Inappropriate antibiotic use is a major public health concern and demonstrates the need for quality improvement initiatives in the delivery of health care.16 Each year nearly 2 million patients in the United States acquire an infection in the hospital, and about 90,000 of them die from these infections.7 More than 70% of the bacteria that cause hospital‐acquired infections are resistant to at least one commonly used drug.7 Persons infected with drug‐resistant organisms have longer hospital stays and higher mortality rates.7

Inappropriate antibiotic use in the inpatient hospital setting can be classified into 5 categories. First, antibiotics may be given for illnesses for which they are not indicated (eg, viral infections). Second, broad‐spectrum antibiotics (such as piperacillin‐tazobactam and quinolones) may be overused in the empiric treatment of common infections.8 Overuse of broad‐spectrum drugs increases selective pressure for antimicrobial resistance and exposes patients to the side effects of some of these drugs, such as Clostridium difficile colitis.8 Third, clinicians occasionally prescribe intravenous (IV) antibiotics when the efficacy of oral agents would be similar. Inappropriate intravenous therapy increases the cost of care and also exposes the patient to the risk of intravenous catheters.8 Fourth, when the correct antibiotic choice is made, inappropriate antibiotic dosage, schedule, and/or duration of treatment can threaten patient safety.8 Fifth, bug‐drug mismatch occurs when susceptibility studies indicate that the drug being used is ineffective or only marginally effective.8 Beyond antimicrobial resistance and safety, these practices also usually increase costs to both the patient and the hospital.7, 910

Influencing providers' prescribing patterns is difficult.11 In this project we assessed the prescribing patterns of hospitalists in an active inpatient environment and then developed an intervention to improve the providers' use of antibiotics. The intervention utilized public health methodologyprior to implementation, we defined the problem, determined its magnitude, identified a behavior change model, and constructed a conceptual framework that identifyied the key determinants. A pilot academic detailing project addressing many determinants was developed, implemented, and evaluated.

Conceptual Model

To change prescribing behaviors is to change learned behaviors. Changing behavior is a complex process affected by several factors including beliefs, expectations, motivations, and the psychosocial environments of the target groups.12 Each of these factors must be considered when attempting to bring about behavior changes. In doing so, a theory that can be depicted in a model often emerges.13 This approach is widely used in understanding and developing public health interventions.

Formulating the Model

In any public health intervention, recognizing and engaging key stakeholders is a critical step. We identified the following stakeholders: (1) hospitalist practitioners and other prescribing providers including residents and infectious disease specialists; (2) nurses; (3) administrators who are focused on cost effectiveness; (4) patients and their families, who want to get well affordably, without side effects; (5) pharmacists; (6) risk management; and (7) society, which is fearful of the propagation of resistant microbes. In consulting with some of the stakeholders, 4 factors that influence hospitalists' prescribing patterns became apparent. These are practitioner factors, environmental factors, perceived rewards, and perceived threats (Fig. 1).

Figure 1

The practitioner factors shaping prescribing are: (1) knowledge of current best care; (2) self‐efficacy, which determines whether a provider is confident in his or her knowledge to adequately treat a specific infection; (3) habit, which causes providers to pick from a narrow repertoire of antibiotics when treating an infection; and (4) fear of liability, which forces some providers to be cautious. Four environmental factors affecting antibiotic prescriptions are: (1) published guidelines regarding organisms' sensitivity to antibiotics; (2) patient‐driven factors such as affordability, compliance with dosing regimens, side effects, and interactions between the antibiotics and other medications; (3) peer influence, in that providers are reluctant to change a prescription started by another provider (eg, emergency room physician); and (4) the formulary of the hospital, as it forces providers to prescribe within specific parameters. The perceived rewards of specific prescribing practices may include improving patient safety and reducing antibiotic resistance and costs, whereas the perceived threats are increasing antimicrobial resistance, having adverse patient outcomes, and increasing costs and hospital length of stay. We selected a high‐yield, low‐effort intervention in order to have an impact on some of the factors underlying hospitalists' prescribing patterns.

METHODS

Data Collection

We collected and assessed prescription patterns over 3 periods: preintervention, interim, and postintervention.

Detailing Procedures

After the review, a profile was assembled for each of the CIMS providers. The study team detailers (a physician and a pharmacist) met with the individual providers for 30 to 45 minutes. Each hospitalist participant completed a short survey that collected demographic information and was asked about the rationale for his or her antibiotic prescribing pattern. Next, the appraisal of the provider's prescribing pattern was reviewed. This review included looking at the costs of the prescribed antibiotics compared with those of the appropriate alternatives and a reexamination of the guidelines for the selected target drugspiperacillin‐tazobactam, vancomycin, and extended‐spectrum quinolones. These 3 antibiotics were picked because our providers had been particularly vulnerable to inappropriately prescribing them. The hospitalists were provided an antibiotic guide developed specifically for this project and based on the Johns Hopkins Antibiotic Guide14 that summarizes the consensus guidelines.

Data Analysis

The primary outcome variable was the aggregate proportion of inappropriate antibiotic prescribed (as defined earlier) before the intervention, during the interim between obtaining informed consent and intervening on all study subjects, and after the intervention. The percentage of appropriate prescriptions versus total not appropriate prescriptions (combining of the effective but inappropriate and inappropriate categories) were compared across the 3 periods. Ninety‐five percent confidence intervals for comparisons of the proportions were determined using Stata 9.0 (College Station, TX). The difference between the proportions of total not appropriate prescriptions before and after academic detailing was computed in Stata using Fisher's exact test to assess significance.

RESULTS

Demographic information and professional characteristics of the 17 providers are shown in Table 1. Their mean age was 36 years, and 76% were female. The top 4 reasons the providers gave for their prescribing practices were: (1) published guidelines, (2) easier dosing schedule for patient when discharged, (3) continuing an antibiotic course initiated in the emergency room, and (4) broad‐spectrum antibiotics cover all possible microbes.

DISCUSSION

We have demonstrated that academic detailing had a positive impact on the prescribing patterns of hospitalists. The aggregated improvement in antibiotic prescribing patterns can be attributed to improvement in the prescribing patterns of almost every hospitalist practitioner (Table 2). This study focused on aggregate prescriptions as the primary outcome measure because the hospitalists at JHBMC, like at many other institutions, function as a team, with a patient routinely having multiple providers over the course of the hospital stay. The improved prescribing patterns noted during the interim period suggest that the mere knowledge of a project can have an impact on providers. Providers informed the investigators that they were more thoughtful about their choice of antibiotics when they knew that they were being studied. The further statistically significant improvement in prescribing patterns with the intervention shows that the academic detailing itself was successful.

The greatest absolute change in practice was seen in effective but inappropriate prescribing (from 30.4% to 6%), whereas inappropriate prescribing only decreased from 26.6% to 20.6%. Although we aimed to have an impact on all inappropriate antibiotic prescribing patterns, we specifically reviewed the prescribing guidelines for piperacillin‐tazobactam, extended‐spectrum quinolones, and vancomycin. These 3 antibiotics were targeted because our providers had been particularly susceptible to inappropriately prescribing them. The focus on these antibiotics may have resulted in the larger absolute change noted in effective but inappropriate prescribing. We did not collect any data to determine if having an impact on effective but inappropriate prescribing changed the clinical course of the patients, such as shortening their hospital stays. Anecdotal evidence, however, suggests that it does. At our institution it is not uncommon for patients to be kept in the hospital for an extra day to ensure they are stable when transitioned from extended‐spectrum to narrower‐spectrum antibiotics prior to discharge. The effect of reducing effective but inappropriate prescriptions on the clinical course of patients could be an outcome measure assessed by a future, larger study.

Our one‐on‐one appraisal of each provider's prescribing patterns included a review of the cost of the prescribed antibiotics compared with that of the appropriate alternatives. Although decisions on antibiotic choice should be driven by clinical guidelines and appropriateness rather than price, we believed it was relevant to include education about costs and pricing so that providers would be reminded to ascertain whether patients would be able to afford their antibiotics. Antibiotic resistance is influenced by a patient's failure to complete the course of treatment, and noncompliance may be caused by an inability to afford the medication. Often, there are affordable, appropriate alternatives to the newest and most expensive drugs.

A hospitalist‐based academic detailing approach to improving antibiotic prescribing may have far‐reaching benefits and influence. First, it has the potential to affect other practitioners by setting an example and role modeling. In addition to that with their immediate peer group, hospitalists have close and repeated contact with house officers and emergency room physicians and often act as consultants to physicians in other departments such as surgery and psychiatry. Furthermore, some community hospitals have no infectious disease specialists readily available. So this represents an opportunity for hospitalists to promote quality in antibiotic prescribing. Practice‐based learning was very effective because it brought the practitioners face to face with their prescribing patterns. Although intellectually everyone agreed that antibiotics are often misused, this approach forced the providers to stop and reflect on their individual practices. This peer‐delivered intervention allowed for a collaborative approach to solving the problem; the peer (detailer) was approachable, nonjudgmental, and available for further discussion and guidance.

The public health quality improvement approach that we used for our intervention helped us to realize and appreciate the factors underlying prescribing patterns. Only by understanding the motivations for prescribing patterns can we hope to make sustainable changes. This coincides with our previous assertion that hospitalists are engaging in some public health practice.16 In pubic health, the programs, services, and institutions involved emphasize the prevention of disease and the health needs of the population as a whole.17 Hospitalist teams aim to make sure that the high‐quality services needed for protecting the health of their community (hospitalized patients) are available and that this population receives proper consideration in the allocation of resources. Antibiotic optimization is a key role that could fall within the mantra of public health practice for the hospitalist.

Several limitations of this pilot should be considered. First, the intervention is labor intensive. However, it is essential to use the problem‐solving paradigm and incorporate behavior change theories in order to identify interventions that can lead to sustainable change. Second, this was not a randomized controlled trial, and it is possible that there might have been some contamination by external forces. However, in reviewing the educational events at our institution, the press, and articles published during the study period, we could not identify any external factors that would have influenced antibiotic prescribing patterns. It would not have been possible to conduct a randomized trial at our institution because the hospitalists work so closely together that we could not ensure complete separation if the subjects were randomized. There would have been contamination from the intervention group to the control group. A trial with randomization at the institution level is the next step. Third, the number of months retrospectively reviewed in order to identify 20 prescriptions of a provider varied. This study assumed there were no other differences during those months that could have affected provider prescribing behavior; this may have introduced some bias. Fourth, the sustainability of this intervention's positive impact is unknown. We assessed outcome soon after the intervention, and it is unknown whether continual booster sessions are required to maintain the positive impact on prescribing patterns.

This pilot was a good starting place to show that behavior change can be realized with a well‐conceived and methodically executed intervention, even among the busiest of physicians. Audit and feedback, or practice‐based learning, appears to be a powerful educational intervention among professionals who take great pride in their work.

Inappropriate antibiotic use is a major public health concern and demonstrates the need for quality improvement initiatives in the delivery of health care.16 Each year nearly 2 million patients in the United States acquire an infection in the hospital, and about 90,000 of them die from these infections.7 More than 70% of the bacteria that cause hospital‐acquired infections are resistant to at least one commonly used drug.7 Persons infected with drug‐resistant organisms have longer hospital stays and higher mortality rates.7

Inappropriate antibiotic use in the inpatient hospital setting can be classified into 5 categories. First, antibiotics may be given for illnesses for which they are not indicated (eg, viral infections). Second, broad‐spectrum antibiotics (such as piperacillin‐tazobactam and quinolones) may be overused in the empiric treatment of common infections.8 Overuse of broad‐spectrum drugs increases selective pressure for antimicrobial resistance and exposes patients to the side effects of some of these drugs, such as Clostridium difficile colitis.8 Third, clinicians occasionally prescribe intravenous (IV) antibiotics when the efficacy of oral agents would be similar. Inappropriate intravenous therapy increases the cost of care and also exposes the patient to the risk of intravenous catheters.8 Fourth, when the correct antibiotic choice is made, inappropriate antibiotic dosage, schedule, and/or duration of treatment can threaten patient safety.8 Fifth, bug‐drug mismatch occurs when susceptibility studies indicate that the drug being used is ineffective or only marginally effective.8 Beyond antimicrobial resistance and safety, these practices also usually increase costs to both the patient and the hospital.7, 910

Influencing providers' prescribing patterns is difficult.11 In this project we assessed the prescribing patterns of hospitalists in an active inpatient environment and then developed an intervention to improve the providers' use of antibiotics. The intervention utilized public health methodologyprior to implementation, we defined the problem, determined its magnitude, identified a behavior change model, and constructed a conceptual framework that identifyied the key determinants. A pilot academic detailing project addressing many determinants was developed, implemented, and evaluated.

Conceptual Model

To change prescribing behaviors is to change learned behaviors. Changing behavior is a complex process affected by several factors including beliefs, expectations, motivations, and the psychosocial environments of the target groups.12 Each of these factors must be considered when attempting to bring about behavior changes. In doing so, a theory that can be depicted in a model often emerges.13 This approach is widely used in understanding and developing public health interventions.

Formulating the Model

In any public health intervention, recognizing and engaging key stakeholders is a critical step. We identified the following stakeholders: (1) hospitalist practitioners and other prescribing providers including residents and infectious disease specialists; (2) nurses; (3) administrators who are focused on cost effectiveness; (4) patients and their families, who want to get well affordably, without side effects; (5) pharmacists; (6) risk management; and (7) society, which is fearful of the propagation of resistant microbes. In consulting with some of the stakeholders, 4 factors that influence hospitalists' prescribing patterns became apparent. These are practitioner factors, environmental factors, perceived rewards, and perceived threats (Fig. 1).

Figure 1

The practitioner factors shaping prescribing are: (1) knowledge of current best care; (2) self‐efficacy, which determines whether a provider is confident in his or her knowledge to adequately treat a specific infection; (3) habit, which causes providers to pick from a narrow repertoire of antibiotics when treating an infection; and (4) fear of liability, which forces some providers to be cautious. Four environmental factors affecting antibiotic prescriptions are: (1) published guidelines regarding organisms' sensitivity to antibiotics; (2) patient‐driven factors such as affordability, compliance with dosing regimens, side effects, and interactions between the antibiotics and other medications; (3) peer influence, in that providers are reluctant to change a prescription started by another provider (eg, emergency room physician); and (4) the formulary of the hospital, as it forces providers to prescribe within specific parameters. The perceived rewards of specific prescribing practices may include improving patient safety and reducing antibiotic resistance and costs, whereas the perceived threats are increasing antimicrobial resistance, having adverse patient outcomes, and increasing costs and hospital length of stay. We selected a high‐yield, low‐effort intervention in order to have an impact on some of the factors underlying hospitalists' prescribing patterns.

