Hospital‐acquired venous thromboembolic events (VTEs) in medically ill patients account for a significant percentage of in‐hospital mortality.1 There have been reports that 50%75% of symptomatic VTEs related to hospitalization occur in medical patients.2, 3 The incidence of hospital‐acquired pulmonary embolism (PE) and deep vein thrombosis (DVT) has been reported nationally as 0.4% and 1.3% of all hospital admissions, respectively.4 The incidence of hospital‐acquired VTEs in patients not on prophylaxis reported in the MEDENOX trial was approximately 15%, although these were predominantly asymptomatic cases.5

Several studies including the MEDENOX and PREVENT trials have supported the use of low‐molecular‐weight heparin (LMWH) and unfractionated heparin (UFH) in the prevention of VTEs.57 Based on this evidence, the American College of Chest Physicians (ACCP) developed guidelines for the use of LMWH and UFH in the prevention of VTEs in patients with acute medical illnesses.8 Despite the promulgation of these guidelines, several studies have indicated that the use of these medications remains suboptimal. Two recent studies showed that the rate of thromboprophylaxis in hospitalized medically ill patients at risk for VTEs was 15%16%.9, 10

A review by Kakkar et al. stated that the underutilization of thromboprophylaxis may be a result of lack of awareness by physicians, disagreement with the guidelines, and lack of outcome data.11 Studies have demonstrated improvement in the use of thromboprophylaxis in hospitalized patients using several strategies. One of these strategies was the use of a hospitalwide clinical pharmacy education program, resulting in the thromboprophylaxis rate improving from 11% to 44% (P < .001).12 Another strategy studied was the use of a combination of physician education, a decision support tool, and regular audit and feedback, which resulted in the rate of thromboprophylaxis increasing from 43% to 85% after 18 months.13 The use of computer alert programs was also studied and was shown to increase the use of thromboprophylaxis among hospitalized patients from 13% to 23.6% (P < 0.001).14

The objective of this article is to report the impact of a continuous quality improvement (CQI) project on adherence to the DVT prophylaxis guidelines as well as the subsequent incidence of hospital‐acquired DVT in medical patients at a large urban teaching hospital between 2002 and 2005.

METHODS

In November 2002, the Kings County Hospital Center Department of Medicine embarked on a CQI project to increase adherence to thromboprophylaxis guidelines. A 3‐tiered approach of provider education, provider reminders with decision support, and audit and feedback was taken. This cycle was repeated each month with the start of a new group of house staff and faculty attendings. This 3‐tiered approach was developed, implemented, and maintained over a 4‐year period from 2002 to 2005. The measured outcomes were rate of DVT prophylaxis and incidence of hospital‐acquired DVT.

Provider Education and Reminders with Decision Support

The first approach was the inclusion of DVT prophylaxis in the Assessment and Plan section of the preprinted admission database. This section on DVT prophylaxis, which required a physician to indicate if a patient required prophylaxis and the type of prophylaxis chosen, was initiated in November 2002 (Fig. 1, arrow A).

Figure 1

In December 2002, DVT prophylaxis was included in the learning goals and objectives handout given to house staff at the start of each inpatient rotation (Fig. 1, arrow B).

Pocket DVT prophylaxis guideline cards that outlined the indications and suggested regimens for DVT prophylaxis in the medically ill patient were issued to members of the house staff and their supervising faculty attendings at the start of each month beginning in March 2003 (Fig. 1, arrow C).

Preprinted admission orders were developed in July 2004 (Fig. 1, arrow F) that included DVT prophylaxis options. The admitting physician was required to indicate the need for DVT prophylaxis and type of prophylaxis chosen for every patient admitted.

In February 2005, a standardized DVT algorithm was included in the preprinted admission database (Figs. 13). The admitting physician was required to follow this algorithm to assess for indications for DVT prophylaxis and, if needed, type of prophylaxis chosen. The signatures of the house officer who completed the form and the supervising attending were required.

Figure 2

Figure 3

RESULTS

Table 1 provides cumulative yearly data for DVT prophylaxis rate, number of hospital‐acquired DVTs, and number of discharges from the general medicine house‐staff service for the 4 years of the observational period, 20022005. The baseline DVT prophylaxis rate from May 2002 to December 2002 was 63% (Fig. 4 and Table 1) and increased to 73%, 90%, and 96% over the succeeding years. In 2002, the number of hospital‐acquired DVTs on the general medicine house‐staff service was 14, followed by 16, 7, and 1 for 2003 through 2005 (Table 1). Twenty‐four of these 38 cases had not received DVT prophylaxis.

Figure 4

When adjusted for each 1000 discharges (Fig. 4 and Table 1), the rates of hospital‐acquired DVT significantly decreased, from 2.6 per 1000 discharges (95% CI 1.54.4) in 2002 to 1.1 per 1000 discharges (95% CI 0.52.3, P = .058) in 2004 and to 0.2 per 1000 discharges (95% CI 0.01.1, P = .007) in 2005. During these years, particularly 20042005, the monthly DVT prophylaxis rate in the sample reviewed was consistently 90% or better (Fig. 1).

DISCUSSION

Our study involved active multifaceted interventions with a layered combination of provider education, provider reminders with decision support, and audit and feedback. This layered approach increased the DVT prophylaxis rate in our department, resulting in a significant decline in clinically evident hospital‐acquired DVTs from a baseline rate of 2.6 per 1000 discharges (0.26%) in 2002 to a rate of 0.2 per 1000 discharges (0.02%) in 2005 (95% CI 0.01.1, P = .007). Our baseline rate was low compared to the nationally reported incidence of 1.3%.4 Had the study not been extended over 4 years, it is likely that the statistical significance of this decline would have been missed. The rate of decrease in hospital0acquired DVTs accelerated with each year, showing no decline between 2002 and 2003, a 58% decline from 2003 to 2004, and an 82% decline from 2004 to 2005.

The acceleration in this decline coincided with the use of pocket DVT prophylaxis guideline cards and monthly audits with feedback starting in 2003 and the implementation of preprinted admission orders in 2004. This acceleration peaked in 2005 with the addition of the DVT algorithm to the admission database.

Despite the ACCP guidelines on thromboprophylaxis,8 several studies have suggested that only approximately 15%16% of medically ill patients at increased risk of VTE receive adequate thromboprophylaxis.9, 10 The 3 barriers identified by Kakkar et al. were lack of physician awareness, disagreement with the guidelines, and lack of outcome data.11 Our interventions addressed these barriers. Lack of physician awareness was addressed by provider education and reminder systems with decision support. Audit with team‐specific feedback provided the outcome data needed to demonstrate and reinforce effective change.

A critical analysis by Shojania et al. analyzed the effectiveness of each of these strategies when implemented alone. Provider education was effective in increasing provider knowledge but was generally ineffective when judged on the basis of improving patient outcomes. When provider reminders were well integrated with work flow, they were more likely to be effective. The effectiveness of decision support was variable, as it sometimes brought about change but was less likely to do so in complex situations. Various forms of audit and feedback produced small to modest effective changes.15 A systematic review by Oxman et al. suggested that there were some benefits when these strategies were implemented in a layered manner compared with single‐faceted strategies and that effective interventions were more likely to involve active rather than passive strategies.16

The major strengths of this study were the large number of patients reviewed, the sustained interventions over a 4‐year period, and the consistency of our results. The Department of Risk Management systematically reviewed all discharge ICD‐9 codes and all relevant imaging studies of 32,293 patients for the presence of hospital‐acquired DVTs. The data provided was all‐inclusive regardless of outcome and was acquired continuously over a 4‐year period rather than at a single point.

Our department continues to seek methods to increase the rate of thromboprophylaxis, and so, noting that computer alert programs have been shown to increase the appropriate use of thromboprophylaxis,14 our institution has designed an admission computer order entry program specific for DVT prophylaxis. This program prompts physicians to perform a DVT risk assessment of each admitted patient and to order the appropriate thromboprophylaxis regimen if indicated. This program was implemented in May 2006.

CONCLUSIONS

The multifaceted layered combination of provider education, provider reminders with decision support, and audit and feedback implemented at our institution could be reproduced and utilized at other institutions, particularly teaching hospitals, to address the underutilization of DVT prophylaxis in medically ill patients and to bring about a decrease in hospital‐acquired DVTs.

Hospital‐acquired venous thromboembolic events (VTEs) in medically ill patients account for a significant percentage of in‐hospital mortality.1 There have been reports that 50%75% of symptomatic VTEs related to hospitalization occur in medical patients.2, 3 The incidence of hospital‐acquired pulmonary embolism (PE) and deep vein thrombosis (DVT) has been reported nationally as 0.4% and 1.3% of all hospital admissions, respectively.4 The incidence of hospital‐acquired VTEs in patients not on prophylaxis reported in the MEDENOX trial was approximately 15%, although these were predominantly asymptomatic cases.5

Several studies including the MEDENOX and PREVENT trials have supported the use of low‐molecular‐weight heparin (LMWH) and unfractionated heparin (UFH) in the prevention of VTEs.57 Based on this evidence, the American College of Chest Physicians (ACCP) developed guidelines for the use of LMWH and UFH in the prevention of VTEs in patients with acute medical illnesses.8 Despite the promulgation of these guidelines, several studies have indicated that the use of these medications remains suboptimal. Two recent studies showed that the rate of thromboprophylaxis in hospitalized medically ill patients at risk for VTEs was 15%16%.9, 10

A review by Kakkar et al. stated that the underutilization of thromboprophylaxis may be a result of lack of awareness by physicians, disagreement with the guidelines, and lack of outcome data.11 Studies have demonstrated improvement in the use of thromboprophylaxis in hospitalized patients using several strategies. One of these strategies was the use of a hospitalwide clinical pharmacy education program, resulting in the thromboprophylaxis rate improving from 11% to 44% (P < .001).12 Another strategy studied was the use of a combination of physician education, a decision support tool, and regular audit and feedback, which resulted in the rate of thromboprophylaxis increasing from 43% to 85% after 18 months.13 The use of computer alert programs was also studied and was shown to increase the use of thromboprophylaxis among hospitalized patients from 13% to 23.6% (P < 0.001).14

The objective of this article is to report the impact of a continuous quality improvement (CQI) project on adherence to the DVT prophylaxis guidelines as well as the subsequent incidence of hospital‐acquired DVT in medical patients at a large urban teaching hospital between 2002 and 2005.

METHODS

In November 2002, the Kings County Hospital Center Department of Medicine embarked on a CQI project to increase adherence to thromboprophylaxis guidelines. A 3‐tiered approach of provider education, provider reminders with decision support, and audit and feedback was taken. This cycle was repeated each month with the start of a new group of house staff and faculty attendings. This 3‐tiered approach was developed, implemented, and maintained over a 4‐year period from 2002 to 2005. The measured outcomes were rate of DVT prophylaxis and incidence of hospital‐acquired DVT.

Provider Education and Reminders with Decision Support

The first approach was the inclusion of DVT prophylaxis in the Assessment and Plan section of the preprinted admission database. This section on DVT prophylaxis, which required a physician to indicate if a patient required prophylaxis and the type of prophylaxis chosen, was initiated in November 2002 (Fig. 1, arrow A).

Figure 1

In December 2002, DVT prophylaxis was included in the learning goals and objectives handout given to house staff at the start of each inpatient rotation (Fig. 1, arrow B).

Pocket DVT prophylaxis guideline cards that outlined the indications and suggested regimens for DVT prophylaxis in the medically ill patient were issued to members of the house staff and their supervising faculty attendings at the start of each month beginning in March 2003 (Fig. 1, arrow C).

Preprinted admission orders were developed in July 2004 (Fig. 1, arrow F) that included DVT prophylaxis options. The admitting physician was required to indicate the need for DVT prophylaxis and type of prophylaxis chosen for every patient admitted.

In February 2005, a standardized DVT algorithm was included in the preprinted admission database (Figs. 13). The admitting physician was required to follow this algorithm to assess for indications for DVT prophylaxis and, if needed, type of prophylaxis chosen. The signatures of the house officer who completed the form and the supervising attending were required.

Figure 2

Figure 3

RESULTS

Table 1 provides cumulative yearly data for DVT prophylaxis rate, number of hospital‐acquired DVTs, and number of discharges from the general medicine house‐staff service for the 4 years of the observational period, 20022005. The baseline DVT prophylaxis rate from May 2002 to December 2002 was 63% (Fig. 4 and Table 1) and increased to 73%, 90%, and 96% over the succeeding years. In 2002, the number of hospital‐acquired DVTs on the general medicine house‐staff service was 14, followed by 16, 7, and 1 for 2003 through 2005 (Table 1). Twenty‐four of these 38 cases had not received DVT prophylaxis.

Figure 4

When adjusted for each 1000 discharges (Fig. 4 and Table 1), the rates of hospital‐acquired DVT significantly decreased, from 2.6 per 1000 discharges (95% CI 1.54.4) in 2002 to 1.1 per 1000 discharges (95% CI 0.52.3, P = .058) in 2004 and to 0.2 per 1000 discharges (95% CI 0.01.1, P = .007) in 2005. During these years, particularly 20042005, the monthly DVT prophylaxis rate in the sample reviewed was consistently 90% or better (Fig. 1).

DISCUSSION

Our study involved active multifaceted interventions with a layered combination of provider education, provider reminders with decision support, and audit and feedback. This layered approach increased the DVT prophylaxis rate in our department, resulting in a significant decline in clinically evident hospital‐acquired DVTs from a baseline rate of 2.6 per 1000 discharges (0.26%) in 2002 to a rate of 0.2 per 1000 discharges (0.02%) in 2005 (95% CI 0.01.1, P = .007). Our baseline rate was low compared to the nationally reported incidence of 1.3%.4 Had the study not been extended over 4 years, it is likely that the statistical significance of this decline would have been missed. The rate of decrease in hospital0acquired DVTs accelerated with each year, showing no decline between 2002 and 2003, a 58% decline from 2003 to 2004, and an 82% decline from 2004 to 2005.

The acceleration in this decline coincided with the use of pocket DVT prophylaxis guideline cards and monthly audits with feedback starting in 2003 and the implementation of preprinted admission orders in 2004. This acceleration peaked in 2005 with the addition of the DVT algorithm to the admission database.

Despite the ACCP guidelines on thromboprophylaxis,8 several studies have suggested that only approximately 15%16% of medically ill patients at increased risk of VTE receive adequate thromboprophylaxis.9, 10 The 3 barriers identified by Kakkar et al. were lack of physician awareness, disagreement with the guidelines, and lack of outcome data.11 Our interventions addressed these barriers. Lack of physician awareness was addressed by provider education and reminder systems with decision support. Audit with team‐specific feedback provided the outcome data needed to demonstrate and reinforce effective change.

A critical analysis by Shojania et al. analyzed the effectiveness of each of these strategies when implemented alone. Provider education was effective in increasing provider knowledge but was generally ineffective when judged on the basis of improving patient outcomes. When provider reminders were well integrated with work flow, they were more likely to be effective. The effectiveness of decision support was variable, as it sometimes brought about change but was less likely to do so in complex situations. Various forms of audit and feedback produced small to modest effective changes.15 A systematic review by Oxman et al. suggested that there were some benefits when these strategies were implemented in a layered manner compared with single‐faceted strategies and that effective interventions were more likely to involve active rather than passive strategies.16

The major strengths of this study were the large number of patients reviewed, the sustained interventions over a 4‐year period, and the consistency of our results. The Department of Risk Management systematically reviewed all discharge ICD‐9 codes and all relevant imaging studies of 32,293 patients for the presence of hospital‐acquired DVTs. The data provided was all‐inclusive regardless of outcome and was acquired continuously over a 4‐year period rather than at a single point.

Our department continues to seek methods to increase the rate of thromboprophylaxis, and so, noting that computer alert programs have been shown to increase the appropriate use of thromboprophylaxis,14 our institution has designed an admission computer order entry program specific for DVT prophylaxis. This program prompts physicians to perform a DVT risk assessment of each admitted patient and to order the appropriate thromboprophylaxis regimen if indicated. This program was implemented in May 2006.

CONCLUSIONS

The multifaceted layered combination of provider education, provider reminders with decision support, and audit and feedback implemented at our institution could be reproduced and utilized at other institutions, particularly teaching hospitals, to address the underutilization of DVT prophylaxis in medically ill patients and to bring about a decrease in hospital‐acquired DVTs.

Hospital‐acquired venous thromboembolic events (VTEs) in medically ill patients account for a significant percentage of in‐hospital mortality.1 There have been reports that 50%75% of symptomatic VTEs related to hospitalization occur in medical patients.2, 3 The incidence of hospital‐acquired pulmonary embolism (PE) and deep vein thrombosis (DVT) has been reported nationally as 0.4% and 1.3% of all hospital admissions, respectively.4 The incidence of hospital‐acquired VTEs in patients not on prophylaxis reported in the MEDENOX trial was approximately 15%, although these were predominantly asymptomatic cases.5

Several studies including the MEDENOX and PREVENT trials have supported the use of low‐molecular‐weight heparin (LMWH) and unfractionated heparin (UFH) in the prevention of VTEs.57 Based on this evidence, the American College of Chest Physicians (ACCP) developed guidelines for the use of LMWH and UFH in the prevention of VTEs in patients with acute medical illnesses.8 Despite the promulgation of these guidelines, several studies have indicated that the use of these medications remains suboptimal. Two recent studies showed that the rate of thromboprophylaxis in hospitalized medically ill patients at risk for VTEs was 15%16%.9, 10

A review by Kakkar et al. stated that the underutilization of thromboprophylaxis may be a result of lack of awareness by physicians, disagreement with the guidelines, and lack of outcome data.11 Studies have demonstrated improvement in the use of thromboprophylaxis in hospitalized patients using several strategies. One of these strategies was the use of a hospitalwide clinical pharmacy education program, resulting in the thromboprophylaxis rate improving from 11% to 44% (P < .001).12 Another strategy studied was the use of a combination of physician education, a decision support tool, and regular audit and feedback, which resulted in the rate of thromboprophylaxis increasing from 43% to 85% after 18 months.13 The use of computer alert programs was also studied and was shown to increase the use of thromboprophylaxis among hospitalized patients from 13% to 23.6% (P < 0.001).14

The objective of this article is to report the impact of a continuous quality improvement (CQI) project on adherence to the DVT prophylaxis guidelines as well as the subsequent incidence of hospital‐acquired DVT in medical patients at a large urban teaching hospital between 2002 and 2005.

METHODS

In November 2002, the Kings County Hospital Center Department of Medicine embarked on a CQI project to increase adherence to thromboprophylaxis guidelines. A 3‐tiered approach of provider education, provider reminders with decision support, and audit and feedback was taken. This cycle was repeated each month with the start of a new group of house staff and faculty attendings. This 3‐tiered approach was developed, implemented, and maintained over a 4‐year period from 2002 to 2005. The measured outcomes were rate of DVT prophylaxis and incidence of hospital‐acquired DVT.

Provider Education and Reminders with Decision Support

The first approach was the inclusion of DVT prophylaxis in the Assessment and Plan section of the preprinted admission database. This section on DVT prophylaxis, which required a physician to indicate if a patient required prophylaxis and the type of prophylaxis chosen, was initiated in November 2002 (Fig. 1, arrow A).

Figure 1

In December 2002, DVT prophylaxis was included in the learning goals and objectives handout given to house staff at the start of each inpatient rotation (Fig. 1, arrow B).

Pocket DVT prophylaxis guideline cards that outlined the indications and suggested regimens for DVT prophylaxis in the medically ill patient were issued to members of the house staff and their supervising faculty attendings at the start of each month beginning in March 2003 (Fig. 1, arrow C).

Preprinted admission orders were developed in July 2004 (Fig. 1, arrow F) that included DVT prophylaxis options. The admitting physician was required to indicate the need for DVT prophylaxis and type of prophylaxis chosen for every patient admitted.

In February 2005, a standardized DVT algorithm was included in the preprinted admission database (Figs. 13). The admitting physician was required to follow this algorithm to assess for indications for DVT prophylaxis and, if needed, type of prophylaxis chosen. The signatures of the house officer who completed the form and the supervising attending were required.

Figure 2

Figure 3

RESULTS

Table 1 provides cumulative yearly data for DVT prophylaxis rate, number of hospital‐acquired DVTs, and number of discharges from the general medicine house‐staff service for the 4 years of the observational period, 20022005. The baseline DVT prophylaxis rate from May 2002 to December 2002 was 63% (Fig. 4 and Table 1) and increased to 73%, 90%, and 96% over the succeeding years. In 2002, the number of hospital‐acquired DVTs on the general medicine house‐staff service was 14, followed by 16, 7, and 1 for 2003 through 2005 (Table 1). Twenty‐four of these 38 cases had not received DVT prophylaxis.

Figure 4

When adjusted for each 1000 discharges (Fig. 4 and Table 1), the rates of hospital‐acquired DVT significantly decreased, from 2.6 per 1000 discharges (95% CI 1.54.4) in 2002 to 1.1 per 1000 discharges (95% CI 0.52.3, P = .058) in 2004 and to 0.2 per 1000 discharges (95% CI 0.01.1, P = .007) in 2005. During these years, particularly 20042005, the monthly DVT prophylaxis rate in the sample reviewed was consistently 90% or better (Fig. 1).

DISCUSSION

Our study involved active multifaceted interventions with a layered combination of provider education, provider reminders with decision support, and audit and feedback. This layered approach increased the DVT prophylaxis rate in our department, resulting in a significant decline in clinically evident hospital‐acquired DVTs from a baseline rate of 2.6 per 1000 discharges (0.26%) in 2002 to a rate of 0.2 per 1000 discharges (0.02%) in 2005 (95% CI 0.01.1, P = .007). Our baseline rate was low compared to the nationally reported incidence of 1.3%.4 Had the study not been extended over 4 years, it is likely that the statistical significance of this decline would have been missed. The rate of decrease in hospital0acquired DVTs accelerated with each year, showing no decline between 2002 and 2003, a 58% decline from 2003 to 2004, and an 82% decline from 2004 to 2005.

