define the strengths and limitations of various decision support approaches to optimally improve patient safety.

We sought to evaluate the effectiveness and safety of a computerized decision support application, which was designed as part of a quality improvement initiative to improve rates of VTE prophylaxis rates on the medicine services at 2 hospital sites.

METHODS

RESULTS

Table 1 compares the effectiveness of the decision support module intervention in medicine intervention and in nonmedicine (nonintervention) services. Among medicine service patients, any VTE prophylaxis ordering increased from 61.9% to 82.1% (P < 0.001), and pharmacologic VTE prophylaxis increased from 59.0% to 74.5% (P < 0.001). Smaller but significant increases were observed on nonmedicine services. Hospital‐acquired VTE incidence on medicine services decreased significantly, from 0.65% to 0.42% (P = 0.008) and nonsignificantly on nonmedicine services.

Table 2 shows ordering patterns for major VTE prophylaxis modalities. Among eligible medicine service patients, rates of low molecular weight heparin prophylaxis increased from 13.0% to 23.7% (P < 0.001), and of unfractionated heparin prophylaxis from 35.1% to 40.7% (P < 0.001). On nonmedicine services, there was no significant change in low molecular weight heparin use, and unfractionated heparin use increased significantly from 37.2% to 40.9% (P < 0.001). Proportions of patients receiving mechanical prophylaxis or not receiving prophylaxis decreased significantly by 37.8% on medicine services and by 9.8% on nonmedicine services Table 3 shows the safety of the decision support module. Bleeding rates increased on medicine services from 2.9% to 4.0% (P < 0.001) and on nonmedicine services from 7.7% to 8.6% (P = 0.043). Nonsignificant changes in thrombocytopenia rates were observed on both services.

DISCUSSION

Following implementation of a computerized decision support application to improve VTE prophylaxis on 2 hospital medicine services, we observed a significant increase in the rate of overall and pharmacologic VTE prophylaxis use and a significant decrease in the incidence of hospital‐acquired VTE. Changes were of greater magnitude and significance on medicine services where the intervention was deployed.

Rates of any VTE prophylaxis and pharmacologic VTE prophylaxis ordering on medicine services increased significantly by 32.7% and 26.3%, respectively. These rates increased on nonmedicine comparison services by a more modest 4.4% for any VTE prophylaxis and 6.7% for pharmacologic VTE prophylaxis. Although the medicine service intervention was designed to be agnostic to the type of prophylactic heparin preparation, the intervention resulted in a significant 82.4% increase in low molecular weight heparin use and a significant 16.0% increase in unfractionated heparin use. With respect to outcomes, we observed a 34.5% decrease (P < 0.001) in hospital‐acquired VTE incidence on medicine services and a nonsignificant decrease on nonmedicine services.

In assessing intervention safety, increased usage of VTE prophylaxis was not accompanied by an increase in thrombocytopenia, but was associated with an increase in bleeding from 2.9% to 4.0% (P < 0.001) on medicine services and from 7.7% to 8.6% (P = 0.043) on non‐medicine services. As our intervention was a quality improvement project, we conducted a brief post hoc analysis to evaluate the increased bleeding rate on the medicine service following intervention. A random sample of 50 records of medicine patients who had received VTE prophylaxis and had a subsequent bleeding event was reviewed. Findings are summarized in Table 4. Prophylaxis was used appropriately in 100% of cases. Bleeding episodes were minor in that no case required more than 2 U of packed red blood cells. The most common clinical scenario was a patient with baseline anemia, typically with chronic kidney disease, who had a slight decrease in hematocrit of unclear etiology requiring 1 U of blood.

Although the intervention occurred on medicine services, favorable albeit smaller changes were observed on nonmedicine services. We expected this favorable secular trend because of VTE prophylaxis awareness efforts across the organization as a whole. There was also ongoing focus on VTE prevention and outcomes by policymakers, regulatory agencies, and professional societies during the time period of study.[3, 4, 5, 12] Public reporting of CMS inpatient surgical VTE prophylaxis measures was required throughout the study period.[13] Changes observed on medicine services occurred during a period where there were no publicly reported measures of VTE prophylaxis for inpatient medicine services.

Our study had several limitations. We derived our eligibility criteria for VTE prophylaxis based on administrative data. To address this, we incorporated accepted standardized definitions,[11] used clinical data elements in our queries beyond ICD‐9 codes (eg, platelet count), and applied pertinent exclusion criteria (eg, length of stay 1 day or less). VTE events that were present on admission were excluded from analyses. However, as these community‐acquired VTE events may be caused by inadequate VTE prophylaxis during a prior hospitalization, the overall true incidence of hospital‐acquired VTE was likely underestimated.

With respect to the hospital‐acquired VTE outcome, we did not distinguish superficial from deep VTE. A consistent AHRQ definition of 13 ICD‐9 VTE codes was used to identify clinically significant VTE events for the periods before and after the intervention. Although the present on admission code identified VTE events that were hospital acquired, 1 new acute VTE ICD‐9 code was added in October 2009, allowing for more specific coding of acute, isolated, upper extremity VTE. Accordingly, our postintervention hospital‐acquired VTE rate may have slightly underestimated the true hospital‐acquired VTE incidence by omitting some coded acute, isolated, upper extremity VTE cases (if not coded using the prior Other VTE codes). In a study in a teaching hospital setting, isolated upper extremity VTE accounted for up to 21% of all symptomatic VTE events among adults.[14]

With respect to VTE prophylaxis, the study evaluated use in a dichotomous fashion but did not assess appropriateness, or adequacy of dosing of pharmacologic agents. We did not employ the intervention in a randomized fashion on the medicine service. As our project was a quality improvement intervention, we used a concurrent control group of nonmedicine service patients to assess potential secular trend bias.

With respect to the safety of the intervention, the record review we performed supported the appropriateness of prophylaxis use following the intervention, but was not designed to establish whether the increase in prophylaxis use was the proximate cause of bleeding events observed. Similarly, as specific testing for heparin‐induced thrombocytopenia was not used, the lack of significant change in thrombocytopenia rates before and after the intervention cannot directly establish the intervention's safety. Finally, our study also included only in‐hospital end points.

The rate of VTE prophylaxis use in hospitals has been noted to be disappointing.[15] Two large multinational studies found that VTE prophylaxis rates in at‐risk hospitalized medical patients in the United States were 48% and 52%.[9, 16] Amin and colleagues found the overall rate of VTE prophylaxis among 227 US hospitals to be 62%.[17] Accordingly, our intervention, which resulted in an 82% compliance rate on a large medical service and was associated with a significantly reduced VTE incidence, appears to be highly effective. Our results are likely more favorable in that beyond length of stay criteria, we did not exclude less acutely ill medical patients from analyses.

Michota summarized quality improvement studies for VTE prevention.[10] Among 9 studies attempting to improve VTE prophylaxis, 2 used electronic decision support as a primary strategy, and only 1, by Kucher et al., used a computerized approach on a medical service.[18] This study showed significant improvement in VTE prophylaxis and incidence in patients randomized to a provider computer program. The intervention was complex, requiring specification of 8 patient‐level risk factors via a customized database, and the physician to recommend specific prophylactic regimens accordingly. Our findings, using a more basic approach, similarly support the effectiveness of using automated decision support, which can be readily modified as evidence‐based guidelines evolve.

Overall adoption of information technology systems in US hospitals is low: only 7.6% of hospitals have a basic system, and 17% have computerized physician order entry.[19] As hospitals have been financially incentivized to adopt such systems, our relatively simple intervention may prove to be readily generalizable across varied vendor systems.[20] The intervention involved order sets triggered by automated logic, corollary information, and a hard stop to prompt VTE prophylaxis. Within the context of intensified emphasis on reducing harm in the inpatient setting and various pay for performance programs, our intervention is also of importance to payers.[3, 5] Using national data in year 2000, Zhan and Miller calculated the excess charges per case associated with VTE to be $21,709.[21]

In conclusion, a relatively simple automated clinical decision support application significantly improved rates of VTE prophylaxis and was associated with significantly lower hospital‐acquired VTE incidence in hospitalized medicine patients, with a reasonable safety profile.

Over 900,000 incident and recurrent venous thromboembolism (VTE) events occur in the United States each year, resulting in nearly 300,000 fatalities.[1] VTE, including deep vein thrombosis (DVT) and pulmonary embolism (PE), is among the most common causes of death in the United States, with more people dying annually from VTE than motor vehicle accidents and breast cancer.[2]

Accordingly, healthcare policy makers and regulators have placed greater emphasis on VTE prevention, including use of VTE prophylaxis measures in the Centers for Medicare and Medicaid Services (CMS) value‐based purchasing (pay for performance) program and the Joint Commission's adoption of a national hospital patient safety goal related to anticoagulation therapy.[3, 4] Beginning in 2008, VTE events following hip and knee procedures were included as 1 of 10 hospital‐acquired conditions for which CMS would not pay for associated additional costs of care.[5]

A typical 300‐bed hospital can expect roughly 150 cases of hospital‐acquired VTE annually.[6] Up to 75% of these cases will occur on the medicine service, where nearly every patient has 1 or more VTE risk factor.[7] Although effective preventive modalities exist, prophylaxis rates among medical patients have been noted to be <50%.[8, 9] While quality improvement interventions have been shown to be effective in improving compliance with VTE prophylaxis, there are few studies describing effectiveness of these interventions in electronic health record (EHR) environments.[10] As EHR implementation accelerates, it will be essential to define the strengths and limitations of various decision support approaches to optimally improve patient safety.

We sought to evaluate the effectiveness and safety of a computerized decision support application, which was designed as part of a quality improvement initiative to improve rates of VTE prophylaxis rates on the medicine services at 2 hospital sites.