METHODS

Data Collection

We collected and assessed prescription patterns over 3 periods: preintervention, interim, and postintervention.

Detailing Procedures

After the review, a profile was assembled for each of the CIMS providers. The study team detailers (a physician and a pharmacist) met with the individual providers for 30 to 45 minutes. Each hospitalist participant completed a short survey that collected demographic information and was asked about the rationale for his or her antibiotic prescribing pattern. Next, the appraisal of the provider's prescribing pattern was reviewed. This review included looking at the costs of the prescribed antibiotics compared with those of the appropriate alternatives and a reexamination of the guidelines for the selected target drugspiperacillin‐tazobactam, vancomycin, and extended‐spectrum quinolones. These 3 antibiotics were picked because our providers had been particularly vulnerable to inappropriately prescribing them. The hospitalists were provided an antibiotic guide developed specifically for this project and based on the Johns Hopkins Antibiotic Guide14 that summarizes the consensus guidelines.

Data Analysis

The primary outcome variable was the aggregate proportion of inappropriate antibiotic prescribed (as defined earlier) before the intervention, during the interim between obtaining informed consent and intervening on all study subjects, and after the intervention. The percentage of appropriate prescriptions versus total not appropriate prescriptions (combining of the effective but inappropriate and inappropriate categories) were compared across the 3 periods. Ninety‐five percent confidence intervals for comparisons of the proportions were determined using Stata 9.0 (College Station, TX). The difference between the proportions of total not appropriate prescriptions before and after academic detailing was computed in Stata using Fisher's exact test to assess significance.

RESULTS

Demographic information and professional characteristics of the 17 providers are shown in Table 1. Their mean age was 36 years, and 76% were female. The top 4 reasons the providers gave for their prescribing practices were: (1) published guidelines, (2) easier dosing schedule for patient when discharged, (3) continuing an antibiotic course initiated in the emergency room, and (4) broad‐spectrum antibiotics cover all possible microbes.

DISCUSSION

We have demonstrated that academic detailing had a positive impact on the prescribing patterns of hospitalists. The aggregated improvement in antibiotic prescribing patterns can be attributed to improvement in the prescribing patterns of almost every hospitalist practitioner (Table 2). This study focused on aggregate prescriptions as the primary outcome measure because the hospitalists at JHBMC, like at many other institutions, function as a team, with a patient routinely having multiple providers over the course of the hospital stay. The improved prescribing patterns noted during the interim period suggest that the mere knowledge of a project can have an impact on providers. Providers informed the investigators that they were more thoughtful about their choice of antibiotics when they knew that they were being studied. The further statistically significant improvement in prescribing patterns with the intervention shows that the academic detailing itself was successful.

The greatest absolute change in practice was seen in effective but inappropriate prescribing (from 30.4% to 6%), whereas inappropriate prescribing only decreased from 26.6% to 20.6%. Although we aimed to have an impact on all inappropriate antibiotic prescribing patterns, we specifically reviewed the prescribing guidelines for piperacillin‐tazobactam, extended‐spectrum quinolones, and vancomycin. These 3 antibiotics were targeted because our providers had been particularly susceptible to inappropriately prescribing them. The focus on these antibiotics may have resulted in the larger absolute change noted in effective but inappropriate prescribing. We did not collect any data to determine if having an impact on effective but inappropriate prescribing changed the clinical course of the patients, such as shortening their hospital stays. Anecdotal evidence, however, suggests that it does. At our institution it is not uncommon for patients to be kept in the hospital for an extra day to ensure they are stable when transitioned from extended‐spectrum to narrower‐spectrum antibiotics prior to discharge. The effect of reducing effective but inappropriate prescriptions on the clinical course of patients could be an outcome measure assessed by a future, larger study.

Our one‐on‐one appraisal of each provider's prescribing patterns included a review of the cost of the prescribed antibiotics compared with that of the appropriate alternatives. Although decisions on antibiotic choice should be driven by clinical guidelines and appropriateness rather than price, we believed it was relevant to include education about costs and pricing so that providers would be reminded to ascertain whether patients would be able to afford their antibiotics. Antibiotic resistance is influenced by a patient's failure to complete the course of treatment, and noncompliance may be caused by an inability to afford the medication. Often, there are affordable, appropriate alternatives to the newest and most expensive drugs.

A hospitalist‐based academic detailing approach to improving antibiotic prescribing may have far‐reaching benefits and influence. First, it has the potential to affect other practitioners by setting an example and role modeling. In addition to that with their immediate peer group, hospitalists have close and repeated contact with house officers and emergency room physicians and often act as consultants to physicians in other departments such as surgery and psychiatry. Furthermore, some community hospitals have no infectious disease specialists readily available. So this represents an opportunity for hospitalists to promote quality in antibiotic prescribing. Practice‐based learning was very effective because it brought the practitioners face to face with their prescribing patterns. Although intellectually everyone agreed that antibiotics are often misused, this approach forced the providers to stop and reflect on their individual practices. This peer‐delivered intervention allowed for a collaborative approach to solving the problem; the peer (detailer) was approachable, nonjudgmental, and available for further discussion and guidance.

The public health quality improvement approach that we used for our intervention helped us to realize and appreciate the factors underlying prescribing patterns. Only by understanding the motivations for prescribing patterns can we hope to make sustainable changes. This coincides with our previous assertion that hospitalists are engaging in some public health practice.16 In pubic health, the programs, services, and institutions involved emphasize the prevention of disease and the health needs of the population as a whole.17 Hospitalist teams aim to make sure that the high‐quality services needed for protecting the health of their community (hospitalized patients) are available and that this population receives proper consideration in the allocation of resources. Antibiotic optimization is a key role that could fall within the mantra of public health practice for the hospitalist.

Several limitations of this pilot should be considered. First, the intervention is labor intensive. However, it is essential to use the problem‐solving paradigm and incorporate behavior change theories in order to identify interventions that can lead to sustainable change. Second, this was not a randomized controlled trial, and it is possible that there might have been some contamination by external forces. However, in reviewing the educational events at our institution, the press, and articles published during the study period, we could not identify any external factors that would have influenced antibiotic prescribing patterns. It would not have been possible to conduct a randomized trial at our institution because the hospitalists work so closely together that we could not ensure complete separation if the subjects were randomized. There would have been contamination from the intervention group to the control group. A trial with randomization at the institution level is the next step. Third, the number of months retrospectively reviewed in order to identify 20 prescriptions of a provider varied. This study assumed there were no other differences during those months that could have affected provider prescribing behavior; this may have introduced some bias. Fourth, the sustainability of this intervention's positive impact is unknown. We assessed outcome soon after the intervention, and it is unknown whether continual booster sessions are required to maintain the positive impact on prescribing patterns.

This pilot was a good starting place to show that behavior change can be realized with a well‐conceived and methodically executed intervention, even among the busiest of physicians. Audit and feedback, or practice‐based learning, appears to be a powerful educational intervention among professionals who take great pride in their work.

Inappropriate antibiotic use is a major public health concern and demonstrates the need for quality improvement initiatives in the delivery of health care.16 Each year nearly 2 million patients in the United States acquire an infection in the hospital, and about 90,000 of them die from these infections.7 More than 70% of the bacteria that cause hospital‐acquired infections are resistant to at least one commonly used drug.7 Persons infected with drug‐resistant organisms have longer hospital stays and higher mortality rates.7

Inappropriate antibiotic use in the inpatient hospital setting can be classified into 5 categories. First, antibiotics may be given for illnesses for which they are not indicated (eg, viral infections). Second, broad‐spectrum antibiotics (such as piperacillin‐tazobactam and quinolones) may be overused in the empiric treatment of common infections.8 Overuse of broad‐spectrum drugs increases selective pressure for antimicrobial resistance and exposes patients to the side effects of some of these drugs, such as Clostridium difficile colitis.8 Third, clinicians occasionally prescribe intravenous (IV) antibiotics when the efficacy of oral agents would be similar. Inappropriate intravenous therapy increases the cost of care and also exposes the patient to the risk of intravenous catheters.8 Fourth, when the correct antibiotic choice is made, inappropriate antibiotic dosage, schedule, and/or duration of treatment can threaten patient safety.8 Fifth, bug‐drug mismatch occurs when susceptibility studies indicate that the drug being used is ineffective or only marginally effective.8 Beyond antimicrobial resistance and safety, these practices also usually increase costs to both the patient and the hospital.7, 910

Influencing providers' prescribing patterns is difficult.11 In this project we assessed the prescribing patterns of hospitalists in an active inpatient environment and then developed an intervention to improve the providers' use of antibiotics. The intervention utilized public health methodologyprior to implementation, we defined the problem, determined its magnitude, identified a behavior change model, and constructed a conceptual framework that identifyied the key determinants. A pilot academic detailing project addressing many determinants was developed, implemented, and evaluated.

Conceptual Model

To change prescribing behaviors is to change learned behaviors. Changing behavior is a complex process affected by several factors including beliefs, expectations, motivations, and the psychosocial environments of the target groups.12 Each of these factors must be considered when attempting to bring about behavior changes. In doing so, a theory that can be depicted in a model often emerges.13 This approach is widely used in understanding and developing public health interventions.

Formulating the Model

In any public health intervention, recognizing and engaging key stakeholders is a critical step. We identified the following stakeholders: (1) hospitalist practitioners and other prescribing providers including residents and infectious disease specialists; (2) nurses; (3) administrators who are focused on cost effectiveness; (4) patients and their families, who want to get well affordably, without side effects; (5) pharmacists; (6) risk management; and (7) society, which is fearful of the propagation of resistant microbes. In consulting with some of the stakeholders, 4 factors that influence hospitalists' prescribing patterns became apparent. These are practitioner factors, environmental factors, perceived rewards, and perceived threats (Fig. 1).

Figure 1

The practitioner factors shaping prescribing are: (1) knowledge of current best care; (2) self‐efficacy, which determines whether a provider is confident in his or her knowledge to adequately treat a specific infection; (3) habit, which causes providers to pick from a narrow repertoire of antibiotics when treating an infection; and (4) fear of liability, which forces some providers to be cautious. Four environmental factors affecting antibiotic prescriptions are: (1) published guidelines regarding organisms' sensitivity to antibiotics; (2) patient‐driven factors such as affordability, compliance with dosing regimens, side effects, and interactions between the antibiotics and other medications; (3) peer influence, in that providers are reluctant to change a prescription started by another provider (eg, emergency room physician); and (4) the formulary of the hospital, as it forces providers to prescribe within specific parameters. The perceived rewards of specific prescribing practices may include improving patient safety and reducing antibiotic resistance and costs, whereas the perceived threats are increasing antimicrobial resistance, having adverse patient outcomes, and increasing costs and hospital length of stay. We selected a high‐yield, low‐effort intervention in order to have an impact on some of the factors underlying hospitalists' prescribing patterns.

METHODS

Data Collection

We collected and assessed prescription patterns over 3 periods: preintervention, interim, and postintervention.

Detailing Procedures

After the review, a profile was assembled for each of the CIMS providers. The study team detailers (a physician and a pharmacist) met with the individual providers for 30 to 45 minutes. Each hospitalist participant completed a short survey that collected demographic information and was asked about the rationale for his or her antibiotic prescribing pattern. Next, the appraisal of the provider's prescribing pattern was reviewed. This review included looking at the costs of the prescribed antibiotics compared with those of the appropriate alternatives and a reexamination of the guidelines for the selected target drugspiperacillin‐tazobactam, vancomycin, and extended‐spectrum quinolones. These 3 antibiotics were picked because our providers had been particularly vulnerable to inappropriately prescribing them. The hospitalists were provided an antibiotic guide developed specifically for this project and based on the Johns Hopkins Antibiotic Guide14 that summarizes the consensus guidelines.

Data Analysis

The primary outcome variable was the aggregate proportion of inappropriate antibiotic prescribed (as defined earlier) before the intervention, during the interim between obtaining informed consent and intervening on all study subjects, and after the intervention. The percentage of appropriate prescriptions versus total not appropriate prescriptions (combining of the effective but inappropriate and inappropriate categories) were compared across the 3 periods. Ninety‐five percent confidence intervals for comparisons of the proportions were determined using Stata 9.0 (College Station, TX). The difference between the proportions of total not appropriate prescriptions before and after academic detailing was computed in Stata using Fisher's exact test to assess significance.

RESULTS

Demographic information and professional characteristics of the 17 providers are shown in Table 1. Their mean age was 36 years, and 76% were female. The top 4 reasons the providers gave for their prescribing practices were: (1) published guidelines, (2) easier dosing schedule for patient when discharged, (3) continuing an antibiotic course initiated in the emergency room, and (4) broad‐spectrum antibiotics cover all possible microbes.

DISCUSSION

We have demonstrated that academic detailing had a positive impact on the prescribing patterns of hospitalists. The aggregated improvement in antibiotic prescribing patterns can be attributed to improvement in the prescribing patterns of almost every hospitalist practitioner (Table 2). This study focused on aggregate prescriptions as the primary outcome measure because the hospitalists at JHBMC, like at many other institutions, function as a team, with a patient routinely having multiple providers over the course of the hospital stay. The improved prescribing patterns noted during the interim period suggest that the mere knowledge of a project can have an impact on providers. Providers informed the investigators that they were more thoughtful about their choice of antibiotics when they knew that they were being studied. The further statistically significant improvement in prescribing patterns with the intervention shows that the academic detailing itself was successful.

The greatest absolute change in practice was seen in effective but inappropriate prescribing (from 30.4% to 6%), whereas inappropriate prescribing only decreased from 26.6% to 20.6%. Although we aimed to have an impact on all inappropriate antibiotic prescribing patterns, we specifically reviewed the prescribing guidelines for piperacillin‐tazobactam, extended‐spectrum quinolones, and vancomycin. These 3 antibiotics were targeted because our providers had been particularly susceptible to inappropriately prescribing them. The focus on these antibiotics may have resulted in the larger absolute change noted in effective but inappropriate prescribing. We did not collect any data to determine if having an impact on effective but inappropriate prescribing changed the clinical course of the patients, such as shortening their hospital stays. Anecdotal evidence, however, suggests that it does. At our institution it is not uncommon for patients to be kept in the hospital for an extra day to ensure they are stable when transitioned from extended‐spectrum to narrower‐spectrum antibiotics prior to discharge. The effect of reducing effective but inappropriate prescriptions on the clinical course of patients could be an outcome measure assessed by a future, larger study.

