A 49‐year‐old woman with a history of asthma and a pneumatocele secondary to a past pneumonia (Fig. 1, prior baseline chest X‐ray) presented with fever, cough, and shortness of breath. This was her third admission for similar symptoms. In the past, these symptoms had resolved with antibiotics and percutaneous drainage of the infected pneumatocele. Her admission chest X‐ray revealed a fluid‐filled pneumatocele (Fig. 2). Given the recurrent nature of her disease, a right middle lobectomy was performed. Her recovery after surgery was uneventful, and she continues to do well 4 months later.

Figure 1

Figure 2

Pneumatoceles are air‐filled cysts that occur because of trauma or inflammation in the lung parenchyma. The most common cause is pneumonia and they occur most frequently in children. Complications of pneumatocele include pneumothorax and, as in this patient, secondary infection. In most instances the pneumatocele resolves after treating the underlying pneumonia, but surgical intervention is sometimes required to have a definitive cure.

A 49‐year‐old woman with a history of asthma and a pneumatocele secondary to a past pneumonia (Fig. 1, prior baseline chest X‐ray) presented with fever, cough, and shortness of breath. This was her third admission for similar symptoms. In the past, these symptoms had resolved with antibiotics and percutaneous drainage of the infected pneumatocele. Her admission chest X‐ray revealed a fluid‐filled pneumatocele (Fig. 2). Given the recurrent nature of her disease, a right middle lobectomy was performed. Her recovery after surgery was uneventful, and she continues to do well 4 months later.

Figure 1

Figure 2

Pneumatoceles are air‐filled cysts that occur because of trauma or inflammation in the lung parenchyma. The most common cause is pneumonia and they occur most frequently in children. Complications of pneumatocele include pneumothorax and, as in this patient, secondary infection. In most instances the pneumatocele resolves after treating the underlying pneumonia, but surgical intervention is sometimes required to have a definitive cure.

A 49‐year‐old woman with a history of asthma and a pneumatocele secondary to a past pneumonia (Fig. 1, prior baseline chest X‐ray) presented with fever, cough, and shortness of breath. This was her third admission for similar symptoms. In the past, these symptoms had resolved with antibiotics and percutaneous drainage of the infected pneumatocele. Her admission chest X‐ray revealed a fluid‐filled pneumatocele (Fig. 2). Given the recurrent nature of her disease, a right middle lobectomy was performed. Her recovery after surgery was uneventful, and she continues to do well 4 months later.

Figure 1

Figure 2

Pneumatoceles are air‐filled cysts that occur because of trauma or inflammation in the lung parenchyma. The most common cause is pneumonia and they occur most frequently in children. Complications of pneumatocele include pneumothorax and, as in this patient, secondary infection. In most instances the pneumatocele resolves after treating the underlying pneumonia, but surgical intervention is sometimes required to have a definitive cure.

Functional decline, defined as loss of the ability to care for oneself, commonly occurs during hospitalization, being experienced by up to 65% of older adults.14 Frequently, recovery of functional ability does not occur by the time of discharge from the hospital, despite resolution of the medical condition responsible for admission to the hospital.1, 5 Causes of the declining ability to perform activities of daily living (ADLs) are multiple and include both acute illness and adverse events associated with hospitalization.4, 6, 7 The functional decline experienced by older persons during hospitalization is a strong predictor of length of stay, nursing home placement, and mortality.8 Loss of ambulatory ability specifically has been noted to occur in 17%‐65% of hospitalized older patients, usually within days of admission, with few recovering this ability prior to discharge.2, 4, 5, 9 Importantly, ambulatory ability is significantly associated with a decline in other ADLs.9

Although several studies have explored risk factors associated with general functional decline during hospitalization,4, 7, 10, 11 only one study specifically examined risk factors associated with loss of ambulatory ability. In a cohort of participants who were largely independently performing ADLs on admission to the hospital, Mahoney et al. found age 85 years, white race, use of a walker, and functional impairment prior to being hospitalized were significant predictors of newly having walking dependence.12

Ambulatory ability could also be affected by a variety of other factors not examined in the Mahoney et al. study; these include severity of illness, bed rest, and hospital‐related treatments such as restraints or urinary catheters. In addition, little is known about predictors of recovery of ambulatory ability in patients expected to have activity limitations on admission who are dependent in most or all ADLs. The deconditioning associated with bed rest and reduced mobility has been described as one of the most predictable causes of functional decline, including loss of ambulatory ability, observed in older hospitalized patients.13 In one study, patients whose activity was limited to a bed or chair during hospitalization were 5.6 times more likely to develop functional decline than those who walked at all, even after controlling for other covariates including severity of illness and comorbidity.14 Those patients with both activity limitations and dependence in most or all ADLs represent an important subset of all hospitalized older patients who might be expected to be at higher risk of developing new ambulatory dependence. The ability to identify, at admission, those patients who will recover ambulatory ability may have important implications for discharge planning as well as for the development of preventive strategies.

The objective of the present study was to define patient demographic, illness severity, comorbid illness, and hospital‐related variables that are independent and significant predictors of regaining ambulatory ability prior to hospital discharge in a cohort of patients who had significant activity limitations and functional impairment at the time of admission to the hospital.

METHODS

RESULTS

For this study, 286 participants met all eligibility criteria, with 119 (42%) regaining ambulatory ability during hospitalization. Mean age of study participants was 73 9 years, with 12% of participants more than 85 years old. On admission, 214 patients (75%) were dependent in all 7 ADLs. Mean length of stay ( SD) was 12.3 6.5 days, with a range of 1‐31 days. Table 1 presents the cohort characteristics and the unadjusted effects of each variable for predicting those who did and did not recover ambulatory ability by characteristic. The P value of a variable in Table 1 indicates how significant that variable was, as determined with a simple logistic regression analysis. The unadjusted odds ratio of each variable is presented in Table 2 in order to facilitate comparisons with the adjusted odds ratios from the multivariate models. Analysis of the effects of the unadjusted single variables showed that age was a significant predictor, with older patients less likely to regain ambulatory ability. Several global health measures and hospital‐related factors were also significantly different between those who recovered and those who did not recover ambulatory ability, as summarized in Table 1. Importantly, length of stay was not significantly associated with recovery.

Table 2 shows the effects of adjusting the model for important covariables in a sequential fashion. Model 1 shows that longer length of stay and older age are associated with reduced odds of regaining ambulatory ability after adjusting for other demographic variables. However, age and length of stay were no longer significant after adjusting for global health measures and specific medical problems (Model 2). As demonstrated by the full model (Model 4), the participants who regained ambulatory ability were more likely to be unmarried, have a longer physician‐rated life expectancy, not have a diagnosis of deep vein thrombosis, not have physical restraints or a urinary catheter in use, and have greater bed mobility at admission.

Predictors that remained significant in the final multivariate model (Model 4) were summed in order to determine the proportion of patients who had one predictor versus those who had more than one predictor. Figure 1 shows the percentage of those who recovered ambulatory ability according to number of predictors, ranging from one to all 6 predictors. The results demonstrate a graded relationship, with number of predictors positively correlated with percentage of those who recovered ambulatory ability (P < .0001).

Figure 1

Patients who died prior to discharge were more likely to be male and have altered mental status, a urinary catheter, restraints, a shorter physician‐rated life expectancy, lower admission bed mobility, and increased severity of illness compared to those who survived. When the 34 patients who died were excluded from the multivariate analysis, the factors predicting ambulatory recovery were unchanged from those reported in Model 4 of Table 2.

DISCUSSION

In this study of older hospitalized patients, approximately 40% were able to regain their ambulatory ability despite being limited to the bed or chair on admission, having significant severity of illness, high level of use of restraints, and functional dependence on admission. Predictors of regaining ambulatory ability were identifiable at hospital admission. These predictors also were correlated with recovery of ambulatory ability in a graded fashion, lending support for the direct relationship between these predictors and recovery.

Physician‐rated life expectancy, a simple assessment that combines a physician's medical knowledge and clinical acumen, was demonstrated in our study to be a better predictor of recovery of ambulatory ability than more elaborate measures like the APACHE II17 and Charlson Comorbidity Index.16 This assessment can easily be done at the bedside and may help to guide discharge planning for the hospitalist physician. Nurse‐rated bed mobility at time of admission, which may reflect aspects of illness severity and cognitive status, was also able to predict recovery.

Of the 6 factors found to be independent predictors of recovery of ambulatory ability, 4 were related to mobility: lack of a DVT diagnosis, absence of a urinary catheter, absence of restraints, and nurse‐rated bed mobility at admission. In the group of patients initially expected to be confined to a bed or chair, those with additional mobility‐reducing factors, such as catheters and restraints, were less likely to recover, even after controlling for illness severity and comorbidity.

Marital status was not found to predict recovery of ambulatory ability in the simple unadjusted bivariate analysis, but after adjusting for other demographic, global health, and disease‐related variables, those who were unmarried were more likely to recover ambulatory ability. These interesting covariate‐adjusted effects for marital status have not been previously reported in the literature. One large study of the impact of marital status on hospital outcomes demonstrated those who were unmarried were more likely to require discharge to a nursing home and had slightly higher hospital costs and longer length of stay.19 Our findings may be related to such patients lacking support other than from themselves, with the possibility of being discharged to a nursing home an incentive to get up and walk.

Although age, race, and previous ADL status were found to be significant predictors in a previous study,12 we did not find this in our sample. This may be because, unlike in other studies, most of our patients had significant functional impairment on admission. The importance of age in our cohort disappeared when illness severity and comorbidities were added to the model.

The strengths of this study include having comprehensive patient‐related data on demographic, illness severity, comorbidity, and hospital‐related factors available, which enabled detailed analyses of predictors for regaining ambulatory ability. In particular, the ability to examine such factors as bed rest and hospital‐related treatments like restraint and catheter use, enabled this study to add significantly to the available knowledge of predictors of ambulatory recovery. The use of nurse interviews to obtain patient‐related data has been demonstrated in previous studies to be a preferred method of collecting data when compared to patient self‐report.20, 21 Examination of these factors in a cohort of patients who would be expected to be at very high risk for remaining bed‐ or chair bound, given their admission activity limitations and functional dependence, is also noteworthy.

Several important limitations deserve comment. Since the data were collected, average length of hospital stay generally has decreased. However, the patient population we studied continue to experience longer hospital stays than functionally intact patients. One recent study demonstrated that the length of hospital stay of patients who were dependent in one or more ADLs on admission was 35% longer than that of those not ADL dependent at admission.22 Seventy‐five percent of those in the present study cohort were dependent in all 7 ADLs and had a mean length of stay of 12.3 days. Despite the longer mean length of stay, 70% of those in the study cohort were discharged within 2 weeks of admission. In the university hospital where this research was conducted, mean length of stay ( SD) in 2004 was 6 8.6 days, but 10% of patients 55 years of age and older remained in the hospital for more than 2 weeks.23 This suggests there continue to be long‐stay patients in the current hospital environment, to which these findings may apply.