The acceleration in this decline coincided with the use of pocket DVT prophylaxis guideline cards and monthly audits with feedback starting in 2003 and the implementation of preprinted admission orders in 2004. This acceleration peaked in 2005 with the addition of the DVT algorithm to the admission database.

Despite the ACCP guidelines on thromboprophylaxis,8 several studies have suggested that only approximately 15%16% of medically ill patients at increased risk of VTE receive adequate thromboprophylaxis.9, 10 The 3 barriers identified by Kakkar et al. were lack of physician awareness, disagreement with the guidelines, and lack of outcome data.11 Our interventions addressed these barriers. Lack of physician awareness was addressed by provider education and reminder systems with decision support. Audit with team‐specific feedback provided the outcome data needed to demonstrate and reinforce effective change.

A critical analysis by Shojania et al. analyzed the effectiveness of each of these strategies when implemented alone. Provider education was effective in increasing provider knowledge but was generally ineffective when judged on the basis of improving patient outcomes. When provider reminders were well integrated with work flow, they were more likely to be effective. The effectiveness of decision support was variable, as it sometimes brought about change but was less likely to do so in complex situations. Various forms of audit and feedback produced small to modest effective changes.15 A systematic review by Oxman et al. suggested that there were some benefits when these strategies were implemented in a layered manner compared with single‐faceted strategies and that effective interventions were more likely to involve active rather than passive strategies.16

The major strengths of this study were the large number of patients reviewed, the sustained interventions over a 4‐year period, and the consistency of our results. The Department of Risk Management systematically reviewed all discharge ICD‐9 codes and all relevant imaging studies of 32,293 patients for the presence of hospital‐acquired DVTs. The data provided was all‐inclusive regardless of outcome and was acquired continuously over a 4‐year period rather than at a single point.

Our department continues to seek methods to increase the rate of thromboprophylaxis, and so, noting that computer alert programs have been shown to increase the appropriate use of thromboprophylaxis,14 our institution has designed an admission computer order entry program specific for DVT prophylaxis. This program prompts physicians to perform a DVT risk assessment of each admitted patient and to order the appropriate thromboprophylaxis regimen if indicated. This program was implemented in May 2006.

CONCLUSIONS

The multifaceted layered combination of provider education, provider reminders with decision support, and audit and feedback implemented at our institution could be reproduced and utilized at other institutions, particularly teaching hospitals, to address the underutilization of DVT prophylaxis in medically ill patients and to bring about a decrease in hospital‐acquired DVTs.

Clostridium difficileassociated disease (CDAD) is a well‐known complication of hospitalization and is the most frequently identified cause of nosocomial diarrhea that hospitalists encounter. Despite widespread epidemiologic attempts to control the disease, its prevalence and clinical severity appear to be increasing.1 The resulting social and economic consequences are profound. The estimated 3 million inpatient cases of CDAD a year result in an average increase in the length of stay of 3.6 days at a cost in inpatient health care of more than $1 billion.2

Early diagnosis of index cases is crucial. A diagnostic delay can result in a treatment delay for the index case, as well as in a delay in implementing isolation procedures to prevent horizontal transmission. Acquisition of CDAD is time dependent and occurs in 20% to 30% of hospitalized patients at a rate of approximately 8% per week.3, 4 This transmission is primarily a result of environmental contamination with CDAD spores, found on 59% of the hands of hospital personnel caring for infected patients, in 49% of rooms of symptomatic patients, and in 29% of rooms of asymptomatic carriers.5 Despite the need for early diagnosis, a study from the United Kingdom documented that the average time from the onset of diarrhea to sampling of CDAD patients is 4.7 days.6 An additional challenge for early diagnosis is the delay in microbiological confirmation of CDAD in a suspected patient. Cytotoxic assays, which have become the standard diagnostic technique for CDAD, exhibit excellent sensitivity and specificity but have a lengthy processing time, between 2 and 4 days. Although antigen detection assays can be rapidly performed, many have inadequate sensitivity and specificity.7

These issues of diagnostic and treatment delays are compounded in patients with recurrent CDAD. As many as 15%35% of patients with an initial CDAD infection will experience a recurrence, usually within 2 months. At least half these infections are a result of reinfection, not relapse.8 This implies that early detection and strict isolation of infected patients is essential for reducing the exposure of at‐risk patients to the disease. There is evidence that the burden of patients on the same ward simultaneously having CDAD increases a patient's risk of acquiring the disease.9 It is currently unknown if recurrent CDAD cases are diagnosed or treated earlier than initial cases. If not, this is a potentially important patient population for hospitalists to target for aggressive containment strategies. This study sought to determine the mean time to sampling and treatment in patients with recurrent CDAD infection compared with those in patients who are initially infected.

Design

The study cohort consisted of all adult patients more than 18 years old with CDAD (by ICD9 code) who had been hospitalized at Brigham and Women's Hospital between 1997 and 2004. Retrospectively, patients were identified through the Partners Healthcare Research Patient Data Repository (RPDR). The RPDR is a centralized clinical data registry that gathers data from various hospital legacy systems and was used to determine the patient demographics and first date of treatment (with vancomycin or metronidazole). Medical and microbiologic records were reviewed to determine the dates of cytotoxic assay submission and symptom onset. Symptoms were defined as diarrhea, abdominal pain/cramping, or radiological/colonoscopic evidence of colitis. Recurrence was defined as any repeat inpatient CDAD diagnosis within 2 months (regardless of admission diagnosis). Baseline characteristics in the recurrence and no‐recurrence populations were compared by the 2‐sided Student t test or the chi‐square test (for continuous and categorical variables, respectively). Mean time from symptom to sampling, from symptom to treatment, and from sampling to treatment were compared between initial and recurrent disease episodes by the 2‐sided Student t test. All P values < .05 were considered significant. Institutional review board approval was obtained by Partners Healthcare.

RESULTS

Between 1997 and 2004 there were 1309 patients with an ICD9 code for CDAD, 151 of whom (12%) had a recurrence. Of these, 125 had 1 recurrence, 23 had 2 recurrences, and 3 had 3 recurrences. There were no significant differences between the groups in basic demographics (Table 1). The mean time to sampling was not significantly different between initial and recurrent CDAD hospital episodes (Table 2). However, the mean time to treatment (from symptoms and sampling) was shorter in recurrent episodes (Table 2). From 1997 to 2004 there was no significant reduction in time to sampling, but there was a significant reduction in time to treatment, from 3.89 days (19972000) to 2.30 days (2001 2004), P = .0012.

DISCUSSION

Clostridium difficileassociated disease (CDAD) has become a significant nosocomial infection in medical institutions, and recurrent CDAD is emerging as a disease of concern for hospitalists. Diagnostic delays represent a major epidemiologic problem, resulting in both delay of treatment delay of the index case and delay in implementing isolation procedures to prevent horizontal transmission. In this study, patients with recurrent disease did not have stool collected any earlier than did patients with their initial episode of CDAD, and these diagnostic delays did not change in successive eras. Recurrent disease patients did receive treatment earlier than did patients with initial episodes. Although this empiric treatment strategy is encouraging and likely reflects heightened awareness of the disease over time, the 2.5‐day span from symptoms to treatment is still a clinically significant delay. Also of concern is the range of time from symptoms to treatment, as long as 19 days in the recurrent treatment group. Although most patients were treated within 12 days, this variability represents the burden of infectious patients with the potential for infecting others. Targeting recurrent CDAD populations for early diagnosis, treatment, and isolation would almost certainly reduce the morbidity associated with horizontal transmission rates.9

This study had several limitations. Our data found a lower incidence of recurrent CDAD than previously published in the literature. This can be accounted for by the identification of cases by ICD9 code, which previously has been documented to underestimate true disease.10, 11 We also were not able to capture recurrent episodes in outpatients or episodes that occurred after the 2004 cohort, which underestimated the true frequency of recurrence. At worst, this underestimation could bias the results toward the null hypothesis. An additional limitation of the study was the assumption that time to treatment was accurately reflected by time to prescription of either vancomycin or flagyl. Some patients may have been treated by suspending treatment with the offending antibiotic along with watchful waiting, which is a reasonable strategy for patients with mild disease and is endorsed by the American College of Gastroenterology and the Society for Healthcare Epidemiology of America.12, 13 This would overestimate time to treatment for those individuals and would make time to treatment appear longer, but would not affect time to sampling. In addition, the symptoms collected from chart review were assumed to be a result of the patient's CDAD, but there is a chance that these symptoms such as diarrhea, abdominal pain, and cramping may have been a result of a different diagnosis. These data were also limited to a cohort from a single institution and may not reflect the patient characteristics or practice patterns at other institutions.

In conclusion, CDAD is a major contributor to morbidity from nosocomial infections, and recurrent CDAD patients are a likely source of horizontal disease transmission. This study documented that there are significant diagnostic and treatment delays, even in populations with recurrent disease. It is especially important that hospitalists take measures to improve the early diagnosis, treatment, and isolation of these patients in order to improve these deficiencies in care.

Clostridium difficileassociated disease (CDAD) is a well‐known complication of hospitalization and is the most frequently identified cause of nosocomial diarrhea that hospitalists encounter. Despite widespread epidemiologic attempts to control the disease, its prevalence and clinical severity appear to be increasing.1 The resulting social and economic consequences are profound. The estimated 3 million inpatient cases of CDAD a year result in an average increase in the length of stay of 3.6 days at a cost in inpatient health care of more than $1 billion.2

Early diagnosis of index cases is crucial. A diagnostic delay can result in a treatment delay for the index case, as well as in a delay in implementing isolation procedures to prevent horizontal transmission. Acquisition of CDAD is time dependent and occurs in 20% to 30% of hospitalized patients at a rate of approximately 8% per week.3, 4 This transmission is primarily a result of environmental contamination with CDAD spores, found on 59% of the hands of hospital personnel caring for infected patients, in 49% of rooms of symptomatic patients, and in 29% of rooms of asymptomatic carriers.5 Despite the need for early diagnosis, a study from the United Kingdom documented that the average time from the onset of diarrhea to sampling of CDAD patients is 4.7 days.6 An additional challenge for early diagnosis is the delay in microbiological confirmation of CDAD in a suspected patient. Cytotoxic assays, which have become the standard diagnostic technique for CDAD, exhibit excellent sensitivity and specificity but have a lengthy processing time, between 2 and 4 days. Although antigen detection assays can be rapidly performed, many have inadequate sensitivity and specificity.7

These issues of diagnostic and treatment delays are compounded in patients with recurrent CDAD. As many as 15%35% of patients with an initial CDAD infection will experience a recurrence, usually within 2 months. At least half these infections are a result of reinfection, not relapse.8 This implies that early detection and strict isolation of infected patients is essential for reducing the exposure of at‐risk patients to the disease. There is evidence that the burden of patients on the same ward simultaneously having CDAD increases a patient's risk of acquiring the disease.9 It is currently unknown if recurrent CDAD cases are diagnosed or treated earlier than initial cases. If not, this is a potentially important patient population for hospitalists to target for aggressive containment strategies. This study sought to determine the mean time to sampling and treatment in patients with recurrent CDAD infection compared with those in patients who are initially infected.

Design

The study cohort consisted of all adult patients more than 18 years old with CDAD (by ICD9 code) who had been hospitalized at Brigham and Women's Hospital between 1997 and 2004. Retrospectively, patients were identified through the Partners Healthcare Research Patient Data Repository (RPDR). The RPDR is a centralized clinical data registry that gathers data from various hospital legacy systems and was used to determine the patient demographics and first date of treatment (with vancomycin or metronidazole). Medical and microbiologic records were reviewed to determine the dates of cytotoxic assay submission and symptom onset. Symptoms were defined as diarrhea, abdominal pain/cramping, or radiological/colonoscopic evidence of colitis. Recurrence was defined as any repeat inpatient CDAD diagnosis within 2 months (regardless of admission diagnosis). Baseline characteristics in the recurrence and no‐recurrence populations were compared by the 2‐sided Student t test or the chi‐square test (for continuous and categorical variables, respectively). Mean time from symptom to sampling, from symptom to treatment, and from sampling to treatment were compared between initial and recurrent disease episodes by the 2‐sided Student t test. All P values < .05 were considered significant. Institutional review board approval was obtained by Partners Healthcare.

RESULTS

Between 1997 and 2004 there were 1309 patients with an ICD9 code for CDAD, 151 of whom (12%) had a recurrence. Of these, 125 had 1 recurrence, 23 had 2 recurrences, and 3 had 3 recurrences. There were no significant differences between the groups in basic demographics (Table 1). The mean time to sampling was not significantly different between initial and recurrent CDAD hospital episodes (Table 2). However, the mean time to treatment (from symptoms and sampling) was shorter in recurrent episodes (Table 2). From 1997 to 2004 there was no significant reduction in time to sampling, but there was a significant reduction in time to treatment, from 3.89 days (19972000) to 2.30 days (2001 2004), P = .0012.

DISCUSSION

Clostridium difficileassociated disease (CDAD) has become a significant nosocomial infection in medical institutions, and recurrent CDAD is emerging as a disease of concern for hospitalists. Diagnostic delays represent a major epidemiologic problem, resulting in both delay of treatment delay of the index case and delay in implementing isolation procedures to prevent horizontal transmission. In this study, patients with recurrent disease did not have stool collected any earlier than did patients with their initial episode of CDAD, and these diagnostic delays did not change in successive eras. Recurrent disease patients did receive treatment earlier than did patients with initial episodes. Although this empiric treatment strategy is encouraging and likely reflects heightened awareness of the disease over time, the 2.5‐day span from symptoms to treatment is still a clinically significant delay. Also of concern is the range of time from symptoms to treatment, as long as 19 days in the recurrent treatment group. Although most patients were treated within 12 days, this variability represents the burden of infectious patients with the potential for infecting others. Targeting recurrent CDAD populations for early diagnosis, treatment, and isolation would almost certainly reduce the morbidity associated with horizontal transmission rates.9

This study had several limitations. Our data found a lower incidence of recurrent CDAD than previously published in the literature. This can be accounted for by the identification of cases by ICD9 code, which previously has been documented to underestimate true disease.10, 11 We also were not able to capture recurrent episodes in outpatients or episodes that occurred after the 2004 cohort, which underestimated the true frequency of recurrence. At worst, this underestimation could bias the results toward the null hypothesis. An additional limitation of the study was the assumption that time to treatment was accurately reflected by time to prescription of either vancomycin or flagyl. Some patients may have been treated by suspending treatment with the offending antibiotic along with watchful waiting, which is a reasonable strategy for patients with mild disease and is endorsed by the American College of Gastroenterology and the Society for Healthcare Epidemiology of America.12, 13 This would overestimate time to treatment for those individuals and would make time to treatment appear longer, but would not affect time to sampling. In addition, the symptoms collected from chart review were assumed to be a result of the patient's CDAD, but there is a chance that these symptoms such as diarrhea, abdominal pain, and cramping may have been a result of a different diagnosis. These data were also limited to a cohort from a single institution and may not reflect the patient characteristics or practice patterns at other institutions.

In conclusion, CDAD is a major contributor to morbidity from nosocomial infections, and recurrent CDAD patients are a likely source of horizontal disease transmission. This study documented that there are significant diagnostic and treatment delays, even in populations with recurrent disease. It is especially important that hospitalists take measures to improve the early diagnosis, treatment, and isolation of these patients in order to improve these deficiencies in care.

Clostridium difficileassociated disease (CDAD) is a well‐known complication of hospitalization and is the most frequently identified cause of nosocomial diarrhea that hospitalists encounter. Despite widespread epidemiologic attempts to control the disease, its prevalence and clinical severity appear to be increasing.1 The resulting social and economic consequences are profound. The estimated 3 million inpatient cases of CDAD a year result in an average increase in the length of stay of 3.6 days at a cost in inpatient health care of more than $1 billion.2

Early diagnosis of index cases is crucial. A diagnostic delay can result in a treatment delay for the index case, as well as in a delay in implementing isolation procedures to prevent horizontal transmission. Acquisition of CDAD is time dependent and occurs in 20% to 30% of hospitalized patients at a rate of approximately 8% per week.3, 4 This transmission is primarily a result of environmental contamination with CDAD spores, found on 59% of the hands of hospital personnel caring for infected patients, in 49% of rooms of symptomatic patients, and in 29% of rooms of asymptomatic carriers.5 Despite the need for early diagnosis, a study from the United Kingdom documented that the average time from the onset of diarrhea to sampling of CDAD patients is 4.7 days.6 An additional challenge for early diagnosis is the delay in microbiological confirmation of CDAD in a suspected patient. Cytotoxic assays, which have become the standard diagnostic technique for CDAD, exhibit excellent sensitivity and specificity but have a lengthy processing time, between 2 and 4 days. Although antigen detection assays can be rapidly performed, many have inadequate sensitivity and specificity.7

These issues of diagnostic and treatment delays are compounded in patients with recurrent CDAD. As many as 15%35% of patients with an initial CDAD infection will experience a recurrence, usually within 2 months. At least half these infections are a result of reinfection, not relapse.8 This implies that early detection and strict isolation of infected patients is essential for reducing the exposure of at‐risk patients to the disease. There is evidence that the burden of patients on the same ward simultaneously having CDAD increases a patient's risk of acquiring the disease.9 It is currently unknown if recurrent CDAD cases are diagnosed or treated earlier than initial cases. If not, this is a potentially important patient population for hospitalists to target for aggressive containment strategies. This study sought to determine the mean time to sampling and treatment in patients with recurrent CDAD infection compared with those in patients who are initially infected.

Design

The study cohort consisted of all adult patients more than 18 years old with CDAD (by ICD9 code) who had been hospitalized at Brigham and Women's Hospital between 1997 and 2004. Retrospectively, patients were identified through the Partners Healthcare Research Patient Data Repository (RPDR). The RPDR is a centralized clinical data registry that gathers data from various hospital legacy systems and was used to determine the patient demographics and first date of treatment (with vancomycin or metronidazole). Medical and microbiologic records were reviewed to determine the dates of cytotoxic assay submission and symptom onset. Symptoms were defined as diarrhea, abdominal pain/cramping, or radiological/colonoscopic evidence of colitis. Recurrence was defined as any repeat inpatient CDAD diagnosis within 2 months (regardless of admission diagnosis). Baseline characteristics in the recurrence and no‐recurrence populations were compared by the 2‐sided Student t test or the chi‐square test (for continuous and categorical variables, respectively). Mean time from symptom to sampling, from symptom to treatment, and from sampling to treatment were compared between initial and recurrent disease episodes by the 2‐sided Student t test. All P values < .05 were considered significant. Institutional review board approval was obtained by Partners Healthcare.

RESULTS

Between 1997 and 2004 there were 1309 patients with an ICD9 code for CDAD, 151 of whom (12%) had a recurrence. Of these, 125 had 1 recurrence, 23 had 2 recurrences, and 3 had 3 recurrences. There were no significant differences between the groups in basic demographics (Table 1). The mean time to sampling was not significantly different between initial and recurrent CDAD hospital episodes (Table 2). However, the mean time to treatment (from symptoms and sampling) was shorter in recurrent episodes (Table 2). From 1997 to 2004 there was no significant reduction in time to sampling, but there was a significant reduction in time to treatment, from 3.89 days (19972000) to 2.30 days (2001 2004), P = .0012.

DISCUSSION

Clostridium difficileassociated disease (CDAD) has become a significant nosocomial infection in medical institutions, and recurrent CDAD is emerging as a disease of concern for hospitalists. Diagnostic delays represent a major epidemiologic problem, resulting in both delay of treatment delay of the index case and delay in implementing isolation procedures to prevent horizontal transmission. In this study, patients with recurrent disease did not have stool collected any earlier than did patients with their initial episode of CDAD, and these diagnostic delays did not change in successive eras. Recurrent disease patients did receive treatment earlier than did patients with initial episodes. Although this empiric treatment strategy is encouraging and likely reflects heightened awareness of the disease over time, the 2.5‐day span from symptoms to treatment is still a clinically significant delay. Also of concern is the range of time from symptoms to treatment, as long as 19 days in the recurrent treatment group. Although most patients were treated within 12 days, this variability represents the burden of infectious patients with the potential for infecting others. Targeting recurrent CDAD populations for early diagnosis, treatment, and isolation would almost certainly reduce the morbidity associated with horizontal transmission rates.9

This study had several limitations. Our data found a lower incidence of recurrent CDAD than previously published in the literature. This can be accounted for by the identification of cases by ICD9 code, which previously has been documented to underestimate true disease.10, 11 We also were not able to capture recurrent episodes in outpatients or episodes that occurred after the 2004 cohort, which underestimated the true frequency of recurrence. At worst, this underestimation could bias the results toward the null hypothesis. An additional limitation of the study was the assumption that time to treatment was accurately reflected by time to prescription of either vancomycin or flagyl. Some patients may have been treated by suspending treatment with the offending antibiotic along with watchful waiting, which is a reasonable strategy for patients with mild disease and is endorsed by the American College of Gastroenterology and the Society for Healthcare Epidemiology of America.12, 13 This would overestimate time to treatment for those individuals and would make time to treatment appear longer, but would not affect time to sampling. In addition, the symptoms collected from chart review were assumed to be a result of the patient's CDAD, but there is a chance that these symptoms such as diarrhea, abdominal pain, and cramping may have been a result of a different diagnosis. These data were also limited to a cohort from a single institution and may not reflect the patient characteristics or practice patterns at other institutions.

In conclusion, CDAD is a major contributor to morbidity from nosocomial infections, and recurrent CDAD patients are a likely source of horizontal disease transmission. This study documented that there are significant diagnostic and treatment delays, even in populations with recurrent disease. It is especially important that hospitalists take measures to improve the early diagnosis, treatment, and isolation of these patients in order to improve these deficiencies in care.