METHODS

RESULTS

Table 1 compares the effectiveness of the decision support module intervention in medicine intervention and in nonmedicine (nonintervention) services. Among medicine service patients, any VTE prophylaxis ordering increased from 61.9% to 82.1% (P < 0.001), and pharmacologic VTE prophylaxis increased from 59.0% to 74.5% (P < 0.001). Smaller but significant increases were observed on nonmedicine services. Hospital‐acquired VTE incidence on medicine services decreased significantly, from 0.65% to 0.42% (P = 0.008) and nonsignificantly on nonmedicine services.

Table 2 shows ordering patterns for major VTE prophylaxis modalities. Among eligible medicine service patients, rates of low molecular weight heparin prophylaxis increased from 13.0% to 23.7% (P < 0.001), and of unfractionated heparin prophylaxis from 35.1% to 40.7% (P < 0.001). On nonmedicine services, there was no significant change in low molecular weight heparin use, and unfractionated heparin use increased significantly from 37.2% to 40.9% (P < 0.001). Proportions of patients receiving mechanical prophylaxis or not receiving prophylaxis decreased significantly by 37.8% on medicine services and by 9.8% on nonmedicine services Table 3 shows the safety of the decision support module. Bleeding rates increased on medicine services from 2.9% to 4.0% (P < 0.001) and on nonmedicine services from 7.7% to 8.6% (P = 0.043). Nonsignificant changes in thrombocytopenia rates were observed on both services.

DISCUSSION

Following implementation of a computerized decision support application to improve VTE prophylaxis on 2 hospital medicine services, we observed a significant increase in the rate of overall and pharmacologic VTE prophylaxis use and a significant decrease in the incidence of hospital‐acquired VTE. Changes were of greater magnitude and significance on medicine services where the intervention was deployed.

Rates of any VTE prophylaxis and pharmacologic VTE prophylaxis ordering on medicine services increased significantly by 32.7% and 26.3%, respectively. These rates increased on nonmedicine comparison services by a more modest 4.4% for any VTE prophylaxis and 6.7% for pharmacologic VTE prophylaxis. Although the medicine service intervention was designed to be agnostic to the type of prophylactic heparin preparation, the intervention resulted in a significant 82.4% increase in low molecular weight heparin use and a significant 16.0% increase in unfractionated heparin use. With respect to outcomes, we observed a 34.5% decrease (P < 0.001) in hospital‐acquired VTE incidence on medicine services and a nonsignificant decrease on nonmedicine services.

In assessing intervention safety, increased usage of VTE prophylaxis was not accompanied by an increase in thrombocytopenia, but was associated with an increase in bleeding from 2.9% to 4.0% (P < 0.001) on medicine services and from 7.7% to 8.6% (P = 0.043) on non‐medicine services. As our intervention was a quality improvement project, we conducted a brief post hoc analysis to evaluate the increased bleeding rate on the medicine service following intervention. A random sample of 50 records of medicine patients who had received VTE prophylaxis and had a subsequent bleeding event was reviewed. Findings are summarized in Table 4. Prophylaxis was used appropriately in 100% of cases. Bleeding episodes were minor in that no case required more than 2 U of packed red blood cells. The most common clinical scenario was a patient with baseline anemia, typically with chronic kidney disease, who had a slight decrease in hematocrit of unclear etiology requiring 1 U of blood.

Although the intervention occurred on medicine services, favorable albeit smaller changes were observed on nonmedicine services. We expected this favorable secular trend because of VTE prophylaxis awareness efforts across the organization as a whole. There was also ongoing focus on VTE prevention and outcomes by policymakers, regulatory agencies, and professional societies during the time period of study.[3, 4, 5, 12] Public reporting of CMS inpatient surgical VTE prophylaxis measures was required throughout the study period.[13] Changes observed on medicine services occurred during a period where there were no publicly reported measures of VTE prophylaxis for inpatient medicine services.

Our study had several limitations. We derived our eligibility criteria for VTE prophylaxis based on administrative data. To address this, we incorporated accepted standardized definitions,[11] used clinical data elements in our queries beyond ICD‐9 codes (eg, platelet count), and applied pertinent exclusion criteria (eg, length of stay 1 day or less). VTE events that were present on admission were excluded from analyses. However, as these community‐acquired VTE events may be caused by inadequate VTE prophylaxis during a prior hospitalization, the overall true incidence of hospital‐acquired VTE was likely underestimated.

With respect to the hospital‐acquired VTE outcome, we did not distinguish superficial from deep VTE. A consistent AHRQ definition of 13 ICD‐9 VTE codes was used to identify clinically significant VTE events for the periods before and after the intervention. Although the present on admission code identified VTE events that were hospital acquired, 1 new acute VTE ICD‐9 code was added in October 2009, allowing for more specific coding of acute, isolated, upper extremity VTE. Accordingly, our postintervention hospital‐acquired VTE rate may have slightly underestimated the true hospital‐acquired VTE incidence by omitting some coded acute, isolated, upper extremity VTE cases (if not coded using the prior Other VTE codes). In a study in a teaching hospital setting, isolated upper extremity VTE accounted for up to 21% of all symptomatic VTE events among adults.[14]

With respect to VTE prophylaxis, the study evaluated use in a dichotomous fashion but did not assess appropriateness, or adequacy of dosing of pharmacologic agents. We did not employ the intervention in a randomized fashion on the medicine service. As our project was a quality improvement intervention, we used a concurrent control group of nonmedicine service patients to assess potential secular trend bias.

With respect to the safety of the intervention, the record review we performed supported the appropriateness of prophylaxis use following the intervention, but was not designed to establish whether the increase in prophylaxis use was the proximate cause of bleeding events observed. Similarly, as specific testing for heparin‐induced thrombocytopenia was not used, the lack of significant change in thrombocytopenia rates before and after the intervention cannot directly establish the intervention's safety. Finally, our study also included only in‐hospital end points.

The rate of VTE prophylaxis use in hospitals has been noted to be disappointing.[15] Two large multinational studies found that VTE prophylaxis rates in at‐risk hospitalized medical patients in the United States were 48% and 52%.[9, 16] Amin and colleagues found the overall rate of VTE prophylaxis among 227 US hospitals to be 62%.[17] Accordingly, our intervention, which resulted in an 82% compliance rate on a large medical service and was associated with a significantly reduced VTE incidence, appears to be highly effective. Our results are likely more favorable in that beyond length of stay criteria, we did not exclude less acutely ill medical patients from analyses.

Michota summarized quality improvement studies for VTE prevention.[10] Among 9 studies attempting to improve VTE prophylaxis, 2 used electronic decision support as a primary strategy, and only 1, by Kucher et al., used a computerized approach on a medical service.[18] This study showed significant improvement in VTE prophylaxis and incidence in patients randomized to a provider computer program. The intervention was complex, requiring specification of 8 patient‐level risk factors via a customized database, and the physician to recommend specific prophylactic regimens accordingly. Our findings, using a more basic approach, similarly support the effectiveness of using automated decision support, which can be readily modified as evidence‐based guidelines evolve.

Overall adoption of information technology systems in US hospitals is low: only 7.6% of hospitals have a basic system, and 17% have computerized physician order entry.[19] As hospitals have been financially incentivized to adopt such systems, our relatively simple intervention may prove to be readily generalizable across varied vendor systems.[20] The intervention involved order sets triggered by automated logic, corollary information, and a hard stop to prompt VTE prophylaxis. Within the context of intensified emphasis on reducing harm in the inpatient setting and various pay for performance programs, our intervention is also of importance to payers.[3, 5] Using national data in year 2000, Zhan and Miller calculated the excess charges per case associated with VTE to be $21,709.[21]

In conclusion, a relatively simple automated clinical decision support application significantly improved rates of VTE prophylaxis and was associated with significantly lower hospital‐acquired VTE incidence in hospitalized medicine patients, with a reasonable safety profile.

Over 900,000 incident and recurrent venous thromboembolism (VTE) events occur in the United States each year, resulting in nearly 300,000 fatalities.[1] VTE, including deep vein thrombosis (DVT) and pulmonary embolism (PE), is among the most common causes of death in the United States, with more people dying annually from VTE than motor vehicle accidents and breast cancer.[2]

Accordingly, healthcare policy makers and regulators have placed greater emphasis on VTE prevention, including use of VTE prophylaxis measures in the Centers for Medicare and Medicaid Services (CMS) value‐based purchasing (pay for performance) program and the Joint Commission's adoption of a national hospital patient safety goal related to anticoagulation therapy.[3, 4] Beginning in 2008, VTE events following hip and knee procedures were included as 1 of 10 hospital‐acquired conditions for which CMS would not pay for associated additional costs of care.[5]

A typical 300‐bed hospital can expect roughly 150 cases of hospital‐acquired VTE annually.[6] Up to 75% of these cases will occur on the medicine service, where nearly every patient has 1 or more VTE risk factor.[7] Although effective preventive modalities exist, prophylaxis rates among medical patients have been noted to be <50%.[8, 9] While quality improvement interventions have been shown to be effective in improving compliance with VTE prophylaxis, there are few studies describing effectiveness of these interventions in electronic health record (EHR) environments.[10] As EHR implementation accelerates, it will be essential to define the strengths and limitations of various decision support approaches to optimally improve patient safety.

We sought to evaluate the effectiveness and safety of a computerized decision support application, which was designed as part of a quality improvement initiative to improve rates of VTE prophylaxis rates on the medicine services at 2 hospital sites.

METHODS

RESULTS

Table 1 compares the effectiveness of the decision support module intervention in medicine intervention and in nonmedicine (nonintervention) services. Among medicine service patients, any VTE prophylaxis ordering increased from 61.9% to 82.1% (P < 0.001), and pharmacologic VTE prophylaxis increased from 59.0% to 74.5% (P < 0.001). Smaller but significant increases were observed on nonmedicine services. Hospital‐acquired VTE incidence on medicine services decreased significantly, from 0.65% to 0.42% (P = 0.008) and nonsignificantly on nonmedicine services.