Standards of practice, such as for use of restraints, have also changed. In 1992, between 7.4% and 17% of all hospitalized medical patients were restrained, according to a literature review.24 A 1998 survey of 3 hospitals found the prevalence of restraints still ranged from 3.9% to 8.2% and noted that among the most common reasons reported for using restraints were to prevent patient disruption of therapy, to confine confused patients, and to reduce the number of falls.25 Thus, our study cohort would be more likely to be restrained, even in the current hospital environment, given that 43% of the cohort had altered mental status and that most were at risk for falls because of their poor functional status.26

Nevertheless, even though the use of restraints has declined since the data were collected for this study, this should affect neither the internal validity of the results nor the ability to address the question of what factors predict recovery of ambulatory ability. Indeed, the inclusion of patients on whom restraints are frequently used emphasizes the need for continued diligence in creating a restraint‐free environment in our hospitals. Data about the use of physical therapy services were not available in the study. Therefore, it is unknown to what extent the use of these services encouraged ambulation.

In this observational study, we found 6 factors associated with regaining ambulatory ability among hospitalized patients who had significant activity limitations and functional dependence on admission. These findings suggest predictors easily assessed by the hospitalist physician can help to identify those patients most likely to recover ambulatory ability prior to discharge. It also demonstrates the importance of mobility in maintaining function, given that many of the predictors are factors that either impede mobility such as restraints and urinary catheters or measure mobility such as admission bed mobility. Last, recognizing physician‐rated life expectancy as a strong independent predictor of recovery of ambulatory ability should encourage hospitalist physicians to continue to use their greatest tool, their clinical judgment, to determine who will recover ambulatory ability.

As most of these predictors can be identified on admission or shortly thereafter, these factors may be useful in helping physicians and other health care providers to predict the potential patients have to recover ambulatory ability. This information may help physicians identify patients who might benefit from early mobility programs, placement on hospital units where mobility will be enhanced, or the early initiation of discharge planning for those patients identified as unlikely to regain ambulation ability. In addition, addressing factors that are potentially modifiable, such as low bed mobility and the use of urinary catheters and restraints may not only improve the chance of recovering ambulatory ability but would also improve the quality of care provided to older patients.

Functional decline, defined as loss of the ability to care for oneself, commonly occurs during hospitalization, being experienced by up to 65% of older adults.14 Frequently, recovery of functional ability does not occur by the time of discharge from the hospital, despite resolution of the medical condition responsible for admission to the hospital.1, 5 Causes of the declining ability to perform activities of daily living (ADLs) are multiple and include both acute illness and adverse events associated with hospitalization.4, 6, 7 The functional decline experienced by older persons during hospitalization is a strong predictor of length of stay, nursing home placement, and mortality.8 Loss of ambulatory ability specifically has been noted to occur in 17%‐65% of hospitalized older patients, usually within days of admission, with few recovering this ability prior to discharge.2, 4, 5, 9 Importantly, ambulatory ability is significantly associated with a decline in other ADLs.9

Although several studies have explored risk factors associated with general functional decline during hospitalization,4, 7, 10, 11 only one study specifically examined risk factors associated with loss of ambulatory ability. In a cohort of participants who were largely independently performing ADLs on admission to the hospital, Mahoney et al. found age 85 years, white race, use of a walker, and functional impairment prior to being hospitalized were significant predictors of newly having walking dependence.12

Ambulatory ability could also be affected by a variety of other factors not examined in the Mahoney et al. study; these include severity of illness, bed rest, and hospital‐related treatments such as restraints or urinary catheters. In addition, little is known about predictors of recovery of ambulatory ability in patients expected to have activity limitations on admission who are dependent in most or all ADLs. The deconditioning associated with bed rest and reduced mobility has been described as one of the most predictable causes of functional decline, including loss of ambulatory ability, observed in older hospitalized patients.13 In one study, patients whose activity was limited to a bed or chair during hospitalization were 5.6 times more likely to develop functional decline than those who walked at all, even after controlling for other covariates including severity of illness and comorbidity.14 Those patients with both activity limitations and dependence in most or all ADLs represent an important subset of all hospitalized older patients who might be expected to be at higher risk of developing new ambulatory dependence. The ability to identify, at admission, those patients who will recover ambulatory ability may have important implications for discharge planning as well as for the development of preventive strategies.

The objective of the present study was to define patient demographic, illness severity, comorbid illness, and hospital‐related variables that are independent and significant predictors of regaining ambulatory ability prior to hospital discharge in a cohort of patients who had significant activity limitations and functional impairment at the time of admission to the hospital.

METHODS

RESULTS

For this study, 286 participants met all eligibility criteria, with 119 (42%) regaining ambulatory ability during hospitalization. Mean age of study participants was 73 9 years, with 12% of participants more than 85 years old. On admission, 214 patients (75%) were dependent in all 7 ADLs. Mean length of stay ( SD) was 12.3 6.5 days, with a range of 1‐31 days. Table 1 presents the cohort characteristics and the unadjusted effects of each variable for predicting those who did and did not recover ambulatory ability by characteristic. The P value of a variable in Table 1 indicates how significant that variable was, as determined with a simple logistic regression analysis. The unadjusted odds ratio of each variable is presented in Table 2 in order to facilitate comparisons with the adjusted odds ratios from the multivariate models. Analysis of the effects of the unadjusted single variables showed that age was a significant predictor, with older patients less likely to regain ambulatory ability. Several global health measures and hospital‐related factors were also significantly different between those who recovered and those who did not recover ambulatory ability, as summarized in Table 1. Importantly, length of stay was not significantly associated with recovery.

Table 2 shows the effects of adjusting the model for important covariables in a sequential fashion. Model 1 shows that longer length of stay and older age are associated with reduced odds of regaining ambulatory ability after adjusting for other demographic variables. However, age and length of stay were no longer significant after adjusting for global health measures and specific medical problems (Model 2). As demonstrated by the full model (Model 4), the participants who regained ambulatory ability were more likely to be unmarried, have a longer physician‐rated life expectancy, not have a diagnosis of deep vein thrombosis, not have physical restraints or a urinary catheter in use, and have greater bed mobility at admission.

Predictors that remained significant in the final multivariate model (Model 4) were summed in order to determine the proportion of patients who had one predictor versus those who had more than one predictor. Figure 1 shows the percentage of those who recovered ambulatory ability according to number of predictors, ranging from one to all 6 predictors. The results demonstrate a graded relationship, with number of predictors positively correlated with percentage of those who recovered ambulatory ability (P < .0001).

Figure 1

Patients who died prior to discharge were more likely to be male and have altered mental status, a urinary catheter, restraints, a shorter physician‐rated life expectancy, lower admission bed mobility, and increased severity of illness compared to those who survived. When the 34 patients who died were excluded from the multivariate analysis, the factors predicting ambulatory recovery were unchanged from those reported in Model 4 of Table 2.

DISCUSSION

In this study of older hospitalized patients, approximately 40% were able to regain their ambulatory ability despite being limited to the bed or chair on admission, having significant severity of illness, high level of use of restraints, and functional dependence on admission. Predictors of regaining ambulatory ability were identifiable at hospital admission. These predictors also were correlated with recovery of ambulatory ability in a graded fashion, lending support for the direct relationship between these predictors and recovery.

Physician‐rated life expectancy, a simple assessment that combines a physician's medical knowledge and clinical acumen, was demonstrated in our study to be a better predictor of recovery of ambulatory ability than more elaborate measures like the APACHE II17 and Charlson Comorbidity Index.16 This assessment can easily be done at the bedside and may help to guide discharge planning for the hospitalist physician. Nurse‐rated bed mobility at time of admission, which may reflect aspects of illness severity and cognitive status, was also able to predict recovery.

Of the 6 factors found to be independent predictors of recovery of ambulatory ability, 4 were related to mobility: lack of a DVT diagnosis, absence of a urinary catheter, absence of restraints, and nurse‐rated bed mobility at admission. In the group of patients initially expected to be confined to a bed or chair, those with additional mobility‐reducing factors, such as catheters and restraints, were less likely to recover, even after controlling for illness severity and comorbidity.

Marital status was not found to predict recovery of ambulatory ability in the simple unadjusted bivariate analysis, but after adjusting for other demographic, global health, and disease‐related variables, those who were unmarried were more likely to recover ambulatory ability. These interesting covariate‐adjusted effects for marital status have not been previously reported in the literature. One large study of the impact of marital status on hospital outcomes demonstrated those who were unmarried were more likely to require discharge to a nursing home and had slightly higher hospital costs and longer length of stay.19 Our findings may be related to such patients lacking support other than from themselves, with the possibility of being discharged to a nursing home an incentive to get up and walk.

Although age, race, and previous ADL status were found to be significant predictors in a previous study,12 we did not find this in our sample. This may be because, unlike in other studies, most of our patients had significant functional impairment on admission. The importance of age in our cohort disappeared when illness severity and comorbidities were added to the model.

The strengths of this study include having comprehensive patient‐related data on demographic, illness severity, comorbidity, and hospital‐related factors available, which enabled detailed analyses of predictors for regaining ambulatory ability. In particular, the ability to examine such factors as bed rest and hospital‐related treatments like restraint and catheter use, enabled this study to add significantly to the available knowledge of predictors of ambulatory recovery. The use of nurse interviews to obtain patient‐related data has been demonstrated in previous studies to be a preferred method of collecting data when compared to patient self‐report.20, 21 Examination of these factors in a cohort of patients who would be expected to be at very high risk for remaining bed‐ or chair bound, given their admission activity limitations and functional dependence, is also noteworthy.

Several important limitations deserve comment. Since the data were collected, average length of hospital stay generally has decreased. However, the patient population we studied continue to experience longer hospital stays than functionally intact patients. One recent study demonstrated that the length of hospital stay of patients who were dependent in one or more ADLs on admission was 35% longer than that of those not ADL dependent at admission.22 Seventy‐five percent of those in the present study cohort were dependent in all 7 ADLs and had a mean length of stay of 12.3 days. Despite the longer mean length of stay, 70% of those in the study cohort were discharged within 2 weeks of admission. In the university hospital where this research was conducted, mean length of stay ( SD) in 2004 was 6 8.6 days, but 10% of patients 55 years of age and older remained in the hospital for more than 2 weeks.23 This suggests there continue to be long‐stay patients in the current hospital environment, to which these findings may apply.