Influenza is estimated to infect up to 40%‐54% of school‐age children during an influenza season.1, 2 Influenza infections result in missed school days, visits to primary care providers, prescriptions for antibiotics, hospitalizations, and secondary infections.310 Although influenza‐related mortality of pediatric patients is rare, influenza‐related morbidity of children and those secondarily infected is common and partially preventable.11

For at‐risk individuals, immunization against influenza is the best method of preventing morbidity. The Centers for Disease Control's Advisory Committee on Immunization Practice (CDC/ACIP) considers children with asthma at high risk of morbidity from influenza and since 1964 has recommended delivery of inactivated influenza vaccination to this population of patients. Even with this recommendation, the current level of adherence to yearly influenza vaccination of children with asthma is approximately 29%, according to the CDC analysis of the National Health Interview Survey.12 Although pediatric providers may prefer to provide preventive care in an ambulatory setting, the CDC/ACIP emphasizes the need for vaccination delivery wherever possible, including in acute‐care hospitals.11 Particularly in an acute‐care setting, vaccinating patients who are having an exacerbation of their asthma and who may have a compromised immune response may raise concerns. However, the vaccine has been shown to be efficacious even with pediatric asthmatics on a short course of corticosteroids.13, 14 Utilization of the inpatient visit as an opportunity to implement evidence‐based preventive care to pediatric asthmatics could have a beneficial effect on patient outcomes. Group‐based settings (school or health fair) are more cost effective than individual‐initiated settings (primary care) for influenza vaccination of pediatric patients.15 However, we found no studies investigating the effectiveness of vaccinating high‐risk patients in an acute‐care setting.

This study used established modeling techniques (decision and cost‐effectiveness analyses) to determine the potential clinical and cost benefits of vaccinating children against influenza during an asthma exacerbation in an acute‐care setting. We used decision analysis to determine the effectiveness of the intervention in improving the up‐to‐date (UTD) status of a hypothetical population. Based on this improvement in vaccination delivery, we completed a cost‐effectiveness analysis to determine direct and indirect costs, improvement in clinical outcomes attributable to influenza (clinic visits, antibiotic use, hospitalization, and secondary infections), and cost savings attributable to the intervention.

METHODS

Decision Analysis

We generated a decision tree to represent an inpatient intervention to assess and deliver influenza vaccination to children with asthma in acute‐care hospitals (Fig. 1). The design of the decision tree intentionally began with a decision node in order to introduce different assessment rates. This approach was used in order to accurately represent an inpatient intervention (where 100% assessment is not always attained) and to assess how less‐than‐perfect rates of assessment of influenza vaccine status would affect the success of the intervention. Once the decision tree was created, we surveyed the literature to obtain data on the assumptions of the decision tree (See Table 1 for assumptions of decision analysis with corresponding numbering in Figure 1 Decision Tree) including percentage of children with asthma who receive the influenza vaccine (29%),12 percentages of patients and caregivers who would agree to be vaccinated in an acute‐care settings (71%),16 and rates of response to vaccine reminder recall systems in a primary care setting (30%)17 in order to account for populations that need a second dose of vaccine to become UTD. To estimate the percentage of children who would require a second vaccination at 4 weeks, we utilized the Healthcare Cost and Utilization Project Kids Inpatient Database (HCUP KID).18 Seventy‐three percent of children younger than 9 years old in the HCUP KID were discharged with a diagnosis of asthma. These patients would require a second vaccination after 4 weeks if they had not previously been immunized. After all the estimates were placed in the decision tree, a hypothetical cohort of children with asthma was introduced, and improvement in UTD status after the intervention was determined.

Figure 1

Table 1.Key Assumptions for the Model

		Reference
Decision analysis
1. Percentage of asthmatics UTD with influenza vaccination without increased inpatient assessment	29%	12
2. Percentage of caregivers who agree with vaccination in an acute‐care setting	71%	16
3. Percentage of asthma exacerbation discharges <9 years old	73%	18
4. Percentage of caregivers who respond to reminder‐recall systems for outpatient vaccination	30%	17
Cost‐effectiveness analysis
5. Prevalence of influenza in school‐age children over 1 season	45%	1
6. Vaccine efficacy		15
Best case	43%
Base case	56%
Worst case	75%
7. Risk of secondary transmission	18%	31, 32
8. Missed school days (per child per year) attributed to influenza	0.79	29, 30
9. Percentage of caregivers who miss work to care for sick child	53%	6

RESULTS

With existing data showing that only 29% of those with asthma are UTD on influenza vaccine in a given year, our decision analysis demonstrated that even modest increases in the rate of screening for influenza vaccine status among hospitalized patients with asthma can result in clinically significant increases in the rate of vaccine delivery. For example, screening of just 20% of hospitalized patients with asthma would result in 35% of the children ultimately being up to date for that influenza season, a 20% overall improvement. In the same manner, a 40% assessment rate would result in 41% of children ultimately becoming UTD, a 60% assessment rate in 47% ultimately becoming UTD, an 80% assessment rate in 53% ultimately becoming UTD, and a 100% assessment in 59% ultimately becoming UTD, which is double the baseline rate (Table 2 ).

This cost analysis demonstrated cost savings of $5.45 per child assessed and $9.19 per child vaccinated (Table 3). The cost savings per child assessed was lower than per child vaccinated because of including the costs of assessment without delivery of vaccination. This estimated savings is lower than the previously published $34.79 per vaccination of school‐age children in a group setting but more comparable to the cost savings of $3.99 in an individual‐initiated setting.15

The cost savings with this intervention depends on the vaccine's efficacy, and efficacy can vary as a function of whether it is an epidemic season and the degree of matching between the prevalent influenza strains and the vaccine strains. At higher predicted levels of vaccination efficacy, UTD patients have improved, less negative outcomes from influenza, and the intervention results in higher cost savings (cost savings with base‐case vaccine efficacy of $5.45/child assessed increases to $11.93/child assessed with best‐case vaccine efficacy). No current research accurately predicts vaccine efficacy for each outcome in pediatric patients with asthma. Therefore, this study design varied vaccine efficacy during sensitivity analyses among 3 potential vaccine efficacy scenarios (best‐, base‐, and worst‐case scenarios).15 Varying the vaccine efficacy to the best‐case scenario resulted in a cost savings of $11.93 per child assessed and $20.13 per child vaccinated. Dropping the efficacy to the worst‐case scenario resulted in cost savings of $1.01 per child assessed and $1.71 per child vaccinated. If the vaccine efficacy was 40%, the intervention was cost neutral. It should be remembered that the goal of the health care system is to generate good health. A project that is cost neutral (cost of vaccination intervention = cost savings from illness prevention), such as this intervention with a vaccine efficacy of 40%, may still be considered cost effective because of the improvement in clinical outcomes. At a vaccine efficacy of 40%, this model predicts that clinic visits will decrease by 19/1000 children, unwarranted antibiotics will decrease by 12/1000 children, hospitalizations will decrease by 4/10,000 children, and secondary infections of adult caregivers will decrease by 17/1000 children, suggesting an overall benefit to the health care system and to patients.

The clinical improvement predicted by this model with base‐case vaccine efficacy and 100% assessment include: a decrease in clinic visits of 27/1000 children, decrease in antibiotic use of 75/1000 children, decrease in hospitalizations of 6/10,000 children, decrease in missed school days of 132/1000 children, and a decrease in secondary infections of adult caregivers of 23/1000 children assessed.

The total cost of vaccination in this intervention was estimated as $10,593 per 1000 children. A hospital that discharges fewer than 1000 children with asthma per year would accrue less direct cost yearly with this intervention ($1059 per 100 children). A portion of this cost could be recuperated through reimbursement for influenza vaccination via insurance payers and/or government programs to reimburse for childhood vaccinations.

The results from the investigation into the effect of year‐to‐year variation in influenza prevalence and morbidity (by using the lowest and highest estimates found in the literature) demonstrated the model to be sensitive to these estimates, but the intervention maintained a cost savings. In a year with low influenza prevalence and morbidity, the cost savings decreases to $1.89 per child assessed and $3.20 per child vaccinated. In a year with high influenza prevalence and morbidity, the cost savings would increase to $8.75 per child assessed and $14.77 per child vaccinated. This finding of cost savings even with low influenza prevalence and morbidity is consistent with previous studies of influenza vaccination of pediatric patients in group‐based settings.15

For all other estimates, traditional sensitivity analysis was performed, and the results continued to show cost savings during this procedure, suggesting that the conclusions based on this model are generalizable and robust.

DISCUSSION

Universal screening for influenza vaccine status and then delivery of the vaccine to those not UTD among hospitalized children with asthma has the potential to increase the percentage of these children receiving the influenza vaccine and to reduce costs. This model suggests that with 100% assessment of this difficult‐to‐reach population for being UTD with the influenza vaccine, it would be possible to achieve a vaccination rate of 59%, doubling the current yearly receipt of influenza vaccination of children admitted secondary to asthma. However, universal screening is not imperative to achieve significant clinical improvement and cost savings. The cost‐effectiveness analysis demonstrated that the cost savings would be $5.45 per child assessed and $9.19 per child vaccinated.

The cost savings in this analysis ($9.19/child vaccinated) is favorable but lower than a previous analysis of influenza vaccination of children in an unspecified group‐based setting ($34.79/child vaccinated).15 Multiple factors contributed to our lower cost savings. First, our model accounts for the cost of children becoming partially vaccinated, incurring the full cost for 1 vaccination without the resulting clinical benefit. Second, the model accommodates for nursing assessment without vaccination delivery. Most importantly, this model estimates direct cost of vaccination delivery more conservatively than other group‐based estimates ($18.84 vs. $4.31).15 Previous cost‐effective analyses of influenza vaccination referenced 2 articles from the mid‐1990s that used a direct cost for 1 vaccination of $4 for group‐based vaccination in an HMO34 and $10 for individual‐initiated vaccination.33 Our estimate of vaccine cost is more conservative than these studies but more likely to represent the cost of vaccination assessment and delivery in an acute‐care hospital setting.

As the model was created, it became apparent that vaccination in the inpatient setting would accrue less indirect costs as compared with an individual‐initiated outpatient setting. This is a result of there not being any incrementally increased loss of work by the family/caregiver in order to be vaccinated above that lost because of hospitalization of the child for asthma. This finding supports that there is cost savings by vaccinating pediatric patients while hospitalized and is consistent with previous literature on this subject.15

The use of modeling techniques to evaluate inpatient interventions has many benefits. Modeling techniques permit the generation of synthetic trials by utilizing the combined published data from multiple studies. This permits the investigator to apply findings to other populations with different risks or prognoses and to other settings, to extend the impact over time, and to display multiple outcomes together. This technique has the potential to be used in investigations of the theoretical benefit of a specific quality improvement intervention in an inpatient setting without the extensive cost of a clinical trial. In fact, preliminary analysis with modeling could improve the cost efficiency of quality improvement research. Modeling studies could demonstrate which interventions are most likely to generate cost savings and direct clinical investigation and funding.

Limitations of this project included the possibility of vaccine efficacy changing from year to year depending on the similarity of the viruses in the vaccine with the predominant infectious strains circulating that year, as observed in previous studies.11 In a year with a poor match, the cost savings would be reduced compared with that in years with a good match. In addition, influenza prevalence and morbidity affects this model and the cost savings of the intervention, but the analysis using the lowest estimates of prevalence and morbidity continued to demonstrate that the intervention was saving costs. This continued cost savings for a group‐based vaccination intervention during years when there is low prevalence and morbidity is consistent with previous reports in the literature for pediatric patients.15 Modeling techniques decrease the effect of year‐to‐year variation in vaccination match, prevalence, and morbidity and estimate results based on an average over multiple influenza seasons.

For children with asthma, the data are not well established about influenza vaccination's ability to improve specific clinical outcomes attributed to influenza illness (missed school days, clinic visits, antibiotic prescriptions, hospitalizations, and secondary infections). Acknowledging the limitations of these data, our model estimates the intervention's improvement in clinical outcomes by using published data on vaccine efficacy and varying the vaccine efficacy from worst case to base case to best case. In addition, with a vaccine efficacy of 40%, this intervention would be cost neutral. Regardless of concerns about vaccine efficacy, the CDC/ACIP recommends finding missed opportunities for vaccination to improve vaccination delivery and the ultimate UTD status of high‐risk children. Recently, the CDC/ACIP stated that influenza vaccination coverage among children with asthma is inadequate and that opportunities for vaccination during health‐care provider visits likely are being missed.12 This intervention to assess and deliver influenza vaccination to pediatric patients while hospitalized for an asthma exacerbation would allow all hospitals that treat children to participate in reaching the goal of the CDC/ACIP to reduce missed opportunities for influenza vaccination.

The direct cost of inpatient vaccination in this intervention would fall primarily on the hospital implementing the intervention ($10,593 per 1000 children). However, the Vaccines for Children program exists to offset the cost of immunizations to both patients and providers for the uninsured and children receiving Medicaid.11 If properly implemented by participating hospitals, this program would improve reimbursement for vaccination and therefore shift the cost away from the individual hospital. For example, if the intervention cost borne by a hospital were decreased by the direct cost of vaccination alone ($6.75), this model predicts the cost of a program per 1000 children would decrease from $10,593 to $7190. This cost reduction would depend on the percentage of patients uninsured or on Medicaid at a given hospital. Another approach to shifting cost from individual hospitals would be for policy makers to consider influenza vaccination of these high‐risk pediatric patients as a performance/quality measure and associate the measure with improved reimbursement to hospitals in order to compensate for the cost of the vaccination intervention.

CONCLUSIONS

Influenza immunization is an accepted method of prevention and is underutilized in children with asthma.12 This model suggests that an inpatient program to deliver influenza vaccination to hospitalized pediatric patients with asthma would be beneficial by producing better health outcomes and reducing health care costs. Further research should be performed to verify the assumptions used in this analysis for children with asthma.

Influenza is estimated to infect up to 40%‐54% of school‐age children during an influenza season.1, 2 Influenza infections result in missed school days, visits to primary care providers, prescriptions for antibiotics, hospitalizations, and secondary infections.310 Although influenza‐related mortality of pediatric patients is rare, influenza‐related morbidity of children and those secondarily infected is common and partially preventable.11

For at‐risk individuals, immunization against influenza is the best method of preventing morbidity. The Centers for Disease Control's Advisory Committee on Immunization Practice (CDC/ACIP) considers children with asthma at high risk of morbidity from influenza and since 1964 has recommended delivery of inactivated influenza vaccination to this population of patients. Even with this recommendation, the current level of adherence to yearly influenza vaccination of children with asthma is approximately 29%, according to the CDC analysis of the National Health Interview Survey.12 Although pediatric providers may prefer to provide preventive care in an ambulatory setting, the CDC/ACIP emphasizes the need for vaccination delivery wherever possible, including in acute‐care hospitals.11 Particularly in an acute‐care setting, vaccinating patients who are having an exacerbation of their asthma and who may have a compromised immune response may raise concerns. However, the vaccine has been shown to be efficacious even with pediatric asthmatics on a short course of corticosteroids.13, 14 Utilization of the inpatient visit as an opportunity to implement evidence‐based preventive care to pediatric asthmatics could have a beneficial effect on patient outcomes. Group‐based settings (school or health fair) are more cost effective than individual‐initiated settings (primary care) for influenza vaccination of pediatric patients.15 However, we found no studies investigating the effectiveness of vaccinating high‐risk patients in an acute‐care setting.

This study used established modeling techniques (decision and cost‐effectiveness analyses) to determine the potential clinical and cost benefits of vaccinating children against influenza during an asthma exacerbation in an acute‐care setting. We used decision analysis to determine the effectiveness of the intervention in improving the up‐to‐date (UTD) status of a hypothetical population. Based on this improvement in vaccination delivery, we completed a cost‐effectiveness analysis to determine direct and indirect costs, improvement in clinical outcomes attributable to influenza (clinic visits, antibiotic use, hospitalization, and secondary infections), and cost savings attributable to the intervention.

METHODS

Decision Analysis

We generated a decision tree to represent an inpatient intervention to assess and deliver influenza vaccination to children with asthma in acute‐care hospitals (Fig. 1). The design of the decision tree intentionally began with a decision node in order to introduce different assessment rates. This approach was used in order to accurately represent an inpatient intervention (where 100% assessment is not always attained) and to assess how less‐than‐perfect rates of assessment of influenza vaccine status would affect the success of the intervention. Once the decision tree was created, we surveyed the literature to obtain data on the assumptions of the decision tree (See Table 1 for assumptions of decision analysis with corresponding numbering in Figure 1 Decision Tree) including percentage of children with asthma who receive the influenza vaccine (29%),12 percentages of patients and caregivers who would agree to be vaccinated in an acute‐care settings (71%),16 and rates of response to vaccine reminder recall systems in a primary care setting (30%)17 in order to account for populations that need a second dose of vaccine to become UTD. To estimate the percentage of children who would require a second vaccination at 4 weeks, we utilized the Healthcare Cost and Utilization Project Kids Inpatient Database (HCUP KID).18 Seventy‐three percent of children younger than 9 years old in the HCUP KID were discharged with a diagnosis of asthma. These patients would require a second vaccination after 4 weeks if they had not previously been immunized. After all the estimates were placed in the decision tree, a hypothetical cohort of children with asthma was introduced, and improvement in UTD status after the intervention was determined.

Figure 1

Table 1.Key Assumptions for the Model

		Reference
Decision analysis
1. Percentage of asthmatics UTD with influenza vaccination without increased inpatient assessment	29%	12
2. Percentage of caregivers who agree with vaccination in an acute‐care setting	71%	16
3. Percentage of asthma exacerbation discharges <9 years old	73%	18
4. Percentage of caregivers who respond to reminder‐recall systems for outpatient vaccination	30%	17
Cost‐effectiveness analysis
5. Prevalence of influenza in school‐age children over 1 season	45%	1
6. Vaccine efficacy		15
Best case	43%
Base case	56%
Worst case	75%
7. Risk of secondary transmission	18%	31, 32
8. Missed school days (per child per year) attributed to influenza	0.79	29, 30
9. Percentage of caregivers who miss work to care for sick child	53%	6

RESULTS

With existing data showing that only 29% of those with asthma are UTD on influenza vaccine in a given year, our decision analysis demonstrated that even modest increases in the rate of screening for influenza vaccine status among hospitalized patients with asthma can result in clinically significant increases in the rate of vaccine delivery. For example, screening of just 20% of hospitalized patients with asthma would result in 35% of the children ultimately being up to date for that influenza season, a 20% overall improvement. In the same manner, a 40% assessment rate would result in 41% of children ultimately becoming UTD, a 60% assessment rate in 47% ultimately becoming UTD, an 80% assessment rate in 53% ultimately becoming UTD, and a 100% assessment in 59% ultimately becoming UTD, which is double the baseline rate (Table 2 ).

This cost analysis demonstrated cost savings of $5.45 per child assessed and $9.19 per child vaccinated (Table 3). The cost savings per child assessed was lower than per child vaccinated because of including the costs of assessment without delivery of vaccination. This estimated savings is lower than the previously published $34.79 per vaccination of school‐age children in a group setting but more comparable to the cost savings of $3.99 in an individual‐initiated setting.15

The cost savings with this intervention depends on the vaccine's efficacy, and efficacy can vary as a function of whether it is an epidemic season and the degree of matching between the prevalent influenza strains and the vaccine strains. At higher predicted levels of vaccination efficacy, UTD patients have improved, less negative outcomes from influenza, and the intervention results in higher cost savings (cost savings with base‐case vaccine efficacy of $5.45/child assessed increases to $11.93/child assessed with best‐case vaccine efficacy). No current research accurately predicts vaccine efficacy for each outcome in pediatric patients with asthma. Therefore, this study design varied vaccine efficacy during sensitivity analyses among 3 potential vaccine efficacy scenarios (best‐, base‐, and worst‐case scenarios).15 Varying the vaccine efficacy to the best‐case scenario resulted in a cost savings of $11.93 per child assessed and $20.13 per child vaccinated. Dropping the efficacy to the worst‐case scenario resulted in cost savings of $1.01 per child assessed and $1.71 per child vaccinated. If the vaccine efficacy was 40%, the intervention was cost neutral. It should be remembered that the goal of the health care system is to generate good health. A project that is cost neutral (cost of vaccination intervention = cost savings from illness prevention), such as this intervention with a vaccine efficacy of 40%, may still be considered cost effective because of the improvement in clinical outcomes. At a vaccine efficacy of 40%, this model predicts that clinic visits will decrease by 19/1000 children, unwarranted antibiotics will decrease by 12/1000 children, hospitalizations will decrease by 4/10,000 children, and secondary infections of adult caregivers will decrease by 17/1000 children, suggesting an overall benefit to the health care system and to patients.

The clinical improvement predicted by this model with base‐case vaccine efficacy and 100% assessment include: a decrease in clinic visits of 27/1000 children, decrease in antibiotic use of 75/1000 children, decrease in hospitalizations of 6/10,000 children, decrease in missed school days of 132/1000 children, and a decrease in secondary infections of adult caregivers of 23/1000 children assessed.

The total cost of vaccination in this intervention was estimated as $10,593 per 1000 children. A hospital that discharges fewer than 1000 children with asthma per year would accrue less direct cost yearly with this intervention ($1059 per 100 children). A portion of this cost could be recuperated through reimbursement for influenza vaccination via insurance payers and/or government programs to reimburse for childhood vaccinations.

The results from the investigation into the effect of year‐to‐year variation in influenza prevalence and morbidity (by using the lowest and highest estimates found in the literature) demonstrated the model to be sensitive to these estimates, but the intervention maintained a cost savings. In a year with low influenza prevalence and morbidity, the cost savings decreases to $1.89 per child assessed and $3.20 per child vaccinated. In a year with high influenza prevalence and morbidity, the cost savings would increase to $8.75 per child assessed and $14.77 per child vaccinated. This finding of cost savings even with low influenza prevalence and morbidity is consistent with previous studies of influenza vaccination of pediatric patients in group‐based settings.15

For all other estimates, traditional sensitivity analysis was performed, and the results continued to show cost savings during this procedure, suggesting that the conclusions based on this model are generalizable and robust.