Table 2 shows ordering patterns for major VTE prophylaxis modalities. Among eligible medicine service patients, rates of low molecular weight heparin prophylaxis increased from 13.0% to 23.7% (P < 0.001), and of unfractionated heparin prophylaxis from 35.1% to 40.7% (P < 0.001). On nonmedicine services, there was no significant change in low molecular weight heparin use, and unfractionated heparin use increased significantly from 37.2% to 40.9% (P < 0.001). Proportions of patients receiving mechanical prophylaxis or not receiving prophylaxis decreased significantly by 37.8% on medicine services and by 9.8% on nonmedicine services Table 3 shows the safety of the decision support module. Bleeding rates increased on medicine services from 2.9% to 4.0% (P < 0.001) and on nonmedicine services from 7.7% to 8.6% (P = 0.043). Nonsignificant changes in thrombocytopenia rates were observed on both services.

DISCUSSION

Following implementation of a computerized decision support application to improve VTE prophylaxis on 2 hospital medicine services, we observed a significant increase in the rate of overall and pharmacologic VTE prophylaxis use and a significant decrease in the incidence of hospital‐acquired VTE. Changes were of greater magnitude and significance on medicine services where the intervention was deployed.

Rates of any VTE prophylaxis and pharmacologic VTE prophylaxis ordering on medicine services increased significantly by 32.7% and 26.3%, respectively. These rates increased on nonmedicine comparison services by a more modest 4.4% for any VTE prophylaxis and 6.7% for pharmacologic VTE prophylaxis. Although the medicine service intervention was designed to be agnostic to the type of prophylactic heparin preparation, the intervention resulted in a significant 82.4% increase in low molecular weight heparin use and a significant 16.0% increase in unfractionated heparin use. With respect to outcomes, we observed a 34.5% decrease (P < 0.001) in hospital‐acquired VTE incidence on medicine services and a nonsignificant decrease on nonmedicine services.

In assessing intervention safety, increased usage of VTE prophylaxis was not accompanied by an increase in thrombocytopenia, but was associated with an increase in bleeding from 2.9% to 4.0% (P < 0.001) on medicine services and from 7.7% to 8.6% (P = 0.043) on non‐medicine services. As our intervention was a quality improvement project, we conducted a brief post hoc analysis to evaluate the increased bleeding rate on the medicine service following intervention. A random sample of 50 records of medicine patients who had received VTE prophylaxis and had a subsequent bleeding event was reviewed. Findings are summarized in Table 4. Prophylaxis was used appropriately in 100% of cases. Bleeding episodes were minor in that no case required more than 2 U of packed red blood cells. The most common clinical scenario was a patient with baseline anemia, typically with chronic kidney disease, who had a slight decrease in hematocrit of unclear etiology requiring 1 U of blood.

Although the intervention occurred on medicine services, favorable albeit smaller changes were observed on nonmedicine services. We expected this favorable secular trend because of VTE prophylaxis awareness efforts across the organization as a whole. There was also ongoing focus on VTE prevention and outcomes by policymakers, regulatory agencies, and professional societies during the time period of study.[3, 4, 5, 12] Public reporting of CMS inpatient surgical VTE prophylaxis measures was required throughout the study period.[13] Changes observed on medicine services occurred during a period where there were no publicly reported measures of VTE prophylaxis for inpatient medicine services.

Our study had several limitations. We derived our eligibility criteria for VTE prophylaxis based on administrative data. To address this, we incorporated accepted standardized definitions,[11] used clinical data elements in our queries beyond ICD‐9 codes (eg, platelet count), and applied pertinent exclusion criteria (eg, length of stay 1 day or less). VTE events that were present on admission were excluded from analyses. However, as these community‐acquired VTE events may be caused by inadequate VTE prophylaxis during a prior hospitalization, the overall true incidence of hospital‐acquired VTE was likely underestimated.

With respect to the hospital‐acquired VTE outcome, we did not distinguish superficial from deep VTE. A consistent AHRQ definition of 13 ICD‐9 VTE codes was used to identify clinically significant VTE events for the periods before and after the intervention. Although the present on admission code identified VTE events that were hospital acquired, 1 new acute VTE ICD‐9 code was added in October 2009, allowing for more specific coding of acute, isolated, upper extremity VTE. Accordingly, our postintervention hospital‐acquired VTE rate may have slightly underestimated the true hospital‐acquired VTE incidence by omitting some coded acute, isolated, upper extremity VTE cases (if not coded using the prior Other VTE codes). In a study in a teaching hospital setting, isolated upper extremity VTE accounted for up to 21% of all symptomatic VTE events among adults.[14]

With respect to VTE prophylaxis, the study evaluated use in a dichotomous fashion but did not assess appropriateness, or adequacy of dosing of pharmacologic agents. We did not employ the intervention in a randomized fashion on the medicine service. As our project was a quality improvement intervention, we used a concurrent control group of nonmedicine service patients to assess potential secular trend bias.

With respect to the safety of the intervention, the record review we performed supported the appropriateness of prophylaxis use following the intervention, but was not designed to establish whether the increase in prophylaxis use was the proximate cause of bleeding events observed. Similarly, as specific testing for heparin‐induced thrombocytopenia was not used, the lack of significant change in thrombocytopenia rates before and after the intervention cannot directly establish the intervention's safety. Finally, our study also included only in‐hospital end points.

The rate of VTE prophylaxis use in hospitals has been noted to be disappointing.[15] Two large multinational studies found that VTE prophylaxis rates in at‐risk hospitalized medical patients in the United States were 48% and 52%.[9, 16] Amin and colleagues found the overall rate of VTE prophylaxis among 227 US hospitals to be 62%.[17] Accordingly, our intervention, which resulted in an 82% compliance rate on a large medical service and was associated with a significantly reduced VTE incidence, appears to be highly effective. Our results are likely more favorable in that beyond length of stay criteria, we did not exclude less acutely ill medical patients from analyses.

Michota summarized quality improvement studies for VTE prevention.[10] Among 9 studies attempting to improve VTE prophylaxis, 2 used electronic decision support as a primary strategy, and only 1, by Kucher et al., used a computerized approach on a medical service.[18] This study showed significant improvement in VTE prophylaxis and incidence in patients randomized to a provider computer program. The intervention was complex, requiring specification of 8 patient‐level risk factors via a customized database, and the physician to recommend specific prophylactic regimens accordingly. Our findings, using a more basic approach, similarly support the effectiveness of using automated decision support, which can be readily modified as evidence‐based guidelines evolve.

Overall adoption of information technology systems in US hospitals is low: only 7.6% of hospitals have a basic system, and 17% have computerized physician order entry.[19] As hospitals have been financially incentivized to adopt such systems, our relatively simple intervention may prove to be readily generalizable across varied vendor systems.[20] The intervention involved order sets triggered by automated logic, corollary information, and a hard stop to prompt VTE prophylaxis. Within the context of intensified emphasis on reducing harm in the inpatient setting and various pay for performance programs, our intervention is also of importance to payers.[3, 5] Using national data in year 2000, Zhan and Miller calculated the excess charges per case associated with VTE to be $21,709.[21]

In conclusion, a relatively simple automated clinical decision support application significantly improved rates of VTE prophylaxis and was associated with significantly lower hospital‐acquired VTE incidence in hospitalized medicine patients, with a reasonable safety profile.

Community‐acquired pneumonia (CAP) is the most common lower respiratory tract infection in adults and a leading cause of infection‐related deaths in the United States.[1] According to a survey, pneumonia was the most common reason for hospital admissions through the emergency department in 2003.[2] CAP is associated with significant morbidity and mortality among those sick enough to require hospitalization. In a prospective study, hospital mortality rates ranged from 5% to 18% and length of stay from 9 to 23 days depending on patient location (intensive care unit [ICU] vs elsewhere) and severity of illness.[3]

Empirical evidence suggests that host inflammatory response contributes significantly to lung injury in pneumonia.[4] Studies have demonstrated reduction in the host inflammatory response as well as in mortality among animals with bacterial pneumonia when exposed to glucocorticoids.[5, 6] Furthermore, the efficacy of adjunctive steroid therapy in severe pneumonia caused by Pneumocystis jirovecii[7] and in pneumococcal meningitis[8, 9] is already established. However, due to equivocal, and at times conflicting, human clinical trial data on the impact of steroid therapy in CAP, the 2007 consensus guidelines (jointly published by the Infectious Diseases Society of America and American Thoracic Society) do not provide recommendations for or against use of steroids in CAP, except in the setting of hypotension secondary to adrenal insufficiency.[10]

In their meta‐analysis, Chen et al. analyzed data from 6 randomized clinical trials (RCTs) published between 1972 and 2007 (including 2 on pediatric patients) and concluded that adding steroids to current standard of care was not beneficial.[11] Earlier, Lamontagne et al.'s meta‐analysis included RCTs on hospitalized CAP patients as well as those on patients with acute lung injury (ALI) or acute respiratory distress syndrome (ARDS) from any cause.[12] They concluded that low‐dose corticosteroid therapy reduced all‐cause in‐hospital mortality in this mixed patient population (relative risk [RR]: 0.68 [95% confidence interval (CI): 0.49 to 0.96]). Recently, data from a number of additional RCTs have become available.[13, 14, 15, 16, 17] Therefore, an updated review of RCTs evaluating the role of adjunctive steroid therapy among adults hospitalized with CAP was warranted.

MATERIALS AND METHODS

We conducted this systematic review and meta‐analysis in accordance with the recommendations published in the Cochrane Handbook for Systematic Reviews of Interventions[18] and reported our findings according to the Preferred Reporting Items for Systematic Reviews and Meta‐analyses guidelines.[19] The overall quality of evidence was judged using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework.[20]

RESULTS

Eight RCTs, comprising 1119 subjects, were eventually chosen.[14]. Seven shortlisted studies were excluded due to methodological limitations, failure to fully meet PICO criteria, or gross insufficiency of descriptive data on subjects or methodology.[18, 31, 32, 33, 34, 35, 36] Figure 1 illustrates the study selection process.