Standards of practice, such as for use of restraints, have also changed. In 1992, between 7.4% and 17% of all hospitalized medical patients were restrained, according to a literature review.24 A 1998 survey of 3 hospitals found the prevalence of restraints still ranged from 3.9% to 8.2% and noted that among the most common reasons reported for using restraints were to prevent patient disruption of therapy, to confine confused patients, and to reduce the number of falls.25 Thus, our study cohort would be more likely to be restrained, even in the current hospital environment, given that 43% of the cohort had altered mental status and that most were at risk for falls because of their poor functional status.26

Nevertheless, even though the use of restraints has declined since the data were collected for this study, this should affect neither the internal validity of the results nor the ability to address the question of what factors predict recovery of ambulatory ability. Indeed, the inclusion of patients on whom restraints are frequently used emphasizes the need for continued diligence in creating a restraint‐free environment in our hospitals. Data about the use of physical therapy services were not available in the study. Therefore, it is unknown to what extent the use of these services encouraged ambulation.

In this observational study, we found 6 factors associated with regaining ambulatory ability among hospitalized patients who had significant activity limitations and functional dependence on admission. These findings suggest predictors easily assessed by the hospitalist physician can help to identify those patients most likely to recover ambulatory ability prior to discharge. It also demonstrates the importance of mobility in maintaining function, given that many of the predictors are factors that either impede mobility such as restraints and urinary catheters or measure mobility such as admission bed mobility. Last, recognizing physician‐rated life expectancy as a strong independent predictor of recovery of ambulatory ability should encourage hospitalist physicians to continue to use their greatest tool, their clinical judgment, to determine who will recover ambulatory ability.

As most of these predictors can be identified on admission or shortly thereafter, these factors may be useful in helping physicians and other health care providers to predict the potential patients have to recover ambulatory ability. This information may help physicians identify patients who might benefit from early mobility programs, placement on hospital units where mobility will be enhanced, or the early initiation of discharge planning for those patients identified as unlikely to regain ambulation ability. In addition, addressing factors that are potentially modifiable, such as low bed mobility and the use of urinary catheters and restraints may not only improve the chance of recovering ambulatory ability but would also improve the quality of care provided to older patients.

Functional decline, defined as loss of the ability to care for oneself, commonly occurs during hospitalization, being experienced by up to 65% of older adults.14 Frequently, recovery of functional ability does not occur by the time of discharge from the hospital, despite resolution of the medical condition responsible for admission to the hospital.1, 5 Causes of the declining ability to perform activities of daily living (ADLs) are multiple and include both acute illness and adverse events associated with hospitalization.4, 6, 7 The functional decline experienced by older persons during hospitalization is a strong predictor of length of stay, nursing home placement, and mortality.8 Loss of ambulatory ability specifically has been noted to occur in 17%‐65% of hospitalized older patients, usually within days of admission, with few recovering this ability prior to discharge.2, 4, 5, 9 Importantly, ambulatory ability is significantly associated with a decline in other ADLs.9

Although several studies have explored risk factors associated with general functional decline during hospitalization,4, 7, 10, 11 only one study specifically examined risk factors associated with loss of ambulatory ability. In a cohort of participants who were largely independently performing ADLs on admission to the hospital, Mahoney et al. found age 85 years, white race, use of a walker, and functional impairment prior to being hospitalized were significant predictors of newly having walking dependence.12

Ambulatory ability could also be affected by a variety of other factors not examined in the Mahoney et al. study; these include severity of illness, bed rest, and hospital‐related treatments such as restraints or urinary catheters. In addition, little is known about predictors of recovery of ambulatory ability in patients expected to have activity limitations on admission who are dependent in most or all ADLs. The deconditioning associated with bed rest and reduced mobility has been described as one of the most predictable causes of functional decline, including loss of ambulatory ability, observed in older hospitalized patients.13 In one study, patients whose activity was limited to a bed or chair during hospitalization were 5.6 times more likely to develop functional decline than those who walked at all, even after controlling for other covariates including severity of illness and comorbidity.14 Those patients with both activity limitations and dependence in most or all ADLs represent an important subset of all hospitalized older patients who might be expected to be at higher risk of developing new ambulatory dependence. The ability to identify, at admission, those patients who will recover ambulatory ability may have important implications for discharge planning as well as for the development of preventive strategies.

The objective of the present study was to define patient demographic, illness severity, comorbid illness, and hospital‐related variables that are independent and significant predictors of regaining ambulatory ability prior to hospital discharge in a cohort of patients who had significant activity limitations and functional impairment at the time of admission to the hospital.

METHODS

RESULTS

For this study, 286 participants met all eligibility criteria, with 119 (42%) regaining ambulatory ability during hospitalization. Mean age of study participants was 73 9 years, with 12% of participants more than 85 years old. On admission, 214 patients (75%) were dependent in all 7 ADLs. Mean length of stay ( SD) was 12.3 6.5 days, with a range of 1‐31 days. Table 1 presents the cohort characteristics and the unadjusted effects of each variable for predicting those who did and did not recover ambulatory ability by characteristic. The P value of a variable in Table 1 indicates how significant that variable was, as determined with a simple logistic regression analysis. The unadjusted odds ratio of each variable is presented in Table 2 in order to facilitate comparisons with the adjusted odds ratios from the multivariate models. Analysis of the effects of the unadjusted single variables showed that age was a significant predictor, with older patients less likely to regain ambulatory ability. Several global health measures and hospital‐related factors were also significantly different between those who recovered and those who did not recover ambulatory ability, as summarized in Table 1. Importantly, length of stay was not significantly associated with recovery.

Table 2 shows the effects of adjusting the model for important covariables in a sequential fashion. Model 1 shows that longer length of stay and older age are associated with reduced odds of regaining ambulatory ability after adjusting for other demographic variables. However, age and length of stay were no longer significant after adjusting for global health measures and specific medical problems (Model 2). As demonstrated by the full model (Model 4), the participants who regained ambulatory ability were more likely to be unmarried, have a longer physician‐rated life expectancy, not have a diagnosis of deep vein thrombosis, not have physical restraints or a urinary catheter in use, and have greater bed mobility at admission.

Predictors that remained significant in the final multivariate model (Model 4) were summed in order to determine the proportion of patients who had one predictor versus those who had more than one predictor. Figure 1 shows the percentage of those who recovered ambulatory ability according to number of predictors, ranging from one to all 6 predictors. The results demonstrate a graded relationship, with number of predictors positively correlated with percentage of those who recovered ambulatory ability (P < .0001).

Figure 1

Patients who died prior to discharge were more likely to be male and have altered mental status, a urinary catheter, restraints, a shorter physician‐rated life expectancy, lower admission bed mobility, and increased severity of illness compared to those who survived. When the 34 patients who died were excluded from the multivariate analysis, the factors predicting ambulatory recovery were unchanged from those reported in Model 4 of Table 2.

DISCUSSION

In this study of older hospitalized patients, approximately 40% were able to regain their ambulatory ability despite being limited to the bed or chair on admission, having significant severity of illness, high level of use of restraints, and functional dependence on admission. Predictors of regaining ambulatory ability were identifiable at hospital admission. These predictors also were correlated with recovery of ambulatory ability in a graded fashion, lending support for the direct relationship between these predictors and recovery.

Physician‐rated life expectancy, a simple assessment that combines a physician's medical knowledge and clinical acumen, was demonstrated in our study to be a better predictor of recovery of ambulatory ability than more elaborate measures like the APACHE II17 and Charlson Comorbidity Index.16 This assessment can easily be done at the bedside and may help to guide discharge planning for the hospitalist physician. Nurse‐rated bed mobility at time of admission, which may reflect aspects of illness severity and cognitive status, was also able to predict recovery.

Of the 6 factors found to be independent predictors of recovery of ambulatory ability, 4 were related to mobility: lack of a DVT diagnosis, absence of a urinary catheter, absence of restraints, and nurse‐rated bed mobility at admission. In the group of patients initially expected to be confined to a bed or chair, those with additional mobility‐reducing factors, such as catheters and restraints, were less likely to recover, even after controlling for illness severity and comorbidity.

Marital status was not found to predict recovery of ambulatory ability in the simple unadjusted bivariate analysis, but after adjusting for other demographic, global health, and disease‐related variables, those who were unmarried were more likely to recover ambulatory ability. These interesting covariate‐adjusted effects for marital status have not been previously reported in the literature. One large study of the impact of marital status on hospital outcomes demonstrated those who were unmarried were more likely to require discharge to a nursing home and had slightly higher hospital costs and longer length of stay.19 Our findings may be related to such patients lacking support other than from themselves, with the possibility of being discharged to a nursing home an incentive to get up and walk.

Although age, race, and previous ADL status were found to be significant predictors in a previous study,12 we did not find this in our sample. This may be because, unlike in other studies, most of our patients had significant functional impairment on admission. The importance of age in our cohort disappeared when illness severity and comorbidities were added to the model.

The strengths of this study include having comprehensive patient‐related data on demographic, illness severity, comorbidity, and hospital‐related factors available, which enabled detailed analyses of predictors for regaining ambulatory ability. In particular, the ability to examine such factors as bed rest and hospital‐related treatments like restraint and catheter use, enabled this study to add significantly to the available knowledge of predictors of ambulatory recovery. The use of nurse interviews to obtain patient‐related data has been demonstrated in previous studies to be a preferred method of collecting data when compared to patient self‐report.20, 21 Examination of these factors in a cohort of patients who would be expected to be at very high risk for remaining bed‐ or chair bound, given their admission activity limitations and functional dependence, is also noteworthy.

Several important limitations deserve comment. Since the data were collected, average length of hospital stay generally has decreased. However, the patient population we studied continue to experience longer hospital stays than functionally intact patients. One recent study demonstrated that the length of hospital stay of patients who were dependent in one or more ADLs on admission was 35% longer than that of those not ADL dependent at admission.22 Seventy‐five percent of those in the present study cohort were dependent in all 7 ADLs and had a mean length of stay of 12.3 days. Despite the longer mean length of stay, 70% of those in the study cohort were discharged within 2 weeks of admission. In the university hospital where this research was conducted, mean length of stay ( SD) in 2004 was 6 8.6 days, but 10% of patients 55 years of age and older remained in the hospital for more than 2 weeks.23 This suggests there continue to be long‐stay patients in the current hospital environment, to which these findings may apply.