DISCUSSION

Universal screening for influenza vaccine status and then delivery of the vaccine to those not UTD among hospitalized children with asthma has the potential to increase the percentage of these children receiving the influenza vaccine and to reduce costs. This model suggests that with 100% assessment of this difficult‐to‐reach population for being UTD with the influenza vaccine, it would be possible to achieve a vaccination rate of 59%, doubling the current yearly receipt of influenza vaccination of children admitted secondary to asthma. However, universal screening is not imperative to achieve significant clinical improvement and cost savings. The cost‐effectiveness analysis demonstrated that the cost savings would be $5.45 per child assessed and $9.19 per child vaccinated.

The cost savings in this analysis ($9.19/child vaccinated) is favorable but lower than a previous analysis of influenza vaccination of children in an unspecified group‐based setting ($34.79/child vaccinated).15 Multiple factors contributed to our lower cost savings. First, our model accounts for the cost of children becoming partially vaccinated, incurring the full cost for 1 vaccination without the resulting clinical benefit. Second, the model accommodates for nursing assessment without vaccination delivery. Most importantly, this model estimates direct cost of vaccination delivery more conservatively than other group‐based estimates ($18.84 vs. $4.31).15 Previous cost‐effective analyses of influenza vaccination referenced 2 articles from the mid‐1990s that used a direct cost for 1 vaccination of $4 for group‐based vaccination in an HMO34 and $10 for individual‐initiated vaccination.33 Our estimate of vaccine cost is more conservative than these studies but more likely to represent the cost of vaccination assessment and delivery in an acute‐care hospital setting.

As the model was created, it became apparent that vaccination in the inpatient setting would accrue less indirect costs as compared with an individual‐initiated outpatient setting. This is a result of there not being any incrementally increased loss of work by the family/caregiver in order to be vaccinated above that lost because of hospitalization of the child for asthma. This finding supports that there is cost savings by vaccinating pediatric patients while hospitalized and is consistent with previous literature on this subject.15

The use of modeling techniques to evaluate inpatient interventions has many benefits. Modeling techniques permit the generation of synthetic trials by utilizing the combined published data from multiple studies. This permits the investigator to apply findings to other populations with different risks or prognoses and to other settings, to extend the impact over time, and to display multiple outcomes together. This technique has the potential to be used in investigations of the theoretical benefit of a specific quality improvement intervention in an inpatient setting without the extensive cost of a clinical trial. In fact, preliminary analysis with modeling could improve the cost efficiency of quality improvement research. Modeling studies could demonstrate which interventions are most likely to generate cost savings and direct clinical investigation and funding.

Limitations of this project included the possibility of vaccine efficacy changing from year to year depending on the similarity of the viruses in the vaccine with the predominant infectious strains circulating that year, as observed in previous studies.11 In a year with a poor match, the cost savings would be reduced compared with that in years with a good match. In addition, influenza prevalence and morbidity affects this model and the cost savings of the intervention, but the analysis using the lowest estimates of prevalence and morbidity continued to demonstrate that the intervention was saving costs. This continued cost savings for a group‐based vaccination intervention during years when there is low prevalence and morbidity is consistent with previous reports in the literature for pediatric patients.15 Modeling techniques decrease the effect of year‐to‐year variation in vaccination match, prevalence, and morbidity and estimate results based on an average over multiple influenza seasons.

For children with asthma, the data are not well established about influenza vaccination's ability to improve specific clinical outcomes attributed to influenza illness (missed school days, clinic visits, antibiotic prescriptions, hospitalizations, and secondary infections). Acknowledging the limitations of these data, our model estimates the intervention's improvement in clinical outcomes by using published data on vaccine efficacy and varying the vaccine efficacy from worst case to base case to best case. In addition, with a vaccine efficacy of 40%, this intervention would be cost neutral. Regardless of concerns about vaccine efficacy, the CDC/ACIP recommends finding missed opportunities for vaccination to improve vaccination delivery and the ultimate UTD status of high‐risk children. Recently, the CDC/ACIP stated that influenza vaccination coverage among children with asthma is inadequate and that opportunities for vaccination during health‐care provider visits likely are being missed.12 This intervention to assess and deliver influenza vaccination to pediatric patients while hospitalized for an asthma exacerbation would allow all hospitals that treat children to participate in reaching the goal of the CDC/ACIP to reduce missed opportunities for influenza vaccination.

The direct cost of inpatient vaccination in this intervention would fall primarily on the hospital implementing the intervention ($10,593 per 1000 children). However, the Vaccines for Children program exists to offset the cost of immunizations to both patients and providers for the uninsured and children receiving Medicaid.11 If properly implemented by participating hospitals, this program would improve reimbursement for vaccination and therefore shift the cost away from the individual hospital. For example, if the intervention cost borne by a hospital were decreased by the direct cost of vaccination alone ($6.75), this model predicts the cost of a program per 1000 children would decrease from $10,593 to $7190. This cost reduction would depend on the percentage of patients uninsured or on Medicaid at a given hospital. Another approach to shifting cost from individual hospitals would be for policy makers to consider influenza vaccination of these high‐risk pediatric patients as a performance/quality measure and associate the measure with improved reimbursement to hospitals in order to compensate for the cost of the vaccination intervention.

CONCLUSIONS

Influenza immunization is an accepted method of prevention and is underutilized in children with asthma.12 This model suggests that an inpatient program to deliver influenza vaccination to hospitalized pediatric patients with asthma would be beneficial by producing better health outcomes and reducing health care costs. Further research should be performed to verify the assumptions used in this analysis for children with asthma.

Influenza is estimated to infect up to 40%‐54% of school‐age children during an influenza season.1, 2 Influenza infections result in missed school days, visits to primary care providers, prescriptions for antibiotics, hospitalizations, and secondary infections.310 Although influenza‐related mortality of pediatric patients is rare, influenza‐related morbidity of children and those secondarily infected is common and partially preventable.11

For at‐risk individuals, immunization against influenza is the best method of preventing morbidity. The Centers for Disease Control's Advisory Committee on Immunization Practice (CDC/ACIP) considers children with asthma at high risk of morbidity from influenza and since 1964 has recommended delivery of inactivated influenza vaccination to this population of patients. Even with this recommendation, the current level of adherence to yearly influenza vaccination of children with asthma is approximately 29%, according to the CDC analysis of the National Health Interview Survey.12 Although pediatric providers may prefer to provide preventive care in an ambulatory setting, the CDC/ACIP emphasizes the need for vaccination delivery wherever possible, including in acute‐care hospitals.11 Particularly in an acute‐care setting, vaccinating patients who are having an exacerbation of their asthma and who may have a compromised immune response may raise concerns. However, the vaccine has been shown to be efficacious even with pediatric asthmatics on a short course of corticosteroids.13, 14 Utilization of the inpatient visit as an opportunity to implement evidence‐based preventive care to pediatric asthmatics could have a beneficial effect on patient outcomes. Group‐based settings (school or health fair) are more cost effective than individual‐initiated settings (primary care) for influenza vaccination of pediatric patients.15 However, we found no studies investigating the effectiveness of vaccinating high‐risk patients in an acute‐care setting.

This study used established modeling techniques (decision and cost‐effectiveness analyses) to determine the potential clinical and cost benefits of vaccinating children against influenza during an asthma exacerbation in an acute‐care setting. We used decision analysis to determine the effectiveness of the intervention in improving the up‐to‐date (UTD) status of a hypothetical population. Based on this improvement in vaccination delivery, we completed a cost‐effectiveness analysis to determine direct and indirect costs, improvement in clinical outcomes attributable to influenza (clinic visits, antibiotic use, hospitalization, and secondary infections), and cost savings attributable to the intervention.

METHODS

Decision Analysis

We generated a decision tree to represent an inpatient intervention to assess and deliver influenza vaccination to children with asthma in acute‐care hospitals (Fig. 1). The design of the decision tree intentionally began with a decision node in order to introduce different assessment rates. This approach was used in order to accurately represent an inpatient intervention (where 100% assessment is not always attained) and to assess how less‐than‐perfect rates of assessment of influenza vaccine status would affect the success of the intervention. Once the decision tree was created, we surveyed the literature to obtain data on the assumptions of the decision tree (See Table 1 for assumptions of decision analysis with corresponding numbering in Figure 1 Decision Tree) including percentage of children with asthma who receive the influenza vaccine (29%),12 percentages of patients and caregivers who would agree to be vaccinated in an acute‐care settings (71%),16 and rates of response to vaccine reminder recall systems in a primary care setting (30%)17 in order to account for populations that need a second dose of vaccine to become UTD. To estimate the percentage of children who would require a second vaccination at 4 weeks, we utilized the Healthcare Cost and Utilization Project Kids Inpatient Database (HCUP KID).18 Seventy‐three percent of children younger than 9 years old in the HCUP KID were discharged with a diagnosis of asthma. These patients would require a second vaccination after 4 weeks if they had not previously been immunized. After all the estimates were placed in the decision tree, a hypothetical cohort of children with asthma was introduced, and improvement in UTD status after the intervention was determined.

Figure 1

Table 1.Key Assumptions for the Model

		Reference
Decision analysis
1. Percentage of asthmatics UTD with influenza vaccination without increased inpatient assessment	29%	12
2. Percentage of caregivers who agree with vaccination in an acute‐care setting	71%	16
3. Percentage of asthma exacerbation discharges <9 years old	73%	18
4. Percentage of caregivers who respond to reminder‐recall systems for outpatient vaccination	30%	17
Cost‐effectiveness analysis
5. Prevalence of influenza in school‐age children over 1 season	45%	1
6. Vaccine efficacy		15
Best case	43%
Base case	56%
Worst case	75%
7. Risk of secondary transmission	18%	31, 32
8. Missed school days (per child per year) attributed to influenza	0.79	29, 30
9. Percentage of caregivers who miss work to care for sick child	53%	6

RESULTS

With existing data showing that only 29% of those with asthma are UTD on influenza vaccine in a given year, our decision analysis demonstrated that even modest increases in the rate of screening for influenza vaccine status among hospitalized patients with asthma can result in clinically significant increases in the rate of vaccine delivery. For example, screening of just 20% of hospitalized patients with asthma would result in 35% of the children ultimately being up to date for that influenza season, a 20% overall improvement. In the same manner, a 40% assessment rate would result in 41% of children ultimately becoming UTD, a 60% assessment rate in 47% ultimately becoming UTD, an 80% assessment rate in 53% ultimately becoming UTD, and a 100% assessment in 59% ultimately becoming UTD, which is double the baseline rate (Table 2 ).

This cost analysis demonstrated cost savings of $5.45 per child assessed and $9.19 per child vaccinated (Table 3). The cost savings per child assessed was lower than per child vaccinated because of including the costs of assessment without delivery of vaccination. This estimated savings is lower than the previously published $34.79 per vaccination of school‐age children in a group setting but more comparable to the cost savings of $3.99 in an individual‐initiated setting.15

The cost savings with this intervention depends on the vaccine's efficacy, and efficacy can vary as a function of whether it is an epidemic season and the degree of matching between the prevalent influenza strains and the vaccine strains. At higher predicted levels of vaccination efficacy, UTD patients have improved, less negative outcomes from influenza, and the intervention results in higher cost savings (cost savings with base‐case vaccine efficacy of $5.45/child assessed increases to $11.93/child assessed with best‐case vaccine efficacy). No current research accurately predicts vaccine efficacy for each outcome in pediatric patients with asthma. Therefore, this study design varied vaccine efficacy during sensitivity analyses among 3 potential vaccine efficacy scenarios (best‐, base‐, and worst‐case scenarios).15 Varying the vaccine efficacy to the best‐case scenario resulted in a cost savings of $11.93 per child assessed and $20.13 per child vaccinated. Dropping the efficacy to the worst‐case scenario resulted in cost savings of $1.01 per child assessed and $1.71 per child vaccinated. If the vaccine efficacy was 40%, the intervention was cost neutral. It should be remembered that the goal of the health care system is to generate good health. A project that is cost neutral (cost of vaccination intervention = cost savings from illness prevention), such as this intervention with a vaccine efficacy of 40%, may still be considered cost effective because of the improvement in clinical outcomes. At a vaccine efficacy of 40%, this model predicts that clinic visits will decrease by 19/1000 children, unwarranted antibiotics will decrease by 12/1000 children, hospitalizations will decrease by 4/10,000 children, and secondary infections of adult caregivers will decrease by 17/1000 children, suggesting an overall benefit to the health care system and to patients.

The clinical improvement predicted by this model with base‐case vaccine efficacy and 100% assessment include: a decrease in clinic visits of 27/1000 children, decrease in antibiotic use of 75/1000 children, decrease in hospitalizations of 6/10,000 children, decrease in missed school days of 132/1000 children, and a decrease in secondary infections of adult caregivers of 23/1000 children assessed.

The total cost of vaccination in this intervention was estimated as $10,593 per 1000 children. A hospital that discharges fewer than 1000 children with asthma per year would accrue less direct cost yearly with this intervention ($1059 per 100 children). A portion of this cost could be recuperated through reimbursement for influenza vaccination via insurance payers and/or government programs to reimburse for childhood vaccinations.

The results from the investigation into the effect of year‐to‐year variation in influenza prevalence and morbidity (by using the lowest and highest estimates found in the literature) demonstrated the model to be sensitive to these estimates, but the intervention maintained a cost savings. In a year with low influenza prevalence and morbidity, the cost savings decreases to $1.89 per child assessed and $3.20 per child vaccinated. In a year with high influenza prevalence and morbidity, the cost savings would increase to $8.75 per child assessed and $14.77 per child vaccinated. This finding of cost savings even with low influenza prevalence and morbidity is consistent with previous studies of influenza vaccination of pediatric patients in group‐based settings.15

For all other estimates, traditional sensitivity analysis was performed, and the results continued to show cost savings during this procedure, suggesting that the conclusions based on this model are generalizable and robust.

DISCUSSION

Universal screening for influenza vaccine status and then delivery of the vaccine to those not UTD among hospitalized children with asthma has the potential to increase the percentage of these children receiving the influenza vaccine and to reduce costs. This model suggests that with 100% assessment of this difficult‐to‐reach population for being UTD with the influenza vaccine, it would be possible to achieve a vaccination rate of 59%, doubling the current yearly receipt of influenza vaccination of children admitted secondary to asthma. However, universal screening is not imperative to achieve significant clinical improvement and cost savings. The cost‐effectiveness analysis demonstrated that the cost savings would be $5.45 per child assessed and $9.19 per child vaccinated.

The cost savings in this analysis ($9.19/child vaccinated) is favorable but lower than a previous analysis of influenza vaccination of children in an unspecified group‐based setting ($34.79/child vaccinated).15 Multiple factors contributed to our lower cost savings. First, our model accounts for the cost of children becoming partially vaccinated, incurring the full cost for 1 vaccination without the resulting clinical benefit. Second, the model accommodates for nursing assessment without vaccination delivery. Most importantly, this model estimates direct cost of vaccination delivery more conservatively than other group‐based estimates ($18.84 vs. $4.31).15 Previous cost‐effective analyses of influenza vaccination referenced 2 articles from the mid‐1990s that used a direct cost for 1 vaccination of $4 for group‐based vaccination in an HMO34 and $10 for individual‐initiated vaccination.33 Our estimate of vaccine cost is more conservative than these studies but more likely to represent the cost of vaccination assessment and delivery in an acute‐care hospital setting.

As the model was created, it became apparent that vaccination in the inpatient setting would accrue less indirect costs as compared with an individual‐initiated outpatient setting. This is a result of there not being any incrementally increased loss of work by the family/caregiver in order to be vaccinated above that lost because of hospitalization of the child for asthma. This finding supports that there is cost savings by vaccinating pediatric patients while hospitalized and is consistent with previous literature on this subject.15

The use of modeling techniques to evaluate inpatient interventions has many benefits. Modeling techniques permit the generation of synthetic trials by utilizing the combined published data from multiple studies. This permits the investigator to apply findings to other populations with different risks or prognoses and to other settings, to extend the impact over time, and to display multiple outcomes together. This technique has the potential to be used in investigations of the theoretical benefit of a specific quality improvement intervention in an inpatient setting without the extensive cost of a clinical trial. In fact, preliminary analysis with modeling could improve the cost efficiency of quality improvement research. Modeling studies could demonstrate which interventions are most likely to generate cost savings and direct clinical investigation and funding.

Limitations of this project included the possibility of vaccine efficacy changing from year to year depending on the similarity of the viruses in the vaccine with the predominant infectious strains circulating that year, as observed in previous studies.11 In a year with a poor match, the cost savings would be reduced compared with that in years with a good match. In addition, influenza prevalence and morbidity affects this model and the cost savings of the intervention, but the analysis using the lowest estimates of prevalence and morbidity continued to demonstrate that the intervention was saving costs. This continued cost savings for a group‐based vaccination intervention during years when there is low prevalence and morbidity is consistent with previous reports in the literature for pediatric patients.15 Modeling techniques decrease the effect of year‐to‐year variation in vaccination match, prevalence, and morbidity and estimate results based on an average over multiple influenza seasons.

For children with asthma, the data are not well established about influenza vaccination's ability to improve specific clinical outcomes attributed to influenza illness (missed school days, clinic visits, antibiotic prescriptions, hospitalizations, and secondary infections). Acknowledging the limitations of these data, our model estimates the intervention's improvement in clinical outcomes by using published data on vaccine efficacy and varying the vaccine efficacy from worst case to base case to best case. In addition, with a vaccine efficacy of 40%, this intervention would be cost neutral. Regardless of concerns about vaccine efficacy, the CDC/ACIP recommends finding missed opportunities for vaccination to improve vaccination delivery and the ultimate UTD status of high‐risk children. Recently, the CDC/ACIP stated that influenza vaccination coverage among children with asthma is inadequate and that opportunities for vaccination during health‐care provider visits likely are being missed.12 This intervention to assess and deliver influenza vaccination to pediatric patients while hospitalized for an asthma exacerbation would allow all hospitals that treat children to participate in reaching the goal of the CDC/ACIP to reduce missed opportunities for influenza vaccination.

The direct cost of inpatient vaccination in this intervention would fall primarily on the hospital implementing the intervention ($10,593 per 1000 children). However, the Vaccines for Children program exists to offset the cost of immunizations to both patients and providers for the uninsured and children receiving Medicaid.11 If properly implemented by participating hospitals, this program would improve reimbursement for vaccination and therefore shift the cost away from the individual hospital. For example, if the intervention cost borne by a hospital were decreased by the direct cost of vaccination alone ($6.75), this model predicts the cost of a program per 1000 children would decrease from $10,593 to $7190. This cost reduction would depend on the percentage of patients uninsured or on Medicaid at a given hospital. Another approach to shifting cost from individual hospitals would be for policy makers to consider influenza vaccination of these high‐risk pediatric patients as a performance/quality measure and associate the measure with improved reimbursement to hospitals in order to compensate for the cost of the vaccination intervention.

CONCLUSIONS

Influenza immunization is an accepted method of prevention and is underutilized in children with asthma.12 This model suggests that an inpatient program to deliver influenza vaccination to hospitalized pediatric patients with asthma would be beneficial by producing better health outcomes and reducing health care costs. Further research should be performed to verify the assumptions used in this analysis for children with asthma.

A 26‐year‐old woman was brought to the emergency department following several episodes of seizures. The patient's friend witnessed several 15‐minute episodes of sudden jerks and tremors of her right arm during which the patient bit her tongue, had word‐finding difficulty, had horizontal eye deviation, and was incontinent of urine. She became unresponsive during the episodes, with incomplete recovery of consciousness between attacks. She was afebrile. Her neurologic exam 4 hours after several seizures revealed word‐finding difficulty and right arm weakness. A complete blood count, chemistry panel including renal and liver function tests, urine toxicology screen, and computed tomography (CT) of the head were normal. After a loading dose of fosphenytoin, the patient did not experience further seizures and was discharged on a maintenance dose of phenytoin.

Over the next week, the patient continued to note a sensation of heaviness in her right arm and felt fatigued. The patient's mother brought her back to the emergency department after witnessing a similar seizure episode that persisted for an hour. On arrival, the patient was no longer seizing.

Although it can sometimes be difficult to differentiate between seizure, stroke, syncope, and other causes of transient loss of consciousness, this constellation of symptoms strongly points to a seizure. I would classify the patient's focal arm movements associated with impaired consciousness as partial complex seizures. One of the first considerations is determining whether the seizure is caused by a systemic process or by an intrinsic central nervous system disorder. Common systemic illnesses include infections, metabolic disturbances, toxins, and malignancies, none of which are evident on the preliminary evaluation. The absence of fever is important as is the time frame (now extending over 1 week) in excluding acute bacterial meningitis. A negative urine toxicology is very helpful but does not exclude the possibility that the seizure is from unmeasured drug intoxication, for example, tricyclic antidepressants, or from drug withdrawal, for example, benzodiazepines, barbiturates, ethanol, and antiepileptic drugs. The persistent right arm heaviness and right arm jerking during the seizures suggest a left cortical focus that the CT scan did not detect. Without a clear diagnosis and with recurrent seizures despite antiepileptic drugs, hospitalization is warranted.

The patient experienced migraine headaches each month during menses. There was no family history of seizures. Her only medication was phenytoin. The mother was unaware of any use of tobacco, alcohol, or recreational drugs. The patient was raised in New Jersey and moved to the San Francisco Bay area 9 months ago. She had no pets and had traveled to Florida and Montreal in the past 6 months. She was a graduate student in performing arts. During the preceding 2 weeks she had been under significant stress and had not slept much in preparation for an upcoming production. The patient's mother was not aware of any head trauma, recent illness, fevers, chills, weight loss, photosensitivity, arthralgias, nausea, vomiting, or diarrhea.