Figure 1

Table 1 summarizes the baseline characteristics of patient populations from each study. Mean ages in 7 RCTs were between 60 years and 80 years. In Marik et al., the mean age of the intervention group was 31.7 years, whereas that of the control group was 40.6 years (P value not reported).[30] Three RCTs included ICU patients only,[17, 28, 30] whereas 4 only included general medical ward patients.[14, 15, 29, 31] Disease severity scores at admission were similar between the 2 groups in all RCTs except Sabry and Omar,[17] which was the only clinical trial to use a chest radiograph score. Only Sabry and Omar,[17] and Mikami et al.[29] excluded chronic obstructive pulmonary disease patients. Where possible, the serum C‐reactive protein (CRP) value on day one was subtracted from that on day eight to generate a one week delta CRP.

The mean ICU length of stay was 12.7 days for the steroid group and 12.3 days for the control group. The mean hospital lengths of stay were 10.2 days and 13.6 days, respectively. Quality of the studies was moderate (see Supporting Information, Appendix I, in the online version of this article). Kappa score was >0.90.

Meta‐analysis

Figure 2 illustrates the results of our meta‐analyses. Although adjunctive steroid therapy had no effect on hospital mortality or ICU length of stay, it was associated with reduced hospital length of stay (RR: 1.21 days [95% CI: 2.12 to 0.29]). Of note, Mikami et al.[29] did not report mortality in their article, whereas in McHardy and Schonell,[31] using the factorial design, each of the 2 treatment groups were further subdivided into those patients who received 1 g of ampicillin and those who received 2 g of ampicillin (Figure 2A).

Figure 2

Analysis of other outcomes was limited by the fact that data were pooled from only a few studies. These included the need for and duration of mechanical ventilation, development of new ARDS and ICU admission rate, neither of which was associated with steroid therapy. However, steroid use was associated with lower incidence of delayed shock (ie, shock occurring after enrollment (RR: 0.12 [95% CI: 0.03 to 0.41]) and lower incidence of persistent chest x‐ray abnormalities at 1 week (RR: 0.13 [95% CI: 0.06 to 0.27]).

DISCUSSION

In this meta‐analysis of 8 RCTs, we found no significant association between steroid therapy and our primary outcome of interest (hospital mortality). However, length of hospital stay was shorter in the steroid group. These findings were not altered in various sensitivity and subgroup analyses. Although adverse effects of steroid therapy were not consistently reported, most of the RCTs reported that hyperglycemia was either no more common in the steroid group or did not require additional treatment.

Previous meta‐analyses have also concluded that adding corticosteroids to conventional therapy does not impact mortality among adults hospitalized with CAP.[12, 22] This may or may not be a consequence of inadequate statistical power. Although Lamontagne et al.[13] reported that low‐dose corticosteroid therapy (2 mg/kg/day or less of methylprednisolone or equivalent) was associated with reduced hospital mortality (RR: 0.68 [95% CI: 0.49 to 0.96]), this result was obtained by pooling data from 5 RCTs on adults hospitalized with CAP and 4 on adults with ALI/ARDS from any cause. In a subgroup analysis of RCTs conducted only on CAP patients, no impact on mortality was found. Of note, all RCTs involving CAP patients had used low‐dose steroids; the 3 RCTs using high‐dose steroids were carried out on ALI/ARDS patients.[36, 37, 38] Similarly, all RCTs in our meta‐analysis were also characterized by steroid doses under 2 mg/kg/day of methylprednisolone or equivalent.

Our study is the first to demonstrate decreased length of hospital stay in this patient population. Importantly, each of the 5 studies that reported this outcome (including 3 relatively recent RCTs) showed the same trend. However, it is not inconceivable that steroid use led to a quicker decline in cytokine levels resulting in an earlier resolution of fever and hence earlier discharge without a faster cure per se. The two studies whose data permitted calculation of delta CRP also demonstrated a faster CRP decline in the steroid group (Table 1).

Our analysis also suggested reduced incidence of delayed shock. However, these data were pooled from only 2 RCTs,[17, 28] and each of them used hydrocortisone, whose direct mineralocorticoid effect is an obvious confounder. Similarly, according to data pooled from 2 RCTs, steroid use was associated with fewer cases of persistent chest x‐ray abnormalities by day 8. Of note, although calculation of the I² statistic was not possible because of too few studies, visual inspection of the forest plots suggested low levels of heterogeneity.

It is plausible that the impact of adjunctive steroids in CAP may vary based on the causative pathogen. This pathogen‐specific association has been observed in patients with bacterial meningitis, where most of the benefit is seemingly limited to pneumococcal meningitis.[9, 10] Unfortunately, as demonstrated by Snijders et al.,[16] establishing microbiologic etiology in CAP can be difficult, and most patients are treated empirically.

Our analysis showed no difference in duration of ventilation among patients who required ventilatory support on admission. However, only 2 studies reported this outcome.[17, 28] Second, in Confalonieri et al.,[28] the steroid group had a more severe baseline inflammatory response as illustrated by higher serum CRP levels (P = 0.04). Moreover, while mechanical ventilation was defined as either invasive or noninvasive ventilation, the steroid group had a higher number of patients who required noninvasive ventilation (P = 0.03), thus introducing selection bias. This study had additional areas of concern too, including a mortality of 0 among its 46 ICU patients, in contrast to established mortality rates of up to around 20%.[4] Unlike this study, Sabry and Omar[17] reported that none of their patients was on noninvasive ventilation. It may be pertinent to compare our findings with those of Steinberg et al.,[40] who studied patients with ARDS (pneumonia being the most common cause) who received methylprednisolone. This group had an early increase in ventilator‐free days, but that effect became less pronounced (though still significant) when the study end point was prolonged from 30 to 90 days.[41]

The 2 studies that were published before 2000 (McHardy and Schonell,[31] and Marik et al.[30]) were excluded in our a priori sensitivity analysis. A number of considerations led to this decision. First, standards of care for inpatient management of pneumoniaincluding pharmacologic therapies and ventilation strategieshave changed considerably over time. For instance, newer generation macrolides became available for clinical use in the early 1990s and meropenem in 1996.[41] Therefore, it would be hard to assume constancy of effect from that time period. Furthermore, the study by McHardy and Schonell[31] suffered from significant differences in the baseline characteristics of its 2 arms. There was incomplete randomization; patients with diabetes were excluded from only the steroid arm. Another issue with Marik et al.[30] was the considerably younger age of participants compared to other studies (Table 1).

Community‐acquired pneumonia (CAP) is the most common lower respiratory tract infection in adults and a leading cause of infection‐related deaths in the United States.[1] According to a survey, pneumonia was the most common reason for hospital admissions through the emergency department in 2003.[2] CAP is associated with significant morbidity and mortality among those sick enough to require hospitalization. In a prospective study, hospital mortality rates ranged from 5% to 18% and length of stay from 9 to 23 days depending on patient location (intensive care unit [ICU] vs elsewhere) and severity of illness.[3]

Empirical evidence suggests that host inflammatory response contributes significantly to lung injury in pneumonia.[4] Studies have demonstrated reduction in the host inflammatory response as well as in mortality among animals with bacterial pneumonia when exposed to glucocorticoids.[5, 6] Furthermore, the efficacy of adjunctive steroid therapy in severe pneumonia caused by Pneumocystis jirovecii[7] and in pneumococcal meningitis[8, 9] is already established. However, due to equivocal, and at times conflicting, human clinical trial data on the impact of steroid therapy in CAP, the 2007 consensus guidelines (jointly published by the Infectious Diseases Society of America and American Thoracic Society) do not provide recommendations for or against use of steroids in CAP, except in the setting of hypotension secondary to adrenal insufficiency.[10]

In their meta‐analysis, Chen et al. analyzed data from 6 randomized clinical trials (RCTs) published between 1972 and 2007 (including 2 on pediatric patients) and concluded that adding steroids to current standard of care was not beneficial.[11] Earlier, Lamontagne et al.'s meta‐analysis included RCTs on hospitalized CAP patients as well as those on patients with acute lung injury (ALI) or acute respiratory distress syndrome (ARDS) from any cause.[12] They concluded that low‐dose corticosteroid therapy reduced all‐cause in‐hospital mortality in this mixed patient population (relative risk [RR]: 0.68 [95% confidence interval (CI): 0.49 to 0.96]). Recently, data from a number of additional RCTs have become available.[13, 14, 15, 16, 17] Therefore, an updated review of RCTs evaluating the role of adjunctive steroid therapy among adults hospitalized with CAP was warranted.

MATERIALS AND METHODS

We conducted this systematic review and meta‐analysis in accordance with the recommendations published in the Cochrane Handbook for Systematic Reviews of Interventions[18] and reported our findings according to the Preferred Reporting Items for Systematic Reviews and Meta‐analyses guidelines.[19] The overall quality of evidence was judged using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework.[20]

RESULTS

Eight RCTs, comprising 1119 subjects, were eventually chosen.[14]. Seven shortlisted studies were excluded due to methodological limitations, failure to fully meet PICO criteria, or gross insufficiency of descriptive data on subjects or methodology.[18, 31, 32, 33, 34, 35, 36] Figure 1 illustrates the study selection process.

Figure 1

Table 1 summarizes the baseline characteristics of patient populations from each study. Mean ages in 7 RCTs were between 60 years and 80 years. In Marik et al., the mean age of the intervention group was 31.7 years, whereas that of the control group was 40.6 years (P value not reported).[30] Three RCTs included ICU patients only,[17, 28, 30] whereas 4 only included general medical ward patients.[14, 15, 29, 31] Disease severity scores at admission were similar between the 2 groups in all RCTs except Sabry and Omar,[17] which was the only clinical trial to use a chest radiograph score. Only Sabry and Omar,[17] and Mikami et al.[29] excluded chronic obstructive pulmonary disease patients. Where possible, the serum C‐reactive protein (CRP) value on day one was subtracted from that on day eight to generate a one week delta CRP.