Standards of practice, such as for use of restraints, have also changed. In 1992, between 7.4% and 17% of all hospitalized medical patients were restrained, according to a literature review.24 A 1998 survey of 3 hospitals found the prevalence of restraints still ranged from 3.9% to 8.2% and noted that among the most common reasons reported for using restraints were to prevent patient disruption of therapy, to confine confused patients, and to reduce the number of falls.25 Thus, our study cohort would be more likely to be restrained, even in the current hospital environment, given that 43% of the cohort had altered mental status and that most were at risk for falls because of their poor functional status.26

Nevertheless, even though the use of restraints has declined since the data were collected for this study, this should affect neither the internal validity of the results nor the ability to address the question of what factors predict recovery of ambulatory ability. Indeed, the inclusion of patients on whom restraints are frequently used emphasizes the need for continued diligence in creating a restraint‐free environment in our hospitals. Data about the use of physical therapy services were not available in the study. Therefore, it is unknown to what extent the use of these services encouraged ambulation.

In this observational study, we found 6 factors associated with regaining ambulatory ability among hospitalized patients who had significant activity limitations and functional dependence on admission. These findings suggest predictors easily assessed by the hospitalist physician can help to identify those patients most likely to recover ambulatory ability prior to discharge. It also demonstrates the importance of mobility in maintaining function, given that many of the predictors are factors that either impede mobility such as restraints and urinary catheters or measure mobility such as admission bed mobility. Last, recognizing physician‐rated life expectancy as a strong independent predictor of recovery of ambulatory ability should encourage hospitalist physicians to continue to use their greatest tool, their clinical judgment, to determine who will recover ambulatory ability.

As most of these predictors can be identified on admission or shortly thereafter, these factors may be useful in helping physicians and other health care providers to predict the potential patients have to recover ambulatory ability. This information may help physicians identify patients who might benefit from early mobility programs, placement on hospital units where mobility will be enhanced, or the early initiation of discharge planning for those patients identified as unlikely to regain ambulation ability. In addition, addressing factors that are potentially modifiable, such as low bed mobility and the use of urinary catheters and restraints may not only improve the chance of recovering ambulatory ability but would also improve the quality of care provided to older patients.

• CLINICAL QUESTION: Is noninvasive positive pressure ventilation effective in managing patients with acute cardiogenic pulmonary edema?

• BOTTOM LINE: Patients with acute cardiogenic pulmonary edema treated with noninvasive positive pressure ventilation (NIPPV) are less likely than those receiving standard care to die in the hospital or to require mechanical ventilation. (LOE = 1a)

• REFERENCE: Peter JV, Moran JL, Phillips‐Hughes J, Graham P, Bersten AD. Effect of non‐invasive positive pressure ventilation (NIPPV) on mortality in patients with acute cardiogenic pulmonary oedema: a meta‐analysis. Lancet 2006;367:1155‐1163.

• STUDY DESIGN: Meta‐analysis (randomized controlled trials)

• SETTING: Inpatient (any location)

• SYNOPSIS: This team systematically reviewed multiple databases, using a sensible search strategy,q to find 23 small randomized controlled trials of NIPPV. NIPPV included continuous positive airway pressure or bilevel ventilation. The data were extracted independently by 2 investigators, with discrepancies resolved by consensus. The authors don't report if the decision to include or exclude studies was similarly done independently. Additionally, they don't say if they looked for unpublished studies. The eligible studies included more than 1300 patients with cardiogenic pulmonary edema. NIPPV was more effective than standard care in preventing in‐hospital mortality (11.7% vs 21.3%; number needed to treat [NNT] =11; 95% CI, 7 ‐ 21), with no difference between continuous positive airway pressure and bilevel ventilation. Similarly, patients receiving NIPPV required mechanical ventilation less frequently (11.9% vs 28.1%; NNT = 7; 5 ‐ 9), with no difference in outcomes between continuous positive airway pressure and bilevel ventilation. The data were fairly consistent across the studies. Since there is a possibility of publication bias in favor of positive results, the results of a mega‐trial (if one occurs) may not look this good.

• CLINICAL QUESTION: Is noninvasive positive pressure ventilation effective in managing patients with acute cardiogenic pulmonary edema?

• BOTTOM LINE: Patients with acute cardiogenic pulmonary edema treated with noninvasive positive pressure ventilation (NIPPV) are less likely than those receiving standard care to die in the hospital or to require mechanical ventilation. (LOE = 1a)

• REFERENCE: Peter JV, Moran JL, Phillips‐Hughes J, Graham P, Bersten AD. Effect of non‐invasive positive pressure ventilation (NIPPV) on mortality in patients with acute cardiogenic pulmonary oedema: a meta‐analysis. Lancet 2006;367:1155‐1163.

• STUDY DESIGN: Meta‐analysis (randomized controlled trials)

• SETTING: Inpatient (any location)

• SYNOPSIS: This team systematically reviewed multiple databases, using a sensible search strategy,q to find 23 small randomized controlled trials of NIPPV. NIPPV included continuous positive airway pressure or bilevel ventilation. The data were extracted independently by 2 investigators, with discrepancies resolved by consensus. The authors don't report if the decision to include or exclude studies was similarly done independently. Additionally, they don't say if they looked for unpublished studies. The eligible studies included more than 1300 patients with cardiogenic pulmonary edema. NIPPV was more effective than standard care in preventing in‐hospital mortality (11.7% vs 21.3%; number needed to treat [NNT] =11; 95% CI, 7 ‐ 21), with no difference between continuous positive airway pressure and bilevel ventilation. Similarly, patients receiving NIPPV required mechanical ventilation less frequently (11.9% vs 28.1%; NNT = 7; 5 ‐ 9), with no difference in outcomes between continuous positive airway pressure and bilevel ventilation. The data were fairly consistent across the studies. Since there is a possibility of publication bias in favor of positive results, the results of a mega‐trial (if one occurs) may not look this good.

• CLINICAL QUESTION: Is noninvasive positive pressure ventilation effective in managing patients with acute cardiogenic pulmonary edema?

• BOTTOM LINE: Patients with acute cardiogenic pulmonary edema treated with noninvasive positive pressure ventilation (NIPPV) are less likely than those receiving standard care to die in the hospital or to require mechanical ventilation. (LOE = 1a)

• REFERENCE: Peter JV, Moran JL, Phillips‐Hughes J, Graham P, Bersten AD. Effect of non‐invasive positive pressure ventilation (NIPPV) on mortality in patients with acute cardiogenic pulmonary oedema: a meta‐analysis. Lancet 2006;367:1155‐1163.

• STUDY DESIGN: Meta‐analysis (randomized controlled trials)

• SETTING: Inpatient (any location)

• SYNOPSIS: This team systematically reviewed multiple databases, using a sensible search strategy,q to find 23 small randomized controlled trials of NIPPV. NIPPV included continuous positive airway pressure or bilevel ventilation. The data were extracted independently by 2 investigators, with discrepancies resolved by consensus. The authors don't report if the decision to include or exclude studies was similarly done independently. Additionally, they don't say if they looked for unpublished studies. The eligible studies included more than 1300 patients with cardiogenic pulmonary edema. NIPPV was more effective than standard care in preventing in‐hospital mortality (11.7% vs 21.3%; number needed to treat [NNT] =11; 95% CI, 7 ‐ 21), with no difference between continuous positive airway pressure and bilevel ventilation. Similarly, patients receiving NIPPV required mechanical ventilation less frequently (11.9% vs 28.1%; NNT = 7; 5 ‐ 9), with no difference in outcomes between continuous positive airway pressure and bilevel ventilation. The data were fairly consistent across the studies. Since there is a possibility of publication bias in favor of positive results, the results of a mega‐trial (if one occurs) may not look this good.

• CLINICAL QUESTION: Can noninvasive imaging replace invasive testing in patients with suspected carotid artery disease?

• BOTTOM LINE: Noninvasive testing, especially contrast‐enhanced magnetic resonance angiography (CEMRA), compares very favorably with invasive angiography. Since noninvasive testing appears to be less accurate in patients with less severe stenosis, a reasonable strategy might begin with CEMRA. If CEMRA demonstrates a greater than 70% stenosis, the diagnosis is settled. If the stenosis appears to be less than 70%, invasive angiography might be considered. Of course, this diagnostic approach needs formal evaluation. (LOE = 1a‐).

• REFERENCE: Wardlaw JM, Chappell FM, Best JJ, Wartolowska K, Berry E; NHS Research and Development Health Technology Assessment Carotid Stenosis Imaging Group. Non‐invasive imaging compared with intra‐arterial angiography in the diagnosis of symptomatic carotid stenosis: a meta‐analysis. Lancet 2006;367:1503‐1512.

• STUDY DESIGN: Systematic review

• FUNDING: Government

• SETTING: Various (meta‐analysis)

• SYNOPSIS: Two members of this research team searched MEDLINE and EMBASE and hand‐searched several journals to identify 41 prospective studies of at least 20 patients with suspected carotid artery disease who underwent 1 or more noninvasive test and invasive angiography. Any disagreements were settled by discussion with a third reviewer. Similarly, 2 reviewers extracted the data and discrepancies were arbitrated by a third reviewer. The identified studies included a total of 2541 patients. Overall, CEMRA generally performed best, although computed tomographic angiography, magnetic resonance angiography, and Doppler ultrasound also performed well. The authors found some variability in test performance, mainly due to differences in accuracy on the basis of the degree of stenosis. Nonetheless, depending on the degree of stenosis, CEMRA provided the most diagnostic information (positive likelihood ratio ranged from 13 to 26 and negative likelihood ratio from 0.04 to 0.24). Generally speaking, the noninvasive tests were not as accurate in patients with moderate stenosis (50%‐69%). Since this group also has a narrow surgical risk‐benefit margin, diagnostic certainty is critical. Finally, CEMRA results may be biased because of small study sizes and, as a new technology, the potential for reporting overly positive studies. Stay tuned.

• CLINICAL QUESTION: Can noninvasive imaging replace invasive testing in patients with suspected carotid artery disease?

• BOTTOM LINE: Noninvasive testing, especially contrast‐enhanced magnetic resonance angiography (CEMRA), compares very favorably with invasive angiography. Since noninvasive testing appears to be less accurate in patients with less severe stenosis, a reasonable strategy might begin with CEMRA. If CEMRA demonstrates a greater than 70% stenosis, the diagnosis is settled. If the stenosis appears to be less than 70%, invasive angiography might be considered. Of course, this diagnostic approach needs formal evaluation. (LOE = 1a‐).

• REFERENCE: Wardlaw JM, Chappell FM, Best JJ, Wartolowska K, Berry E; NHS Research and Development Health Technology Assessment Carotid Stenosis Imaging Group. Non‐invasive imaging compared with intra‐arterial angiography in the diagnosis of symptomatic carotid stenosis: a meta‐analysis. Lancet 2006;367:1503‐1512.