Our one‐on‐one appraisal of each provider's prescribing patterns included a review of the cost of the prescribed antibiotics compared with that of the appropriate alternatives. Although decisions on antibiotic choice should be driven by clinical guidelines and appropriateness rather than price, we believed it was relevant to include education about costs and pricing so that providers would be reminded to ascertain whether patients would be able to afford their antibiotics. Antibiotic resistance is influenced by a patient's failure to complete the course of treatment, and noncompliance may be caused by an inability to afford the medication. Often, there are affordable, appropriate alternatives to the newest and most expensive drugs.

A hospitalist‐based academic detailing approach to improving antibiotic prescribing may have far‐reaching benefits and influence. First, it has the potential to affect other practitioners by setting an example and role modeling. In addition to that with their immediate peer group, hospitalists have close and repeated contact with house officers and emergency room physicians and often act as consultants to physicians in other departments such as surgery and psychiatry. Furthermore, some community hospitals have no infectious disease specialists readily available. So this represents an opportunity for hospitalists to promote quality in antibiotic prescribing. Practice‐based learning was very effective because it brought the practitioners face to face with their prescribing patterns. Although intellectually everyone agreed that antibiotics are often misused, this approach forced the providers to stop and reflect on their individual practices. This peer‐delivered intervention allowed for a collaborative approach to solving the problem; the peer (detailer) was approachable, nonjudgmental, and available for further discussion and guidance.

The public health quality improvement approach that we used for our intervention helped us to realize and appreciate the factors underlying prescribing patterns. Only by understanding the motivations for prescribing patterns can we hope to make sustainable changes. This coincides with our previous assertion that hospitalists are engaging in some public health practice.16 In pubic health, the programs, services, and institutions involved emphasize the prevention of disease and the health needs of the population as a whole.17 Hospitalist teams aim to make sure that the high‐quality services needed for protecting the health of their community (hospitalized patients) are available and that this population receives proper consideration in the allocation of resources. Antibiotic optimization is a key role that could fall within the mantra of public health practice for the hospitalist.

Several limitations of this pilot should be considered. First, the intervention is labor intensive. However, it is essential to use the problem‐solving paradigm and incorporate behavior change theories in order to identify interventions that can lead to sustainable change. Second, this was not a randomized controlled trial, and it is possible that there might have been some contamination by external forces. However, in reviewing the educational events at our institution, the press, and articles published during the study period, we could not identify any external factors that would have influenced antibiotic prescribing patterns. It would not have been possible to conduct a randomized trial at our institution because the hospitalists work so closely together that we could not ensure complete separation if the subjects were randomized. There would have been contamination from the intervention group to the control group. A trial with randomization at the institution level is the next step. Third, the number of months retrospectively reviewed in order to identify 20 prescriptions of a provider varied. This study assumed there were no other differences during those months that could have affected provider prescribing behavior; this may have introduced some bias. Fourth, the sustainability of this intervention's positive impact is unknown. We assessed outcome soon after the intervention, and it is unknown whether continual booster sessions are required to maintain the positive impact on prescribing patterns.

This pilot was a good starting place to show that behavior change can be realized with a well‐conceived and methodically executed intervention, even among the busiest of physicians. Audit and feedback, or practice‐based learning, appears to be a powerful educational intervention among professionals who take great pride in their work.

Given recent changes in the goals and objectives of residency training as well as changes in the functioning of teaching hospitals, traditional educational formats may need to be supplemented or replaced.1 The Accreditation Council for Graduate Medical Education (ACGME) is promoting changes in resident education with the goal of not only enhancing trainee competency using innovative methods but also of demonstrating that these educational innovations result in enhanced quality of patient care and improved patient safety.1 A challenging aspect of these initiatives is that programs are working to implement them at a time when there are greater nonteaching demands on faculty time, mandated resident work‐hour limitations have been instituted, in some states by law, and resident patient care and educational activities are prone to disruptions inherent in caring for patients in a complex health system. Various solutions have been proposed including increased incorporation of self‐directed learning as a means of meeting modern resident educational challenges, yet the ideal tools with which to accomplish this are unknown.

Computer‐based instruction in medicine has been available since the 1960s, and although its use had initially been more widespread in medical student, nursing, allied health professional, and patient education,2, 3 it is being increasingly incorporated into resident education as well. Some studies have shown that for medical students, computer‐based teaching is at least as effective in improving knowledge as conventional lectures4 and that learners' satisfaction with computer‐based formats appears comparable with that of traditional didactic lectures.5 In recent years computer‐based teaching has been applied to resident education in various fields including surgery and surgical subspecialties, pediatrics, and obstetrics and gynecology.612 Little is known, however, about how computer‐based educational methods affect resident knowledge and especially how these methods might affect clinical practice.

Venous thromboembolism (VTE) is a common and hazardous complication of acute inpatient hospitalization.13 Recognizing that errors in proper prescribing and monitoring of anticoagulants are a major cause of acute inpatient morbidity and mortality,14 we began an initiative to educate our residents and improve their patient care practices regarding the proper use of anticoagulants. To accomplish this, we developed a CD‐ROM‐based learning module with the aim of increasing resident knowledge of anticoagulation as well as compliance with national standards for VTE prevention. In this study we assessed the impact of the CD‐ROM intervention on resident knowledge and their appropriate use of VTE prophylaxis.

METHODS

The study was approved by the institutional review board. With the participation of faculty educators in the departments of medicine, surgery, and neurology, one of the authors (H.K.) coordinated the development of a CD‐ROM containing concise modules on core topics in anticoagulation (Table 1). The presenters for these topics included the director of clinical hematology, 2 cardiologists including the director of the coronary care unit, the director of the medical intensive care unit, the director of cerebrovascular diseases, and 2 vascular surgeons, one of whom serves as vice chair of surgery. These modules, each lasting about 1 hour, had audio and slide components detailing the proper indications, monitoring, and efficacy of anticoagulants in atrial fibrillation, acute ischemic stroke, acute coronary syndromes, and VTE prevention in acutely ill hospitalized patients. The guidelines presented were based on the sixth (2000) ACCP guidelines for antithrombotic therapy for the prevention and treatment of thrombosis.15 The content of the CD‐ROM was reviewed for accuracy by the authors, though none of them were speakers. We asked that before all current residents in the departments of cardiothoracic surgery, emergency medicine, otolaryngology, internal medicine, neurosurgery, dental medicine, neurology, obstetrics and gynecology, orthopedics, surgery, and urology viewed the CD‐ROM, they complete a pretest to determine their baseline knowledge of this subject. After completing the pretest, the residents were required to view the CD‐ROM and retake the same test. We then compared pre‐ and posttest scores.

To determine whether an increase in knowledge was secondary to the CD‐ROM intervention or simply a consequence of acquired clinical experience during training, we compared test scores of residents who did and did not receive the CD‐ROM intervention. In the academic year following our initial testing, we asked the incoming categorical medical PGY‐1 classes at our hospital and at a comparable local tertiary‐care hospital in our health system 2 miles away to take an anticoagulation pretest (different from the examination given for the initial testing) during their PGY‐1 orientations. The 2 institutions are comparable in many ways including in patient demographics and size and most residents come from the same medical schools, have a similar rotation structure, and use a comparable curriculum under a unified graduate medical education office. The CD‐ROMs were only given to categorical PGY‐1 residents at our institution. Both groups then retook the same test (posttest) 3 months into their clinical training, a time chosen because it is when all PGY‐1s would be expected to have gained significant clinical experience on the medical wards and or in the intensive care units. The exam questions were generated by one of the authors (B.M.) and covered all the topics in the CD‐ROM.

An Anticoagulation Steering Committee was formed to assess whether the CD‐ROM intervention affected our residents' patient care practices. None of the members of this committee were authors of this work. Members of this committee reviewed inpatient charts and documented resident compliance with VTE prevention standards during periods before and after they had viewed the CD‐ROMs. We chose this particular portion of the CD‐ROM because at both the test and control hospitals, initiatives were underway using order sets to improve anticoagulation in cardiac, neurological, and surgical patients but not in VTE prophylaxis. Charts from the same 2 nursing units on the medical service were reviewed in each period and included patients with a discharge diagnosis of congestive heart failure, any oncologic diagnosis, or sepsis. The chart review tool was developed by the anticoagulation committee and included a thrombosis risk factor assessment section as well as a list of contraindications to anticoagulation to determine if anticoagulation was appropriately implemented. Charts were reviewed for compliance with VTE prophylaxis after the CD‐ROM intervention (given in July 2004) in August 2004. To have a comparable pre‐CD‐ROM comparison, charts of patients with the diagnoses stated above were reviewed from August of the preceding year. The same month was chosen in the previous year to minimize any impact of resident experience, which would likely be a confounding factor if charts from May or June of the academic year were used as a control, for example. To determine whether an improvement in adherence to VTE prophylaxis standards was sustained, an additional chart review was carried out 7 months after the initial CD‐ROM viewing. The same group of observers, none of whom were authors, did all the chart assessments.

RESULTS

Assessment of Independent Effect of CD‐ROM Intervention

To determine what independent effect the CD‐ROM intervention might have, given that scores may improve with the acquisition of clinical experience alone, in July 2004 we tested internal medicine categorical PGY‐1s at our institution and at another tertiary‐care hospital, as described in the Methods section. The results of testing both groups are shown in Figure 1. Nineteen medical PGY‐1s at our hospital (hospital A) completed the anticoagulation pretest, and 16 completed the posttest. Twenty‐two medical PGY‐1s completed the pretest, and 17 completed the posttest at our neighboring hospital (hospital B). Although posttest scores were higher at both institutions, the increase in scores at our institution, which received the CD‐ROM intervention, was statistically significant, whereas the increase for the group not receiving the intervention was not significant. These findings suggest that the CD‐ROM intervention may have had an independent effect on resident knowledge of anticoagulation.

Figure 1

Effect of CD‐ROM Intervention on Resident Use of VTE Prophylaxis

Appropriate use of VTE prophylaxis by residents was assessed at 3 points, as detailed in the Methods section: 1 year before the CD‐ROM intervention (baseline), immediately after the CD‐ROM intervention, and 7 months after the CD‐ROM intervention. VTE prophylaxis, one element of the CD‐ROM, was chosen as a surrogate marker for the impact of the CD‐ROM initiative. A review of 40 charts of patients with the specified diagnoses (100% of the patients with the specified diagnoses, which represented about one third of admissions to the unit) before the CD‐ROM intervention revealed that 30 patients (75%) received appropriate VTE prophylaxis. A review of 38 charts after the CD‐ROM intervention showed that 36 patients (95%) received appropriate prophylaxis; similar findings were obtained 7 months after the CD‐ROM intervention (33 of 35 patients, 94%, P = .0107). These findings, which are shown in Figure 2, suggest that the CD‐ROM intervention enhanced resident compliance with VTE prophylaxis guidelines and that this effect was sustained for at least 7 months.

Figure 2

DISCUSSION

Residency training is facing challenges on several fronts. In addition to substantially changed educational requirements, strict limits on the amount of time that trainees can spend in the hospital have resulted from ACGME requirements and several state laws. Residents who are on night rotations or were on call the night before often miss educational conferences or must choose between attending patient carerelated activities and educational sessions. Time constraints on faculty have compounded this problem, and for residents to effectively learn, the focus of graduate medical education may need to shift somewhat from teaching medical information toward teaching the practice of self‐directed learning, with CD‐ROMs one such mode by which this can take place. Accomplishing this will require novel teaching approaches, and residency programs will need to document their effectiveness.

In this study we demonstrated that our residents increased their knowledge and improved their patient care practices using a CD‐ROM‐based educational tool. Residents frequently make use of computer‐based educational resources in the form of journals, textbooks, informational databases such as comprehensive drug listings, and personal digital assistantbased tools. Advantages of the computer‐based learning format include increased accessibility and flexibility in viewing the material. Residents have the option of repeated screening as desired and of viewing the CD‐ROM in segments if necessary. Although residents often must choose to attend a scheduled traditional lecture or engage in a patient carerelated activity, the CD‐ROM format allows the resident to choose the ideal time and setting to engage in structured educational activities. Other advantages of the CD‐ROM format would be ease of monitoring for accuracy, applicability, and comprehensiveness as well as more flexibility in faculty time commitments. It should be noted that we have no information about how much time residents devoted to the CD‐ROM program and how often they may have returned to the module for review. It should also be noted that although there have been some reports suggesting that CD‐ROM‐based education may play a useful role in student and perhaps resident education,1618 there is no evidence to date demonstrating that widespread use of CD‐ROMs in residency training can differentially affect resident behavior compared with the use of traditional methods.

A number of variables could have affected our results. For overall test scores, the response rate was less than 50%, with variability between departments suggesting that perhaps it was more motivated residents who participated and were therefore more likely to demonstrate improvement. Although our data comparing institutions with and without the CD‐ROM intervention suggested that the CD‐ROM intervention had a discernable effect on resident knowledge, we must also consider the possibility that the 2 groups might not have been comparable, as attitudes, expectations, and other variables might have differed. All the residents were categorical trainees, and given the similarities in many aspects of the training programs in these 2 tertiary‐care hospitals, as described in the Methods section, it is hoped that any such differences were minor. Nevertheless, this must be considered a limitation of our study. Also of note, the number of trainees was small, as was the patient population studied with VTE prophylaxis; hence, we recognize that our work can best be regarded as a pilot study using an alternative learning method. We also realize that giving a group of residents a test followed by distribution of a CD‐ROM might have suggested that we were directing them toward a goal, and this may have affected the results. Heightened awareness of the importance of anticoagulation from the introduction of new guidelines and other variables also could have affected our findings. The taking of an examination itself might also have had an impact on knowledge that could affect subsequent test scores. An additional point to consider is that if knowledge and patient care did improve, we do not know whether this affects residents acquiring other knowledge or whether this will translate into improved patient care in other areas.