The mean ICU length of stay was 12.7 days for the steroid group and 12.3 days for the control group. The mean hospital lengths of stay were 10.2 days and 13.6 days, respectively. Quality of the studies was moderate (see Supporting Information, Appendix I, in the online version of this article). Kappa score was >0.90.

Meta‐analysis

Figure 2 illustrates the results of our meta‐analyses. Although adjunctive steroid therapy had no effect on hospital mortality or ICU length of stay, it was associated with reduced hospital length of stay (RR: 1.21 days [95% CI: 2.12 to 0.29]). Of note, Mikami et al.[29] did not report mortality in their article, whereas in McHardy and Schonell,[31] using the factorial design, each of the 2 treatment groups were further subdivided into those patients who received 1 g of ampicillin and those who received 2 g of ampicillin (Figure 2A).

Figure 2

Analysis of other outcomes was limited by the fact that data were pooled from only a few studies. These included the need for and duration of mechanical ventilation, development of new ARDS and ICU admission rate, neither of which was associated with steroid therapy. However, steroid use was associated with lower incidence of delayed shock (ie, shock occurring after enrollment (RR: 0.12 [95% CI: 0.03 to 0.41]) and lower incidence of persistent chest x‐ray abnormalities at 1 week (RR: 0.13 [95% CI: 0.06 to 0.27]).

DISCUSSION

In this meta‐analysis of 8 RCTs, we found no significant association between steroid therapy and our primary outcome of interest (hospital mortality). However, length of hospital stay was shorter in the steroid group. These findings were not altered in various sensitivity and subgroup analyses. Although adverse effects of steroid therapy were not consistently reported, most of the RCTs reported that hyperglycemia was either no more common in the steroid group or did not require additional treatment.

Previous meta‐analyses have also concluded that adding corticosteroids to conventional therapy does not impact mortality among adults hospitalized with CAP.[12, 22] This may or may not be a consequence of inadequate statistical power. Although Lamontagne et al.[13] reported that low‐dose corticosteroid therapy (2 mg/kg/day or less of methylprednisolone or equivalent) was associated with reduced hospital mortality (RR: 0.68 [95% CI: 0.49 to 0.96]), this result was obtained by pooling data from 5 RCTs on adults hospitalized with CAP and 4 on adults with ALI/ARDS from any cause. In a subgroup analysis of RCTs conducted only on CAP patients, no impact on mortality was found. Of note, all RCTs involving CAP patients had used low‐dose steroids; the 3 RCTs using high‐dose steroids were carried out on ALI/ARDS patients.[36, 37, 38] Similarly, all RCTs in our meta‐analysis were also characterized by steroid doses under 2 mg/kg/day of methylprednisolone or equivalent.

Our study is the first to demonstrate decreased length of hospital stay in this patient population. Importantly, each of the 5 studies that reported this outcome (including 3 relatively recent RCTs) showed the same trend. However, it is not inconceivable that steroid use led to a quicker decline in cytokine levels resulting in an earlier resolution of fever and hence earlier discharge without a faster cure per se. The two studies whose data permitted calculation of delta CRP also demonstrated a faster CRP decline in the steroid group (Table 1).

Our analysis also suggested reduced incidence of delayed shock. However, these data were pooled from only 2 RCTs,[17, 28] and each of them used hydrocortisone, whose direct mineralocorticoid effect is an obvious confounder. Similarly, according to data pooled from 2 RCTs, steroid use was associated with fewer cases of persistent chest x‐ray abnormalities by day 8. Of note, although calculation of the I² statistic was not possible because of too few studies, visual inspection of the forest plots suggested low levels of heterogeneity.

It is plausible that the impact of adjunctive steroids in CAP may vary based on the causative pathogen. This pathogen‐specific association has been observed in patients with bacterial meningitis, where most of the benefit is seemingly limited to pneumococcal meningitis.[9, 10] Unfortunately, as demonstrated by Snijders et al.,[16] establishing microbiologic etiology in CAP can be difficult, and most patients are treated empirically.

Our analysis showed no difference in duration of ventilation among patients who required ventilatory support on admission. However, only 2 studies reported this outcome.[17, 28] Second, in Confalonieri et al.,[28] the steroid group had a more severe baseline inflammatory response as illustrated by higher serum CRP levels (P = 0.04). Moreover, while mechanical ventilation was defined as either invasive or noninvasive ventilation, the steroid group had a higher number of patients who required noninvasive ventilation (P = 0.03), thus introducing selection bias. This study had additional areas of concern too, including a mortality of 0 among its 46 ICU patients, in contrast to established mortality rates of up to around 20%.[4] Unlike this study, Sabry and Omar[17] reported that none of their patients was on noninvasive ventilation. It may be pertinent to compare our findings with those of Steinberg et al.,[40] who studied patients with ARDS (pneumonia being the most common cause) who received methylprednisolone. This group had an early increase in ventilator‐free days, but that effect became less pronounced (though still significant) when the study end point was prolonged from 30 to 90 days.[41]

The 2 studies that were published before 2000 (McHardy and Schonell,[31] and Marik et al.[30]) were excluded in our a priori sensitivity analysis. A number of considerations led to this decision. First, standards of care for inpatient management of pneumoniaincluding pharmacologic therapies and ventilation strategieshave changed considerably over time. For instance, newer generation macrolides became available for clinical use in the early 1990s and meropenem in 1996.[41] Therefore, it would be hard to assume constancy of effect from that time period. Furthermore, the study by McHardy and Schonell[31] suffered from significant differences in the baseline characteristics of its 2 arms. There was incomplete randomization; patients with diabetes were excluded from only the steroid arm. Another issue with Marik et al.[30] was the considerably younger age of participants compared to other studies (Table 1).

Community‐acquired pneumonia (CAP) is the most common lower respiratory tract infection in adults and a leading cause of infection‐related deaths in the United States.[1] According to a survey, pneumonia was the most common reason for hospital admissions through the emergency department in 2003.[2] CAP is associated with significant morbidity and mortality among those sick enough to require hospitalization. In a prospective study, hospital mortality rates ranged from 5% to 18% and length of stay from 9 to 23 days depending on patient location (intensive care unit [ICU] vs elsewhere) and severity of illness.[3]

Empirical evidence suggests that host inflammatory response contributes significantly to lung injury in pneumonia.[4] Studies have demonstrated reduction in the host inflammatory response as well as in mortality among animals with bacterial pneumonia when exposed to glucocorticoids.[5, 6] Furthermore, the efficacy of adjunctive steroid therapy in severe pneumonia caused by Pneumocystis jirovecii[7] and in pneumococcal meningitis[8, 9] is already established. However, due to equivocal, and at times conflicting, human clinical trial data on the impact of steroid therapy in CAP, the 2007 consensus guidelines (jointly published by the Infectious Diseases Society of America and American Thoracic Society) do not provide recommendations for or against use of steroids in CAP, except in the setting of hypotension secondary to adrenal insufficiency.[10]

In their meta‐analysis, Chen et al. analyzed data from 6 randomized clinical trials (RCTs) published between 1972 and 2007 (including 2 on pediatric patients) and concluded that adding steroids to current standard of care was not beneficial.[11] Earlier, Lamontagne et al.'s meta‐analysis included RCTs on hospitalized CAP patients as well as those on patients with acute lung injury (ALI) or acute respiratory distress syndrome (ARDS) from any cause.[12] They concluded that low‐dose corticosteroid therapy reduced all‐cause in‐hospital mortality in this mixed patient population (relative risk [RR]: 0.68 [95% confidence interval (CI): 0.49 to 0.96]). Recently, data from a number of additional RCTs have become available.[13, 14, 15, 16, 17] Therefore, an updated review of RCTs evaluating the role of adjunctive steroid therapy among adults hospitalized with CAP was warranted.

MATERIALS AND METHODS

We conducted this systematic review and meta‐analysis in accordance with the recommendations published in the Cochrane Handbook for Systematic Reviews of Interventions[18] and reported our findings according to the Preferred Reporting Items for Systematic Reviews and Meta‐analyses guidelines.[19] The overall quality of evidence was judged using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework.[20]

RESULTS

Eight RCTs, comprising 1119 subjects, were eventually chosen.[14]. Seven shortlisted studies were excluded due to methodological limitations, failure to fully meet PICO criteria, or gross insufficiency of descriptive data on subjects or methodology.[18, 31, 32, 33, 34, 35, 36] Figure 1 illustrates the study selection process.

Figure 1

Table 1 summarizes the baseline characteristics of patient populations from each study. Mean ages in 7 RCTs were between 60 years and 80 years. In Marik et al., the mean age of the intervention group was 31.7 years, whereas that of the control group was 40.6 years (P value not reported).[30] Three RCTs included ICU patients only,[17, 28, 30] whereas 4 only included general medical ward patients.[14, 15, 29, 31] Disease severity scores at admission were similar between the 2 groups in all RCTs except Sabry and Omar,[17] which was the only clinical trial to use a chest radiograph score. Only Sabry and Omar,[17] and Mikami et al.[29] excluded chronic obstructive pulmonary disease patients. Where possible, the serum C‐reactive protein (CRP) value on day one was subtracted from that on day eight to generate a one week delta CRP.

The mean ICU length of stay was 12.7 days for the steroid group and 12.3 days for the control group. The mean hospital lengths of stay were 10.2 days and 13.6 days, respectively. Quality of the studies was moderate (see Supporting Information, Appendix I, in the online version of this article). Kappa score was >0.90.

Meta‐analysis

Figure 2 illustrates the results of our meta‐analyses. Although adjunctive steroid therapy had no effect on hospital mortality or ICU length of stay, it was associated with reduced hospital length of stay (RR: 1.21 days [95% CI: 2.12 to 0.29]). Of note, Mikami et al.[29] did not report mortality in their article, whereas in McHardy and Schonell,[31] using the factorial design, each of the 2 treatment groups were further subdivided into those patients who received 1 g of ampicillin and those who received 2 g of ampicillin (Figure 2A).