• STUDY DESIGN: Systematic review

• FUNDING: Government

• SETTING: Various (meta‐analysis)

• SYNOPSIS: Two members of this research team searched MEDLINE and EMBASE and hand‐searched several journals to identify 41 prospective studies of at least 20 patients with suspected carotid artery disease who underwent 1 or more noninvasive test and invasive angiography. Any disagreements were settled by discussion with a third reviewer. Similarly, 2 reviewers extracted the data and discrepancies were arbitrated by a third reviewer. The identified studies included a total of 2541 patients. Overall, CEMRA generally performed best, although computed tomographic angiography, magnetic resonance angiography, and Doppler ultrasound also performed well. The authors found some variability in test performance, mainly due to differences in accuracy on the basis of the degree of stenosis. Nonetheless, depending on the degree of stenosis, CEMRA provided the most diagnostic information (positive likelihood ratio ranged from 13 to 26 and negative likelihood ratio from 0.04 to 0.24). Generally speaking, the noninvasive tests were not as accurate in patients with moderate stenosis (50%‐69%). Since this group also has a narrow surgical risk‐benefit margin, diagnostic certainty is critical. Finally, CEMRA results may be biased because of small study sizes and, as a new technology, the potential for reporting overly positive studies. Stay tuned.

• CLINICAL QUESTION: Can noninvasive imaging replace invasive testing in patients with suspected carotid artery disease?

• BOTTOM LINE: Noninvasive testing, especially contrast‐enhanced magnetic resonance angiography (CEMRA), compares very favorably with invasive angiography. Since noninvasive testing appears to be less accurate in patients with less severe stenosis, a reasonable strategy might begin with CEMRA. If CEMRA demonstrates a greater than 70% stenosis, the diagnosis is settled. If the stenosis appears to be less than 70%, invasive angiography might be considered. Of course, this diagnostic approach needs formal evaluation. (LOE = 1a‐).

• REFERENCE: Wardlaw JM, Chappell FM, Best JJ, Wartolowska K, Berry E; NHS Research and Development Health Technology Assessment Carotid Stenosis Imaging Group. Non‐invasive imaging compared with intra‐arterial angiography in the diagnosis of symptomatic carotid stenosis: a meta‐analysis. Lancet 2006;367:1503‐1512.

• STUDY DESIGN: Systematic review

• FUNDING: Government

• SETTING: Various (meta‐analysis)

• SYNOPSIS: Two members of this research team searched MEDLINE and EMBASE and hand‐searched several journals to identify 41 prospective studies of at least 20 patients with suspected carotid artery disease who underwent 1 or more noninvasive test and invasive angiography. Any disagreements were settled by discussion with a third reviewer. Similarly, 2 reviewers extracted the data and discrepancies were arbitrated by a third reviewer. The identified studies included a total of 2541 patients. Overall, CEMRA generally performed best, although computed tomographic angiography, magnetic resonance angiography, and Doppler ultrasound also performed well. The authors found some variability in test performance, mainly due to differences in accuracy on the basis of the degree of stenosis. Nonetheless, depending on the degree of stenosis, CEMRA provided the most diagnostic information (positive likelihood ratio ranged from 13 to 26 and negative likelihood ratio from 0.04 to 0.24). Generally speaking, the noninvasive tests were not as accurate in patients with moderate stenosis (50%‐69%). Since this group also has a narrow surgical risk‐benefit margin, diagnostic certainty is critical. Finally, CEMRA results may be biased because of small study sizes and, as a new technology, the potential for reporting overly positive studies. Stay tuned.

Of the many skills a hospitalist nurtures and develops, seeing and processing visual images is paramount. From the moment we enter the hospital each day, we look at patient faces, trying to fathom their levels of pain, of illness, and of response to therapy. We look at their rashes, facial droops, surgical wounds, and neck vein elevation. When we are done visually scrutinizing our patients, we inspect their rhythm strips or electrocardiograms and then move on to their X‐rays and advanced imaging.

Although we are, in a sense, slaves to the images before us, we also enjoy medicine for the challenge that these images providethere is always something novel to see. For the experienced clinician, a deep sense of satisfaction perfuses our limbic system with the quick recognition of the delta wave of Wolf‐Parkinson‐White on an EKG or the first vesicle of a zoster outbreak in a patient with initially unexplained cutaneous pain. What we recognize easily tends to come from having seen something beforefor better or worse, we depend on pattern recognition.

Unfortunately, we are all busier than we like and receive more journals than we have time to read. To make JHM even more germane and stimulating, we are initiating Hospital Images Dx. The main goal of Hospital Images Dx will be to show interesting images that a hospitalist might encounter, both the common and the obscure. Images, whether subtle or awe‐inspiring, should generally be able to speak the proverbial thousand words. To supplement those thousand words spoken by the submitted image, accompanying text will be limited to 250 words. The text should give a brief clinical summary of the patient's problem, relevant adjunct data, and 1 or 2 succinct teaching points related to the image. Our goal, simply stated, is for the reader to walk away after reading Hospital Images Dx with an image and a couple of key teaching points stored away for a rainy day.

We anticipate a substantial number of exciting submissions to Hospital Images Dx. Health care providers clearly get excited about the things they see as well as about the recordsthe imagesthat document both mundane and unusual encounters with a patient or a patient's data. We hope to tap into this enthusiasm and to teach a few things along the way. The editors look forward to receiving your Hospital Images Dx submission soon!

Of the many skills a hospitalist nurtures and develops, seeing and processing visual images is paramount. From the moment we enter the hospital each day, we look at patient faces, trying to fathom their levels of pain, of illness, and of response to therapy. We look at their rashes, facial droops, surgical wounds, and neck vein elevation. When we are done visually scrutinizing our patients, we inspect their rhythm strips or electrocardiograms and then move on to their X‐rays and advanced imaging.

Although we are, in a sense, slaves to the images before us, we also enjoy medicine for the challenge that these images providethere is always something novel to see. For the experienced clinician, a deep sense of satisfaction perfuses our limbic system with the quick recognition of the delta wave of Wolf‐Parkinson‐White on an EKG or the first vesicle of a zoster outbreak in a patient with initially unexplained cutaneous pain. What we recognize easily tends to come from having seen something beforefor better or worse, we depend on pattern recognition.

Unfortunately, we are all busier than we like and receive more journals than we have time to read. To make JHM even more germane and stimulating, we are initiating Hospital Images Dx. The main goal of Hospital Images Dx will be to show interesting images that a hospitalist might encounter, both the common and the obscure. Images, whether subtle or awe‐inspiring, should generally be able to speak the proverbial thousand words. To supplement those thousand words spoken by the submitted image, accompanying text will be limited to 250 words. The text should give a brief clinical summary of the patient's problem, relevant adjunct data, and 1 or 2 succinct teaching points related to the image. Our goal, simply stated, is for the reader to walk away after reading Hospital Images Dx with an image and a couple of key teaching points stored away for a rainy day.

We anticipate a substantial number of exciting submissions to Hospital Images Dx. Health care providers clearly get excited about the things they see as well as about the recordsthe imagesthat document both mundane and unusual encounters with a patient or a patient's data. We hope to tap into this enthusiasm and to teach a few things along the way. The editors look forward to receiving your Hospital Images Dx submission soon!

Of the many skills a hospitalist nurtures and develops, seeing and processing visual images is paramount. From the moment we enter the hospital each day, we look at patient faces, trying to fathom their levels of pain, of illness, and of response to therapy. We look at their rashes, facial droops, surgical wounds, and neck vein elevation. When we are done visually scrutinizing our patients, we inspect their rhythm strips or electrocardiograms and then move on to their X‐rays and advanced imaging.

Although we are, in a sense, slaves to the images before us, we also enjoy medicine for the challenge that these images providethere is always something novel to see. For the experienced clinician, a deep sense of satisfaction perfuses our limbic system with the quick recognition of the delta wave of Wolf‐Parkinson‐White on an EKG or the first vesicle of a zoster outbreak in a patient with initially unexplained cutaneous pain. What we recognize easily tends to come from having seen something beforefor better or worse, we depend on pattern recognition.

Unfortunately, we are all busier than we like and receive more journals than we have time to read. To make JHM even more germane and stimulating, we are initiating Hospital Images Dx. The main goal of Hospital Images Dx will be to show interesting images that a hospitalist might encounter, both the common and the obscure. Images, whether subtle or awe‐inspiring, should generally be able to speak the proverbial thousand words. To supplement those thousand words spoken by the submitted image, accompanying text will be limited to 250 words. The text should give a brief clinical summary of the patient's problem, relevant adjunct data, and 1 or 2 succinct teaching points related to the image. Our goal, simply stated, is for the reader to walk away after reading Hospital Images Dx with an image and a couple of key teaching points stored away for a rainy day.

We anticipate a substantial number of exciting submissions to Hospital Images Dx. Health care providers clearly get excited about the things they see as well as about the recordsthe imagesthat document both mundane and unusual encounters with a patient or a patient's data. We hope to tap into this enthusiasm and to teach a few things along the way. The editors look forward to receiving your Hospital Images Dx submission soon!

Because as many as 90% of all blood cultures grow no organisms1 developing rules that predict which patients are at the lowest risk of having bacteremia could improve the utilization of this test and markedly reduce its cost. Of the approximate 10% of cultures that do grow organisms, only about half represent true bacteremia (ie, true positives), whereas the other half are considered contaminants (ie, false positives)2; the latter are known to increase both the cost and duration of care.3 Accordingly, reducing the number of contaminants could also reduce the cost of care. We assessed which of these two strategies would be the most cost effective. Although only 6% of the blood cultures obtained at our hospital represented contaminants, their associated cost was more than twice that associated with the 87% of cultures that were true negatives.

METHODS

We conducted a retrospective review of microbiological results and hospital records of patients for whom blood cultures were obtained in January 2002 at Denver Health Medical Center, a 400‐bed university‐affiliated public safety net hospital. The study was given exempt status by the Colorado Multiple Institutional Review Board. Patients were identified using a preexisting laboratory database.

We adopted the definitions used by Bates et al.3 for the inclusion and exclusion criteria and the definition of a blood culture episode so that we could apply the financial data presented by these authors to our results. Briefly, a blood culture set was defined as a single venipuncture, regardless of the number of bottles sent for culturing, and a blood culture episode was defined as the 48‐hour period beginning when a blood culture was drawn. All sets within the same 48‐hour period were considered part of the same episode. Cultures that grew bacteria were classified as either true positive, representing bacteremia, or false positive, representing contaminants. Determination of whether a patient had a true‐positive culture versus a contaminant was made in a weekly conference attended by the chief of the Infectious Disease Division, an Infectious Disease fellow, and at least one microbiologist, during which the species of organism cultured and the associated clinical data for each patient were considered. Organisms considered to indicate false positives included diphtheroids, Bacillus sp, Propionibacterium sp, coagulase‐negative staphylococci, and micrococci. All other organisms were considered true positives in the setting of appropriate clinical criteria as specified by the CDC guidelines.4 Hospital charges and lengths of stay were obtained from our institutional database.