Although CD‐ROM‐based learning could serve a useful function in the increasingly complex environment of residency training, this learning method also has disadvantages, including not providing personal contact or having the capability of question‐and‐answer sessions between teacher and resident. This could be overcome by providing time for faculty‐precepted question‐and‐answer sessions or perhaps creating a Web‐based venue for questions to be submitted and answered. In addition, the CD‐ROMs themselves can be designed in an interactive format in which residents can provide answers to clinical questions with feedback based on their selections provided as part of the CD‐ROM program.

In summary, the CD‐ROM‐based program in this study appears to have had an effect on not only knowledge but also patient care practice and suggests that this type of format could serve a useful role in residency training. Studies of additional interventions such as this one might allow for more extensive evaluation of the utility of CD‐ROM‐based learning as a residency training tool.

Given recent changes in the goals and objectives of residency training as well as changes in the functioning of teaching hospitals, traditional educational formats may need to be supplemented or replaced.1 The Accreditation Council for Graduate Medical Education (ACGME) is promoting changes in resident education with the goal of not only enhancing trainee competency using innovative methods but also of demonstrating that these educational innovations result in enhanced quality of patient care and improved patient safety.1 A challenging aspect of these initiatives is that programs are working to implement them at a time when there are greater nonteaching demands on faculty time, mandated resident work‐hour limitations have been instituted, in some states by law, and resident patient care and educational activities are prone to disruptions inherent in caring for patients in a complex health system. Various solutions have been proposed including increased incorporation of self‐directed learning as a means of meeting modern resident educational challenges, yet the ideal tools with which to accomplish this are unknown.

Computer‐based instruction in medicine has been available since the 1960s, and although its use had initially been more widespread in medical student, nursing, allied health professional, and patient education,2, 3 it is being increasingly incorporated into resident education as well. Some studies have shown that for medical students, computer‐based teaching is at least as effective in improving knowledge as conventional lectures4 and that learners' satisfaction with computer‐based formats appears comparable with that of traditional didactic lectures.5 In recent years computer‐based teaching has been applied to resident education in various fields including surgery and surgical subspecialties, pediatrics, and obstetrics and gynecology.612 Little is known, however, about how computer‐based educational methods affect resident knowledge and especially how these methods might affect clinical practice.

Venous thromboembolism (VTE) is a common and hazardous complication of acute inpatient hospitalization.13 Recognizing that errors in proper prescribing and monitoring of anticoagulants are a major cause of acute inpatient morbidity and mortality,14 we began an initiative to educate our residents and improve their patient care practices regarding the proper use of anticoagulants. To accomplish this, we developed a CD‐ROM‐based learning module with the aim of increasing resident knowledge of anticoagulation as well as compliance with national standards for VTE prevention. In this study we assessed the impact of the CD‐ROM intervention on resident knowledge and their appropriate use of VTE prophylaxis.

METHODS

The study was approved by the institutional review board. With the participation of faculty educators in the departments of medicine, surgery, and neurology, one of the authors (H.K.) coordinated the development of a CD‐ROM containing concise modules on core topics in anticoagulation (Table 1). The presenters for these topics included the director of clinical hematology, 2 cardiologists including the director of the coronary care unit, the director of the medical intensive care unit, the director of cerebrovascular diseases, and 2 vascular surgeons, one of whom serves as vice chair of surgery. These modules, each lasting about 1 hour, had audio and slide components detailing the proper indications, monitoring, and efficacy of anticoagulants in atrial fibrillation, acute ischemic stroke, acute coronary syndromes, and VTE prevention in acutely ill hospitalized patients. The guidelines presented were based on the sixth (2000) ACCP guidelines for antithrombotic therapy for the prevention and treatment of thrombosis.15 The content of the CD‐ROM was reviewed for accuracy by the authors, though none of them were speakers. We asked that before all current residents in the departments of cardiothoracic surgery, emergency medicine, otolaryngology, internal medicine, neurosurgery, dental medicine, neurology, obstetrics and gynecology, orthopedics, surgery, and urology viewed the CD‐ROM, they complete a pretest to determine their baseline knowledge of this subject. After completing the pretest, the residents were required to view the CD‐ROM and retake the same test. We then compared pre‐ and posttest scores.

To determine whether an increase in knowledge was secondary to the CD‐ROM intervention or simply a consequence of acquired clinical experience during training, we compared test scores of residents who did and did not receive the CD‐ROM intervention. In the academic year following our initial testing, we asked the incoming categorical medical PGY‐1 classes at our hospital and at a comparable local tertiary‐care hospital in our health system 2 miles away to take an anticoagulation pretest (different from the examination given for the initial testing) during their PGY‐1 orientations. The 2 institutions are comparable in many ways including in patient demographics and size and most residents come from the same medical schools, have a similar rotation structure, and use a comparable curriculum under a unified graduate medical education office. The CD‐ROMs were only given to categorical PGY‐1 residents at our institution. Both groups then retook the same test (posttest) 3 months into their clinical training, a time chosen because it is when all PGY‐1s would be expected to have gained significant clinical experience on the medical wards and or in the intensive care units. The exam questions were generated by one of the authors (B.M.) and covered all the topics in the CD‐ROM.

An Anticoagulation Steering Committee was formed to assess whether the CD‐ROM intervention affected our residents' patient care practices. None of the members of this committee were authors of this work. Members of this committee reviewed inpatient charts and documented resident compliance with VTE prevention standards during periods before and after they had viewed the CD‐ROMs. We chose this particular portion of the CD‐ROM because at both the test and control hospitals, initiatives were underway using order sets to improve anticoagulation in cardiac, neurological, and surgical patients but not in VTE prophylaxis. Charts from the same 2 nursing units on the medical service were reviewed in each period and included patients with a discharge diagnosis of congestive heart failure, any oncologic diagnosis, or sepsis. The chart review tool was developed by the anticoagulation committee and included a thrombosis risk factor assessment section as well as a list of contraindications to anticoagulation to determine if anticoagulation was appropriately implemented. Charts were reviewed for compliance with VTE prophylaxis after the CD‐ROM intervention (given in July 2004) in August 2004. To have a comparable pre‐CD‐ROM comparison, charts of patients with the diagnoses stated above were reviewed from August of the preceding year. The same month was chosen in the previous year to minimize any impact of resident experience, which would likely be a confounding factor if charts from May or June of the academic year were used as a control, for example. To determine whether an improvement in adherence to VTE prophylaxis standards was sustained, an additional chart review was carried out 7 months after the initial CD‐ROM viewing. The same group of observers, none of whom were authors, did all the chart assessments.

RESULTS

Assessment of Independent Effect of CD‐ROM Intervention

To determine what independent effect the CD‐ROM intervention might have, given that scores may improve with the acquisition of clinical experience alone, in July 2004 we tested internal medicine categorical PGY‐1s at our institution and at another tertiary‐care hospital, as described in the Methods section. The results of testing both groups are shown in Figure 1. Nineteen medical PGY‐1s at our hospital (hospital A) completed the anticoagulation pretest, and 16 completed the posttest. Twenty‐two medical PGY‐1s completed the pretest, and 17 completed the posttest at our neighboring hospital (hospital B). Although posttest scores were higher at both institutions, the increase in scores at our institution, which received the CD‐ROM intervention, was statistically significant, whereas the increase for the group not receiving the intervention was not significant. These findings suggest that the CD‐ROM intervention may have had an independent effect on resident knowledge of anticoagulation.

Figure 1

Effect of CD‐ROM Intervention on Resident Use of VTE Prophylaxis

Appropriate use of VTE prophylaxis by residents was assessed at 3 points, as detailed in the Methods section: 1 year before the CD‐ROM intervention (baseline), immediately after the CD‐ROM intervention, and 7 months after the CD‐ROM intervention. VTE prophylaxis, one element of the CD‐ROM, was chosen as a surrogate marker for the impact of the CD‐ROM initiative. A review of 40 charts of patients with the specified diagnoses (100% of the patients with the specified diagnoses, which represented about one third of admissions to the unit) before the CD‐ROM intervention revealed that 30 patients (75%) received appropriate VTE prophylaxis. A review of 38 charts after the CD‐ROM intervention showed that 36 patients (95%) received appropriate prophylaxis; similar findings were obtained 7 months after the CD‐ROM intervention (33 of 35 patients, 94%, P = .0107). These findings, which are shown in Figure 2, suggest that the CD‐ROM intervention enhanced resident compliance with VTE prophylaxis guidelines and that this effect was sustained for at least 7 months.

Figure 2

DISCUSSION

Residency training is facing challenges on several fronts. In addition to substantially changed educational requirements, strict limits on the amount of time that trainees can spend in the hospital have resulted from ACGME requirements and several state laws. Residents who are on night rotations or were on call the night before often miss educational conferences or must choose between attending patient carerelated activities and educational sessions. Time constraints on faculty have compounded this problem, and for residents to effectively learn, the focus of graduate medical education may need to shift somewhat from teaching medical information toward teaching the practice of self‐directed learning, with CD‐ROMs one such mode by which this can take place. Accomplishing this will require novel teaching approaches, and residency programs will need to document their effectiveness.

In this study we demonstrated that our residents increased their knowledge and improved their patient care practices using a CD‐ROM‐based educational tool. Residents frequently make use of computer‐based educational resources in the form of journals, textbooks, informational databases such as comprehensive drug listings, and personal digital assistantbased tools. Advantages of the computer‐based learning format include increased accessibility and flexibility in viewing the material. Residents have the option of repeated screening as desired and of viewing the CD‐ROM in segments if necessary. Although residents often must choose to attend a scheduled traditional lecture or engage in a patient carerelated activity, the CD‐ROM format allows the resident to choose the ideal time and setting to engage in structured educational activities. Other advantages of the CD‐ROM format would be ease of monitoring for accuracy, applicability, and comprehensiveness as well as more flexibility in faculty time commitments. It should be noted that we have no information about how much time residents devoted to the CD‐ROM program and how often they may have returned to the module for review. It should also be noted that although there have been some reports suggesting that CD‐ROM‐based education may play a useful role in student and perhaps resident education,1618 there is no evidence to date demonstrating that widespread use of CD‐ROMs in residency training can differentially affect resident behavior compared with the use of traditional methods.

A number of variables could have affected our results. For overall test scores, the response rate was less than 50%, with variability between departments suggesting that perhaps it was more motivated residents who participated and were therefore more likely to demonstrate improvement. Although our data comparing institutions with and without the CD‐ROM intervention suggested that the CD‐ROM intervention had a discernable effect on resident knowledge, we must also consider the possibility that the 2 groups might not have been comparable, as attitudes, expectations, and other variables might have differed. All the residents were categorical trainees, and given the similarities in many aspects of the training programs in these 2 tertiary‐care hospitals, as described in the Methods section, it is hoped that any such differences were minor. Nevertheless, this must be considered a limitation of our study. Also of note, the number of trainees was small, as was the patient population studied with VTE prophylaxis; hence, we recognize that our work can best be regarded as a pilot study using an alternative learning method. We also realize that giving a group of residents a test followed by distribution of a CD‐ROM might have suggested that we were directing them toward a goal, and this may have affected the results. Heightened awareness of the importance of anticoagulation from the introduction of new guidelines and other variables also could have affected our findings. The taking of an examination itself might also have had an impact on knowledge that could affect subsequent test scores. An additional point to consider is that if knowledge and patient care did improve, we do not know whether this affects residents acquiring other knowledge or whether this will translate into improved patient care in other areas.

Although CD‐ROM‐based learning could serve a useful function in the increasingly complex environment of residency training, this learning method also has disadvantages, including not providing personal contact or having the capability of question‐and‐answer sessions between teacher and resident. This could be overcome by providing time for faculty‐precepted question‐and‐answer sessions or perhaps creating a Web‐based venue for questions to be submitted and answered. In addition, the CD‐ROMs themselves can be designed in an interactive format in which residents can provide answers to clinical questions with feedback based on their selections provided as part of the CD‐ROM program.

In summary, the CD‐ROM‐based program in this study appears to have had an effect on not only knowledge but also patient care practice and suggests that this type of format could serve a useful role in residency training. Studies of additional interventions such as this one might allow for more extensive evaluation of the utility of CD‐ROM‐based learning as a residency training tool.

Given recent changes in the goals and objectives of residency training as well as changes in the functioning of teaching hospitals, traditional educational formats may need to be supplemented or replaced.1 The Accreditation Council for Graduate Medical Education (ACGME) is promoting changes in resident education with the goal of not only enhancing trainee competency using innovative methods but also of demonstrating that these educational innovations result in enhanced quality of patient care and improved patient safety.1 A challenging aspect of these initiatives is that programs are working to implement them at a time when there are greater nonteaching demands on faculty time, mandated resident work‐hour limitations have been instituted, in some states by law, and resident patient care and educational activities are prone to disruptions inherent in caring for patients in a complex health system. Various solutions have been proposed including increased incorporation of self‐directed learning as a means of meeting modern resident educational challenges, yet the ideal tools with which to accomplish this are unknown.

Computer‐based instruction in medicine has been available since the 1960s, and although its use had initially been more widespread in medical student, nursing, allied health professional, and patient education,2, 3 it is being increasingly incorporated into resident education as well. Some studies have shown that for medical students, computer‐based teaching is at least as effective in improving knowledge as conventional lectures4 and that learners' satisfaction with computer‐based formats appears comparable with that of traditional didactic lectures.5 In recent years computer‐based teaching has been applied to resident education in various fields including surgery and surgical subspecialties, pediatrics, and obstetrics and gynecology.612 Little is known, however, about how computer‐based educational methods affect resident knowledge and especially how these methods might affect clinical practice.

Venous thromboembolism (VTE) is a common and hazardous complication of acute inpatient hospitalization.13 Recognizing that errors in proper prescribing and monitoring of anticoagulants are a major cause of acute inpatient morbidity and mortality,14 we began an initiative to educate our residents and improve their patient care practices regarding the proper use of anticoagulants. To accomplish this, we developed a CD‐ROM‐based learning module with the aim of increasing resident knowledge of anticoagulation as well as compliance with national standards for VTE prevention. In this study we assessed the impact of the CD‐ROM intervention on resident knowledge and their appropriate use of VTE prophylaxis.