Recent sleep deprivation could provoke seizures in a patient with a latent anatomic focus or metabolic predisposition. Nonadherence to antiepileptic drug therapy is the most common reason for patients to present to the ED with seizures; therefore, I would check a phenytoin level to assess whether she is at a therapeutic level and would consider administering another loading dose. In the absence of immunocompromise or unusual activities or exposures, North American travel does not bring to mind additional etiologies at this time.

On exam, temperature was 37.3C, blood pressure was 148/84 mm Hg, heart rate was 120 per minute, and respiratory rate was 16 per minute. The patient was stuporous and withdrew from painful stimuli. She was unable to speak. Pupils were 4 mm in diameter and reacted to light. No gaze preference or nystagmus was present. There was no meningismus. Deep tendon reflexes were 1+ and symmetrical in both upper and lower extremities. Plantar reflexes were extensor bilaterally. The tone in the right upper extremity was mildly increased compared to the left. The patient demonstrated semipurposeful movement of the limbs, such as reaching for the bed rails with her arms. Examination of the heart, lungs, abdomen, skin, and oropharynx was normal.

The white blood cell count was 22,300/mm³ with 50% neutrophils, 40% lymphocytes, 7% monocytes, and 3% eosinophils. Results of the chemistry panel including electrolytes, glucose, creatinine, and liver enzymes, urinalysis, and thyroid‐stimulating hormone were normal. Serum phenytoin level was 8.1 g/mL. Urine toxicology screen, obtained after the patient had received lorazepam, was positive only for benzodiazepines. A chest radiograph was normal.

The cerebrospinal fluid (CSF) was colorless, containing 35 white blood cells/mm³ (48% lymphocytes, 30% neutrophils, 22% monocytes), 3 red blood cells/mm³, 62 mg/dL protein, and 50 mg/dL glucose. There was no xanthochromia. The CSF was negative for cryptococcal antigen, antibodies to West Nile virus, PCR for herpes simplex viruses‐1 and ‐2, and PCR for Borrelia burgdorferi. CSF bacterial culture, cryptococcal antigen, and AFB stain were negative. The serum antinuclear antibody, rheumatoid factor, and rapid plasma reagin were negative. Serum antibodies to human immunodeficiency virus, hepatitis B and C viruses, Borrelia burgdorferi, and herpes simplex viruses were negative. The erythrocyte sedimentation rate was 25 mm/hr. There was no growth in her blood cultures.

These CSF findings have to be interpreted in light of her clinical picture, as they are congruent with both an aseptic meningitis and encephalitis. In practice, these can be hard to distinguish, but the early and dominant cortical findings (focal neurologic deficits, prominent altered mental status, bilateral extensor plantar reflexes) and absence of meningeal signs favor encephalitis. This CSF profile can be seen in a variety of disease processes causing a meningoencephalitis, including partially treated bacterial meningitis; meningitis due to viruses, fungi, mycobacteria, or atypical bacteria (eg, Listeria); neurosarcoidosis; carcinomatous meningitis; and infection or inflammation from a parameningeal focus in the sinuses, epidural space, or brain parenchyma. Seizure itself can lead to a postictal pleocytosis in the CSF, although this degree of inflammation would be unusual. Many tests can be sent, and the clinicians appropriately focused on some of the most treatable and serious etiologies first. The negative HIV test limits the list of opportunistic pathogens. The negative ANA substantially lowers the likelihood of systemic lupus, an important consideration in a young woman with an inflammatory disorder involving the central nervous system.

Magnetic resonance imaging (MRI) of the brain showed cortical T2 prolongation with significant enhancement with gadolinium in the cortex and leptomeninges of the left parietal and posterotemporal lobes and right cingulate gyrus region (Fig. 1). The patient was admitted to the intensive care unit, and phenytoin and levetiracetam were administered. Over the next several days, she remained afebrile, and her leukocytosis resolved. She continued to have seizures every day despite receiving phenytoin, levetiracetam, and lamotrigine. She was alert and complained about persistent right arm weakness and word‐finding difficulties. Posterior cervical lymphadenopathy at the base of her left occiput was detected on subsequent exam.

Figure 1

An excisional lymph node biopsy demonstrated extensive necrosis without evidence of granulomata, malignancy, or lymphoproliferative disease. Stains and cultures for bacteria, fungi, and mycobacteria were negative. The patient's electroencephalogram captured epileptiform activity over the left hemisphere 2 hours after a cluster of seizures. MR angiography and cerebral angiography demonstrated no abnormalities.

Despite this additional information, there is no distinguishing clue that points to a single diagnosis. This is a 26‐year‐old healthy, seemingly immunocompetent woman who has had a 2‐week progressive and refractory seizure disorder secondary to a multifocal neuroinvasive process with a CSF pleocytosis. She does not have evidence of a systemic underlying disorder, save for nonspecific localized lymphadenopathy and a transient episode of leukocytosis on admission, and has no distinguishing epidemiological factors or exposures.

Despite my initial concerns for infectious meningoencephalitis, the negative stains, serologies, and cultures of the blood, CSF, and lymph nodes in the setting of a normal immune system and no suspect exposure substantially lower this probability. Arthropod‐borne viruses are still possible, especially West Nile virus, because the serological tests are less sensitive early in the illness, acknowledging that the absence of fever, weakness, and known mosquito bites detracts from this diagnosis. Pathogens that cause regional lymphadenopathy and encephalitis such as Bartonella remain possibilities, as the history of exposure to a kitten can be easily overlooked.

Rheumatologic disorders merit close attention in a young woman, but the negative ANA makes lupus cerebritis unlikely, and the 2 angiograms did not detect evidence of vasculitis. Finally, there is the question of malignancy and other miscellaneous infiltrative disorders (such as sarcoid), which are of importance here because of the multifocal cortical involvement on imaging.

At this point, I would resample the CSF for viral etiologies (eg, West Nile virus) and cytology and would send serum Bartonella serologies. If these studies were negative, a brain biopsy, primarily to exclude malignancy but also to uncover an unsuspected process, would be indicated. I cannot make a definitive diagnosis or find a perfect fit here, but in the absence of strong evidence of an infection, I am concerned about a malignancy, perhaps a low‐grade primary brain tumor.

Brain biopsy of the leptomeninges and cortex of the left parietal lobe showed multiple blood vessels infiltrated by lymphocytes, neutrophils, and eosinophils (Fig. 2). The pattern of inflammation was consistent with primary angiitis of the central nervous system (PACNS). The patient received 1 g of intravenous methylprednisolone on 3 consecutive days, followed by oral prednisone and cyclophosphamide. The seizures ceased, and she made steady progress with rehabilitation therapy. Four months after discharge a cerebral angiogram (done to ensure there was no interval evidence of vasculitis prior to tapering therapy) demonstrated patency of all major intracranial arteries and venous sinuses.

Figure 2

COMMENTARY

When a patient presents with symptoms or signs referable to the central nervous system (CNS), hospitalists must simultaneously consider primary neurologic disorders and systemic diseases that involve the CNS. Initial evaluation includes a thorough history and physical examination, basic lab studies, routine CSF analysis, and neuroimaging (often a CT scan of the head). Complicated neurologic cases may warrant more elaborate testing including EEG, brain MRI, cerebral angiography, and specialized blood and CSF studies. Clinicians may still find themselves faced with a patient who has clear CNS dysfunction but no obvious diagnosis despite an exhaustive and expensive evaluation. Several disorders match this profile including intravascular lymphoma, prion diseases, paraneoplastic syndromes, and cerebritis. Primary angiitis of the central nervous system (PACNS), a rare disorder characterized by inflammation of the medium‐sized and small arteries of the CNS, is among these disorders. Although the aforementioned diseases may sometimes have suggestive or even pathognomonic features (eg, the string of beads angiographic appearance in vasculitides), they are challenging to diagnose when such findings are absent.

Like any vasculitis of the CNS, PACNS may present with a wide spectrum of clinical features.12 Although headache and altered mental status are the most common complaints, paresis, seizures, ataxia, visual changes, and aphasia have all been described. The onset of symptoms ranges from acute to chronic, and neurologic deficits can be focal or diffuse. Systemic manifestations such as fever and weight loss are rare. The average age of onset is 42 years, with no significant sex preponderance. The histopathology of PACNS is granulomatous inflammation of arteries in the parenchyma and leptomeninges of the brain and less commonly in the spinal cord. The narrowing of the affected vessels causes cerebral ischemia and the associated neurologic deficits. The trigger for this focal inflammation is unknown.

After common disorders have been excluded in cases of CNS dysfunction, compatible CSF findings and imaging results may prompt consideration of PACNS. CSF analysis in patients with PACNS typically demonstrates a lymphocytic pleocytosis. MRI abnormalities in PACNS include multiple infarcts in the cortex, deep white matter, or leptomeninges.34 Less specific findings are contrast enhancement in the leptomeninges and white matter disease, both of which may direct the site for meningeal and brain biopsy.

Both brain MRA and cerebral angiography have a limited role in the diagnosis of vasculitis within the CNS. In 18 patients with CNS vasculitis due to autoimmune disease, all had parenchymal abnormalities on MRA but only 65% had evidence of vasculitis on angiography. In 2 retrospective studies of patients with suspected PACNS, abnormal angiograms had a specificity less than 30% for PACNS, whereas brain biopsies had a negative predictive value of 70%.57 Although in practice patients with compatible clinical features are sometimes diagnosed with CNS vasculitis on the basis of angiographic findings, brain biopsy is necessary to differentiate vasculitis from other vasculopathies and to establish a definitive diagnosis.

Before a diagnosis of PACNS is made, care must be taken to exclude infections, neoplasms, and autoimmune processes that cause angiitis of the CNS (Table 1). The presence of any extracranial abnormalities (which were not present in this case) should prompt consideration of an underlying systemic disorder causing a secondary CNS vasculitis and should cast doubt on the diagnosis of PACNS. Meningovascular syphilis and tuberculosis are among the long list of infections that may cause inflammation of the CNS vasculature. Autoimmune disorders that may cause vasculitis inside the brain include polyarteritis nodosa and Wegener's granulomatosis. Reversible cerebral vasoconstrictive disease, which is most commonly seen in women ages 20 to 50, and sympathomimetic toxins such as cocaine and amphetamine may exhibit clinical and angiographic abnormalities indistinguishable from PACNS.89

There are no prospective trials investigating PACNS treatment. Aggressive immunosuppression with cyclophosphamide and glucocorticoids is the mainstay of treatment. The duration of treatment varies with the severity of the disease and response to therapy. One study suggests that treatment should be continued for 6 to 12 months.10 Neurologic deficits may remain irreversible because of scarring of the affected vessels. Serial brain MRI examinations are often used to follow radiographic resolution during and after the therapy, although radiographic changes do not predict clinical response.11 New abnormalities on MRI, however, delay any tapering of treatment. The availability of neuroimaging studies and immunosuppressive therapy has improved the prognosis of PACNS. One study reported a favorable outcome with a 29% relapse rate and a 10% mortality rate in 54 patients over a mean follow‐up period of 35 months.12

PACNS remains a challenging diagnosis because of its rarity, the wide range of neurologic manifestations, and the difficulty in establishing a diagnosis noninvasively. It is an extremely uncommon disease but should be considered in patients with unexplained neurologic deficits referable to the CNS alone after an exhaustive workup. Ultimately, the diagnosis is made by a thorough history and physical examination, exclusion of underlying conditions (particularly systemic vasculitides and infections), and histological confirmation.

A 26‐year‐old woman was brought to the emergency department following several episodes of seizures. The patient's friend witnessed several 15‐minute episodes of sudden jerks and tremors of her right arm during which the patient bit her tongue, had word‐finding difficulty, had horizontal eye deviation, and was incontinent of urine. She became unresponsive during the episodes, with incomplete recovery of consciousness between attacks. She was afebrile. Her neurologic exam 4 hours after several seizures revealed word‐finding difficulty and right arm weakness. A complete blood count, chemistry panel including renal and liver function tests, urine toxicology screen, and computed tomography (CT) of the head were normal. After a loading dose of fosphenytoin, the patient did not experience further seizures and was discharged on a maintenance dose of phenytoin.

Over the next week, the patient continued to note a sensation of heaviness in her right arm and felt fatigued. The patient's mother brought her back to the emergency department after witnessing a similar seizure episode that persisted for an hour. On arrival, the patient was no longer seizing.

Although it can sometimes be difficult to differentiate between seizure, stroke, syncope, and other causes of transient loss of consciousness, this constellation of symptoms strongly points to a seizure. I would classify the patient's focal arm movements associated with impaired consciousness as partial complex seizures. One of the first considerations is determining whether the seizure is caused by a systemic process or by an intrinsic central nervous system disorder. Common systemic illnesses include infections, metabolic disturbances, toxins, and malignancies, none of which are evident on the preliminary evaluation. The absence of fever is important as is the time frame (now extending over 1 week) in excluding acute bacterial meningitis. A negative urine toxicology is very helpful but does not exclude the possibility that the seizure is from unmeasured drug intoxication, for example, tricyclic antidepressants, or from drug withdrawal, for example, benzodiazepines, barbiturates, ethanol, and antiepileptic drugs. The persistent right arm heaviness and right arm jerking during the seizures suggest a left cortical focus that the CT scan did not detect. Without a clear diagnosis and with recurrent seizures despite antiepileptic drugs, hospitalization is warranted.

The patient experienced migraine headaches each month during menses. There was no family history of seizures. Her only medication was phenytoin. The mother was unaware of any use of tobacco, alcohol, or recreational drugs. The patient was raised in New Jersey and moved to the San Francisco Bay area 9 months ago. She had no pets and had traveled to Florida and Montreal in the past 6 months. She was a graduate student in performing arts. During the preceding 2 weeks she had been under significant stress and had not slept much in preparation for an upcoming production. The patient's mother was not aware of any head trauma, recent illness, fevers, chills, weight loss, photosensitivity, arthralgias, nausea, vomiting, or diarrhea.

Recent sleep deprivation could provoke seizures in a patient with a latent anatomic focus or metabolic predisposition. Nonadherence to antiepileptic drug therapy is the most common reason for patients to present to the ED with seizures; therefore, I would check a phenytoin level to assess whether she is at a therapeutic level and would consider administering another loading dose. In the absence of immunocompromise or unusual activities or exposures, North American travel does not bring to mind additional etiologies at this time.

On exam, temperature was 37.3C, blood pressure was 148/84 mm Hg, heart rate was 120 per minute, and respiratory rate was 16 per minute. The patient was stuporous and withdrew from painful stimuli. She was unable to speak. Pupils were 4 mm in diameter and reacted to light. No gaze preference or nystagmus was present. There was no meningismus. Deep tendon reflexes were 1+ and symmetrical in both upper and lower extremities. Plantar reflexes were extensor bilaterally. The tone in the right upper extremity was mildly increased compared to the left. The patient demonstrated semipurposeful movement of the limbs, such as reaching for the bed rails with her arms. Examination of the heart, lungs, abdomen, skin, and oropharynx was normal.

The white blood cell count was 22,300/mm³ with 50% neutrophils, 40% lymphocytes, 7% monocytes, and 3% eosinophils. Results of the chemistry panel including electrolytes, glucose, creatinine, and liver enzymes, urinalysis, and thyroid‐stimulating hormone were normal. Serum phenytoin level was 8.1 g/mL. Urine toxicology screen, obtained after the patient had received lorazepam, was positive only for benzodiazepines. A chest radiograph was normal.

The cerebrospinal fluid (CSF) was colorless, containing 35 white blood cells/mm³ (48% lymphocytes, 30% neutrophils, 22% monocytes), 3 red blood cells/mm³, 62 mg/dL protein, and 50 mg/dL glucose. There was no xanthochromia. The CSF was negative for cryptococcal antigen, antibodies to West Nile virus, PCR for herpes simplex viruses‐1 and ‐2, and PCR for Borrelia burgdorferi. CSF bacterial culture, cryptococcal antigen, and AFB stain were negative. The serum antinuclear antibody, rheumatoid factor, and rapid plasma reagin were negative. Serum antibodies to human immunodeficiency virus, hepatitis B and C viruses, Borrelia burgdorferi, and herpes simplex viruses were negative. The erythrocyte sedimentation rate was 25 mm/hr. There was no growth in her blood cultures.

These CSF findings have to be interpreted in light of her clinical picture, as they are congruent with both an aseptic meningitis and encephalitis. In practice, these can be hard to distinguish, but the early and dominant cortical findings (focal neurologic deficits, prominent altered mental status, bilateral extensor plantar reflexes) and absence of meningeal signs favor encephalitis. This CSF profile can be seen in a variety of disease processes causing a meningoencephalitis, including partially treated bacterial meningitis; meningitis due to viruses, fungi, mycobacteria, or atypical bacteria (eg, Listeria); neurosarcoidosis; carcinomatous meningitis; and infection or inflammation from a parameningeal focus in the sinuses, epidural space, or brain parenchyma. Seizure itself can lead to a postictal pleocytosis in the CSF, although this degree of inflammation would be unusual. Many tests can be sent, and the clinicians appropriately focused on some of the most treatable and serious etiologies first. The negative HIV test limits the list of opportunistic pathogens. The negative ANA substantially lowers the likelihood of systemic lupus, an important consideration in a young woman with an inflammatory disorder involving the central nervous system.

Magnetic resonance imaging (MRI) of the brain showed cortical T2 prolongation with significant enhancement with gadolinium in the cortex and leptomeninges of the left parietal and posterotemporal lobes and right cingulate gyrus region (Fig. 1). The patient was admitted to the intensive care unit, and phenytoin and levetiracetam were administered. Over the next several days, she remained afebrile, and her leukocytosis resolved. She continued to have seizures every day despite receiving phenytoin, levetiracetam, and lamotrigine. She was alert and complained about persistent right arm weakness and word‐finding difficulties. Posterior cervical lymphadenopathy at the base of her left occiput was detected on subsequent exam.

Figure 1

An excisional lymph node biopsy demonstrated extensive necrosis without evidence of granulomata, malignancy, or lymphoproliferative disease. Stains and cultures for bacteria, fungi, and mycobacteria were negative. The patient's electroencephalogram captured epileptiform activity over the left hemisphere 2 hours after a cluster of seizures. MR angiography and cerebral angiography demonstrated no abnormalities.

Despite this additional information, there is no distinguishing clue that points to a single diagnosis. This is a 26‐year‐old healthy, seemingly immunocompetent woman who has had a 2‐week progressive and refractory seizure disorder secondary to a multifocal neuroinvasive process with a CSF pleocytosis. She does not have evidence of a systemic underlying disorder, save for nonspecific localized lymphadenopathy and a transient episode of leukocytosis on admission, and has no distinguishing epidemiological factors or exposures.

Despite my initial concerns for infectious meningoencephalitis, the negative stains, serologies, and cultures of the blood, CSF, and lymph nodes in the setting of a normal immune system and no suspect exposure substantially lower this probability. Arthropod‐borne viruses are still possible, especially West Nile virus, because the serological tests are less sensitive early in the illness, acknowledging that the absence of fever, weakness, and known mosquito bites detracts from this diagnosis. Pathogens that cause regional lymphadenopathy and encephalitis such as Bartonella remain possibilities, as the history of exposure to a kitten can be easily overlooked.

Rheumatologic disorders merit close attention in a young woman, but the negative ANA makes lupus cerebritis unlikely, and the 2 angiograms did not detect evidence of vasculitis. Finally, there is the question of malignancy and other miscellaneous infiltrative disorders (such as sarcoid), which are of importance here because of the multifocal cortical involvement on imaging.

At this point, I would resample the CSF for viral etiologies (eg, West Nile virus) and cytology and would send serum Bartonella serologies. If these studies were negative, a brain biopsy, primarily to exclude malignancy but also to uncover an unsuspected process, would be indicated. I cannot make a definitive diagnosis or find a perfect fit here, but in the absence of strong evidence of an infection, I am concerned about a malignancy, perhaps a low‐grade primary brain tumor.

Brain biopsy of the leptomeninges and cortex of the left parietal lobe showed multiple blood vessels infiltrated by lymphocytes, neutrophils, and eosinophils (Fig. 2). The pattern of inflammation was consistent with primary angiitis of the central nervous system (PACNS). The patient received 1 g of intravenous methylprednisolone on 3 consecutive days, followed by oral prednisone and cyclophosphamide. The seizures ceased, and she made steady progress with rehabilitation therapy. Four months after discharge a cerebral angiogram (done to ensure there was no interval evidence of vasculitis prior to tapering therapy) demonstrated patency of all major intracranial arteries and venous sinuses.

Figure 2

COMMENTARY

When a patient presents with symptoms or signs referable to the central nervous system (CNS), hospitalists must simultaneously consider primary neurologic disorders and systemic diseases that involve the CNS. Initial evaluation includes a thorough history and physical examination, basic lab studies, routine CSF analysis, and neuroimaging (often a CT scan of the head). Complicated neurologic cases may warrant more elaborate testing including EEG, brain MRI, cerebral angiography, and specialized blood and CSF studies. Clinicians may still find themselves faced with a patient who has clear CNS dysfunction but no obvious diagnosis despite an exhaustive and expensive evaluation. Several disorders match this profile including intravascular lymphoma, prion diseases, paraneoplastic syndromes, and cerebritis. Primary angiitis of the central nervous system (PACNS), a rare disorder characterized by inflammation of the medium‐sized and small arteries of the CNS, is among these disorders. Although the aforementioned diseases may sometimes have suggestive or even pathognomonic features (eg, the string of beads angiographic appearance in vasculitides), they are challenging to diagnose when such findings are absent.