The cost associated with a true‐negative blood culture was determined by summing the charges for phlebotomy and microbiological testing obtained from the January 2005 Denver Health hospital charge master and applying the cost‐to‐charge ratio reported on the Medicare Cost Report for inpatient services (not including the costs of physician salaries and benefits).

The cost of a false positive was determined two ways: (a) adjusting the data reported by Bates et al.3 for changes in the Consumer Price Index5 and (b) comparing the actual hospital charges of the patients in our sample who had false‐positive cultures with those who did not (adjusting both by the hospital's inpatient cost‐to‐charge ratio, again not including the cost of physician salaries and benefits).

The length of stay and cost of care for patients with true‐ and false‐positive blood cultures were compared by chi‐square analysis. P < .05 was considered statistically significant. The data were not normally distributed and, as such, are presented as medians and interquartile ranges.

RESULTS

Table 1 summarizes the interpretation of the 939 blood cultures drawn in January 2002. Only 6 culture sets (0.6%) could not be classified. The positive predictive value of a positive blood culture was only 53%.

Laboratory charges for patients with true‐negative and false‐positive blood cultures in January 2002 are shown in Table 2. Annualized, the associated charges were $1,781,292, and the costs were $748,143.

Bates et al.3 found that false‐positive blood cultures increased the length of hospital stay by 4.5 days and increased total charges by $4385 over those for patients with no contaminants. This adjusted to $6878 in 2005 according to the Consumer Price Index.5 After grouping our blood cultures into episodes as defined by Bates et al. (Table 3), we had 41 episodes of contaminated blood cultures that would annualize to charges of $3,383,976 and costs of $1,421,270 after applying the cost‐to‐charge ratio.

The median length of hospital stay and total charges for the patients with true‐negative and false‐positive blood cultures at Denver Health in January 2002 are summarized in Table 4. Using this approach, patients with false‐positive blood cultures at our institution added 1450‐2200 extra hospital days and accrued additional charges of $4,305,000 and costs of $1,808,100.

DISCUSSION

The important finding of this study is that, despite there being nearly 15 times as many true‐negative blood cultures as false‐positive ones, the savings generated by reducing contaminants would be approximately twice that saved by reducing the true negatives (eg, a 50% reduction in the rate of contamination would reduce the total number of false‐positive episodes by 246 annually, saving $710,635‐$904,050, whereas reducing the true negatives by 50% would only save approximately $375,000.

There is no independent gold standard for evaluating the operating characteristics of a blood culture.6 Data from a series of repeated blood cultures represent the closest surrogate. Weinstein et al.7 drew at least 3 sets of cultures from 282 bacteremic patients and noted that bacteremia was documented in 91.5% of the first cultures, in 99.3% in 1 of the first 2 cultures, and in 99.6% in 1 of the first 3 cultures. Because 2 blood culture sets are drawn routinely, the difference between those 2 (if negative) and a third (if it represents a true positive) is 0.3% and would represent a false‐negative culture rate. Given that the true‐negative rate of blood cultures is 87%‐90%,1 the potential 0.3% false‐negative rate would not affect our analysis, and as such, we chose to ignore it. Accordingly, all sets of blood cultures with no growth were classified as true negatives.

Although we cannot show a cause‐and‐effect relationship between false‐positive cultures and the charges associated resulting from them, a recent study suggested that much of the excess length of stay of such patients is attributable to the false‐positive culture itself.8

Because health care costs have exceeded increases in general goods and services, adjusting the results of Bates et al.3 using the Consumer Price Index likely underestimated the projected cost of the false‐positive cultures. This limitation likely accounts for the observation that the difference in actual charges for our patients between those who did and those who did not have false‐positive blood cultures was greater than the cost of these false‐positive cultures as estimated by extrapolating from the data of Bates et al.3 Given the magnitude of the financial difference we observed, however, we suggest that this difference is not of sufficient size to alter our conclusion.

Physicians working at Denver Health are directly employed by the hospital, and the cost of physician salaries and benefits is included in the cost‐to‐charge ratio reported in our Medicare Cost Report. For purposes of this study, however, we elected to utilize a cost‐to‐charge ratio that was exclusive of physician salaries and benefits (ie, 0.42 rather than 0.66) because most hospitals in the United States do not employ their physicians. Accordingly, the costs we present underestimate the true cost to our institution by approximately 32% but are more representative of the costs of services provided by most hospitals in the United States.

Recent studies have shown that the rate of false‐positive cultures is higher when blood is drawn from indwelling catheters than when it is obtained by peripheral venipuncture.9, 10 The rates we cite from the literature2 and from our own institution (Table 1) are aggregate data that include samples drawn from both sites. Separating these would not alter our conclusion that a 50% reduction in false positives would save approximately twice as much as a 50% reduction in false negatives. These studies do, however, identify an important method for reducing false positives: sampling by venipuncture whenever possible, and only drawing through a catheter under very limited circumstances.

There are additional factors that favor a strategy of reducing contaminants over one that attempts to reduce the number of true‐negative cultures. First, reducing the total number of true‐negative blood cultures by 50% would require a very ambitious prediction rule that did not reduce the number of true positives to any meaningful extent. Prediction rules to reduce blood culture testing have been developed for patients with community‐acquired pneumonia, but the rules only reduced the number of cultures by 37% and, more importantly, left 11% of true bacteremias undetected.8 Reducing contaminants would have no effect on the detection of true positives, whereas any prediction rule would inevitably increase the risk of missing true bacteremia in at least a fraction of patients. Second, methods aimed at reducing contaminants can be implemented immediately, whereas deriving a prediction rule would take years to develop and test before it could be utilized. Third, implementing prediction rules may be difficult because many physicians prefer to rely on their clinical impressions.11

Reducing contaminants would require improving the technique by which blood cultures are obtained, with the objective of shifting a portion of false positives to true negatives. This might be accomplished in many ways: increasing the time spent on antiseptic scrubbing, improving the ways in which antiseptic devices are used, waiting for the antiseptic to air‐dry completely, choosing the antiseptic that is most effective in trials, drawing blood by venipuncture instead of through an indwelling catheter, limiting the number of venipuncture attempts before requiring a second site to be prepared, requiring all cultures be drawn by trained phlebotomists, and reducing phlebotomist turnover, among others. Denver Health has a 4‐page set of directions for phlebotomists to follow when obtaining blood cultures. Accordingly, there are numerous places the process could break down. Although having 2 phlebotomists involved (ie, one to perform the procedure and the other to observe and guide the first, assuring that all the appropriate steps are followed) might be considered an extraordinary step, our findings suggest the potential saving to the institution could far outweigh the additional personnel expense resulting from such an approach. Other potential solutions we have considered but not tested include providing a monthly salary bonus to the phlebotomist with the lowest contamination rate or giving bonuses to every phlebotomist who achieves a zero contamination rate.

In summary, we have concluded that the resource utilization associated with obtaining blood cultures can best be improved by reducing the small percentage of cultures that represent contaminants rather than by developing rules to reduce the much larger number of true negatives. The magnitude of the potential savings resulting from reducing contaminants is sufficiently large to warrant expending additional resources to accomplish this task.

Because as many as 90% of all blood cultures grow no organisms1 developing rules that predict which patients are at the lowest risk of having bacteremia could improve the utilization of this test and markedly reduce its cost. Of the approximate 10% of cultures that do grow organisms, only about half represent true bacteremia (ie, true positives), whereas the other half are considered contaminants (ie, false positives)2; the latter are known to increase both the cost and duration of care.3 Accordingly, reducing the number of contaminants could also reduce the cost of care. We assessed which of these two strategies would be the most cost effective. Although only 6% of the blood cultures obtained at our hospital represented contaminants, their associated cost was more than twice that associated with the 87% of cultures that were true negatives.

METHODS

We conducted a retrospective review of microbiological results and hospital records of patients for whom blood cultures were obtained in January 2002 at Denver Health Medical Center, a 400‐bed university‐affiliated public safety net hospital. The study was given exempt status by the Colorado Multiple Institutional Review Board. Patients were identified using a preexisting laboratory database.

We adopted the definitions used by Bates et al.3 for the inclusion and exclusion criteria and the definition of a blood culture episode so that we could apply the financial data presented by these authors to our results. Briefly, a blood culture set was defined as a single venipuncture, regardless of the number of bottles sent for culturing, and a blood culture episode was defined as the 48‐hour period beginning when a blood culture was drawn. All sets within the same 48‐hour period were considered part of the same episode. Cultures that grew bacteria were classified as either true positive, representing bacteremia, or false positive, representing contaminants. Determination of whether a patient had a true‐positive culture versus a contaminant was made in a weekly conference attended by the chief of the Infectious Disease Division, an Infectious Disease fellow, and at least one microbiologist, during which the species of organism cultured and the associated clinical data for each patient were considered. Organisms considered to indicate false positives included diphtheroids, Bacillus sp, Propionibacterium sp, coagulase‐negative staphylococci, and micrococci. All other organisms were considered true positives in the setting of appropriate clinical criteria as specified by the CDC guidelines.4 Hospital charges and lengths of stay were obtained from our institutional database.

The cost associated with a true‐negative blood culture was determined by summing the charges for phlebotomy and microbiological testing obtained from the January 2005 Denver Health hospital charge master and applying the cost‐to‐charge ratio reported on the Medicare Cost Report for inpatient services (not including the costs of physician salaries and benefits).

The cost of a false positive was determined two ways: (a) adjusting the data reported by Bates et al.3 for changes in the Consumer Price Index5 and (b) comparing the actual hospital charges of the patients in our sample who had false‐positive cultures with those who did not (adjusting both by the hospital's inpatient cost‐to‐charge ratio, again not including the cost of physician salaries and benefits).

The length of stay and cost of care for patients with true‐ and false‐positive blood cultures were compared by chi‐square analysis. P < .05 was considered statistically significant. The data were not normally distributed and, as such, are presented as medians and interquartile ranges.

RESULTS

Table 1 summarizes the interpretation of the 939 blood cultures drawn in January 2002. Only 6 culture sets (0.6%) could not be classified. The positive predictive value of a positive blood culture was only 53%.

Laboratory charges for patients with true‐negative and false‐positive blood cultures in January 2002 are shown in Table 2. Annualized, the associated charges were $1,781,292, and the costs were $748,143.

Bates et al.3 found that false‐positive blood cultures increased the length of hospital stay by 4.5 days and increased total charges by $4385 over those for patients with no contaminants. This adjusted to $6878 in 2005 according to the Consumer Price Index.5 After grouping our blood cultures into episodes as defined by Bates et al. (Table 3), we had 41 episodes of contaminated blood cultures that would annualize to charges of $3,383,976 and costs of $1,421,270 after applying the cost‐to‐charge ratio.