METHODS

The study was approved by the institutional review board. With the participation of faculty educators in the departments of medicine, surgery, and neurology, one of the authors (H.K.) coordinated the development of a CD‐ROM containing concise modules on core topics in anticoagulation (Table 1). The presenters for these topics included the director of clinical hematology, 2 cardiologists including the director of the coronary care unit, the director of the medical intensive care unit, the director of cerebrovascular diseases, and 2 vascular surgeons, one of whom serves as vice chair of surgery. These modules, each lasting about 1 hour, had audio and slide components detailing the proper indications, monitoring, and efficacy of anticoagulants in atrial fibrillation, acute ischemic stroke, acute coronary syndromes, and VTE prevention in acutely ill hospitalized patients. The guidelines presented were based on the sixth (2000) ACCP guidelines for antithrombotic therapy for the prevention and treatment of thrombosis.15 The content of the CD‐ROM was reviewed for accuracy by the authors, though none of them were speakers. We asked that before all current residents in the departments of cardiothoracic surgery, emergency medicine, otolaryngology, internal medicine, neurosurgery, dental medicine, neurology, obstetrics and gynecology, orthopedics, surgery, and urology viewed the CD‐ROM, they complete a pretest to determine their baseline knowledge of this subject. After completing the pretest, the residents were required to view the CD‐ROM and retake the same test. We then compared pre‐ and posttest scores.

To determine whether an increase in knowledge was secondary to the CD‐ROM intervention or simply a consequence of acquired clinical experience during training, we compared test scores of residents who did and did not receive the CD‐ROM intervention. In the academic year following our initial testing, we asked the incoming categorical medical PGY‐1 classes at our hospital and at a comparable local tertiary‐care hospital in our health system 2 miles away to take an anticoagulation pretest (different from the examination given for the initial testing) during their PGY‐1 orientations. The 2 institutions are comparable in many ways including in patient demographics and size and most residents come from the same medical schools, have a similar rotation structure, and use a comparable curriculum under a unified graduate medical education office. The CD‐ROMs were only given to categorical PGY‐1 residents at our institution. Both groups then retook the same test (posttest) 3 months into their clinical training, a time chosen because it is when all PGY‐1s would be expected to have gained significant clinical experience on the medical wards and or in the intensive care units. The exam questions were generated by one of the authors (B.M.) and covered all the topics in the CD‐ROM.

An Anticoagulation Steering Committee was formed to assess whether the CD‐ROM intervention affected our residents' patient care practices. None of the members of this committee were authors of this work. Members of this committee reviewed inpatient charts and documented resident compliance with VTE prevention standards during periods before and after they had viewed the CD‐ROMs. We chose this particular portion of the CD‐ROM because at both the test and control hospitals, initiatives were underway using order sets to improve anticoagulation in cardiac, neurological, and surgical patients but not in VTE prophylaxis. Charts from the same 2 nursing units on the medical service were reviewed in each period and included patients with a discharge diagnosis of congestive heart failure, any oncologic diagnosis, or sepsis. The chart review tool was developed by the anticoagulation committee and included a thrombosis risk factor assessment section as well as a list of contraindications to anticoagulation to determine if anticoagulation was appropriately implemented. Charts were reviewed for compliance with VTE prophylaxis after the CD‐ROM intervention (given in July 2004) in August 2004. To have a comparable pre‐CD‐ROM comparison, charts of patients with the diagnoses stated above were reviewed from August of the preceding year. The same month was chosen in the previous year to minimize any impact of resident experience, which would likely be a confounding factor if charts from May or June of the academic year were used as a control, for example. To determine whether an improvement in adherence to VTE prophylaxis standards was sustained, an additional chart review was carried out 7 months after the initial CD‐ROM viewing. The same group of observers, none of whom were authors, did all the chart assessments.

RESULTS

Assessment of Independent Effect of CD‐ROM Intervention

To determine what independent effect the CD‐ROM intervention might have, given that scores may improve with the acquisition of clinical experience alone, in July 2004 we tested internal medicine categorical PGY‐1s at our institution and at another tertiary‐care hospital, as described in the Methods section. The results of testing both groups are shown in Figure 1. Nineteen medical PGY‐1s at our hospital (hospital A) completed the anticoagulation pretest, and 16 completed the posttest. Twenty‐two medical PGY‐1s completed the pretest, and 17 completed the posttest at our neighboring hospital (hospital B). Although posttest scores were higher at both institutions, the increase in scores at our institution, which received the CD‐ROM intervention, was statistically significant, whereas the increase for the group not receiving the intervention was not significant. These findings suggest that the CD‐ROM intervention may have had an independent effect on resident knowledge of anticoagulation.

Figure 1

Effect of CD‐ROM Intervention on Resident Use of VTE Prophylaxis

Appropriate use of VTE prophylaxis by residents was assessed at 3 points, as detailed in the Methods section: 1 year before the CD‐ROM intervention (baseline), immediately after the CD‐ROM intervention, and 7 months after the CD‐ROM intervention. VTE prophylaxis, one element of the CD‐ROM, was chosen as a surrogate marker for the impact of the CD‐ROM initiative. A review of 40 charts of patients with the specified diagnoses (100% of the patients with the specified diagnoses, which represented about one third of admissions to the unit) before the CD‐ROM intervention revealed that 30 patients (75%) received appropriate VTE prophylaxis. A review of 38 charts after the CD‐ROM intervention showed that 36 patients (95%) received appropriate prophylaxis; similar findings were obtained 7 months after the CD‐ROM intervention (33 of 35 patients, 94%, P = .0107). These findings, which are shown in Figure 2, suggest that the CD‐ROM intervention enhanced resident compliance with VTE prophylaxis guidelines and that this effect was sustained for at least 7 months.

Figure 2

DISCUSSION

Residency training is facing challenges on several fronts. In addition to substantially changed educational requirements, strict limits on the amount of time that trainees can spend in the hospital have resulted from ACGME requirements and several state laws. Residents who are on night rotations or were on call the night before often miss educational conferences or must choose between attending patient carerelated activities and educational sessions. Time constraints on faculty have compounded this problem, and for residents to effectively learn, the focus of graduate medical education may need to shift somewhat from teaching medical information toward teaching the practice of self‐directed learning, with CD‐ROMs one such mode by which this can take place. Accomplishing this will require novel teaching approaches, and residency programs will need to document their effectiveness.

In this study we demonstrated that our residents increased their knowledge and improved their patient care practices using a CD‐ROM‐based educational tool. Residents frequently make use of computer‐based educational resources in the form of journals, textbooks, informational databases such as comprehensive drug listings, and personal digital assistantbased tools. Advantages of the computer‐based learning format include increased accessibility and flexibility in viewing the material. Residents have the option of repeated screening as desired and of viewing the CD‐ROM in segments if necessary. Although residents often must choose to attend a scheduled traditional lecture or engage in a patient carerelated activity, the CD‐ROM format allows the resident to choose the ideal time and setting to engage in structured educational activities. Other advantages of the CD‐ROM format would be ease of monitoring for accuracy, applicability, and comprehensiveness as well as more flexibility in faculty time commitments. It should be noted that we have no information about how much time residents devoted to the CD‐ROM program and how often they may have returned to the module for review. It should also be noted that although there have been some reports suggesting that CD‐ROM‐based education may play a useful role in student and perhaps resident education,1618 there is no evidence to date demonstrating that widespread use of CD‐ROMs in residency training can differentially affect resident behavior compared with the use of traditional methods.

A number of variables could have affected our results. For overall test scores, the response rate was less than 50%, with variability between departments suggesting that perhaps it was more motivated residents who participated and were therefore more likely to demonstrate improvement. Although our data comparing institutions with and without the CD‐ROM intervention suggested that the CD‐ROM intervention had a discernable effect on resident knowledge, we must also consider the possibility that the 2 groups might not have been comparable, as attitudes, expectations, and other variables might have differed. All the residents were categorical trainees, and given the similarities in many aspects of the training programs in these 2 tertiary‐care hospitals, as described in the Methods section, it is hoped that any such differences were minor. Nevertheless, this must be considered a limitation of our study. Also of note, the number of trainees was small, as was the patient population studied with VTE prophylaxis; hence, we recognize that our work can best be regarded as a pilot study using an alternative learning method. We also realize that giving a group of residents a test followed by distribution of a CD‐ROM might have suggested that we were directing them toward a goal, and this may have affected the results. Heightened awareness of the importance of anticoagulation from the introduction of new guidelines and other variables also could have affected our findings. The taking of an examination itself might also have had an impact on knowledge that could affect subsequent test scores. An additional point to consider is that if knowledge and patient care did improve, we do not know whether this affects residents acquiring other knowledge or whether this will translate into improved patient care in other areas.

Although CD‐ROM‐based learning could serve a useful function in the increasingly complex environment of residency training, this learning method also has disadvantages, including not providing personal contact or having the capability of question‐and‐answer sessions between teacher and resident. This could be overcome by providing time for faculty‐precepted question‐and‐answer sessions or perhaps creating a Web‐based venue for questions to be submitted and answered. In addition, the CD‐ROMs themselves can be designed in an interactive format in which residents can provide answers to clinical questions with feedback based on their selections provided as part of the CD‐ROM program.

In summary, the CD‐ROM‐based program in this study appears to have had an effect on not only knowledge but also patient care practice and suggests that this type of format could serve a useful role in residency training. Studies of additional interventions such as this one might allow for more extensive evaluation of the utility of CD‐ROM‐based learning as a residency training tool.

Hyperhemolysis syndrome is a form of atypical hemolytic transfusion reaction (HTR). It is characterized by a significant drop in hemoglobin (Hb) after seemingly compatible red blood cell transfusions, suggesting destruction of both transfused and autologous red blood cells. Its pathophysiology is not well understood, and a serologic cause is often not identified.14 In contrast, delayed HTRs are typically characterized by a positive direct antiglobulin test (DAT), suggesting that the patient's red blood cells are coated by immunoglobulin G and/or complement components and by the appearance of previously undetected red blood cell alloantibody or antibodies that developed from a secondary anamnestic response; however, autologous red cells are not destroyed.

CASE

A 48‐year‐old African American woman with sickle cell disease (SCD) was readmitted for pain crisis. Her medical history included stroke, pulmonary hypertension, and congestive heart failure. She had received several transfusions and consequently had developed antibodies to seven clinically significant red blood cell antigens. A week prior to readmission, she was discharged from the hospital with an Hb of 6.9 g/dL after a sickle cell crisis precipitated by pneumonia. She was treated with hydration, pain medications, antibiotics, and a unit of cross‐match‐compatible red blood cells (RBCs) that was antigen negative for her antibodies.

On readmission, she had an Hb of 5.6 g/dL and an uncorrected reticulocyte count of 17.6%. Her reticulocyte production index, a reticulocyte count corrected for the degree of anemia and reticulocyte maturation time, was elevated at 2.6. She was transfused with 1 unit of phenotypically matched and cross‐match‐compatible RBCs. Three hours after transfusion, she developed dark‐colored urine. The transfusion reaction investigation revealed no clerical error or incompatibility, a negative DAT, and an antibody panel identical to that from pretransfusion testing. During hospitalization, the hemolytic anemia worsened (Fig. 1). On the 10th hospital day, she became severely dyspneic as her Hb reached its nadir of 3.6 g/dL despite ongoing erythropoiesis. She developed decompensated heart failure and renal insufficiency, precipitated by the acutely worsening anemia. Along with diuretic and vasodilator therapies, she was treated with methylprednisolone at 125 mg twice daily for 3 days followed by tapering doses of prednisone for 2 weeks, intravenous immunoglobulin (IVIG) at 400 mg/kg a day for 5 days, and 4 cross‐match‐compatible RBC transfusions that were antigen negative for her antibodies. The hemolysis resolved and the patient improved. Throughout hospitalization, her DAT remained negative. The Hb remained stable at 7 g/dL until she was discharged. Ten months of follow‐up showed no new red blood cell antibody in her serum or recurrence of hyperhemolysis syndrome despite receiving subsequent transfusions.

Figure 1

DISCUSSION

Hyperhemolysis syndrome has been described in patients with SCD,14, 6, 7 suggesting that an underlying hemoglobinopathy may be a risk factor; however, a patient with anemia of chronic disease was recently described in the literature to have developed hyperhemolysis syndrome.5 Possible mechanisms include innocent bystander hemolysis through complement‐mediated lysis and/or formation of red blood cell alloantibody or autoantibody;1, 2 and hyperactive macrophages of the reticuloendothelial system that recognize Hb S RBCs of patients with SCD more avidly than normal RBCs because of the exposure of aminophosphatides in the outer layer of the sickled RBC membrane.3 In effect, red blood cells may be destroyed regardless of whether they are autologous or transfused. Additionally, transfusion‐related suppression of erythropoiesis may worsen the severity of anemia.2 Recent studies of patients with SCD suggest that the presence of free plasma Hb, as a consequence of hemolysis, reduces nitric oxide bioavailability, promotes endothelial dysfunction, and contributes to the development of pulmonary hypertension and the varying presentations of vasoocclusion.6 A common observation among patients who experience hyperhemolysis syndrome is that withholding transfusion seems beneficial, probably because immunologic reactions are not exacerbated, whereas treatment with steroids1, 2, 4 and/or IVIG3, 7 resolves hemolysis because of their immunomodulatory effects.

CONCLUSIONS

Hyperhemolysis syndrome is a potentially life‐threatening complication of RBC transfusion. It is important to recognize this syndrome when managing patients with SCD who present with worsening anemia after RBC transfusions. Although further transfusions can exacerbate hemolysis4, 7 and may be relatively contraindicated, in severe and desperate situations, simultaneous treatment with steroids and IVIG, together with RBC transfusions, may be lifesaving.

Hyperhemolysis syndrome is a form of atypical hemolytic transfusion reaction (HTR). It is characterized by a significant drop in hemoglobin (Hb) after seemingly compatible red blood cell transfusions, suggesting destruction of both transfused and autologous red blood cells. Its pathophysiology is not well understood, and a serologic cause is often not identified.14 In contrast, delayed HTRs are typically characterized by a positive direct antiglobulin test (DAT), suggesting that the patient's red blood cells are coated by immunoglobulin G and/or complement components and by the appearance of previously undetected red blood cell alloantibody or antibodies that developed from a secondary anamnestic response; however, autologous red cells are not destroyed.