Figure 2

Analysis of other outcomes was limited by the fact that data were pooled from only a few studies. These included the need for and duration of mechanical ventilation, development of new ARDS and ICU admission rate, neither of which was associated with steroid therapy. However, steroid use was associated with lower incidence of delayed shock (ie, shock occurring after enrollment (RR: 0.12 [95% CI: 0.03 to 0.41]) and lower incidence of persistent chest x‐ray abnormalities at 1 week (RR: 0.13 [95% CI: 0.06 to 0.27]).

DISCUSSION

In this meta‐analysis of 8 RCTs, we found no significant association between steroid therapy and our primary outcome of interest (hospital mortality). However, length of hospital stay was shorter in the steroid group. These findings were not altered in various sensitivity and subgroup analyses. Although adverse effects of steroid therapy were not consistently reported, most of the RCTs reported that hyperglycemia was either no more common in the steroid group or did not require additional treatment.

Previous meta‐analyses have also concluded that adding corticosteroids to conventional therapy does not impact mortality among adults hospitalized with CAP.[12, 22] This may or may not be a consequence of inadequate statistical power. Although Lamontagne et al.[13] reported that low‐dose corticosteroid therapy (2 mg/kg/day or less of methylprednisolone or equivalent) was associated with reduced hospital mortality (RR: 0.68 [95% CI: 0.49 to 0.96]), this result was obtained by pooling data from 5 RCTs on adults hospitalized with CAP and 4 on adults with ALI/ARDS from any cause. In a subgroup analysis of RCTs conducted only on CAP patients, no impact on mortality was found. Of note, all RCTs involving CAP patients had used low‐dose steroids; the 3 RCTs using high‐dose steroids were carried out on ALI/ARDS patients.[36, 37, 38] Similarly, all RCTs in our meta‐analysis were also characterized by steroid doses under 2 mg/kg/day of methylprednisolone or equivalent.

Our study is the first to demonstrate decreased length of hospital stay in this patient population. Importantly, each of the 5 studies that reported this outcome (including 3 relatively recent RCTs) showed the same trend. However, it is not inconceivable that steroid use led to a quicker decline in cytokine levels resulting in an earlier resolution of fever and hence earlier discharge without a faster cure per se. The two studies whose data permitted calculation of delta CRP also demonstrated a faster CRP decline in the steroid group (Table 1).

Our analysis also suggested reduced incidence of delayed shock. However, these data were pooled from only 2 RCTs,[17, 28] and each of them used hydrocortisone, whose direct mineralocorticoid effect is an obvious confounder. Similarly, according to data pooled from 2 RCTs, steroid use was associated with fewer cases of persistent chest x‐ray abnormalities by day 8. Of note, although calculation of the I² statistic was not possible because of too few studies, visual inspection of the forest plots suggested low levels of heterogeneity.

It is plausible that the impact of adjunctive steroids in CAP may vary based on the causative pathogen. This pathogen‐specific association has been observed in patients with bacterial meningitis, where most of the benefit is seemingly limited to pneumococcal meningitis.[9, 10] Unfortunately, as demonstrated by Snijders et al.,[16] establishing microbiologic etiology in CAP can be difficult, and most patients are treated empirically.

Our analysis showed no difference in duration of ventilation among patients who required ventilatory support on admission. However, only 2 studies reported this outcome.[17, 28] Second, in Confalonieri et al.,[28] the steroid group had a more severe baseline inflammatory response as illustrated by higher serum CRP levels (P = 0.04). Moreover, while mechanical ventilation was defined as either invasive or noninvasive ventilation, the steroid group had a higher number of patients who required noninvasive ventilation (P = 0.03), thus introducing selection bias. This study had additional areas of concern too, including a mortality of 0 among its 46 ICU patients, in contrast to established mortality rates of up to around 20%.[4] Unlike this study, Sabry and Omar[17] reported that none of their patients was on noninvasive ventilation. It may be pertinent to compare our findings with those of Steinberg et al.,[40] who studied patients with ARDS (pneumonia being the most common cause) who received methylprednisolone. This group had an early increase in ventilator‐free days, but that effect became less pronounced (though still significant) when the study end point was prolonged from 30 to 90 days.[41]

The 2 studies that were published before 2000 (McHardy and Schonell,[31] and Marik et al.[30]) were excluded in our a priori sensitivity analysis. A number of considerations led to this decision. First, standards of care for inpatient management of pneumoniaincluding pharmacologic therapies and ventilation strategieshave changed considerably over time. For instance, newer generation macrolides became available for clinical use in the early 1990s and meropenem in 1996.[41] Therefore, it would be hard to assume constancy of effect from that time period. Furthermore, the study by McHardy and Schonell[31] suffered from significant differences in the baseline characteristics of its 2 arms. There was incomplete randomization; patients with diabetes were excluded from only the steroid arm. Another issue with Marik et al.[30] was the considerably younger age of participants compared to other studies (Table 1).

Containing the rise of healthcare costs has taken on a new sense of urgency in the wake of the recent economic recession and continued growth in the cost of healthcare. Accordingly, many stakeholders seek solutions to improve value (reducing costs while improving care)[1]; hospital readmissions, which are common and costly,[2] have emerged as a key target. The Centers for Medicare and Medicaid Services (CMS) have instituted several programs intended to reduce readmissions, including funding for community‐based, care‐transition programs; penalties for hospitals with elevated risk‐adjusted readmission rates for selected diagnoses; pioneer Accountable Care Organizations (ACOs) with incentives to reduce global costs of care; and Hospital Engagement Networks (HENs) through the Partnership for Patients.[3] A primary aim of these initiatives is to enhance the quality of care transitions as patients are discharged from the hospital.

Though the recent focus on hospital readmissions has appropriately drawn attention to transitions in care, some have expressed concerns. Among these are questions about: 1) the extent to which readmissions truly reflect the quality of hospital care[4]; 2) the preventability of readmissions[5]; 3) limitations in risk‐adjustment techniques[6]; and 4) best practices for preventing readmissions.[7] We believe these concerns stem in part from deficiencies in the state of the science of transitional care, and that future efforts in this area will be hindered without a clear vision of an ideal transition in care. We propose the key components of an ideal transition in care and discuss the implications of this concept as it pertains to hospital readmissions.

THE IDEAL TRANSITION IN CARE

We propose the key components of an ideal transition in care in Figure 1 and Table 1. Figure 1 represents 10 domains described more fully below as structural supports of the bridge patients must cross from one care environment to another during a care transition. This figure highlights key domains and suggests that lack of a domain makes the bridge weaker and more prone to gaps in care and poor outcomes. It also implies that the more components are missing, the less safe is the bridge or transition. Those domains that mainly take place prior to discharge are placed closer to the hospital side of the bridge, those that mainly take place after discharge are placed closer to the community side of the bridge, while those that take place both prior to and after discharge are in the middle. Table 1 provides descriptions of the key content for each of these domains, as well as guidance about which personnel might be involved and where in the transition process that domain should be implemented. We support these domains with supporting evidence where available.

Figure 1

Our concept of an ideal transition in care began with work by Naylor, who described several important components of a safe transition in care, including complete communication of information, patient education, enlisting the help of social and community supports, ensuring continuity of care, and coordinating care among team members.[8] It is supplemented by the Transitions of Care Consensus Policy Statement proposed by representatives from hospital medicine, primary care, and emergency medicine, which emphasized aspects of timeliness and content of communication between providers.[9] Our present articulation of these key components includes 10 organizing domains.

The Discharge Planning domain highlights the important principle of planning ahead for hospital discharge while the patient is still being treated in the hospital, a paradigm espoused by Project RED[10] and other successful care transitions interventions.[11, 12] Collaborating with the outpatient provider and taking the patient and caregiver's preferences for appointment scheduling into account can help ensure optimal outpatient follow‐up.

Complete Communication of Information refers to the content that should be included in discharge summaries and other means of information transfer from hospital to postdischarge care. The specific content areas are based on the Society of Hospital Medicine and Society of General Internal Medicine Continuity of Care Task Force systematic review and recommendations,[13] which takes into account information requested by primary care physicians after discharge.

Availability, Timeliness, Clarity, and Organization of that information is as important as the content because postdischarge providers must be able to access and quickly understand the information they have been provided before assuming care of the patient.[14, 15]

The Medication Safety domain is of central importance because medications are responsible for most postdischarge adverse events.[16] Taking an accurate medication history,[17] reconciling changes throughout the hospitalization,[18] and communicating the reconciled medication regimen to patients and providers across transitions of care can reduce medication errors and improve patient safety.[19, 20, 21, 22]

The Patient Education and Promotion of Self‐Management domain involves teaching patients and their caregivers about the main hospital diagnoses and instructions for self‐care, including medication changes, appointments, and whom to contact if issues arise. Confirming comprehension of instructions through assessments of acute (delirium) and chronic (dementia) cognitive impairments[23, 24, 25, 26] and teach‐back from the patient (or caregiver) is an important aspect of such counseling, as is providing patients and caregivers with educational materials that are appropriate for their level of health literacy and preferred language.[14] High‐risk patients may benefit from patient coaching to improve their self‐management skills.[12] These recommendations are based on years of health literacy research,[27, 28, 29] and such elements are generally included in effective interventions (including Project RED,[10] Naylor and colleagues' Transitional Care Model,[11] and Coleman and colleagues' Care Transitions Intervention[12]).