The median length of hospital stay and total charges for the patients with true‐negative and false‐positive blood cultures at Denver Health in January 2002 are summarized in Table 4. Using this approach, patients with false‐positive blood cultures at our institution added 1450‐2200 extra hospital days and accrued additional charges of $4,305,000 and costs of $1,808,100.

DISCUSSION

The important finding of this study is that, despite there being nearly 15 times as many true‐negative blood cultures as false‐positive ones, the savings generated by reducing contaminants would be approximately twice that saved by reducing the true negatives (eg, a 50% reduction in the rate of contamination would reduce the total number of false‐positive episodes by 246 annually, saving $710,635‐$904,050, whereas reducing the true negatives by 50% would only save approximately $375,000.

There is no independent gold standard for evaluating the operating characteristics of a blood culture.6 Data from a series of repeated blood cultures represent the closest surrogate. Weinstein et al.7 drew at least 3 sets of cultures from 282 bacteremic patients and noted that bacteremia was documented in 91.5% of the first cultures, in 99.3% in 1 of the first 2 cultures, and in 99.6% in 1 of the first 3 cultures. Because 2 blood culture sets are drawn routinely, the difference between those 2 (if negative) and a third (if it represents a true positive) is 0.3% and would represent a false‐negative culture rate. Given that the true‐negative rate of blood cultures is 87%‐90%,1 the potential 0.3% false‐negative rate would not affect our analysis, and as such, we chose to ignore it. Accordingly, all sets of blood cultures with no growth were classified as true negatives.

Although we cannot show a cause‐and‐effect relationship between false‐positive cultures and the charges associated resulting from them, a recent study suggested that much of the excess length of stay of such patients is attributable to the false‐positive culture itself.8

Because health care costs have exceeded increases in general goods and services, adjusting the results of Bates et al.3 using the Consumer Price Index likely underestimated the projected cost of the false‐positive cultures. This limitation likely accounts for the observation that the difference in actual charges for our patients between those who did and those who did not have false‐positive blood cultures was greater than the cost of these false‐positive cultures as estimated by extrapolating from the data of Bates et al.3 Given the magnitude of the financial difference we observed, however, we suggest that this difference is not of sufficient size to alter our conclusion.

Physicians working at Denver Health are directly employed by the hospital, and the cost of physician salaries and benefits is included in the cost‐to‐charge ratio reported in our Medicare Cost Report. For purposes of this study, however, we elected to utilize a cost‐to‐charge ratio that was exclusive of physician salaries and benefits (ie, 0.42 rather than 0.66) because most hospitals in the United States do not employ their physicians. Accordingly, the costs we present underestimate the true cost to our institution by approximately 32% but are more representative of the costs of services provided by most hospitals in the United States.

Recent studies have shown that the rate of false‐positive cultures is higher when blood is drawn from indwelling catheters than when it is obtained by peripheral venipuncture.9, 10 The rates we cite from the literature2 and from our own institution (Table 1) are aggregate data that include samples drawn from both sites. Separating these would not alter our conclusion that a 50% reduction in false positives would save approximately twice as much as a 50% reduction in false negatives. These studies do, however, identify an important method for reducing false positives: sampling by venipuncture whenever possible, and only drawing through a catheter under very limited circumstances.

There are additional factors that favor a strategy of reducing contaminants over one that attempts to reduce the number of true‐negative cultures. First, reducing the total number of true‐negative blood cultures by 50% would require a very ambitious prediction rule that did not reduce the number of true positives to any meaningful extent. Prediction rules to reduce blood culture testing have been developed for patients with community‐acquired pneumonia, but the rules only reduced the number of cultures by 37% and, more importantly, left 11% of true bacteremias undetected.8 Reducing contaminants would have no effect on the detection of true positives, whereas any prediction rule would inevitably increase the risk of missing true bacteremia in at least a fraction of patients. Second, methods aimed at reducing contaminants can be implemented immediately, whereas deriving a prediction rule would take years to develop and test before it could be utilized. Third, implementing prediction rules may be difficult because many physicians prefer to rely on their clinical impressions.11

Reducing contaminants would require improving the technique by which blood cultures are obtained, with the objective of shifting a portion of false positives to true negatives. This might be accomplished in many ways: increasing the time spent on antiseptic scrubbing, improving the ways in which antiseptic devices are used, waiting for the antiseptic to air‐dry completely, choosing the antiseptic that is most effective in trials, drawing blood by venipuncture instead of through an indwelling catheter, limiting the number of venipuncture attempts before requiring a second site to be prepared, requiring all cultures be drawn by trained phlebotomists, and reducing phlebotomist turnover, among others. Denver Health has a 4‐page set of directions for phlebotomists to follow when obtaining blood cultures. Accordingly, there are numerous places the process could break down. Although having 2 phlebotomists involved (ie, one to perform the procedure and the other to observe and guide the first, assuring that all the appropriate steps are followed) might be considered an extraordinary step, our findings suggest the potential saving to the institution could far outweigh the additional personnel expense resulting from such an approach. Other potential solutions we have considered but not tested include providing a monthly salary bonus to the phlebotomist with the lowest contamination rate or giving bonuses to every phlebotomist who achieves a zero contamination rate.

In summary, we have concluded that the resource utilization associated with obtaining blood cultures can best be improved by reducing the small percentage of cultures that represent contaminants rather than by developing rules to reduce the much larger number of true negatives. The magnitude of the potential savings resulting from reducing contaminants is sufficiently large to warrant expending additional resources to accomplish this task.

Because as many as 90% of all blood cultures grow no organisms1 developing rules that predict which patients are at the lowest risk of having bacteremia could improve the utilization of this test and markedly reduce its cost. Of the approximate 10% of cultures that do grow organisms, only about half represent true bacteremia (ie, true positives), whereas the other half are considered contaminants (ie, false positives)2; the latter are known to increase both the cost and duration of care.3 Accordingly, reducing the number of contaminants could also reduce the cost of care. We assessed which of these two strategies would be the most cost effective. Although only 6% of the blood cultures obtained at our hospital represented contaminants, their associated cost was more than twice that associated with the 87% of cultures that were true negatives.

METHODS

We conducted a retrospective review of microbiological results and hospital records of patients for whom blood cultures were obtained in January 2002 at Denver Health Medical Center, a 400‐bed university‐affiliated public safety net hospital. The study was given exempt status by the Colorado Multiple Institutional Review Board. Patients were identified using a preexisting laboratory database.

We adopted the definitions used by Bates et al.3 for the inclusion and exclusion criteria and the definition of a blood culture episode so that we could apply the financial data presented by these authors to our results. Briefly, a blood culture set was defined as a single venipuncture, regardless of the number of bottles sent for culturing, and a blood culture episode was defined as the 48‐hour period beginning when a blood culture was drawn. All sets within the same 48‐hour period were considered part of the same episode. Cultures that grew bacteria were classified as either true positive, representing bacteremia, or false positive, representing contaminants. Determination of whether a patient had a true‐positive culture versus a contaminant was made in a weekly conference attended by the chief of the Infectious Disease Division, an Infectious Disease fellow, and at least one microbiologist, during which the species of organism cultured and the associated clinical data for each patient were considered. Organisms considered to indicate false positives included diphtheroids, Bacillus sp, Propionibacterium sp, coagulase‐negative staphylococci, and micrococci. All other organisms were considered true positives in the setting of appropriate clinical criteria as specified by the CDC guidelines.4 Hospital charges and lengths of stay were obtained from our institutional database.

The cost associated with a true‐negative blood culture was determined by summing the charges for phlebotomy and microbiological testing obtained from the January 2005 Denver Health hospital charge master and applying the cost‐to‐charge ratio reported on the Medicare Cost Report for inpatient services (not including the costs of physician salaries and benefits).

The cost of a false positive was determined two ways: (a) adjusting the data reported by Bates et al.3 for changes in the Consumer Price Index5 and (b) comparing the actual hospital charges of the patients in our sample who had false‐positive cultures with those who did not (adjusting both by the hospital's inpatient cost‐to‐charge ratio, again not including the cost of physician salaries and benefits).

The length of stay and cost of care for patients with true‐ and false‐positive blood cultures were compared by chi‐square analysis. P < .05 was considered statistically significant. The data were not normally distributed and, as such, are presented as medians and interquartile ranges.

RESULTS

Table 1 summarizes the interpretation of the 939 blood cultures drawn in January 2002. Only 6 culture sets (0.6%) could not be classified. The positive predictive value of a positive blood culture was only 53%.

Laboratory charges for patients with true‐negative and false‐positive blood cultures in January 2002 are shown in Table 2. Annualized, the associated charges were $1,781,292, and the costs were $748,143.

Bates et al.3 found that false‐positive blood cultures increased the length of hospital stay by 4.5 days and increased total charges by $4385 over those for patients with no contaminants. This adjusted to $6878 in 2005 according to the Consumer Price Index.5 After grouping our blood cultures into episodes as defined by Bates et al. (Table 3), we had 41 episodes of contaminated blood cultures that would annualize to charges of $3,383,976 and costs of $1,421,270 after applying the cost‐to‐charge ratio.

The median length of hospital stay and total charges for the patients with true‐negative and false‐positive blood cultures at Denver Health in January 2002 are summarized in Table 4. Using this approach, patients with false‐positive blood cultures at our institution added 1450‐2200 extra hospital days and accrued additional charges of $4,305,000 and costs of $1,808,100.

DISCUSSION

The important finding of this study is that, despite there being nearly 15 times as many true‐negative blood cultures as false‐positive ones, the savings generated by reducing contaminants would be approximately twice that saved by reducing the true negatives (eg, a 50% reduction in the rate of contamination would reduce the total number of false‐positive episodes by 246 annually, saving $710,635‐$904,050, whereas reducing the true negatives by 50% would only save approximately $375,000.

There is no independent gold standard for evaluating the operating characteristics of a blood culture.6 Data from a series of repeated blood cultures represent the closest surrogate. Weinstein et al.7 drew at least 3 sets of cultures from 282 bacteremic patients and noted that bacteremia was documented in 91.5% of the first cultures, in 99.3% in 1 of the first 2 cultures, and in 99.6% in 1 of the first 3 cultures. Because 2 blood culture sets are drawn routinely, the difference between those 2 (if negative) and a third (if it represents a true positive) is 0.3% and would represent a false‐negative culture rate. Given that the true‐negative rate of blood cultures is 87%‐90%,1 the potential 0.3% false‐negative rate would not affect our analysis, and as such, we chose to ignore it. Accordingly, all sets of blood cultures with no growth were classified as true negatives.