Like any vasculitis of the CNS, PACNS may present with a wide spectrum of clinical features.12 Although headache and altered mental status are the most common complaints, paresis, seizures, ataxia, visual changes, and aphasia have all been described. The onset of symptoms ranges from acute to chronic, and neurologic deficits can be focal or diffuse. Systemic manifestations such as fever and weight loss are rare. The average age of onset is 42 years, with no significant sex preponderance. The histopathology of PACNS is granulomatous inflammation of arteries in the parenchyma and leptomeninges of the brain and less commonly in the spinal cord. The narrowing of the affected vessels causes cerebral ischemia and the associated neurologic deficits. The trigger for this focal inflammation is unknown.

After common disorders have been excluded in cases of CNS dysfunction, compatible CSF findings and imaging results may prompt consideration of PACNS. CSF analysis in patients with PACNS typically demonstrates a lymphocytic pleocytosis. MRI abnormalities in PACNS include multiple infarcts in the cortex, deep white matter, or leptomeninges.34 Less specific findings are contrast enhancement in the leptomeninges and white matter disease, both of which may direct the site for meningeal and brain biopsy.

Both brain MRA and cerebral angiography have a limited role in the diagnosis of vasculitis within the CNS. In 18 patients with CNS vasculitis due to autoimmune disease, all had parenchymal abnormalities on MRA but only 65% had evidence of vasculitis on angiography. In 2 retrospective studies of patients with suspected PACNS, abnormal angiograms had a specificity less than 30% for PACNS, whereas brain biopsies had a negative predictive value of 70%.57 Although in practice patients with compatible clinical features are sometimes diagnosed with CNS vasculitis on the basis of angiographic findings, brain biopsy is necessary to differentiate vasculitis from other vasculopathies and to establish a definitive diagnosis.

Before a diagnosis of PACNS is made, care must be taken to exclude infections, neoplasms, and autoimmune processes that cause angiitis of the CNS (Table 1). The presence of any extracranial abnormalities (which were not present in this case) should prompt consideration of an underlying systemic disorder causing a secondary CNS vasculitis and should cast doubt on the diagnosis of PACNS. Meningovascular syphilis and tuberculosis are among the long list of infections that may cause inflammation of the CNS vasculature. Autoimmune disorders that may cause vasculitis inside the brain include polyarteritis nodosa and Wegener's granulomatosis. Reversible cerebral vasoconstrictive disease, which is most commonly seen in women ages 20 to 50, and sympathomimetic toxins such as cocaine and amphetamine may exhibit clinical and angiographic abnormalities indistinguishable from PACNS.89

There are no prospective trials investigating PACNS treatment. Aggressive immunosuppression with cyclophosphamide and glucocorticoids is the mainstay of treatment. The duration of treatment varies with the severity of the disease and response to therapy. One study suggests that treatment should be continued for 6 to 12 months.10 Neurologic deficits may remain irreversible because of scarring of the affected vessels. Serial brain MRI examinations are often used to follow radiographic resolution during and after the therapy, although radiographic changes do not predict clinical response.11 New abnormalities on MRI, however, delay any tapering of treatment. The availability of neuroimaging studies and immunosuppressive therapy has improved the prognosis of PACNS. One study reported a favorable outcome with a 29% relapse rate and a 10% mortality rate in 54 patients over a mean follow‐up period of 35 months.12

PACNS remains a challenging diagnosis because of its rarity, the wide range of neurologic manifestations, and the difficulty in establishing a diagnosis noninvasively. It is an extremely uncommon disease but should be considered in patients with unexplained neurologic deficits referable to the CNS alone after an exhaustive workup. Ultimately, the diagnosis is made by a thorough history and physical examination, exclusion of underlying conditions (particularly systemic vasculitides and infections), and histological confirmation.

A 26‐year‐old woman was brought to the emergency department following several episodes of seizures. The patient's friend witnessed several 15‐minute episodes of sudden jerks and tremors of her right arm during which the patient bit her tongue, had word‐finding difficulty, had horizontal eye deviation, and was incontinent of urine. She became unresponsive during the episodes, with incomplete recovery of consciousness between attacks. She was afebrile. Her neurologic exam 4 hours after several seizures revealed word‐finding difficulty and right arm weakness. A complete blood count, chemistry panel including renal and liver function tests, urine toxicology screen, and computed tomography (CT) of the head were normal. After a loading dose of fosphenytoin, the patient did not experience further seizures and was discharged on a maintenance dose of phenytoin.

Over the next week, the patient continued to note a sensation of heaviness in her right arm and felt fatigued. The patient's mother brought her back to the emergency department after witnessing a similar seizure episode that persisted for an hour. On arrival, the patient was no longer seizing.

Although it can sometimes be difficult to differentiate between seizure, stroke, syncope, and other causes of transient loss of consciousness, this constellation of symptoms strongly points to a seizure. I would classify the patient's focal arm movements associated with impaired consciousness as partial complex seizures. One of the first considerations is determining whether the seizure is caused by a systemic process or by an intrinsic central nervous system disorder. Common systemic illnesses include infections, metabolic disturbances, toxins, and malignancies, none of which are evident on the preliminary evaluation. The absence of fever is important as is the time frame (now extending over 1 week) in excluding acute bacterial meningitis. A negative urine toxicology is very helpful but does not exclude the possibility that the seizure is from unmeasured drug intoxication, for example, tricyclic antidepressants, or from drug withdrawal, for example, benzodiazepines, barbiturates, ethanol, and antiepileptic drugs. The persistent right arm heaviness and right arm jerking during the seizures suggest a left cortical focus that the CT scan did not detect. Without a clear diagnosis and with recurrent seizures despite antiepileptic drugs, hospitalization is warranted.

The patient experienced migraine headaches each month during menses. There was no family history of seizures. Her only medication was phenytoin. The mother was unaware of any use of tobacco, alcohol, or recreational drugs. The patient was raised in New Jersey and moved to the San Francisco Bay area 9 months ago. She had no pets and had traveled to Florida and Montreal in the past 6 months. She was a graduate student in performing arts. During the preceding 2 weeks she had been under significant stress and had not slept much in preparation for an upcoming production. The patient's mother was not aware of any head trauma, recent illness, fevers, chills, weight loss, photosensitivity, arthralgias, nausea, vomiting, or diarrhea.

Recent sleep deprivation could provoke seizures in a patient with a latent anatomic focus or metabolic predisposition. Nonadherence to antiepileptic drug therapy is the most common reason for patients to present to the ED with seizures; therefore, I would check a phenytoin level to assess whether she is at a therapeutic level and would consider administering another loading dose. In the absence of immunocompromise or unusual activities or exposures, North American travel does not bring to mind additional etiologies at this time.

On exam, temperature was 37.3C, blood pressure was 148/84 mm Hg, heart rate was 120 per minute, and respiratory rate was 16 per minute. The patient was stuporous and withdrew from painful stimuli. She was unable to speak. Pupils were 4 mm in diameter and reacted to light. No gaze preference or nystagmus was present. There was no meningismus. Deep tendon reflexes were 1+ and symmetrical in both upper and lower extremities. Plantar reflexes were extensor bilaterally. The tone in the right upper extremity was mildly increased compared to the left. The patient demonstrated semipurposeful movement of the limbs, such as reaching for the bed rails with her arms. Examination of the heart, lungs, abdomen, skin, and oropharynx was normal.

The white blood cell count was 22,300/mm³ with 50% neutrophils, 40% lymphocytes, 7% monocytes, and 3% eosinophils. Results of the chemistry panel including electrolytes, glucose, creatinine, and liver enzymes, urinalysis, and thyroid‐stimulating hormone were normal. Serum phenytoin level was 8.1 g/mL. Urine toxicology screen, obtained after the patient had received lorazepam, was positive only for benzodiazepines. A chest radiograph was normal.

The cerebrospinal fluid (CSF) was colorless, containing 35 white blood cells/mm³ (48% lymphocytes, 30% neutrophils, 22% monocytes), 3 red blood cells/mm³, 62 mg/dL protein, and 50 mg/dL glucose. There was no xanthochromia. The CSF was negative for cryptococcal antigen, antibodies to West Nile virus, PCR for herpes simplex viruses‐1 and ‐2, and PCR for Borrelia burgdorferi. CSF bacterial culture, cryptococcal antigen, and AFB stain were negative. The serum antinuclear antibody, rheumatoid factor, and rapid plasma reagin were negative. Serum antibodies to human immunodeficiency virus, hepatitis B and C viruses, Borrelia burgdorferi, and herpes simplex viruses were negative. The erythrocyte sedimentation rate was 25 mm/hr. There was no growth in her blood cultures.

These CSF findings have to be interpreted in light of her clinical picture, as they are congruent with both an aseptic meningitis and encephalitis. In practice, these can be hard to distinguish, but the early and dominant cortical findings (focal neurologic deficits, prominent altered mental status, bilateral extensor plantar reflexes) and absence of meningeal signs favor encephalitis. This CSF profile can be seen in a variety of disease processes causing a meningoencephalitis, including partially treated bacterial meningitis; meningitis due to viruses, fungi, mycobacteria, or atypical bacteria (eg, Listeria); neurosarcoidosis; carcinomatous meningitis; and infection or inflammation from a parameningeal focus in the sinuses, epidural space, or brain parenchyma. Seizure itself can lead to a postictal pleocytosis in the CSF, although this degree of inflammation would be unusual. Many tests can be sent, and the clinicians appropriately focused on some of the most treatable and serious etiologies first. The negative HIV test limits the list of opportunistic pathogens. The negative ANA substantially lowers the likelihood of systemic lupus, an important consideration in a young woman with an inflammatory disorder involving the central nervous system.

Magnetic resonance imaging (MRI) of the brain showed cortical T2 prolongation with significant enhancement with gadolinium in the cortex and leptomeninges of the left parietal and posterotemporal lobes and right cingulate gyrus region (Fig. 1). The patient was admitted to the intensive care unit, and phenytoin and levetiracetam were administered. Over the next several days, she remained afebrile, and her leukocytosis resolved. She continued to have seizures every day despite receiving phenytoin, levetiracetam, and lamotrigine. She was alert and complained about persistent right arm weakness and word‐finding difficulties. Posterior cervical lymphadenopathy at the base of her left occiput was detected on subsequent exam.

Figure 1

An excisional lymph node biopsy demonstrated extensive necrosis without evidence of granulomata, malignancy, or lymphoproliferative disease. Stains and cultures for bacteria, fungi, and mycobacteria were negative. The patient's electroencephalogram captured epileptiform activity over the left hemisphere 2 hours after a cluster of seizures. MR angiography and cerebral angiography demonstrated no abnormalities.

Despite this additional information, there is no distinguishing clue that points to a single diagnosis. This is a 26‐year‐old healthy, seemingly immunocompetent woman who has had a 2‐week progressive and refractory seizure disorder secondary to a multifocal neuroinvasive process with a CSF pleocytosis. She does not have evidence of a systemic underlying disorder, save for nonspecific localized lymphadenopathy and a transient episode of leukocytosis on admission, and has no distinguishing epidemiological factors or exposures.

Despite my initial concerns for infectious meningoencephalitis, the negative stains, serologies, and cultures of the blood, CSF, and lymph nodes in the setting of a normal immune system and no suspect exposure substantially lower this probability. Arthropod‐borne viruses are still possible, especially West Nile virus, because the serological tests are less sensitive early in the illness, acknowledging that the absence of fever, weakness, and known mosquito bites detracts from this diagnosis. Pathogens that cause regional lymphadenopathy and encephalitis such as Bartonella remain possibilities, as the history of exposure to a kitten can be easily overlooked.

Rheumatologic disorders merit close attention in a young woman, but the negative ANA makes lupus cerebritis unlikely, and the 2 angiograms did not detect evidence of vasculitis. Finally, there is the question of malignancy and other miscellaneous infiltrative disorders (such as sarcoid), which are of importance here because of the multifocal cortical involvement on imaging.

At this point, I would resample the CSF for viral etiologies (eg, West Nile virus) and cytology and would send serum Bartonella serologies. If these studies were negative, a brain biopsy, primarily to exclude malignancy but also to uncover an unsuspected process, would be indicated. I cannot make a definitive diagnosis or find a perfect fit here, but in the absence of strong evidence of an infection, I am concerned about a malignancy, perhaps a low‐grade primary brain tumor.

Brain biopsy of the leptomeninges and cortex of the left parietal lobe showed multiple blood vessels infiltrated by lymphocytes, neutrophils, and eosinophils (Fig. 2). The pattern of inflammation was consistent with primary angiitis of the central nervous system (PACNS). The patient received 1 g of intravenous methylprednisolone on 3 consecutive days, followed by oral prednisone and cyclophosphamide. The seizures ceased, and she made steady progress with rehabilitation therapy. Four months after discharge a cerebral angiogram (done to ensure there was no interval evidence of vasculitis prior to tapering therapy) demonstrated patency of all major intracranial arteries and venous sinuses.

Figure 2

COMMENTARY

When a patient presents with symptoms or signs referable to the central nervous system (CNS), hospitalists must simultaneously consider primary neurologic disorders and systemic diseases that involve the CNS. Initial evaluation includes a thorough history and physical examination, basic lab studies, routine CSF analysis, and neuroimaging (often a CT scan of the head). Complicated neurologic cases may warrant more elaborate testing including EEG, brain MRI, cerebral angiography, and specialized blood and CSF studies. Clinicians may still find themselves faced with a patient who has clear CNS dysfunction but no obvious diagnosis despite an exhaustive and expensive evaluation. Several disorders match this profile including intravascular lymphoma, prion diseases, paraneoplastic syndromes, and cerebritis. Primary angiitis of the central nervous system (PACNS), a rare disorder characterized by inflammation of the medium‐sized and small arteries of the CNS, is among these disorders. Although the aforementioned diseases may sometimes have suggestive or even pathognomonic features (eg, the string of beads angiographic appearance in vasculitides), they are challenging to diagnose when such findings are absent.

Like any vasculitis of the CNS, PACNS may present with a wide spectrum of clinical features.12 Although headache and altered mental status are the most common complaints, paresis, seizures, ataxia, visual changes, and aphasia have all been described. The onset of symptoms ranges from acute to chronic, and neurologic deficits can be focal or diffuse. Systemic manifestations such as fever and weight loss are rare. The average age of onset is 42 years, with no significant sex preponderance. The histopathology of PACNS is granulomatous inflammation of arteries in the parenchyma and leptomeninges of the brain and less commonly in the spinal cord. The narrowing of the affected vessels causes cerebral ischemia and the associated neurologic deficits. The trigger for this focal inflammation is unknown.

After common disorders have been excluded in cases of CNS dysfunction, compatible CSF findings and imaging results may prompt consideration of PACNS. CSF analysis in patients with PACNS typically demonstrates a lymphocytic pleocytosis. MRI abnormalities in PACNS include multiple infarcts in the cortex, deep white matter, or leptomeninges.34 Less specific findings are contrast enhancement in the leptomeninges and white matter disease, both of which may direct the site for meningeal and brain biopsy.

Both brain MRA and cerebral angiography have a limited role in the diagnosis of vasculitis within the CNS. In 18 patients with CNS vasculitis due to autoimmune disease, all had parenchymal abnormalities on MRA but only 65% had evidence of vasculitis on angiography. In 2 retrospective studies of patients with suspected PACNS, abnormal angiograms had a specificity less than 30% for PACNS, whereas brain biopsies had a negative predictive value of 70%.57 Although in practice patients with compatible clinical features are sometimes diagnosed with CNS vasculitis on the basis of angiographic findings, brain biopsy is necessary to differentiate vasculitis from other vasculopathies and to establish a definitive diagnosis.

Before a diagnosis of PACNS is made, care must be taken to exclude infections, neoplasms, and autoimmune processes that cause angiitis of the CNS (Table 1). The presence of any extracranial abnormalities (which were not present in this case) should prompt consideration of an underlying systemic disorder causing a secondary CNS vasculitis and should cast doubt on the diagnosis of PACNS. Meningovascular syphilis and tuberculosis are among the long list of infections that may cause inflammation of the CNS vasculature. Autoimmune disorders that may cause vasculitis inside the brain include polyarteritis nodosa and Wegener's granulomatosis. Reversible cerebral vasoconstrictive disease, which is most commonly seen in women ages 20 to 50, and sympathomimetic toxins such as cocaine and amphetamine may exhibit clinical and angiographic abnormalities indistinguishable from PACNS.89

There are no prospective trials investigating PACNS treatment. Aggressive immunosuppression with cyclophosphamide and glucocorticoids is the mainstay of treatment. The duration of treatment varies with the severity of the disease and response to therapy. One study suggests that treatment should be continued for 6 to 12 months.10 Neurologic deficits may remain irreversible because of scarring of the affected vessels. Serial brain MRI examinations are often used to follow radiographic resolution during and after the therapy, although radiographic changes do not predict clinical response.11 New abnormalities on MRI, however, delay any tapering of treatment. The availability of neuroimaging studies and immunosuppressive therapy has improved the prognosis of PACNS. One study reported a favorable outcome with a 29% relapse rate and a 10% mortality rate in 54 patients over a mean follow‐up period of 35 months.12

PACNS remains a challenging diagnosis because of its rarity, the wide range of neurologic manifestations, and the difficulty in establishing a diagnosis noninvasively. It is an extremely uncommon disease but should be considered in patients with unexplained neurologic deficits referable to the CNS alone after an exhaustive workup. Ultimately, the diagnosis is made by a thorough history and physical examination, exclusion of underlying conditions (particularly systemic vasculitides and infections), and histological confirmation.

Many hospitalists work with clinical coordinators and case managers.13 The descriptions of these roles often overlap4 and commonly include activities such as obtaining medical records, expediting tests and procedures, coordinating the plan of care with other health care providers, assessing postdischarge needs, completing discharge paperwork, and arranging follow‐up visits.2, 5, 6 Despite the potential to improve patient care and hospital efficiency, few studies have formally evaluated the impact of these roles. Moher et al. found that adding a clinical coordinator to a general medical team decreased length of stay (LOS) and improved patient satisfaction.5 However, this study was conducted at a time when the LOS was routinely longer than it is today. Forster et al. found that adding a clinical coordinator to a general medical team resulted in improved patient satisfaction but did not reduce length of stay or risk of adverse events occurring following hospital discharge.6 Both these studies evaluated the impact of adding a clinical coordinator to resident‐covered medical teams. Yet many hospitalists deliver care without residents, limiting the generalizability of the findings from these studies.

To date, no studies have evaluated the impact of clinical coordinators, case managers, or other nonphysician providers on the hospitalist work experience. This is surprising, as hospital medicine group leaders list daily workload and work hours among their top concerns.7 Clinical coordinators have the potential to improve patient care and hospital efficiency while simultaneously improving the experience of the hospitalists with whom they work. We conducted this study to evaluate the impact of a hospitalistcare coordinator team on hospitalist work experience, patient satisfaction, and hospital efficiency.

METHODS

RESULTS

There were 356 patients cared for by hospitalistHCC teams and 337 patients cared for by control hospitalists. Of the 60 weeks of hospitalist service of the study, hospitalistHCC teams accounted for 31 weeks (52%) and control hospitalists for 29 weeks (48%). Patients cared for by the hospitalistHCC teams were similar in age, sex, ethnicity, payer type, and DRG weight to those cared for by control hospitalists (see Table 2).

Sixty surveys were completed by hospitalists at the end of their week on service (response rate 100%). Of the 31 responses from hospitalists completing a hospitalistHCC team week, 28 (90%) reported that working with an HCC improved their efficiency and 28 (90%) that working with an HCC improved their job satisfaction. The hospitalists indicated that working with an HCC significantly improved the efficiency of most of their activities (see Table 3). Specifically, activities related to communication with nurses and patients and activities involving discharge planning and execution were improved with the use of an HCC. As would be expected, certain other activities did not improve. For example, there were no differences between the groups in the perceived efficiency of performing histories and physicals or placing admission orders. For activities that were significantly different, the Wilcoxon rank sum test and linear regression analysis adjusting for physician clustering showed identical results.

Seventy‐one of 196 eligible patients (36%) completed the discharge satisfaction interview. Of the 71 patients interviewed, 44 (62%) were cared for by hospitalistHCC teams and 27 (38%) were cared for by control hospitalists. Patient satisfaction with the clarity of the verbal and written discharge instructions and overall satisfaction with hospital discharge was similar between the 2 groups (see Table 4).

The unadjusted mean LOS for patients cared for by hospitalistHCC teams was 4.70 4.15 days compared with 5.07 3.99 days for patients cared for by control hospitalists (P = .005; see Table 5). The unadjusted mean cost for patients cared for by hospitalistHCC teams was $10,052.96 $11,708.73 compared with $11,703.19 $20,455.78 for patients cared for by control hospitalists (P = .008). In multivariate analysis using age, sex, ethnicity, payer type, and DRG weight as independent variables and adjusting for physician clustering, LOS remained lower for patients cared for by hospitalistHCC teams; however, this result was not statistically significant (0.28 days, P = .17). Similar multivariate regression analysis showed a trend toward lower cost for patients cared for by the hospitalistHCC teams (585.62, P = .15).