Enlisting the help of Social and Community Supports is an important adjunct to medical care and is the rationale for the recent increase in CMS funding for community‐based, care‐transition programs. These programs are crucial for assisting patients with household activities, meals, and other necessities during the period of recovery, though they should be distinguished from care management or care coordination interventions, which have not been found to be helpful in preventing readmissions unless high touch in nature.[30, 31]

The Advanced Care Planning domain may begin in the hospital or outpatient setting, and involves establishing goals of care and healthcare proxies, as well as engaging with palliative care or hospice services if appropriate. Emerging evidence supports the intuitive conclusion that this approach prevents readmissions, particularly in patients who do not benefit from hospital readmission.[32, 33]

Attention to the Coordinating Care Among Team Members domain is needed to synchronize efforts across settings and providers. Clearly, many healthcare professionals as well as other involved parties can be involved in helping a single patient during transitions in care. It is vital that they coordinate information, assessments, and plans as a team.[34]

We recognize the domain of Monitoring and Managing Symptoms After Discharge as increasingly crucial as reflected in our growing understanding of the reasons for readmission, especially among patients with fragile conditions such as heart failure, chronic lung disease, gastrointestinal disorders, dementia,[23, 24, 25, 26] and vascular disease.[35] Monitoring for new or worsening symptoms; medication side effects, discrepancies, or nonadherence; and other self‐management challenges will allow problems to be detected and addressed early, before they result in unplanned healthcare utilization. It is noteworthy that successful interventions in this regard rely on in‐home evaluation[13, 14, 29] by nurses rather than telemonitoring, which in isolation has not been effective to date.[36, 37]

Finally, optimal Outpatient Follow‐Up with appropriate postdischarge providers is crucial for providing ideal transitions. These appointments need to be prompt[38, 39] (eg, within 7 days if not sooner for high‐risk patients) and with providers who have a longitudinal relationship to the patient, as prior work has shown increased readmissions when the provider is unfamiliar with the patient.[40] The advantages and disadvantages of hospitalist‐run postdischarge clinics as one way to increase access and expedite follow‐up are currently being explored. Although the optimal content of a postdischarge visit has not been defined, logical tasks to be completed are myriad and imply the need for checklists, adequate time, and a multidisciplinary team of providers.[41]

IMPLICATIONS OF THE IDEAL TRANSITION IN CARE

Our conceptualization of an ideal transition in care provides insight for hospital and healthcare system leadership, policymakers, researchers, clinicians, and educators seeking to improve transitions of care and reduce hospital readmissions. In the sections below, we briefly review commonly cited concerns about the recent focus on readmissions as a quality measure, illustrate how the Ideal Transition in Care addresses these concerns, and propose fruitful areas for future work.

NEXT STEPS

For hospital and healthcare system leaders, who need to take action now to avoid financial penalties, we recommend starting with proven, high‐touch interventions such as Project RED and the Care Transitions Intervention, which are durable, cost‐effective, robustly address multiple domains of the Ideal Transition in Care, and have been implemented at numerous sites.[54, 55] Each hospital or group will need to decide on a bundle of interventions and customize them based on local workflow, resources, and culture.

Risk‐stratification, to match the intensity of the intervention to the risk of readmission of the patient, will undoubtedly be a key component for the efficient use of resources. We anticipate future research will allow risk stratification to be a robust part of any implementation plan. However, as noted above, current risk prediction models are imperfect,[6] and more work is needed to determine which patients benefit most from which interventions. Few if any studies have described interventions tailored to risk for this reason.

Based on our ideal transition in care, our collective experience, and published evidence,[7, 10, 11, 12] potential elements to start with include: early discharge planning; medication reconciliation[56]; patient/caregiver education using health literacy principles, cognitive assessments, and teach‐back to confirm understanding; synchronous communication (eg, by phone) between inpatient and postdischarge providers; follow‐up phone calls to patients within 72 hours of discharge; 24/7 availability of a responsible inpatient provider to address questions and problems (both from the patient/caregiver and from postdischarge providers); and prompt appointments for patients discharged home. High‐risk patients will likely require additional interventions, including in‐home assessments, disease‐monitoring programs, and/or patient coaching. Lastly, patients with certain conditions prone to readmission (such as heart failure and chronic obstructive pulmonary disease) may benefit from disease‐specific programs, including patient education, outpatient disease management, and monitoring protocols.

It is likely that the most effective interventions are those that come from combined, coordinated interventions shared between inpatient and outpatient settings, and are intensive in nature. We expect that the more domains in the framework that are addressed, the safer and more seamless transitions in care will be, with improvement in patient outcomes. To the extent that fragmentation of care has been a barrier to the implementation of these types of interventions in the past, ACOs, perhaps with imbedded Patient‐Centered Medical Homes, may be in the best position to take advantage of newly aligned financial incentives to design comprehensive transitional care. Indeed, we anticipate that Figure 1 may provide substrate for a discussion of postdischarge care and division of responsibilities between inpatient and outpatient care teams at the time of transition, so effort is not duplicated and multiple domains are addressed.

Other barriers to implementation of ideal transitions in care will continue to be an issue for most healthcare systems. Financial constraints that have been a barrier up until now will be partially overcome by penalties for high readmission rates and by ACOs, bundled payments, and alternative care contracts (ie, global payments), but the extent to which each institution feels rewarded for investing in transitional interventions will vary greatly. Healthcare leadership that sees the value of improving transitions in care will be critical to overcoming this barrier. Competing demands (such as lowering hospital length of stay and carrying out other patient care responsibilities),[57] lack of coordination and diffusion of responsibility among various clinical personnel, and lack of standards are other barriers[58] that will require clear prioritization from leadership, policy changes, team‐based care, provider education and feedback, and adequate allocation of personnel resources. In short, process redesign using continuous quality improvement efforts and effective tools will be required to maximize the possibility of success.

CONCLUSIONS

Readmissions are costly and undesirable. Intuition suggests they are a marker of poor care and that hospitals should be capable of reducing them, thereby improving care and decreasing costs. In a potential future world of ACOs based on global payments, financial incentives would be aligned for each system to reduce readmissions below their current baseline, therefore obviating the need for external financial rewards and penalties. In the meantime, financial penalties do exist, and controversy exists over their fairness and likelihood of driving appropriate behavior. To address these controversies and promote better transitional care, we call for the development and use of multifaceted, collaborative transitions interventions that span settings, risk‐adjustment models that allow for fairer comparisons among hospitals, better and more widespread measurement of processes of transitional care, a better understanding of what interventions are most effective and in whom, and better guidance in how to implement these interventions. Our conceptualization of an ideal transition of care serves as a guide and provides a common vocabulary for these efforts. Such research is likely to produce the knowledge needed for healthcare systems to improve transitions in care, reduce readmissions, and reduce costs.

Containing the rise of healthcare costs has taken on a new sense of urgency in the wake of the recent economic recession and continued growth in the cost of healthcare. Accordingly, many stakeholders seek solutions to improve value (reducing costs while improving care)[1]; hospital readmissions, which are common and costly,[2] have emerged as a key target. The Centers for Medicare and Medicaid Services (CMS) have instituted several programs intended to reduce readmissions, including funding for community‐based, care‐transition programs; penalties for hospitals with elevated risk‐adjusted readmission rates for selected diagnoses; pioneer Accountable Care Organizations (ACOs) with incentives to reduce global costs of care; and Hospital Engagement Networks (HENs) through the Partnership for Patients.[3] A primary aim of these initiatives is to enhance the quality of care transitions as patients are discharged from the hospital.

Though the recent focus on hospital readmissions has appropriately drawn attention to transitions in care, some have expressed concerns. Among these are questions about: 1) the extent to which readmissions truly reflect the quality of hospital care[4]; 2) the preventability of readmissions[5]; 3) limitations in risk‐adjustment techniques[6]; and 4) best practices for preventing readmissions.[7] We believe these concerns stem in part from deficiencies in the state of the science of transitional care, and that future efforts in this area will be hindered without a clear vision of an ideal transition in care. We propose the key components of an ideal transition in care and discuss the implications of this concept as it pertains to hospital readmissions.

THE IDEAL TRANSITION IN CARE

We propose the key components of an ideal transition in care in Figure 1 and Table 1. Figure 1 represents 10 domains described more fully below as structural supports of the bridge patients must cross from one care environment to another during a care transition. This figure highlights key domains and suggests that lack of a domain makes the bridge weaker and more prone to gaps in care and poor outcomes. It also implies that the more components are missing, the less safe is the bridge or transition. Those domains that mainly take place prior to discharge are placed closer to the hospital side of the bridge, those that mainly take place after discharge are placed closer to the community side of the bridge, while those that take place both prior to and after discharge are in the middle. Table 1 provides descriptions of the key content for each of these domains, as well as guidance about which personnel might be involved and where in the transition process that domain should be implemented. We support these domains with supporting evidence where available.

Figure 1

Our concept of an ideal transition in care began with work by Naylor, who described several important components of a safe transition in care, including complete communication of information, patient education, enlisting the help of social and community supports, ensuring continuity of care, and coordinating care among team members.[8] It is supplemented by the Transitions of Care Consensus Policy Statement proposed by representatives from hospital medicine, primary care, and emergency medicine, which emphasized aspects of timeliness and content of communication between providers.[9] Our present articulation of these key components includes 10 organizing domains.

The Discharge Planning domain highlights the important principle of planning ahead for hospital discharge while the patient is still being treated in the hospital, a paradigm espoused by Project RED[10] and other successful care transitions interventions.[11, 12] Collaborating with the outpatient provider and taking the patient and caregiver's preferences for appointment scheduling into account can help ensure optimal outpatient follow‐up.

Complete Communication of Information refers to the content that should be included in discharge summaries and other means of information transfer from hospital to postdischarge care. The specific content areas are based on the Society of Hospital Medicine and Society of General Internal Medicine Continuity of Care Task Force systematic review and recommendations,[13] which takes into account information requested by primary care physicians after discharge.