Although we cannot show a cause‐and‐effect relationship between false‐positive cultures and the charges associated resulting from them, a recent study suggested that much of the excess length of stay of such patients is attributable to the false‐positive culture itself.8

Because health care costs have exceeded increases in general goods and services, adjusting the results of Bates et al.3 using the Consumer Price Index likely underestimated the projected cost of the false‐positive cultures. This limitation likely accounts for the observation that the difference in actual charges for our patients between those who did and those who did not have false‐positive blood cultures was greater than the cost of these false‐positive cultures as estimated by extrapolating from the data of Bates et al.3 Given the magnitude of the financial difference we observed, however, we suggest that this difference is not of sufficient size to alter our conclusion.

Physicians working at Denver Health are directly employed by the hospital, and the cost of physician salaries and benefits is included in the cost‐to‐charge ratio reported in our Medicare Cost Report. For purposes of this study, however, we elected to utilize a cost‐to‐charge ratio that was exclusive of physician salaries and benefits (ie, 0.42 rather than 0.66) because most hospitals in the United States do not employ their physicians. Accordingly, the costs we present underestimate the true cost to our institution by approximately 32% but are more representative of the costs of services provided by most hospitals in the United States.

Recent studies have shown that the rate of false‐positive cultures is higher when blood is drawn from indwelling catheters than when it is obtained by peripheral venipuncture.9, 10 The rates we cite from the literature2 and from our own institution (Table 1) are aggregate data that include samples drawn from both sites. Separating these would not alter our conclusion that a 50% reduction in false positives would save approximately twice as much as a 50% reduction in false negatives. These studies do, however, identify an important method for reducing false positives: sampling by venipuncture whenever possible, and only drawing through a catheter under very limited circumstances.

There are additional factors that favor a strategy of reducing contaminants over one that attempts to reduce the number of true‐negative cultures. First, reducing the total number of true‐negative blood cultures by 50% would require a very ambitious prediction rule that did not reduce the number of true positives to any meaningful extent. Prediction rules to reduce blood culture testing have been developed for patients with community‐acquired pneumonia, but the rules only reduced the number of cultures by 37% and, more importantly, left 11% of true bacteremias undetected.8 Reducing contaminants would have no effect on the detection of true positives, whereas any prediction rule would inevitably increase the risk of missing true bacteremia in at least a fraction of patients. Second, methods aimed at reducing contaminants can be implemented immediately, whereas deriving a prediction rule would take years to develop and test before it could be utilized. Third, implementing prediction rules may be difficult because many physicians prefer to rely on their clinical impressions.11

Reducing contaminants would require improving the technique by which blood cultures are obtained, with the objective of shifting a portion of false positives to true negatives. This might be accomplished in many ways: increasing the time spent on antiseptic scrubbing, improving the ways in which antiseptic devices are used, waiting for the antiseptic to air‐dry completely, choosing the antiseptic that is most effective in trials, drawing blood by venipuncture instead of through an indwelling catheter, limiting the number of venipuncture attempts before requiring a second site to be prepared, requiring all cultures be drawn by trained phlebotomists, and reducing phlebotomist turnover, among others. Denver Health has a 4‐page set of directions for phlebotomists to follow when obtaining blood cultures. Accordingly, there are numerous places the process could break down. Although having 2 phlebotomists involved (ie, one to perform the procedure and the other to observe and guide the first, assuring that all the appropriate steps are followed) might be considered an extraordinary step, our findings suggest the potential saving to the institution could far outweigh the additional personnel expense resulting from such an approach. Other potential solutions we have considered but not tested include providing a monthly salary bonus to the phlebotomist with the lowest contamination rate or giving bonuses to every phlebotomist who achieves a zero contamination rate.

In summary, we have concluded that the resource utilization associated with obtaining blood cultures can best be improved by reducing the small percentage of cultures that represent contaminants rather than by developing rules to reduce the much larger number of true negatives. The magnitude of the potential savings resulting from reducing contaminants is sufficiently large to warrant expending additional resources to accomplish this task.

In addition to marking my initial transition from student to physician, the commencement address by the medical informatics pioneer, Larry Weed, is my most enduring memory of medical school graduation. A provocative thinker in the field of decision support and structured medical records, Weed was credited by my teachers with developing the organized SOAP note. During his address he depressingly equated all the knowledge we had digested during the preceding mentally strenuous 4 years to shoveling a mountain of manure with a teaspoon. Although I agreed that some of the information I learned seemed to lack relevance (still don't know why I needed to learn the details of the Krebs cycle), as I began caring for patients as an intern, I found that much of it mattered. As I launched into residency training, I also discovered that lifelong learning would be a perpetual component of my medical career.

Despite becoming a passionate advocate of practicing evidence‐based medicine (EBM), I also recognized the impossibility of keeping up with the medical literature, given the relentless arrival of journals in the mail. Learning all the evidence is impossible, so we must develop information management skills and allow others to help us in identifying, reviewing, and summarizing salient and valid clinical information.1 One of my vital goals as editor of the Journal of Hospital Medicine is to ensure we provide our readers with useful clinical information that is concise, easily digested, and usable.

To that end we are introducing Patient‐Oriented Evidence that Matters, or POEMs. As described on the InfoPOEMs website (www.infopoems.com), POEMs have to meet three criteria:

• They address a question that we face as clinicians.

• They measure outcomes that we and our patients care about: symptoms, morbidity, quality of life, and mortality.

• They have the potential to change the way we practice.

We are not the first journal to do this and join the company of the British Medical Journal and the Cleveland Clinic Journal of Medicine.2, 3 Staff physicians at InfoPOEMs screen more than 100 peer‐reviewed medical journals for relevant articles that practicing physicians can use at the point of care, the patient. A trained physician poet then summarizes selected articles in a standardized manner into a POEM. A POEM begins with a clinical question and then provides a bottom line answer from a recently published journal article. This is followed by a structured abstract that includes the study design and setting, followed by a synopsis of the article.

We will start with at least 2 POEMs per issue focused on clinical topics relevant to hospitalists. Soon, an experienced academic hospitalist and knowledgeable expert in EBM, Dr. Jennifer Kleinbart, will be writing POEMs for hospitalists. We look forward to your opinions about whether we should increase this content. Let me know.

In addition to marking my initial transition from student to physician, the commencement address by the medical informatics pioneer, Larry Weed, is my most enduring memory of medical school graduation. A provocative thinker in the field of decision support and structured medical records, Weed was credited by my teachers with developing the organized SOAP note. During his address he depressingly equated all the knowledge we had digested during the preceding mentally strenuous 4 years to shoveling a mountain of manure with a teaspoon. Although I agreed that some of the information I learned seemed to lack relevance (still don't know why I needed to learn the details of the Krebs cycle), as I began caring for patients as an intern, I found that much of it mattered. As I launched into residency training, I also discovered that lifelong learning would be a perpetual component of my medical career.

Despite becoming a passionate advocate of practicing evidence‐based medicine (EBM), I also recognized the impossibility of keeping up with the medical literature, given the relentless arrival of journals in the mail. Learning all the evidence is impossible, so we must develop information management skills and allow others to help us in identifying, reviewing, and summarizing salient and valid clinical information.1 One of my vital goals as editor of the Journal of Hospital Medicine is to ensure we provide our readers with useful clinical information that is concise, easily digested, and usable.

To that end we are introducing Patient‐Oriented Evidence that Matters, or POEMs. As described on the InfoPOEMs website (www.infopoems.com), POEMs have to meet three criteria:

• They address a question that we face as clinicians.

• They measure outcomes that we and our patients care about: symptoms, morbidity, quality of life, and mortality.

• They have the potential to change the way we practice.

We are not the first journal to do this and join the company of the British Medical Journal and the Cleveland Clinic Journal of Medicine.2, 3 Staff physicians at InfoPOEMs screen more than 100 peer‐reviewed medical journals for relevant articles that practicing physicians can use at the point of care, the patient. A trained physician poet then summarizes selected articles in a standardized manner into a POEM. A POEM begins with a clinical question and then provides a bottom line answer from a recently published journal article. This is followed by a structured abstract that includes the study design and setting, followed by a synopsis of the article.

We will start with at least 2 POEMs per issue focused on clinical topics relevant to hospitalists. Soon, an experienced academic hospitalist and knowledgeable expert in EBM, Dr. Jennifer Kleinbart, will be writing POEMs for hospitalists. We look forward to your opinions about whether we should increase this content. Let me know.

In addition to marking my initial transition from student to physician, the commencement address by the medical informatics pioneer, Larry Weed, is my most enduring memory of medical school graduation. A provocative thinker in the field of decision support and structured medical records, Weed was credited by my teachers with developing the organized SOAP note. During his address he depressingly equated all the knowledge we had digested during the preceding mentally strenuous 4 years to shoveling a mountain of manure with a teaspoon. Although I agreed that some of the information I learned seemed to lack relevance (still don't know why I needed to learn the details of the Krebs cycle), as I began caring for patients as an intern, I found that much of it mattered. As I launched into residency training, I also discovered that lifelong learning would be a perpetual component of my medical career.

Despite becoming a passionate advocate of practicing evidence‐based medicine (EBM), I also recognized the impossibility of keeping up with the medical literature, given the relentless arrival of journals in the mail. Learning all the evidence is impossible, so we must develop information management skills and allow others to help us in identifying, reviewing, and summarizing salient and valid clinical information.1 One of my vital goals as editor of the Journal of Hospital Medicine is to ensure we provide our readers with useful clinical information that is concise, easily digested, and usable.

To that end we are introducing Patient‐Oriented Evidence that Matters, or POEMs. As described on the InfoPOEMs website (www.infopoems.com), POEMs have to meet three criteria:

• They address a question that we face as clinicians.

• They measure outcomes that we and our patients care about: symptoms, morbidity, quality of life, and mortality.

• They have the potential to change the way we practice.

We are not the first journal to do this and join the company of the British Medical Journal and the Cleveland Clinic Journal of Medicine.2, 3 Staff physicians at InfoPOEMs screen more than 100 peer‐reviewed medical journals for relevant articles that practicing physicians can use at the point of care, the patient. A trained physician poet then summarizes selected articles in a standardized manner into a POEM. A POEM begins with a clinical question and then provides a bottom line answer from a recently published journal article. This is followed by a structured abstract that includes the study design and setting, followed by a synopsis of the article.

We will start with at least 2 POEMs per issue focused on clinical topics relevant to hospitalists. Soon, an experienced academic hospitalist and knowledgeable expert in EBM, Dr. Jennifer Kleinbart, will be writing POEMs for hospitalists. We look forward to your opinions about whether we should increase this content. Let me know.