DISCUSSION

Our study found that hospitalists working in a team approach with an HCC rated the efficiency of their daily work and their job satisfaction significantly higher than did control hospitalists. Specific areas of improved efficiency included communication activities and activities related to hospital discharge. A prior study conducted by our group found that hospitalists spend a lot of time on indirect patient care activities such as communication and activities related to the discharge process, while spending relatively little time on direct patient care.8 Improving the efficiency of indirect patient care activities of hospitalists is likely to improve their job satisfaction. The importance of improving hospitalist workload and job satisfaction is underscored by the relatively high number of hospitalists at risk for burnout9 and the growing concern about daily workload among hospital medicine group leaders.7

Patient satisfaction was not significantly affected by the use of the hospitalistHCC team in our study. A priori, we postulated that patient satisfaction with the discharge process might improve with the use of the hospitalistHCC team. We therefore limited survey questions to assessing only satisfaction with hospital discharge rather than other aspects of patient hospital care. A recent study reported that patients rated the quality of discharge instructions significantly lower than they rated the overall quality of their hospital stay.10 However, the patients in our study gave high ratings to both discharge instructions and overall satisfaction with hospital discharge. This may explain why we were unable to detect a difference. Our study was limited by the relatively small number of patients we were able to contact to assess satisfaction. Previous studies evaluating the impact of care coordinators either did not assess patient satisfaction with discharge5 or found no difference in satisfaction with hospital discharge.6

Although our study did not find a difference in patient satisfaction with the discharge process, we believe the hospitalistHCC model has the potential to complement efforts to reduce the risk of adverse events as patients transition out of the hospital. It has been reported that 12% of patients have a preventable or ameliorable adverse event in the period immediately following hospital discharge.11, 12 Although Forster et al. did not find a reduction in the risk of adverse events with the addition of a clinical coordinator to a general medical team, they noted incongruence between the coordinator's role and the outcomes measured.6 Similarly, we would need to modify the role of the HCC from a position designed mainly to improve efficiency to one that complements efforts to improve the quality of the discharge process. Possible ways to enhance the HCC role in this regard include increasing the emphasis on and training in patient education skills. Several recently published articles have emphasized the need to redesign the discharge process in an effort to reduce the risk of adverse events following hospital discharge.1315 A modified HCC role might be an essential feature of a redesigned multidisciplinary discharge process.

We were unable to demonstrate improved efficiency for the hospital. Although LOS and cost were lower for patients cared for by the hospitalistHCC teams, the difference was not statistically significant. One possible explanation for why we did not observe a larger reduction in LOS is that our hospitalist service had a lower‐than‐average patient volume during the study period. The lower volume mirrored an unanticipated dip in hospital volume during the same period. Specifically, our service normally discharges an average of 338 patients per month, but during the study period we discharged an average of 235 patients per month. A potential LOS and cost benefit may have been attenuated by the relatively low volume, as hospitalists had ample time to dedicate to communication and coordination of discharge plans.

Our study had several limitations. It was conducted on a nonteaching hospitalist service at a single site. Hospitalist practices vary widely in their staffing and scheduling models. As previously mentioned, we were only able to perform patient satisfaction surveys during the second half of the study period. In addition, hospitalistHCC team patients made up a larger percentage of the patient survey responses (62%) than did control hospitalist patients (38%). This may have affected our ability to detect differences in satisfaction with the hospital discharge process. As also previously noted, our patient volume was lower than normal during the study period. We believe that a higher volume would have magnified differences in hospitalists' perceived efficiency and perhaps resulted in significant improvements in LOS and cost. Finally, the hospital provided funding for only a 12‐week study. This limited our sample size and the power of the study to detect important differences. It is possible that a larger sample size and/or longer study period may have been able to demonstrate a statistically significant improvement in LOS and cost.

Our findings are of particular importance in light of the persistent concerns about hospitalist workload and job satisfaction. Although many hospitalists work with clinical coordinators and case managers, we believe that having the formal structure of a hospitalistcare coordinator team was the key element to improving hospitalist efficiency and satisfaction. We hope that our study is a precursor to research evaluating models of delivering hospital care and their impact on hospitalist work experience, hospital efficiency, and patient outcomes.

Many hospitalists work with clinical coordinators and case managers.13 The descriptions of these roles often overlap4 and commonly include activities such as obtaining medical records, expediting tests and procedures, coordinating the plan of care with other health care providers, assessing postdischarge needs, completing discharge paperwork, and arranging follow‐up visits.2, 5, 6 Despite the potential to improve patient care and hospital efficiency, few studies have formally evaluated the impact of these roles. Moher et al. found that adding a clinical coordinator to a general medical team decreased length of stay (LOS) and improved patient satisfaction.5 However, this study was conducted at a time when the LOS was routinely longer than it is today. Forster et al. found that adding a clinical coordinator to a general medical team resulted in improved patient satisfaction but did not reduce length of stay or risk of adverse events occurring following hospital discharge.6 Both these studies evaluated the impact of adding a clinical coordinator to resident‐covered medical teams. Yet many hospitalists deliver care without residents, limiting the generalizability of the findings from these studies.

To date, no studies have evaluated the impact of clinical coordinators, case managers, or other nonphysician providers on the hospitalist work experience. This is surprising, as hospital medicine group leaders list daily workload and work hours among their top concerns.7 Clinical coordinators have the potential to improve patient care and hospital efficiency while simultaneously improving the experience of the hospitalists with whom they work. We conducted this study to evaluate the impact of a hospitalistcare coordinator team on hospitalist work experience, patient satisfaction, and hospital efficiency.

METHODS

RESULTS

There were 356 patients cared for by hospitalistHCC teams and 337 patients cared for by control hospitalists. Of the 60 weeks of hospitalist service of the study, hospitalistHCC teams accounted for 31 weeks (52%) and control hospitalists for 29 weeks (48%). Patients cared for by the hospitalistHCC teams were similar in age, sex, ethnicity, payer type, and DRG weight to those cared for by control hospitalists (see Table 2).

Sixty surveys were completed by hospitalists at the end of their week on service (response rate 100%). Of the 31 responses from hospitalists completing a hospitalistHCC team week, 28 (90%) reported that working with an HCC improved their efficiency and 28 (90%) that working with an HCC improved their job satisfaction. The hospitalists indicated that working with an HCC significantly improved the efficiency of most of their activities (see Table 3). Specifically, activities related to communication with nurses and patients and activities involving discharge planning and execution were improved with the use of an HCC. As would be expected, certain other activities did not improve. For example, there were no differences between the groups in the perceived efficiency of performing histories and physicals or placing admission orders. For activities that were significantly different, the Wilcoxon rank sum test and linear regression analysis adjusting for physician clustering showed identical results.

Seventy‐one of 196 eligible patients (36%) completed the discharge satisfaction interview. Of the 71 patients interviewed, 44 (62%) were cared for by hospitalistHCC teams and 27 (38%) were cared for by control hospitalists. Patient satisfaction with the clarity of the verbal and written discharge instructions and overall satisfaction with hospital discharge was similar between the 2 groups (see Table 4).

The unadjusted mean LOS for patients cared for by hospitalistHCC teams was 4.70 4.15 days compared with 5.07 3.99 days for patients cared for by control hospitalists (P = .005; see Table 5). The unadjusted mean cost for patients cared for by hospitalistHCC teams was $10,052.96 $11,708.73 compared with $11,703.19 $20,455.78 for patients cared for by control hospitalists (P = .008). In multivariate analysis using age, sex, ethnicity, payer type, and DRG weight as independent variables and adjusting for physician clustering, LOS remained lower for patients cared for by hospitalistHCC teams; however, this result was not statistically significant (0.28 days, P = .17). Similar multivariate regression analysis showed a trend toward lower cost for patients cared for by the hospitalistHCC teams (585.62, P = .15).

DISCUSSION

Our study found that hospitalists working in a team approach with an HCC rated the efficiency of their daily work and their job satisfaction significantly higher than did control hospitalists. Specific areas of improved efficiency included communication activities and activities related to hospital discharge. A prior study conducted by our group found that hospitalists spend a lot of time on indirect patient care activities such as communication and activities related to the discharge process, while spending relatively little time on direct patient care.8 Improving the efficiency of indirect patient care activities of hospitalists is likely to improve their job satisfaction. The importance of improving hospitalist workload and job satisfaction is underscored by the relatively high number of hospitalists at risk for burnout9 and the growing concern about daily workload among hospital medicine group leaders.7

Patient satisfaction was not significantly affected by the use of the hospitalistHCC team in our study. A priori, we postulated that patient satisfaction with the discharge process might improve with the use of the hospitalistHCC team. We therefore limited survey questions to assessing only satisfaction with hospital discharge rather than other aspects of patient hospital care. A recent study reported that patients rated the quality of discharge instructions significantly lower than they rated the overall quality of their hospital stay.10 However, the patients in our study gave high ratings to both discharge instructions and overall satisfaction with hospital discharge. This may explain why we were unable to detect a difference. Our study was limited by the relatively small number of patients we were able to contact to assess satisfaction. Previous studies evaluating the impact of care coordinators either did not assess patient satisfaction with discharge5 or found no difference in satisfaction with hospital discharge.6

Although our study did not find a difference in patient satisfaction with the discharge process, we believe the hospitalistHCC model has the potential to complement efforts to reduce the risk of adverse events as patients transition out of the hospital. It has been reported that 12% of patients have a preventable or ameliorable adverse event in the period immediately following hospital discharge.11, 12 Although Forster et al. did not find a reduction in the risk of adverse events with the addition of a clinical coordinator to a general medical team, they noted incongruence between the coordinator's role and the outcomes measured.6 Similarly, we would need to modify the role of the HCC from a position designed mainly to improve efficiency to one that complements efforts to improve the quality of the discharge process. Possible ways to enhance the HCC role in this regard include increasing the emphasis on and training in patient education skills. Several recently published articles have emphasized the need to redesign the discharge process in an effort to reduce the risk of adverse events following hospital discharge.1315 A modified HCC role might be an essential feature of a redesigned multidisciplinary discharge process.

We were unable to demonstrate improved efficiency for the hospital. Although LOS and cost were lower for patients cared for by the hospitalistHCC teams, the difference was not statistically significant. One possible explanation for why we did not observe a larger reduction in LOS is that our hospitalist service had a lower‐than‐average patient volume during the study period. The lower volume mirrored an unanticipated dip in hospital volume during the same period. Specifically, our service normally discharges an average of 338 patients per month, but during the study period we discharged an average of 235 patients per month. A potential LOS and cost benefit may have been attenuated by the relatively low volume, as hospitalists had ample time to dedicate to communication and coordination of discharge plans.

Our study had several limitations. It was conducted on a nonteaching hospitalist service at a single site. Hospitalist practices vary widely in their staffing and scheduling models. As previously mentioned, we were only able to perform patient satisfaction surveys during the second half of the study period. In addition, hospitalistHCC team patients made up a larger percentage of the patient survey responses (62%) than did control hospitalist patients (38%). This may have affected our ability to detect differences in satisfaction with the hospital discharge process. As also previously noted, our patient volume was lower than normal during the study period. We believe that a higher volume would have magnified differences in hospitalists' perceived efficiency and perhaps resulted in significant improvements in LOS and cost. Finally, the hospital provided funding for only a 12‐week study. This limited our sample size and the power of the study to detect important differences. It is possible that a larger sample size and/or longer study period may have been able to demonstrate a statistically significant improvement in LOS and cost.

Our findings are of particular importance in light of the persistent concerns about hospitalist workload and job satisfaction. Although many hospitalists work with clinical coordinators and case managers, we believe that having the formal structure of a hospitalistcare coordinator team was the key element to improving hospitalist efficiency and satisfaction. We hope that our study is a precursor to research evaluating models of delivering hospital care and their impact on hospitalist work experience, hospital efficiency, and patient outcomes.

Many hospitalists work with clinical coordinators and case managers.13 The descriptions of these roles often overlap4 and commonly include activities such as obtaining medical records, expediting tests and procedures, coordinating the plan of care with other health care providers, assessing postdischarge needs, completing discharge paperwork, and arranging follow‐up visits.2, 5, 6 Despite the potential to improve patient care and hospital efficiency, few studies have formally evaluated the impact of these roles. Moher et al. found that adding a clinical coordinator to a general medical team decreased length of stay (LOS) and improved patient satisfaction.5 However, this study was conducted at a time when the LOS was routinely longer than it is today. Forster et al. found that adding a clinical coordinator to a general medical team resulted in improved patient satisfaction but did not reduce length of stay or risk of adverse events occurring following hospital discharge.6 Both these studies evaluated the impact of adding a clinical coordinator to resident‐covered medical teams. Yet many hospitalists deliver care without residents, limiting the generalizability of the findings from these studies.

To date, no studies have evaluated the impact of clinical coordinators, case managers, or other nonphysician providers on the hospitalist work experience. This is surprising, as hospital medicine group leaders list daily workload and work hours among their top concerns.7 Clinical coordinators have the potential to improve patient care and hospital efficiency while simultaneously improving the experience of the hospitalists with whom they work. We conducted this study to evaluate the impact of a hospitalistcare coordinator team on hospitalist work experience, patient satisfaction, and hospital efficiency.

METHODS

RESULTS

There were 356 patients cared for by hospitalistHCC teams and 337 patients cared for by control hospitalists. Of the 60 weeks of hospitalist service of the study, hospitalistHCC teams accounted for 31 weeks (52%) and control hospitalists for 29 weeks (48%). Patients cared for by the hospitalistHCC teams were similar in age, sex, ethnicity, payer type, and DRG weight to those cared for by control hospitalists (see Table 2).

Sixty surveys were completed by hospitalists at the end of their week on service (response rate 100%). Of the 31 responses from hospitalists completing a hospitalistHCC team week, 28 (90%) reported that working with an HCC improved their efficiency and 28 (90%) that working with an HCC improved their job satisfaction. The hospitalists indicated that working with an HCC significantly improved the efficiency of most of their activities (see Table 3). Specifically, activities related to communication with nurses and patients and activities involving discharge planning and execution were improved with the use of an HCC. As would be expected, certain other activities did not improve. For example, there were no differences between the groups in the perceived efficiency of performing histories and physicals or placing admission orders. For activities that were significantly different, the Wilcoxon rank sum test and linear regression analysis adjusting for physician clustering showed identical results.

Seventy‐one of 196 eligible patients (36%) completed the discharge satisfaction interview. Of the 71 patients interviewed, 44 (62%) were cared for by hospitalistHCC teams and 27 (38%) were cared for by control hospitalists. Patient satisfaction with the clarity of the verbal and written discharge instructions and overall satisfaction with hospital discharge was similar between the 2 groups (see Table 4).

The unadjusted mean LOS for patients cared for by hospitalistHCC teams was 4.70 4.15 days compared with 5.07 3.99 days for patients cared for by control hospitalists (P = .005; see Table 5). The unadjusted mean cost for patients cared for by hospitalistHCC teams was $10,052.96 $11,708.73 compared with $11,703.19 $20,455.78 for patients cared for by control hospitalists (P = .008). In multivariate analysis using age, sex, ethnicity, payer type, and DRG weight as independent variables and adjusting for physician clustering, LOS remained lower for patients cared for by hospitalistHCC teams; however, this result was not statistically significant (0.28 days, P = .17). Similar multivariate regression analysis showed a trend toward lower cost for patients cared for by the hospitalistHCC teams (585.62, P = .15).

DISCUSSION

Our study found that hospitalists working in a team approach with an HCC rated the efficiency of their daily work and their job satisfaction significantly higher than did control hospitalists. Specific areas of improved efficiency included communication activities and activities related to hospital discharge. A prior study conducted by our group found that hospitalists spend a lot of time on indirect patient care activities such as communication and activities related to the discharge process, while spending relatively little time on direct patient care.8 Improving the efficiency of indirect patient care activities of hospitalists is likely to improve their job satisfaction. The importance of improving hospitalist workload and job satisfaction is underscored by the relatively high number of hospitalists at risk for burnout9 and the growing concern about daily workload among hospital medicine group leaders.7

Patient satisfaction was not significantly affected by the use of the hospitalistHCC team in our study. A priori, we postulated that patient satisfaction with the discharge process might improve with the use of the hospitalistHCC team. We therefore limited survey questions to assessing only satisfaction with hospital discharge rather than other aspects of patient hospital care. A recent study reported that patients rated the quality of discharge instructions significantly lower than they rated the overall quality of their hospital stay.10 However, the patients in our study gave high ratings to both discharge instructions and overall satisfaction with hospital discharge. This may explain why we were unable to detect a difference. Our study was limited by the relatively small number of patients we were able to contact to assess satisfaction. Previous studies evaluating the impact of care coordinators either did not assess patient satisfaction with discharge5 or found no difference in satisfaction with hospital discharge.6

Although our study did not find a difference in patient satisfaction with the discharge process, we believe the hospitalistHCC model has the potential to complement efforts to reduce the risk of adverse events as patients transition out of the hospital. It has been reported that 12% of patients have a preventable or ameliorable adverse event in the period immediately following hospital discharge.11, 12 Although Forster et al. did not find a reduction in the risk of adverse events with the addition of a clinical coordinator to a general medical team, they noted incongruence between the coordinator's role and the outcomes measured.6 Similarly, we would need to modify the role of the HCC from a position designed mainly to improve efficiency to one that complements efforts to improve the quality of the discharge process. Possible ways to enhance the HCC role in this regard include increasing the emphasis on and training in patient education skills. Several recently published articles have emphasized the need to redesign the discharge process in an effort to reduce the risk of adverse events following hospital discharge.1315 A modified HCC role might be an essential feature of a redesigned multidisciplinary discharge process.

We were unable to demonstrate improved efficiency for the hospital. Although LOS and cost were lower for patients cared for by the hospitalistHCC teams, the difference was not statistically significant. One possible explanation for why we did not observe a larger reduction in LOS is that our hospitalist service had a lower‐than‐average patient volume during the study period. The lower volume mirrored an unanticipated dip in hospital volume during the same period. Specifically, our service normally discharges an average of 338 patients per month, but during the study period we discharged an average of 235 patients per month. A potential LOS and cost benefit may have been attenuated by the relatively low volume, as hospitalists had ample time to dedicate to communication and coordination of discharge plans.

Our study had several limitations. It was conducted on a nonteaching hospitalist service at a single site. Hospitalist practices vary widely in their staffing and scheduling models. As previously mentioned, we were only able to perform patient satisfaction surveys during the second half of the study period. In addition, hospitalistHCC team patients made up a larger percentage of the patient survey responses (62%) than did control hospitalist patients (38%). This may have affected our ability to detect differences in satisfaction with the hospital discharge process. As also previously noted, our patient volume was lower than normal during the study period. We believe that a higher volume would have magnified differences in hospitalists' perceived efficiency and perhaps resulted in significant improvements in LOS and cost. Finally, the hospital provided funding for only a 12‐week study. This limited our sample size and the power of the study to detect important differences. It is possible that a larger sample size and/or longer study period may have been able to demonstrate a statistically significant improvement in LOS and cost.

Our findings are of particular importance in light of the persistent concerns about hospitalist workload and job satisfaction. Although many hospitalists work with clinical coordinators and case managers, we believe that having the formal structure of a hospitalistcare coordinator team was the key element to improving hospitalist efficiency and satisfaction. We hope that our study is a precursor to research evaluating models of delivering hospital care and their impact on hospitalist work experience, hospital efficiency, and patient outcomes.

Medications are central to managing the health of older patients. In 2006, more than 93% of adults 65 years or older reported taking at least 1 medication in the last week, 58% reported taking 5 or more medications, and 18% reported taking 10 or more.1 Medication use by older adults will likely increase further as the U.S. population ages, new drugs are developed, and new therapeutic and preventive uses for medications are discovered.2

Older patients, especially those who are chronically frail or acutely ill, may require special consideration when making prescribing decisions because of age‐related changes in the metabolism and clearance of medications and enhanced pharmacodynamic sensitivities.3 Thus, panels of experts in pharmacology and geriatrics have compiled lists of medications to avoid prescribing for patients 65 years of age or older. The most commonly used list is the Beers criteria, which were introduced in 1991 to serve researchers evaluating prescribing quality in nursing homes. The Beers criteria were updated in 1997 and again in 2003 to include 48 potentially inappropriate medications (PIMs) for which, according to the consensus panel, there are more effective or safer alternatives for older patients.4

Numerous studies in the last 15 years have found that PIMs continue to be used in 12% to 40% of older patients in community and nursing home settings.5 To address the continued use of PIMs, the Centers for Medicare and Medicaid Services incorporated the Beers criteria into federal safety regulations for long‐term care facilities in 1999.6 In 2006, the prescription rate of PIMs was introduced as a Health Plan and Employer Data and Information Set (HEDIS) quality measure for managed care plans.7 Despite adoption of the Beers criteria to monitor prescribing quality and safety in nursing homes and outpatient settings, there has been considerably less study of potentially inappropriate medication use in hospitalized patients.

In this issue of the Journal of Hospital Medicine, Rothberg and colleagues analyzed administrative data from nearly 400 hospitals across the United States and found that nearly half of all older patients hospitalized for 7 common conditions were prescribed at least 1 PIM.8 Thus, the incidence of PIM use in hospitalized older patients far exceeded that reported in most studies of community‐dwelling or nursing home patients. Most notable, however, was the variability found in prescribing rates based on a number of physician and hospital characteristics. For example, although hospitalists and geriatricians were found to be less likely to prescribe PIMs than cardiologists and general internists, among high‐volume cardiologists and internists, PIM prescribing rates ranged widely, from 0% to more than 90%. Similarly, hospitalwide prescribing rates varied by geographic region, and there were 7 hospitals in which not a single PIM was reportedly prescribed.

These findings raise three questions and bring to mind parallels with efforts to control inappropriate antimicrobial use. First question: Can inpatient use of PIMs truly be higher than outpatient use? Yes. The finding that more hospitalized patients are prescribed PIMs than ambulatory patients has face validity for several reasons. First, patients admitted for an acute hospitalization may have more comorbid diseases and take more medications than community‐dwelling older adults. Second, new medications are typically added to treat acutely ill patients on hospitalization. Third, previous studies estimating outpatient PIM use have typically used more narrowly defined lists of PIMs and have not captured over‐the‐counter use of PIMs, particularly antihistamines.9 Diphenhydramine alone accounted for 9% of PIM use in Rothberg's study. Finally, as Rothberg and colleagues point out, this study was limited to certain diagnoses such as acute myocardial infarction that may have protocol‐driven prescribing, which includes PIMs, that may be used only a single time such as promethazine.