CASE

A 48‐year‐old African American woman with sickle cell disease (SCD) was readmitted for pain crisis. Her medical history included stroke, pulmonary hypertension, and congestive heart failure. She had received several transfusions and consequently had developed antibodies to seven clinically significant red blood cell antigens. A week prior to readmission, she was discharged from the hospital with an Hb of 6.9 g/dL after a sickle cell crisis precipitated by pneumonia. She was treated with hydration, pain medications, antibiotics, and a unit of cross‐match‐compatible red blood cells (RBCs) that was antigen negative for her antibodies.

On readmission, she had an Hb of 5.6 g/dL and an uncorrected reticulocyte count of 17.6%. Her reticulocyte production index, a reticulocyte count corrected for the degree of anemia and reticulocyte maturation time, was elevated at 2.6. She was transfused with 1 unit of phenotypically matched and cross‐match‐compatible RBCs. Three hours after transfusion, she developed dark‐colored urine. The transfusion reaction investigation revealed no clerical error or incompatibility, a negative DAT, and an antibody panel identical to that from pretransfusion testing. During hospitalization, the hemolytic anemia worsened (Fig. 1). On the 10th hospital day, she became severely dyspneic as her Hb reached its nadir of 3.6 g/dL despite ongoing erythropoiesis. She developed decompensated heart failure and renal insufficiency, precipitated by the acutely worsening anemia. Along with diuretic and vasodilator therapies, she was treated with methylprednisolone at 125 mg twice daily for 3 days followed by tapering doses of prednisone for 2 weeks, intravenous immunoglobulin (IVIG) at 400 mg/kg a day for 5 days, and 4 cross‐match‐compatible RBC transfusions that were antigen negative for her antibodies. The hemolysis resolved and the patient improved. Throughout hospitalization, her DAT remained negative. The Hb remained stable at 7 g/dL until she was discharged. Ten months of follow‐up showed no new red blood cell antibody in her serum or recurrence of hyperhemolysis syndrome despite receiving subsequent transfusions.

Figure 1

DISCUSSION

Hyperhemolysis syndrome has been described in patients with SCD,14, 6, 7 suggesting that an underlying hemoglobinopathy may be a risk factor; however, a patient with anemia of chronic disease was recently described in the literature to have developed hyperhemolysis syndrome.5 Possible mechanisms include innocent bystander hemolysis through complement‐mediated lysis and/or formation of red blood cell alloantibody or autoantibody;1, 2 and hyperactive macrophages of the reticuloendothelial system that recognize Hb S RBCs of patients with SCD more avidly than normal RBCs because of the exposure of aminophosphatides in the outer layer of the sickled RBC membrane.3 In effect, red blood cells may be destroyed regardless of whether they are autologous or transfused. Additionally, transfusion‐related suppression of erythropoiesis may worsen the severity of anemia.2 Recent studies of patients with SCD suggest that the presence of free plasma Hb, as a consequence of hemolysis, reduces nitric oxide bioavailability, promotes endothelial dysfunction, and contributes to the development of pulmonary hypertension and the varying presentations of vasoocclusion.6 A common observation among patients who experience hyperhemolysis syndrome is that withholding transfusion seems beneficial, probably because immunologic reactions are not exacerbated, whereas treatment with steroids1, 2, 4 and/or IVIG3, 7 resolves hemolysis because of their immunomodulatory effects.

CONCLUSIONS

Hyperhemolysis syndrome is a potentially life‐threatening complication of RBC transfusion. It is important to recognize this syndrome when managing patients with SCD who present with worsening anemia after RBC transfusions. Although further transfusions can exacerbate hemolysis4, 7 and may be relatively contraindicated, in severe and desperate situations, simultaneous treatment with steroids and IVIG, together with RBC transfusions, may be lifesaving.

Hyperhemolysis syndrome is a form of atypical hemolytic transfusion reaction (HTR). It is characterized by a significant drop in hemoglobin (Hb) after seemingly compatible red blood cell transfusions, suggesting destruction of both transfused and autologous red blood cells. Its pathophysiology is not well understood, and a serologic cause is often not identified.14 In contrast, delayed HTRs are typically characterized by a positive direct antiglobulin test (DAT), suggesting that the patient's red blood cells are coated by immunoglobulin G and/or complement components and by the appearance of previously undetected red blood cell alloantibody or antibodies that developed from a secondary anamnestic response; however, autologous red cells are not destroyed.

CASE

A 48‐year‐old African American woman with sickle cell disease (SCD) was readmitted for pain crisis. Her medical history included stroke, pulmonary hypertension, and congestive heart failure. She had received several transfusions and consequently had developed antibodies to seven clinically significant red blood cell antigens. A week prior to readmission, she was discharged from the hospital with an Hb of 6.9 g/dL after a sickle cell crisis precipitated by pneumonia. She was treated with hydration, pain medications, antibiotics, and a unit of cross‐match‐compatible red blood cells (RBCs) that was antigen negative for her antibodies.

On readmission, she had an Hb of 5.6 g/dL and an uncorrected reticulocyte count of 17.6%. Her reticulocyte production index, a reticulocyte count corrected for the degree of anemia and reticulocyte maturation time, was elevated at 2.6. She was transfused with 1 unit of phenotypically matched and cross‐match‐compatible RBCs. Three hours after transfusion, she developed dark‐colored urine. The transfusion reaction investigation revealed no clerical error or incompatibility, a negative DAT, and an antibody panel identical to that from pretransfusion testing. During hospitalization, the hemolytic anemia worsened (Fig. 1). On the 10th hospital day, she became severely dyspneic as her Hb reached its nadir of 3.6 g/dL despite ongoing erythropoiesis. She developed decompensated heart failure and renal insufficiency, precipitated by the acutely worsening anemia. Along with diuretic and vasodilator therapies, she was treated with methylprednisolone at 125 mg twice daily for 3 days followed by tapering doses of prednisone for 2 weeks, intravenous immunoglobulin (IVIG) at 400 mg/kg a day for 5 days, and 4 cross‐match‐compatible RBC transfusions that were antigen negative for her antibodies. The hemolysis resolved and the patient improved. Throughout hospitalization, her DAT remained negative. The Hb remained stable at 7 g/dL until she was discharged. Ten months of follow‐up showed no new red blood cell antibody in her serum or recurrence of hyperhemolysis syndrome despite receiving subsequent transfusions.

Figure 1

DISCUSSION

Hyperhemolysis syndrome has been described in patients with SCD,14, 6, 7 suggesting that an underlying hemoglobinopathy may be a risk factor; however, a patient with anemia of chronic disease was recently described in the literature to have developed hyperhemolysis syndrome.5 Possible mechanisms include innocent bystander hemolysis through complement‐mediated lysis and/or formation of red blood cell alloantibody or autoantibody;1, 2 and hyperactive macrophages of the reticuloendothelial system that recognize Hb S RBCs of patients with SCD more avidly than normal RBCs because of the exposure of aminophosphatides in the outer layer of the sickled RBC membrane.3 In effect, red blood cells may be destroyed regardless of whether they are autologous or transfused. Additionally, transfusion‐related suppression of erythropoiesis may worsen the severity of anemia.2 Recent studies of patients with SCD suggest that the presence of free plasma Hb, as a consequence of hemolysis, reduces nitric oxide bioavailability, promotes endothelial dysfunction, and contributes to the development of pulmonary hypertension and the varying presentations of vasoocclusion.6 A common observation among patients who experience hyperhemolysis syndrome is that withholding transfusion seems beneficial, probably because immunologic reactions are not exacerbated, whereas treatment with steroids1, 2, 4 and/or IVIG3, 7 resolves hemolysis because of their immunomodulatory effects.

CONCLUSIONS

Hyperhemolysis syndrome is a potentially life‐threatening complication of RBC transfusion. It is important to recognize this syndrome when managing patients with SCD who present with worsening anemia after RBC transfusions. Although further transfusions can exacerbate hemolysis4, 7 and may be relatively contraindicated, in severe and desperate situations, simultaneous treatment with steroids and IVIG, together with RBC transfusions, may be lifesaving.

The use of hospitalists, physicians who specialize in inpatient care, has seen a rapid expansion over the last decade.1 Several studies have shown that with hospitalists there is a shorter length of stay (LOS) and decreased utilization of resources and that hospitalists play a positive role in medical education.24 However, only a few studies have examined the specific strategies employed by hospitalists to achieve improved efficiency and outcomes.

Congestive heart failure (CHF) is the most common diagnosis of hospitalized patients older than age 65, with more Medicare spending devoted to patients with CHF than to any other diagnosis‐related group (DRG).5, 6 Over the last 2 decades hospital discharges for congestive heart failure increased by 165%.7 In addition, the rate of hospital readmission of patients with CHF remains high: 2%, 20%, and 50% within 2 days, 1 month, and 6 months, respectively.8

Several previous studies have shown that patients cared for by hospitalists had improved clinical outcomes. Meltzer et al. found that 30‐day mortality of hospitalists' patients was lower than that of non‐hospitalists' patients, 4.2% versus 6.0%, respectively, in the second year of implementation of a hospitalist program.3 A study by Huddleston et al. showed a reduction of 11.8% in the rate of complications experienced by postsurgical orthopedic patients with the involvement of hospitalists in their care in conjunction with the surgeons.4

Many previous studies have pointed to improvements in economic outcomes such as LOS and costs for patients followed by hospitalists. Kulaga et al. showed that patients cared for by hospitalists had reductions of approximately 20% in LOS and 18% in total costs per case compared with those cared for by community‐based physicians.2 Meltzer et al. found a decrease in the average adjusted LOS of 0.49 days in the second year of implementation of a hospitalist program.3 Rifkin et al. found that patients with pneumonia cared for by hospitalists had a mean adjusted LOS of 5.6 days versus 6.5 days for those cared for by non‐hospitalists.9

Few previous studies have looked at specific practice patterns of hospitalists that result in improved efficiency and better outcomes. Rifkin et al., who found that patients with pneumonia cared for by hospitalists had a shorter LOS, suggested this finding was a result of the earlier recognition by hospitalists that patients were stable and more rapid conversion to oral antibiotics.9 Likewise, Stein et al. found that community‐acquired pneumonia patients treated by hospitalists had a shorter LOS than those treated by non‐hospitalists. However, they were unable to assess the differences in patient management that led to this result because of the design of the study.10

Lindenauer et al. compared quality‐of‐care indicators and resource utilization for patients with congestive heart failure treated by hospitalists and non‐hospitalist general internists. They found that patients under the care of hospitalists had a shorter LOS than those cared for by general internists but that the overall costs of care were similar between the groups.11 They compared the quality indicators developed by the Joint Commission on Accreditation of Healthcare Organizations in the Core Measures Initiative, but did not focus on patterns of practices of hospitalists and nonhospitalists. Moreover, they did not look at full‐time hospitalists but focused on physicians who spent at least 25% of their practice caring for inpatients.

We sought to identify distinct, quantifiable practices of full‐time hospitalists in the management of their patients with CHF. We hypothesized that hospitalists would adhere more closely to the current congestive heart failure guidelines and would utilize available resources more judiciously, leading to improved clinical and economic outcomes. To identify these practices, we compared utilization of well‐established therapeutic and diagnostic modalities such as use of ACE‐I, ARB, and beta‐blockers; ordering of chest x‐rays; measurement of brain natriuretic peptide (BNP); and use of medical subspecialty consultants. We also compared standard clinical and economic outcomes such as in‐hospital mortality, readmission rate, LOS, and costs per case between hospitalists and community‐based physicians.

METHODS

RESULTS

DISCUSSION

The use of hospitalists, physicians who specialize in inpatient care, has seen a rapid expansion over the last decade.1 Several studies have shown that with hospitalists there is a shorter length of stay (LOS) and decreased utilization of resources and that hospitalists play a positive role in medical education.24 However, only a few studies have examined the specific strategies employed by hospitalists to achieve improved efficiency and outcomes.

Congestive heart failure (CHF) is the most common diagnosis of hospitalized patients older than age 65, with more Medicare spending devoted to patients with CHF than to any other diagnosis‐related group (DRG).5, 6 Over the last 2 decades hospital discharges for congestive heart failure increased by 165%.7 In addition, the rate of hospital readmission of patients with CHF remains high: 2%, 20%, and 50% within 2 days, 1 month, and 6 months, respectively.8

Several previous studies have shown that patients cared for by hospitalists had improved clinical outcomes. Meltzer et al. found that 30‐day mortality of hospitalists' patients was lower than that of non‐hospitalists' patients, 4.2% versus 6.0%, respectively, in the second year of implementation of a hospitalist program.3 A study by Huddleston et al. showed a reduction of 11.8% in the rate of complications experienced by postsurgical orthopedic patients with the involvement of hospitalists in their care in conjunction with the surgeons.4

Many previous studies have pointed to improvements in economic outcomes such as LOS and costs for patients followed by hospitalists. Kulaga et al. showed that patients cared for by hospitalists had reductions of approximately 20% in LOS and 18% in total costs per case compared with those cared for by community‐based physicians.2 Meltzer et al. found a decrease in the average adjusted LOS of 0.49 days in the second year of implementation of a hospitalist program.3 Rifkin et al. found that patients with pneumonia cared for by hospitalists had a mean adjusted LOS of 5.6 days versus 6.5 days for those cared for by non‐hospitalists.9

Few previous studies have looked at specific practice patterns of hospitalists that result in improved efficiency and better outcomes. Rifkin et al., who found that patients with pneumonia cared for by hospitalists had a shorter LOS, suggested this finding was a result of the earlier recognition by hospitalists that patients were stable and more rapid conversion to oral antibiotics.9 Likewise, Stein et al. found that community‐acquired pneumonia patients treated by hospitalists had a shorter LOS than those treated by non‐hospitalists. However, they were unable to assess the differences in patient management that led to this result because of the design of the study.10

Lindenauer et al. compared quality‐of‐care indicators and resource utilization for patients with congestive heart failure treated by hospitalists and non‐hospitalist general internists. They found that patients under the care of hospitalists had a shorter LOS than those cared for by general internists but that the overall costs of care were similar between the groups.11 They compared the quality indicators developed by the Joint Commission on Accreditation of Healthcare Organizations in the Core Measures Initiative, but did not focus on patterns of practices of hospitalists and nonhospitalists. Moreover, they did not look at full‐time hospitalists but focused on physicians who spent at least 25% of their practice caring for inpatients.