Availability, Timeliness, Clarity, and Organization of that information is as important as the content because postdischarge providers must be able to access and quickly understand the information they have been provided before assuming care of the patient.[14, 15]

The Medication Safety domain is of central importance because medications are responsible for most postdischarge adverse events.[16] Taking an accurate medication history,[17] reconciling changes throughout the hospitalization,[18] and communicating the reconciled medication regimen to patients and providers across transitions of care can reduce medication errors and improve patient safety.[19, 20, 21, 22]

The Patient Education and Promotion of Self‐Management domain involves teaching patients and their caregivers about the main hospital diagnoses and instructions for self‐care, including medication changes, appointments, and whom to contact if issues arise. Confirming comprehension of instructions through assessments of acute (delirium) and chronic (dementia) cognitive impairments[23, 24, 25, 26] and teach‐back from the patient (or caregiver) is an important aspect of such counseling, as is providing patients and caregivers with educational materials that are appropriate for their level of health literacy and preferred language.[14] High‐risk patients may benefit from patient coaching to improve their self‐management skills.[12] These recommendations are based on years of health literacy research,[27, 28, 29] and such elements are generally included in effective interventions (including Project RED,[10] Naylor and colleagues' Transitional Care Model,[11] and Coleman and colleagues' Care Transitions Intervention[12]).

Enlisting the help of Social and Community Supports is an important adjunct to medical care and is the rationale for the recent increase in CMS funding for community‐based, care‐transition programs. These programs are crucial for assisting patients with household activities, meals, and other necessities during the period of recovery, though they should be distinguished from care management or care coordination interventions, which have not been found to be helpful in preventing readmissions unless high touch in nature.[30, 31]

The Advanced Care Planning domain may begin in the hospital or outpatient setting, and involves establishing goals of care and healthcare proxies, as well as engaging with palliative care or hospice services if appropriate. Emerging evidence supports the intuitive conclusion that this approach prevents readmissions, particularly in patients who do not benefit from hospital readmission.[32, 33]

Attention to the Coordinating Care Among Team Members domain is needed to synchronize efforts across settings and providers. Clearly, many healthcare professionals as well as other involved parties can be involved in helping a single patient during transitions in care. It is vital that they coordinate information, assessments, and plans as a team.[34]

We recognize the domain of Monitoring and Managing Symptoms After Discharge as increasingly crucial as reflected in our growing understanding of the reasons for readmission, especially among patients with fragile conditions such as heart failure, chronic lung disease, gastrointestinal disorders, dementia,[23, 24, 25, 26] and vascular disease.[35] Monitoring for new or worsening symptoms; medication side effects, discrepancies, or nonadherence; and other self‐management challenges will allow problems to be detected and addressed early, before they result in unplanned healthcare utilization. It is noteworthy that successful interventions in this regard rely on in‐home evaluation[13, 14, 29] by nurses rather than telemonitoring, which in isolation has not been effective to date.[36, 37]

Finally, optimal Outpatient Follow‐Up with appropriate postdischarge providers is crucial for providing ideal transitions. These appointments need to be prompt[38, 39] (eg, within 7 days if not sooner for high‐risk patients) and with providers who have a longitudinal relationship to the patient, as prior work has shown increased readmissions when the provider is unfamiliar with the patient.[40] The advantages and disadvantages of hospitalist‐run postdischarge clinics as one way to increase access and expedite follow‐up are currently being explored. Although the optimal content of a postdischarge visit has not been defined, logical tasks to be completed are myriad and imply the need for checklists, adequate time, and a multidisciplinary team of providers.[41]

IMPLICATIONS OF THE IDEAL TRANSITION IN CARE

Our conceptualization of an ideal transition in care provides insight for hospital and healthcare system leadership, policymakers, researchers, clinicians, and educators seeking to improve transitions of care and reduce hospital readmissions. In the sections below, we briefly review commonly cited concerns about the recent focus on readmissions as a quality measure, illustrate how the Ideal Transition in Care addresses these concerns, and propose fruitful areas for future work.

NEXT STEPS

For hospital and healthcare system leaders, who need to take action now to avoid financial penalties, we recommend starting with proven, high‐touch interventions such as Project RED and the Care Transitions Intervention, which are durable, cost‐effective, robustly address multiple domains of the Ideal Transition in Care, and have been implemented at numerous sites.[54, 55] Each hospital or group will need to decide on a bundle of interventions and customize them based on local workflow, resources, and culture.

Risk‐stratification, to match the intensity of the intervention to the risk of readmission of the patient, will undoubtedly be a key component for the efficient use of resources. We anticipate future research will allow risk stratification to be a robust part of any implementation plan. However, as noted above, current risk prediction models are imperfect,[6] and more work is needed to determine which patients benefit most from which interventions. Few if any studies have described interventions tailored to risk for this reason.

Based on our ideal transition in care, our collective experience, and published evidence,[7, 10, 11, 12] potential elements to start with include: early discharge planning; medication reconciliation[56]; patient/caregiver education using health literacy principles, cognitive assessments, and teach‐back to confirm understanding; synchronous communication (eg, by phone) between inpatient and postdischarge providers; follow‐up phone calls to patients within 72 hours of discharge; 24/7 availability of a responsible inpatient provider to address questions and problems (both from the patient/caregiver and from postdischarge providers); and prompt appointments for patients discharged home. High‐risk patients will likely require additional interventions, including in‐home assessments, disease‐monitoring programs, and/or patient coaching. Lastly, patients with certain conditions prone to readmission (such as heart failure and chronic obstructive pulmonary disease) may benefit from disease‐specific programs, including patient education, outpatient disease management, and monitoring protocols.

It is likely that the most effective interventions are those that come from combined, coordinated interventions shared between inpatient and outpatient settings, and are intensive in nature. We expect that the more domains in the framework that are addressed, the safer and more seamless transitions in care will be, with improvement in patient outcomes. To the extent that fragmentation of care has been a barrier to the implementation of these types of interventions in the past, ACOs, perhaps with imbedded Patient‐Centered Medical Homes, may be in the best position to take advantage of newly aligned financial incentives to design comprehensive transitional care. Indeed, we anticipate that Figure 1 may provide substrate for a discussion of postdischarge care and division of responsibilities between inpatient and outpatient care teams at the time of transition, so effort is not duplicated and multiple domains are addressed.

Other barriers to implementation of ideal transitions in care will continue to be an issue for most healthcare systems. Financial constraints that have been a barrier up until now will be partially overcome by penalties for high readmission rates and by ACOs, bundled payments, and alternative care contracts (ie, global payments), but the extent to which each institution feels rewarded for investing in transitional interventions will vary greatly. Healthcare leadership that sees the value of improving transitions in care will be critical to overcoming this barrier. Competing demands (such as lowering hospital length of stay and carrying out other patient care responsibilities),[57] lack of coordination and diffusion of responsibility among various clinical personnel, and lack of standards are other barriers[58] that will require clear prioritization from leadership, policy changes, team‐based care, provider education and feedback, and adequate allocation of personnel resources. In short, process redesign using continuous quality improvement efforts and effective tools will be required to maximize the possibility of success.

CONCLUSIONS

Readmissions are costly and undesirable. Intuition suggests they are a marker of poor care and that hospitals should be capable of reducing them, thereby improving care and decreasing costs. In a potential future world of ACOs based on global payments, financial incentives would be aligned for each system to reduce readmissions below their current baseline, therefore obviating the need for external financial rewards and penalties. In the meantime, financial penalties do exist, and controversy exists over their fairness and likelihood of driving appropriate behavior. To address these controversies and promote better transitional care, we call for the development and use of multifaceted, collaborative transitions interventions that span settings, risk‐adjustment models that allow for fairer comparisons among hospitals, better and more widespread measurement of processes of transitional care, a better understanding of what interventions are most effective and in whom, and better guidance in how to implement these interventions. Our conceptualization of an ideal transition of care serves as a guide and provides a common vocabulary for these efforts. Such research is likely to produce the knowledge needed for healthcare systems to improve transitions in care, reduce readmissions, and reduce costs.

Containing the rise of healthcare costs has taken on a new sense of urgency in the wake of the recent economic recession and continued growth in the cost of healthcare. Accordingly, many stakeholders seek solutions to improve value (reducing costs while improving care)[1]; hospital readmissions, which are common and costly,[2] have emerged as a key target. The Centers for Medicare and Medicaid Services (CMS) have instituted several programs intended to reduce readmissions, including funding for community‐based, care‐transition programs; penalties for hospitals with elevated risk‐adjusted readmission rates for selected diagnoses; pioneer Accountable Care Organizations (ACOs) with incentives to reduce global costs of care; and Hospital Engagement Networks (HENs) through the Partnership for Patients.[3] A primary aim of these initiatives is to enhance the quality of care transitions as patients are discharged from the hospital.

Though the recent focus on hospital readmissions has appropriately drawn attention to transitions in care, some have expressed concerns. Among these are questions about: 1) the extent to which readmissions truly reflect the quality of hospital care[4]; 2) the preventability of readmissions[5]; 3) limitations in risk‐adjustment techniques[6]; and 4) best practices for preventing readmissions.[7] We believe these concerns stem in part from deficiencies in the state of the science of transitional care, and that future efforts in this area will be hindered without a clear vision of an ideal transition in care. We propose the key components of an ideal transition in care and discuss the implications of this concept as it pertains to hospital readmissions.

THE IDEAL TRANSITION IN CARE

We propose the key components of an ideal transition in care in Figure 1 and Table 1. Figure 1 represents 10 domains described more fully below as structural supports of the bridge patients must cross from one care environment to another during a care transition. This figure highlights key domains and suggests that lack of a domain makes the bridge weaker and more prone to gaps in care and poor outcomes. It also implies that the more components are missing, the less safe is the bridge or transition. Those domains that mainly take place prior to discharge are placed closer to the hospital side of the bridge, those that mainly take place after discharge are placed closer to the community side of the bridge, while those that take place both prior to and after discharge are in the middle. Table 1 provides descriptions of the key content for each of these domains, as well as guidance about which personnel might be involved and where in the transition process that domain should be implemented. We support these domains with supporting evidence where available.

Figure 1