Second question: Can PIM prescribing truly be so variable across regions, specialties, and individual hospitals and physicians? Yes. Using multivariable modeling, Rothberg and colleagues controlled for many patient, hospital, and physician characteristics and still found significant variation. John Wennberg and others have documented similar variations for a host of medical treatments; but although variation is interesting, it is unwarranted variation that matters for improving health care quality.10 It is not clear how much of the variation in prescribing rates of PIMs is unwarranted.

Some degree of variation in PIM prescribing rates is certainly acceptable. As the creators of the Beers criteria acknowledge, these medications are deemed only potentially inappropriate, and individual treatment decisions should be tailored to individual patients. However, others have taken the term potentially inappropriate one step further by recategorizing Beers medications as always avoid medications, rarely acceptable medications, and medications that indeed have some indications for use in older adults.8

Variation in prescribing practice may also be acceptable when there is not a clear consensus on the superiority of one practice over another. Indeed, the evidence that PIM prescribing causes large numbers of clinically significant adverse drug events and patient harm is weak and largely based on observational studies with inconsistent results. Although some studies demonstrated an epidemiological association between Beers criteria medications and general adverse outcomes (eg, hospitalizations),11 other studies did not.12 A recent systematic review concluded that Beers criteria medications were associated with some adverse health effects, but the studies analyzed were too heterogeneous to support formal meta‐analysis.13 Thus, variability in prescribing rates of Beers medications may simply reflect individual clinical judgment in the absence of conclusive outcomes data.

Third question: Can hospitalists use the findings of Rothberg and colleagues to improve the quality of medication prescribing for older adults in their institutions? Maybe. But hospitalists wishing to reduce PIM use in their institutions should draw lessons from other efforts to modify physician‐prescribing practice such as efforts to reduce inappropriate antimicrobial use. Although national data draw attention to the high frequency of potentially inappropriate medication use in hospitalized patients, the large variation in use across hospitals confirms the need for monitoring in individual facilities. For example, the National Healthcare Safety Network provides national benchmarks of antimicrobial use and resistance, but individual hospitals monitor antibiotic use and resistance in their own institutions to tailor local efforts to improve antimicrobial prescribing.14

Also, initiating a quality improvement effort targeting all 48 Beers criteria medications may be an inefficient use of resources. Using such a composite measure obscures the contribution of the component medications, each of which possesses unique and sometimes controversial profiles of efficacy and harm for older patients. Instead, a targeted intervention addressing the most commonly prescribed Beers medications that have widely accepted alternatives could be more practical. For instance, many antibiotic management programs focus on replacing a popular, extended‐spectrum antimicrobial with a narrow‐spectrum agent as soon as microbiological susceptibly results are available.

Propoxephene is a PIM that may be an attractive target for intervention. Propoxephene was the third most commonly prescribed PIM identified by Rothberg and colleagues, but meta‐analyses of controlled trials have concluded that propoxephene provides inferior analgesia for acute pain compared with that provided by other opioids with similar side effects, and has more adverse effects than nonopioid analgesics.15 Indeed, Rothberg found that just 3 of 48 PIMs (promethazine, diphenhydramine, and propoxyphene), each of which has viable alternative agents, accounted for approximately a quarter of all potentially inappropriate prescribing.

However, not all of the 48 Beers medications have alternatives with strong evidence of superiority. The Beers list includes medications (eg, amiodarone) that may not have equivalent alternative agents. On the other hand, some Beers medications have largely been supplanted (eg, ticlopidine or tripelennamine), and identifying these medications may be an inefficient use of scarce patient safety resources. As with antimicrobial stewardship programs, local surveillance of PIM use should be combined with local consensus on appropriate alternatives to target PIM interventions.

Of course, once specific PIM use is targeted for improvement, a specific intervention must be implemented. Only a handful of studies have examined the effectiveness of interventions (eg, computerized pharmacy alerts) to reduce PIM use, and most of these have focused on the outpatient setting rather than hospitalized patients.3 One study that included hospitalized patients utilized a team approach (geriatricians, nurses, social workers, and pharmacists) and demonstrated a reduction in potentially inappropriate medication use but no reduction in adverse drug reactions during hospitalization.16 In light of the scarcity of controlled intervention trials to reduce PIM use, initiatives to reduce inappropriate antimicrobial prescribing may provide useful insights into the strengths and limitations of approaches such as clinician education, formulary restrictions, pharmacist review, and computer‐based monitoring.17

Finally, any intervention to reduce PIM use should have reasonable expectations. The Beers criteria were developed to improve the effectiveness of medication therapy for older adults as well as to prevent harm, but it is unlikely that reducing PIM use in hospitalized patients will result in improvements that could be measured easily during an initial hospitalization. If preventing drug‐induced harm during the hospitalization of older patients is the primary concern, a shift in focus is required. Safety efforts should be concentrated on identifying and mitigating the most common and severe adverse drug events, rather than focusing efforts on reducing the use of PIMs. National data demonstrate that a handful of drugsinsulin, warfarin, and digoxinmost commonly cause severe adverse events in older outpatients.18 Optimizing the management of these medications may be another approach for improving drug safety in hospitalized patients. Regardless of the focus of a drug safety intervention, the experience of infection control and hospital epidemiology programs suggests that success will require dedicated professionals and the commitment of resources to examine patterns of local use, implement interventions, and monitor outcomes.

Medications are central to managing the health of older patients. In 2006, more than 93% of adults 65 years or older reported taking at least 1 medication in the last week, 58% reported taking 5 or more medications, and 18% reported taking 10 or more.1 Medication use by older adults will likely increase further as the U.S. population ages, new drugs are developed, and new therapeutic and preventive uses for medications are discovered.2

Older patients, especially those who are chronically frail or acutely ill, may require special consideration when making prescribing decisions because of age‐related changes in the metabolism and clearance of medications and enhanced pharmacodynamic sensitivities.3 Thus, panels of experts in pharmacology and geriatrics have compiled lists of medications to avoid prescribing for patients 65 years of age or older. The most commonly used list is the Beers criteria, which were introduced in 1991 to serve researchers evaluating prescribing quality in nursing homes. The Beers criteria were updated in 1997 and again in 2003 to include 48 potentially inappropriate medications (PIMs) for which, according to the consensus panel, there are more effective or safer alternatives for older patients.4

Numerous studies in the last 15 years have found that PIMs continue to be used in 12% to 40% of older patients in community and nursing home settings.5 To address the continued use of PIMs, the Centers for Medicare and Medicaid Services incorporated the Beers criteria into federal safety regulations for long‐term care facilities in 1999.6 In 2006, the prescription rate of PIMs was introduced as a Health Plan and Employer Data and Information Set (HEDIS) quality measure for managed care plans.7 Despite adoption of the Beers criteria to monitor prescribing quality and safety in nursing homes and outpatient settings, there has been considerably less study of potentially inappropriate medication use in hospitalized patients.

In this issue of the Journal of Hospital Medicine, Rothberg and colleagues analyzed administrative data from nearly 400 hospitals across the United States and found that nearly half of all older patients hospitalized for 7 common conditions were prescribed at least 1 PIM.8 Thus, the incidence of PIM use in hospitalized older patients far exceeded that reported in most studies of community‐dwelling or nursing home patients. Most notable, however, was the variability found in prescribing rates based on a number of physician and hospital characteristics. For example, although hospitalists and geriatricians were found to be less likely to prescribe PIMs than cardiologists and general internists, among high‐volume cardiologists and internists, PIM prescribing rates ranged widely, from 0% to more than 90%. Similarly, hospitalwide prescribing rates varied by geographic region, and there were 7 hospitals in which not a single PIM was reportedly prescribed.

These findings raise three questions and bring to mind parallels with efforts to control inappropriate antimicrobial use. First question: Can inpatient use of PIMs truly be higher than outpatient use? Yes. The finding that more hospitalized patients are prescribed PIMs than ambulatory patients has face validity for several reasons. First, patients admitted for an acute hospitalization may have more comorbid diseases and take more medications than community‐dwelling older adults. Second, new medications are typically added to treat acutely ill patients on hospitalization. Third, previous studies estimating outpatient PIM use have typically used more narrowly defined lists of PIMs and have not captured over‐the‐counter use of PIMs, particularly antihistamines.9 Diphenhydramine alone accounted for 9% of PIM use in Rothberg's study. Finally, as Rothberg and colleagues point out, this study was limited to certain diagnoses such as acute myocardial infarction that may have protocol‐driven prescribing, which includes PIMs, that may be used only a single time such as promethazine.

Second question: Can PIM prescribing truly be so variable across regions, specialties, and individual hospitals and physicians? Yes. Using multivariable modeling, Rothberg and colleagues controlled for many patient, hospital, and physician characteristics and still found significant variation. John Wennberg and others have documented similar variations for a host of medical treatments; but although variation is interesting, it is unwarranted variation that matters for improving health care quality.10 It is not clear how much of the variation in prescribing rates of PIMs is unwarranted.

Some degree of variation in PIM prescribing rates is certainly acceptable. As the creators of the Beers criteria acknowledge, these medications are deemed only potentially inappropriate, and individual treatment decisions should be tailored to individual patients. However, others have taken the term potentially inappropriate one step further by recategorizing Beers medications as always avoid medications, rarely acceptable medications, and medications that indeed have some indications for use in older adults.8

Variation in prescribing practice may also be acceptable when there is not a clear consensus on the superiority of one practice over another. Indeed, the evidence that PIM prescribing causes large numbers of clinically significant adverse drug events and patient harm is weak and largely based on observational studies with inconsistent results. Although some studies demonstrated an epidemiological association between Beers criteria medications and general adverse outcomes (eg, hospitalizations),11 other studies did not.12 A recent systematic review concluded that Beers criteria medications were associated with some adverse health effects, but the studies analyzed were too heterogeneous to support formal meta‐analysis.13 Thus, variability in prescribing rates of Beers medications may simply reflect individual clinical judgment in the absence of conclusive outcomes data.

Third question: Can hospitalists use the findings of Rothberg and colleagues to improve the quality of medication prescribing for older adults in their institutions? Maybe. But hospitalists wishing to reduce PIM use in their institutions should draw lessons from other efforts to modify physician‐prescribing practice such as efforts to reduce inappropriate antimicrobial use. Although national data draw attention to the high frequency of potentially inappropriate medication use in hospitalized patients, the large variation in use across hospitals confirms the need for monitoring in individual facilities. For example, the National Healthcare Safety Network provides national benchmarks of antimicrobial use and resistance, but individual hospitals monitor antibiotic use and resistance in their own institutions to tailor local efforts to improve antimicrobial prescribing.14

Also, initiating a quality improvement effort targeting all 48 Beers criteria medications may be an inefficient use of resources. Using such a composite measure obscures the contribution of the component medications, each of which possesses unique and sometimes controversial profiles of efficacy and harm for older patients. Instead, a targeted intervention addressing the most commonly prescribed Beers medications that have widely accepted alternatives could be more practical. For instance, many antibiotic management programs focus on replacing a popular, extended‐spectrum antimicrobial with a narrow‐spectrum agent as soon as microbiological susceptibly results are available.

Propoxephene is a PIM that may be an attractive target for intervention. Propoxephene was the third most commonly prescribed PIM identified by Rothberg and colleagues, but meta‐analyses of controlled trials have concluded that propoxephene provides inferior analgesia for acute pain compared with that provided by other opioids with similar side effects, and has more adverse effects than nonopioid analgesics.15 Indeed, Rothberg found that just 3 of 48 PIMs (promethazine, diphenhydramine, and propoxyphene), each of which has viable alternative agents, accounted for approximately a quarter of all potentially inappropriate prescribing.

However, not all of the 48 Beers medications have alternatives with strong evidence of superiority. The Beers list includes medications (eg, amiodarone) that may not have equivalent alternative agents. On the other hand, some Beers medications have largely been supplanted (eg, ticlopidine or tripelennamine), and identifying these medications may be an inefficient use of scarce patient safety resources. As with antimicrobial stewardship programs, local surveillance of PIM use should be combined with local consensus on appropriate alternatives to target PIM interventions.

Of course, once specific PIM use is targeted for improvement, a specific intervention must be implemented. Only a handful of studies have examined the effectiveness of interventions (eg, computerized pharmacy alerts) to reduce PIM use, and most of these have focused on the outpatient setting rather than hospitalized patients.3 One study that included hospitalized patients utilized a team approach (geriatricians, nurses, social workers, and pharmacists) and demonstrated a reduction in potentially inappropriate medication use but no reduction in adverse drug reactions during hospitalization.16 In light of the scarcity of controlled intervention trials to reduce PIM use, initiatives to reduce inappropriate antimicrobial prescribing may provide useful insights into the strengths and limitations of approaches such as clinician education, formulary restrictions, pharmacist review, and computer‐based monitoring.17

Finally, any intervention to reduce PIM use should have reasonable expectations. The Beers criteria were developed to improve the effectiveness of medication therapy for older adults as well as to prevent harm, but it is unlikely that reducing PIM use in hospitalized patients will result in improvements that could be measured easily during an initial hospitalization. If preventing drug‐induced harm during the hospitalization of older patients is the primary concern, a shift in focus is required. Safety efforts should be concentrated on identifying and mitigating the most common and severe adverse drug events, rather than focusing efforts on reducing the use of PIMs. National data demonstrate that a handful of drugsinsulin, warfarin, and digoxinmost commonly cause severe adverse events in older outpatients.18 Optimizing the management of these medications may be another approach for improving drug safety in hospitalized patients. Regardless of the focus of a drug safety intervention, the experience of infection control and hospital epidemiology programs suggests that success will require dedicated professionals and the commitment of resources to examine patterns of local use, implement interventions, and monitor outcomes.

Medications are central to managing the health of older patients. In 2006, more than 93% of adults 65 years or older reported taking at least 1 medication in the last week, 58% reported taking 5 or more medications, and 18% reported taking 10 or more.1 Medication use by older adults will likely increase further as the U.S. population ages, new drugs are developed, and new therapeutic and preventive uses for medications are discovered.2

Older patients, especially those who are chronically frail or acutely ill, may require special consideration when making prescribing decisions because of age‐related changes in the metabolism and clearance of medications and enhanced pharmacodynamic sensitivities.3 Thus, panels of experts in pharmacology and geriatrics have compiled lists of medications to avoid prescribing for patients 65 years of age or older. The most commonly used list is the Beers criteria, which were introduced in 1991 to serve researchers evaluating prescribing quality in nursing homes. The Beers criteria were updated in 1997 and again in 2003 to include 48 potentially inappropriate medications (PIMs) for which, according to the consensus panel, there are more effective or safer alternatives for older patients.4

Numerous studies in the last 15 years have found that PIMs continue to be used in 12% to 40% of older patients in community and nursing home settings.5 To address the continued use of PIMs, the Centers for Medicare and Medicaid Services incorporated the Beers criteria into federal safety regulations for long‐term care facilities in 1999.6 In 2006, the prescription rate of PIMs was introduced as a Health Plan and Employer Data and Information Set (HEDIS) quality measure for managed care plans.7 Despite adoption of the Beers criteria to monitor prescribing quality and safety in nursing homes and outpatient settings, there has been considerably less study of potentially inappropriate medication use in hospitalized patients.

In this issue of the Journal of Hospital Medicine, Rothberg and colleagues analyzed administrative data from nearly 400 hospitals across the United States and found that nearly half of all older patients hospitalized for 7 common conditions were prescribed at least 1 PIM.8 Thus, the incidence of PIM use in hospitalized older patients far exceeded that reported in most studies of community‐dwelling or nursing home patients. Most notable, however, was the variability found in prescribing rates based on a number of physician and hospital characteristics. For example, although hospitalists and geriatricians were found to be less likely to prescribe PIMs than cardiologists and general internists, among high‐volume cardiologists and internists, PIM prescribing rates ranged widely, from 0% to more than 90%. Similarly, hospitalwide prescribing rates varied by geographic region, and there were 7 hospitals in which not a single PIM was reportedly prescribed.

These findings raise three questions and bring to mind parallels with efforts to control inappropriate antimicrobial use. First question: Can inpatient use of PIMs truly be higher than outpatient use? Yes. The finding that more hospitalized patients are prescribed PIMs than ambulatory patients has face validity for several reasons. First, patients admitted for an acute hospitalization may have more comorbid diseases and take more medications than community‐dwelling older adults. Second, new medications are typically added to treat acutely ill patients on hospitalization. Third, previous studies estimating outpatient PIM use have typically used more narrowly defined lists of PIMs and have not captured over‐the‐counter use of PIMs, particularly antihistamines.9 Diphenhydramine alone accounted for 9% of PIM use in Rothberg's study. Finally, as Rothberg and colleagues point out, this study was limited to certain diagnoses such as acute myocardial infarction that may have protocol‐driven prescribing, which includes PIMs, that may be used only a single time such as promethazine.

Second question: Can PIM prescribing truly be so variable across regions, specialties, and individual hospitals and physicians? Yes. Using multivariable modeling, Rothberg and colleagues controlled for many patient, hospital, and physician characteristics and still found significant variation. John Wennberg and others have documented similar variations for a host of medical treatments; but although variation is interesting, it is unwarranted variation that matters for improving health care quality.10 It is not clear how much of the variation in prescribing rates of PIMs is unwarranted.

Some degree of variation in PIM prescribing rates is certainly acceptable. As the creators of the Beers criteria acknowledge, these medications are deemed only potentially inappropriate, and individual treatment decisions should be tailored to individual patients. However, others have taken the term potentially inappropriate one step further by recategorizing Beers medications as always avoid medications, rarely acceptable medications, and medications that indeed have some indications for use in older adults.8

Variation in prescribing practice may also be acceptable when there is not a clear consensus on the superiority of one practice over another. Indeed, the evidence that PIM prescribing causes large numbers of clinically significant adverse drug events and patient harm is weak and largely based on observational studies with inconsistent results. Although some studies demonstrated an epidemiological association between Beers criteria medications and general adverse outcomes (eg, hospitalizations),11 other studies did not.12 A recent systematic review concluded that Beers criteria medications were associated with some adverse health effects, but the studies analyzed were too heterogeneous to support formal meta‐analysis.13 Thus, variability in prescribing rates of Beers medications may simply reflect individual clinical judgment in the absence of conclusive outcomes data.

Third question: Can hospitalists use the findings of Rothberg and colleagues to improve the quality of medication prescribing for older adults in their institutions? Maybe. But hospitalists wishing to reduce PIM use in their institutions should draw lessons from other efforts to modify physician‐prescribing practice such as efforts to reduce inappropriate antimicrobial use. Although national data draw attention to the high frequency of potentially inappropriate medication use in hospitalized patients, the large variation in use across hospitals confirms the need for monitoring in individual facilities. For example, the National Healthcare Safety Network provides national benchmarks of antimicrobial use and resistance, but individual hospitals monitor antibiotic use and resistance in their own institutions to tailor local efforts to improve antimicrobial prescribing.14

Also, initiating a quality improvement effort targeting all 48 Beers criteria medications may be an inefficient use of resources. Using such a composite measure obscures the contribution of the component medications, each of which possesses unique and sometimes controversial profiles of efficacy and harm for older patients. Instead, a targeted intervention addressing the most commonly prescribed Beers medications that have widely accepted alternatives could be more practical. For instance, many antibiotic management programs focus on replacing a popular, extended‐spectrum antimicrobial with a narrow‐spectrum agent as soon as microbiological susceptibly results are available.

Propoxephene is a PIM that may be an attractive target for intervention. Propoxephene was the third most commonly prescribed PIM identified by Rothberg and colleagues, but meta‐analyses of controlled trials have concluded that propoxephene provides inferior analgesia for acute pain compared with that provided by other opioids with similar side effects, and has more adverse effects than nonopioid analgesics.15 Indeed, Rothberg found that just 3 of 48 PIMs (promethazine, diphenhydramine, and propoxyphene), each of which has viable alternative agents, accounted for approximately a quarter of all potentially inappropriate prescribing.

However, not all of the 48 Beers medications have alternatives with strong evidence of superiority. The Beers list includes medications (eg, amiodarone) that may not have equivalent alternative agents. On the other hand, some Beers medications have largely been supplanted (eg, ticlopidine or tripelennamine), and identifying these medications may be an inefficient use of scarce patient safety resources. As with antimicrobial stewardship programs, local surveillance of PIM use should be combined with local consensus on appropriate alternatives to target PIM interventions.

Of course, once specific PIM use is targeted for improvement, a specific intervention must be implemented. Only a handful of studies have examined the effectiveness of interventions (eg, computerized pharmacy alerts) to reduce PIM use, and most of these have focused on the outpatient setting rather than hospitalized patients.3 One study that included hospitalized patients utilized a team approach (geriatricians, nurses, social workers, and pharmacists) and demonstrated a reduction in potentially inappropriate medication use but no reduction in adverse drug reactions during hospitalization.16 In light of the scarcity of controlled intervention trials to reduce PIM use, initiatives to reduce inappropriate antimicrobial prescribing may provide useful insights into the strengths and limitations of approaches such as clinician education, formulary restrictions, pharmacist review, and computer‐based monitoring.17

Finally, any intervention to reduce PIM use should have reasonable expectations. The Beers criteria were developed to improve the effectiveness of medication therapy for older adults as well as to prevent harm, but it is unlikely that reducing PIM use in hospitalized patients will result in improvements that could be measured easily during an initial hospitalization. If preventing drug‐induced harm during the hospitalization of older patients is the primary concern, a shift in focus is required. Safety efforts should be concentrated on identifying and mitigating the most common and severe adverse drug events, rather than focusing efforts on reducing the use of PIMs. National data demonstrate that a handful of drugsinsulin, warfarin, and digoxinmost commonly cause severe adverse events in older outpatients.18 Optimizing the management of these medications may be another approach for improving drug safety in hospitalized patients. Regardless of the focus of a drug safety intervention, the experience of infection control and hospital epidemiology programs suggests that success will require dedicated professionals and the commitment of resources to examine patterns of local use, implement interventions, and monitor outcomes.