We sought to identify distinct, quantifiable practices of full‐time hospitalists in the management of their patients with CHF. We hypothesized that hospitalists would adhere more closely to the current congestive heart failure guidelines and would utilize available resources more judiciously, leading to improved clinical and economic outcomes. To identify these practices, we compared utilization of well‐established therapeutic and diagnostic modalities such as use of ACE‐I, ARB, and beta‐blockers; ordering of chest x‐rays; measurement of brain natriuretic peptide (BNP); and use of medical subspecialty consultants. We also compared standard clinical and economic outcomes such as in‐hospital mortality, readmission rate, LOS, and costs per case between hospitalists and community‐based physicians.

METHODS

RESULTS

DISCUSSION

The use of hospitalists, physicians who specialize in inpatient care, has seen a rapid expansion over the last decade.1 Several studies have shown that with hospitalists there is a shorter length of stay (LOS) and decreased utilization of resources and that hospitalists play a positive role in medical education.24 However, only a few studies have examined the specific strategies employed by hospitalists to achieve improved efficiency and outcomes.

Congestive heart failure (CHF) is the most common diagnosis of hospitalized patients older than age 65, with more Medicare spending devoted to patients with CHF than to any other diagnosis‐related group (DRG).5, 6 Over the last 2 decades hospital discharges for congestive heart failure increased by 165%.7 In addition, the rate of hospital readmission of patients with CHF remains high: 2%, 20%, and 50% within 2 days, 1 month, and 6 months, respectively.8

Several previous studies have shown that patients cared for by hospitalists had improved clinical outcomes. Meltzer et al. found that 30‐day mortality of hospitalists' patients was lower than that of non‐hospitalists' patients, 4.2% versus 6.0%, respectively, in the second year of implementation of a hospitalist program.3 A study by Huddleston et al. showed a reduction of 11.8% in the rate of complications experienced by postsurgical orthopedic patients with the involvement of hospitalists in their care in conjunction with the surgeons.4

Many previous studies have pointed to improvements in economic outcomes such as LOS and costs for patients followed by hospitalists. Kulaga et al. showed that patients cared for by hospitalists had reductions of approximately 20% in LOS and 18% in total costs per case compared with those cared for by community‐based physicians.2 Meltzer et al. found a decrease in the average adjusted LOS of 0.49 days in the second year of implementation of a hospitalist program.3 Rifkin et al. found that patients with pneumonia cared for by hospitalists had a mean adjusted LOS of 5.6 days versus 6.5 days for those cared for by non‐hospitalists.9

Few previous studies have looked at specific practice patterns of hospitalists that result in improved efficiency and better outcomes. Rifkin et al., who found that patients with pneumonia cared for by hospitalists had a shorter LOS, suggested this finding was a result of the earlier recognition by hospitalists that patients were stable and more rapid conversion to oral antibiotics.9 Likewise, Stein et al. found that community‐acquired pneumonia patients treated by hospitalists had a shorter LOS than those treated by non‐hospitalists. However, they were unable to assess the differences in patient management that led to this result because of the design of the study.10

Lindenauer et al. compared quality‐of‐care indicators and resource utilization for patients with congestive heart failure treated by hospitalists and non‐hospitalist general internists. They found that patients under the care of hospitalists had a shorter LOS than those cared for by general internists but that the overall costs of care were similar between the groups.11 They compared the quality indicators developed by the Joint Commission on Accreditation of Healthcare Organizations in the Core Measures Initiative, but did not focus on patterns of practices of hospitalists and nonhospitalists. Moreover, they did not look at full‐time hospitalists but focused on physicians who spent at least 25% of their practice caring for inpatients.

We sought to identify distinct, quantifiable practices of full‐time hospitalists in the management of their patients with CHF. We hypothesized that hospitalists would adhere more closely to the current congestive heart failure guidelines and would utilize available resources more judiciously, leading to improved clinical and economic outcomes. To identify these practices, we compared utilization of well‐established therapeutic and diagnostic modalities such as use of ACE‐I, ARB, and beta‐blockers; ordering of chest x‐rays; measurement of brain natriuretic peptide (BNP); and use of medical subspecialty consultants. We also compared standard clinical and economic outcomes such as in‐hospital mortality, readmission rate, LOS, and costs per case between hospitalists and community‐based physicians.

METHODS

RESULTS

DISCUSSION

Diabetes mellitus is a common comorbid condition in hospitalized patients. In 2003, diabetes was listed as a diagnosis in 17.2% of hospital discharges in the United States.1 Because these diagnosis codes do not account for undiagnosed diabetes or hospital‐related hyperglycemia, the true prevalence of diabetes or hyperglycemia in hospitalized patients is likely higher and has been estimated to be as great as 38%.2 Hyperglycemia has been associated with adverse outcomes among hospitalized patients, including infectious complications, increased length of stay, and increased mortality.27 However, because hyperglycemia is not usually the primary reason patients with diabetes are hospitalized, its management is often not a focus in the inpatient setting. Sliding‐scale insulin alone continues to be commonly prescribed despite clinical evidence showing it to be ineffective in achieving glycemic control.8, 9

Recent randomized controlled trials have demonstrated that aggressive treatment of inpatient hyperglycemia improves outcomes in surgical and medical intensive care units10, 11 and in patients admitted for myocardial infarction.12, 13 Based on this clinical evidence and strong observational data linking hyperglycemia to poor patient outcomes in the non‐ICU setting,27 the American Diabetes Association (ADA) now advocates good metabolic control, defined as preprandial glucose levels of 90‐130 mg/dL and peak postprandial glucose levels < 180 mg/dL in hospitalized non‐ICU patients with hyperglycemia14 (note that these targets are less aggressive than those for ICU patients, for whom randomized controlled trials showed the benefits of reduced mortality provided by tight glucose control).11 To reach these targets, the ADA and American College of Endocrinology suggest that multidisciplinary teams develop and implement hyperglycemia management guidelines and protocols.15 Protocols should promote the use of continuous intravenous insulin or scheduled subcutaneous insulin as opposed to the use of sliding‐scale insulin alone. Subcutaneous insulin protocols should include target glucose levels; basal, nutritional, and supplemental insulin; and daily adjustments based on previous glucose levels, insulin sensitivity, nutritional intake, illness, and medications.6, 15 To date, few published protocols or algorithms for inpatient subcutaneous insulin have been shown to be effective.16, 17 It is therefore not known how best to design and implement an inpatient diabetes management protocol that is effective, efficient, and self‐perpetuating. The aims of our pilot study were to develop and implement a subcutaneous insulin protocol on a general medicine service, to identify barriers to implementation, and to determine the effect of this protocol on glycemic control.

METHODS

RESULTS

After screening all 785 admissions to the 2 medical teams during the study period, we prospectively identified 109 patients (14%) for the pilot study. Twenty patients were subsequently excluded: 7 patients who were discharged the same day they were identified, 4 who did not have a fasting blood glucose value greater than 140 mg/dL, 4 patients who had type 1 diabetes, 2 patients who were admitted with diabetic ketoacidosis, and 3 patients whose data could not be accessed because of repeated unavailability of the medical record. Characteristics of the remaining 89 study subjects are shown in Table 2 and are compared to 91 baseline subjects. The mean age of the study subjects was 68.7 years; 45% were men. Five patients (6%) did not have a previous diagnosis of diabetes, and 51% were taking insulin prior to admission; the median A1C was 6.8%.

The medical residents agreed, at least in theory, to follow the subcutaneous insulin protocol for 50 patients (56%), partially accepted it for 8 (9%), and declined for 31 (35%). Reasons for declining the protocol included fear of hypoglycemia, severity of patient's other disease states or overall poor health of patient, concern for the effects of renal insufficiency on insulin clearance, concern for the effect of steroid tapers on glucose levels, desire to titrate oral medications, and anticipation of patient's imminent discharge. Other reasons such as the glucose levels are not that bad and let's watch the glucose levels for one more day suggest that some residents did not view hyperglycemia as an acute problem requiring immediate attention.

Regarding insulin‐ordering practices (Table 3), basal insulin was prescribed for 57 patients (64%) in the pilot group compared to 45 patients (49%) in the baseline group (P = .05). Nutritional insulin was prescribed to 12 patients (13%) in the pilot group compared to no patients in the baseline group (P < .001). Oral hypoglycemic agents were prescribed less often in the pilot study than at baseline (20% vs. 38%, P = .01). The use of a standard default sliding scale from the hospital computer order set was high and was not significantly different in the pilot study compared with that at baseline (93% vs. 90%, P = .78). Twenty‐four of the 83 patients in the pilot group (29%) received sliding‐scale insulin without ever receiving basal or nutritional insulin during hospitalization compared to 45 of 91 patients in the baseline group (49%; P = .01 for comparison). Among patients started on basal insulin, 42% (24 of 57) were started after the first full hospital day. The initial basal insulin dose was appropriate according to the protocol (within 20%) in 38 of 57 patients (67%). Only 20 of 61 patients (33%) who had any hypo‐ or hyperglycemia had any change to their insulin regimen made during days 2 through 7 of their hospitalization on GMS, similar to the rate noted at baseline (36%).

Regarding glucose control (Table 3), the mean percentage of glucose readings per patient greater than 180 mg/dL was not significantly different in the pilot study compared to baseline (31.6% vs. 33.3%, P = .85). Despite implementation of the protocol and increased use of basal and nutritional insulin, 76% of patients had at least 1 routine glucose reading greater than 180 mg/dL, and 37% of patients had at least 40% of their routine glucose readings greater than 180 mg/dL, comparable to baseline (74% and 38%, respectively, P = NS for both comparisons). At least 1 hypoglycemic event (glucose reading below 60 mg/dL) occurred in 7% of patients at baseline and 13% during the pilot study (P = .20). Eleven hypoglycemic events in the pilot study were between 50 and 59 mg/dL (55%), 6 were between 40 and 49 mg/dL (30%), 3 were between 30 and 39 mg/dL (15%), and none were less than 30 mg/dL. Nine occurred before breakfast (45%), 5 before dinner (25%), 3 before lunch (15%), and 3 at bedtime (15%).

During the pilot study, the use of basal insulin did improve over the first 5 days of hospitalization (Fig. 1), in both the percentage of patients prescribed any basal insulin and the percentage of each patient's total insulin dose (basal, nutritional, and supplemental) given as basal (both P < .001 for trend). Hyperglycemia rates also improved during hospitalization (Fig. 1), decreasing from 48% on hospital day 1 to 34% on hospital day 5 (P = .004 for trend). These trends were not observed in the baseline group, with hyperglycemia rates of 37% on hospital day 1 and 34% on hospital day 5 (P = .16 for trend).

Figure 1

Patients for whom the resident accepted or partially accepted the protocol had higher use of basal insulin (91% vs. 13%, P < .0001), higher use of nutritional insulin (21% vs. 0%, P = .01), and more frequent dose adjustments (47% vs. 7%, P = .01) compared with patients for whom the resident declined the protocol. However, the rate of hyperglycemia was higher in patients for whom the protocol was accepted or partially accepted than in patients for whom the protocol was declined (37% vs. 20%, P = .02).

DISCUSSION

Our subcutaneous insulin protocol focused on increasing the use of basal and nutritional insulin, avoiding the use of sliding‐scale insulin by itself, and performing daily insulin adjustments in response to the hypo‐ or hyperglycemia of general medical inpatients with diabetes or hyperglycemia.

The most notable finding of our pilot study was that residents were resistant to using the protocol, both in general and in its specific recommendations. Despite receiving education about inpatient diabetes control and protocol recommendations from the team pharmacist, and despite being on a hospitalist‐run medical service, the residents accepted use of the protocol for only half the eligible patients. Patients who were started on basal insulin were often underdosed or started after the first day of hospitalization, and daily dose adjustments were not consistently made despite persistent hypo‐ or hyperglycemia. Although the use of nutritional insulin was greater compared with that in the baseline group, it was still only prescribed for 13% of patients. Use of a standard sliding scale from the hospital computer order set was common in the pilot study and similar to that in the baseline group. These results suggest significant resistance to changing the current standard of practice.

Despite this lack of adherence to the protocol, some modest improvements in processes of care were seen. Basal insulin was ordered more often during the pilot study than at baseline, especially over the course of a hospital stay. Nutritional insulin was also ordered more often during the pilot study than at baseline, but was still infrequent. Oral antihyperglycemic agents were ordered less often during the pilot study than at baseline. This demonstrates that use of the protocol may be able to improve process outcomes. However, the modest improvements in process outcomes could have simply been a result of increased awareness and education, not the protocol itself.

Regarding patient outcomes, the overall hyperglycemia rate did not improve in the pilot study relative to that at baseline. Importantly, hypoglycemia rates did not increase significantly compared with those at baseline. However, because of the small number of hypoglycemia events, the sample size may not have been sufficient to detect a true difference between groups.

The most likely reason that the protocol did not show an effect on glycemic control was that its recommendations were not adhered to. In turn, this may have been a result of incomplete education, training, and implementation measures and/or inherent problems with the protocol that made its recommendations difficult to follow. Another possibility is that the protocol itself may not have been capable of improving glucose control, even when properly used. However, we do know that resident agreement to use the protocol did lead to higher rates of recommended best practices being carried out, such as basal insulin use and daily insulin dose adjustments, and that use of the protocol was associated with improvements in glucose control over the hospital stay. A larger study with a higher degree of protocol adherence would be better able to evaluate the merits of the protocol itself, as would a randomized controlled trial using instrumental variables to measure treatment efficacy. Another possibility explanation for the lack of effect is that glucose control on admission happened to be worse in the pilot group than in the control group: rates of hyperglycemia on day 1 were 48% in the pilot group compared with 37% in the baseline group (Fig. 1). Also, the decreased use of oral agents in the pilot group, a purposeful change to decrease the risk of hypoglycemia, may have counteracted the beneficial effects of more appropriate insulin use. Finally, there were few patients with poorly controlled diabetes at baseline (18 patients with A1C 8.0 in the baseline group and 12 such patients in the pilot group), arguably those most likely to benefit.

There is a pressing need to identify protocols that can improve glucose control in the non‐ICU inpatient setting and successfully implement these protocols with a minimum of resources and effort. To date, most studies that have improved glucose control in the non‐ICU setting have relied on frequent input from diabetologists or nurse‐practitioners.14, 15

The results of this study should be viewed in light of its limitations, including its relatively small sample size (thus limiting our ability to detect possible significant differences between groups) and that it was conducted at a single institution (thus limiting its generalizability). Patients were enrolled on weekdays, so patients admitted and discharged over a weekend or on a holiday may have been missed. Also, because of the nonrandomized design of the study, we cannot exclude the possibility that the improvements noted in the pilot study were a result of the increased education provided or of increased awareness and general improvement in diabetes management over the course of the study. Finally, implementation of the protocol was somewhat labor intensive and required staff support that could be difficult to replicate in other institutions. However, most of the study staff's effort was necessary either to implement the protocol in the absence of an order set or to evaluate barriers to implementation. Widespread implementation of a protocol with an order set, education, and the use of highly reliable tools should be possible with much less effort and resources. The strengths of this study include its prospective data collection methods, which included rigorous inclusion criteria and collection of detailed clinical data.

Our study findings suggest several approaches to improve care in the future. To combat resistance to change, the American Association of Clinical Endocrinologists strongly recommends that each institution ensure that all its clinicians involved agree about general philosophies of diabetes management.19 A more expansive, hospital‐wide educational and promotional plan may increase the initial acceptance of the protocol. Interviews with residents also indicated there was unfamiliarity with diabetes management and significant concerns about the harmful affects of tight glucose control (ie, risk of hypoglycemia), especially in certain patient subgroups. These results confirmed the need for more practical individualized training and sparked the implementation of small‐group, case‐based educational sessions on inpatient diabetes management for all house officers, with a particular focus on patients with multiple comorbidities, on steroid tapers, and/or with renal failure.

The lack of nutritional insulin orders, delays in ordering basal insulin, and use of inadequate doses of insulin may be counteracted by the use of an order set, in our case built into our computer physician order entry (CPOE) system. The use of CPOE also allows reminders to be automatically sent to clinicians if eligible patients are not started on these orders. Clinical inertia (eg, failure to adjust the insulin doses of specific patients despite hyperglycemia) is more difficult to combat but may be addressed through better organization of clinical data, individualized, case‐based education, and CPOE reminders and eventually through culture change.

As a result of our pilot study, additional revisions were made to the protocol in hopes of increasing protocol adherence. For example, for patients eating discrete meals who are not taking insulin at home, the pilot protocol had suggested a starting insulin dose range for basal and nutritional insulin that required 2 separate calculations. The revised protocol was simplified to recommend a total daily insulin dose to be split evenly between basal and nutritional insulin. The daily adjustment instructions were also simplified. The pilot protocol had included a complicated table of adjustment recommendations based on bedside glucose trends. The revised protocol recommends adjusting the new daily dose by adding the total units of insulin given the previous day (including supplemental doses), making minor adjustments for hyper‐ or hypoglycemia and other clinical factors (like renal failure), and splitting this dose evenly between scheduled basal and nutritional insulin. In addition, 3 order sets were built into our computerized physician order entry system to facilitate early and appropriate insulin orders for patients with different diets (discrete meals, continuous tube feeds, and nothing by mouth); 3 different insulin sliding scales were created for patients with different degrees of insulin resistance; a diabetes management page for our electronic medication administration record is being developed to better organize clinical data; and hospital‐wide education and individualized training are ongoing.

In conclusion, the adherence to an inpatient glycemic management protocol that focused on increasing use of basal insulin and performing daily insulin adjustments was only fair. Barriers to successful implementation included clinical inertia regarding individual patients, unfamiliarity with inpatient diabetes management strategies, fear of hypoglycemia, and resistance to changing the current standard of practice. Targeted education, standard order sets, better organization of clinical data, protocol simplification, and institutional culture changes may be necessary for successful protocol implementation and improved inpatient glucose control.

Diabetes mellitus is a common comorbid condition in hospitalized patients. In 2003, diabetes was listed as a diagnosis in 17.2% of hospital discharges in the United States.1 Because these diagnosis codes do not account for undiagnosed diabetes or hospital‐related hyperglycemia, the true prevalence of diabetes or hyperglycemia in hospitalized patients is likely higher and has been estimated to be as great as 38%.2 Hyperglycemia has been associated with adverse outcomes among hospitalized patients, including infectious complications, increased length of stay, and increased mortality.27 However, because hyperglycemia is not usually the primary reason patients with diabetes are hospitalized, its management is often not a focus in the inpatient setting. Sliding‐scale insulin alone continues to be commonly prescribed despite clinical evidence showing it to be ineffective in achieving glycemic control.8, 9

Recent randomized controlled trials have demonstrated that aggressive treatment of inpatient hyperglycemia improves outcomes in surgical and medical intensive care units10, 11 and in patients admitted for myocardial infarction.12, 13 Based on this clinical evidence and strong observational data linking hyperglycemia to poor patient outcomes in the non‐ICU setting,27 the American Diabetes Association (ADA) now advocates good metabolic control, defined as preprandial glucose levels of 90‐130 mg/dL and peak postprandial glucose levels < 180 mg/dL in hospitalized non‐ICU patients with hyperglycemia14 (note that these targets are less aggressive than those for ICU patients, for whom randomized controlled trials showed the benefits of reduced mortality provided by tight glucose control).11 To reach these targets, the ADA and American College of Endocrinology suggest that multidisciplinary teams develop and implement hyperglycemia management guidelines and protocols.15 Protocols should promote the use of continuous intravenous insulin or scheduled subcutaneous insulin as opposed to the use of sliding‐scale insulin alone. Subcutaneous insulin protocols should include target glucose levels; basal, nutritional, and supplemental insulin; and daily adjustments based on previous glucose levels, insulin sensitivity, nutritional intake, illness, and medications.6, 15 To date, few published protocols or algorithms for inpatient subcutaneous insulin have been shown to be effective.16, 17 It is therefore not known how best to design and implement an inpatient diabetes management protocol that is effective, efficient, and self‐perpetuating. The aims of our pilot study were to develop and implement a subcutaneous insulin protocol on a general medicine service, to identify barriers to implementation, and to determine the effect of this protocol on glycemic control.

METHODS

RESULTS

After screening all 785 admissions to the 2 medical teams during the study period, we prospectively identified 109 patients (14%) for the pilot study. Twenty patients were subsequently excluded: 7 patients who were discharged the same day they were identified, 4 who did not have a fasting blood glucose value greater than 140 mg/dL, 4 patients who had type 1 diabetes, 2 patients who were admitted with diabetic ketoacidosis, and 3 patients whose data could not be accessed because of repeated unavailability of the medical record. Characteristics of the remaining 89 study subjects are shown in Table 2 and are compared to 91 baseline subjects. The mean age of the study subjects was 68.7 years; 45% were men. Five patients (6%) did not have a previous diagnosis of diabetes, and 51% were taking insulin prior to admission; the median A1C was 6.8%.

The medical residents agreed, at least in theory, to follow the subcutaneous insulin protocol for 50 patients (56%), partially accepted it for 8 (9%), and declined for 31 (35%). Reasons for declining the protocol included fear of hypoglycemia, severity of patient's other disease states or overall poor health of patient, concern for the effects of renal insufficiency on insulin clearance, concern for the effect of steroid tapers on glucose levels, desire to titrate oral medications, and anticipation of patient's imminent discharge. Other reasons such as the glucose levels are not that bad and let's watch the glucose levels for one more day suggest that some residents did not view hyperglycemia as an acute problem requiring immediate attention.

Regarding insulin‐ordering practices (Table 3), basal insulin was prescribed for 57 patients (64%) in the pilot group compared to 45 patients (49%) in the baseline group (P = .05). Nutritional insulin was prescribed to 12 patients (13%) in the pilot group compared to no patients in the baseline group (P < .001). Oral hypoglycemic agents were prescribed less often in the pilot study than at baseline (20% vs. 38%, P = .01). The use of a standard default sliding scale from the hospital computer order set was high and was not significantly different in the pilot study compared with that at baseline (93% vs. 90%, P = .78). Twenty‐four of the 83 patients in the pilot group (29%) received sliding‐scale insulin without ever receiving basal or nutritional insulin during hospitalization compared to 45 of 91 patients in the baseline group (49%; P = .01 for comparison). Among patients started on basal insulin, 42% (24 of 57) were started after the first full hospital day. The initial basal insulin dose was appropriate according to the protocol (within 20%) in 38 of 57 patients (67%). Only 20 of 61 patients (33%) who had any hypo‐ or hyperglycemia had any change to their insulin regimen made during days 2 through 7 of their hospitalization on GMS, similar to the rate noted at baseline (36%).

Regarding glucose control (Table 3), the mean percentage of glucose readings per patient greater than 180 mg/dL was not significantly different in the pilot study compared to baseline (31.6% vs. 33.3%, P = .85). Despite implementation of the protocol and increased use of basal and nutritional insulin, 76% of patients had at least 1 routine glucose reading greater than 180 mg/dL, and 37% of patients had at least 40% of their routine glucose readings greater than 180 mg/dL, comparable to baseline (74% and 38%, respectively, P = NS for both comparisons). At least 1 hypoglycemic event (glucose reading below 60 mg/dL) occurred in 7% of patients at baseline and 13% during the pilot study (P = .20). Eleven hypoglycemic events in the pilot study were between 50 and 59 mg/dL (55%), 6 were between 40 and 49 mg/dL (30%), 3 were between 30 and 39 mg/dL (15%), and none were less than 30 mg/dL. Nine occurred before breakfast (45%), 5 before dinner (25%), 3 before lunch (15%), and 3 at bedtime (15%).

During the pilot study, the use of basal insulin did improve over the first 5 days of hospitalization (Fig. 1), in both the percentage of patients prescribed any basal insulin and the percentage of each patient's total insulin dose (basal, nutritional, and supplemental) given as basal (both P < .001 for trend). Hyperglycemia rates also improved during hospitalization (Fig. 1), decreasing from 48% on hospital day 1 to 34% on hospital day 5 (P = .004 for trend). These trends were not observed in the baseline group, with hyperglycemia rates of 37% on hospital day 1 and 34% on hospital day 5 (P = .16 for trend).

Figure 1

Patients for whom the resident accepted or partially accepted the protocol had higher use of basal insulin (91% vs. 13%, P < .0001), higher use of nutritional insulin (21% vs. 0%, P = .01), and more frequent dose adjustments (47% vs. 7%, P = .01) compared with patients for whom the resident declined the protocol. However, the rate of hyperglycemia was higher in patients for whom the protocol was accepted or partially accepted than in patients for whom the protocol was declined (37% vs. 20%, P = .02).

DISCUSSION

Our subcutaneous insulin protocol focused on increasing the use of basal and nutritional insulin, avoiding the use of sliding‐scale insulin by itself, and performing daily insulin adjustments in response to the hypo‐ or hyperglycemia of general medical inpatients with diabetes or hyperglycemia.

The most notable finding of our pilot study was that residents were resistant to using the protocol, both in general and in its specific recommendations. Despite receiving education about inpatient diabetes control and protocol recommendations from the team pharmacist, and despite being on a hospitalist‐run medical service, the residents accepted use of the protocol for only half the eligible patients. Patients who were started on basal insulin were often underdosed or started after the first day of hospitalization, and daily dose adjustments were not consistently made despite persistent hypo‐ or hyperglycemia. Although the use of nutritional insulin was greater compared with that in the baseline group, it was still only prescribed for 13% of patients. Use of a standard sliding scale from the hospital computer order set was common in the pilot study and similar to that in the baseline group. These results suggest significant resistance to changing the current standard of practice.

Despite this lack of adherence to the protocol, some modest improvements in processes of care were seen. Basal insulin was ordered more often during the pilot study than at baseline, especially over the course of a hospital stay. Nutritional insulin was also ordered more often during the pilot study than at baseline, but was still infrequent. Oral antihyperglycemic agents were ordered less often during the pilot study than at baseline. This demonstrates that use of the protocol may be able to improve process outcomes. However, the modest improvements in process outcomes could have simply been a result of increased awareness and education, not the protocol itself.

Regarding patient outcomes, the overall hyperglycemia rate did not improve in the pilot study relative to that at baseline. Importantly, hypoglycemia rates did not increase significantly compared with those at baseline. However, because of the small number of hypoglycemia events, the sample size may not have been sufficient to detect a true difference between groups.

The most likely reason that the protocol did not show an effect on glycemic control was that its recommendations were not adhered to. In turn, this may have been a result of incomplete education, training, and implementation measures and/or inherent problems with the protocol that made its recommendations difficult to follow. Another possibility is that the protocol itself may not have been capable of improving glucose control, even when properly used. However, we do know that resident agreement to use the protocol did lead to higher rates of recommended best practices being carried out, such as basal insulin use and daily insulin dose adjustments, and that use of the protocol was associated with improvements in glucose control over the hospital stay. A larger study with a higher degree of protocol adherence would be better able to evaluate the merits of the protocol itself, as would a randomized controlled trial using instrumental variables to measure treatment efficacy. Another possibility explanation for the lack of effect is that glucose control on admission happened to be worse in the pilot group than in the control group: rates of hyperglycemia on day 1 were 48% in the pilot group compared with 37% in the baseline group (Fig. 1). Also, the decreased use of oral agents in the pilot group, a purposeful change to decrease the risk of hypoglycemia, may have counteracted the beneficial effects of more appropriate insulin use. Finally, there were few patients with poorly controlled diabetes at baseline (18 patients with A1C 8.0 in the baseline group and 12 such patients in the pilot group), arguably those most likely to benefit.

There is a pressing need to identify protocols that can improve glucose control in the non‐ICU inpatient setting and successfully implement these protocols with a minimum of resources and effort. To date, most studies that have improved glucose control in the non‐ICU setting have relied on frequent input from diabetologists or nurse‐practitioners.14, 15

The results of this study should be viewed in light of its limitations, including its relatively small sample size (thus limiting our ability to detect possible significant differences between groups) and that it was conducted at a single institution (thus limiting its generalizability). Patients were enrolled on weekdays, so patients admitted and discharged over a weekend or on a holiday may have been missed. Also, because of the nonrandomized design of the study, we cannot exclude the possibility that the improvements noted in the pilot study were a result of the increased education provided or of increased awareness and general improvement in diabetes management over the course of the study. Finally, implementation of the protocol was somewhat labor intensive and required staff support that could be difficult to replicate in other institutions. However, most of the study staff's effort was necessary either to implement the protocol in the absence of an order set or to evaluate barriers to implementation. Widespread implementation of a protocol with an order set, education, and the use of highly reliable tools should be possible with much less effort and resources. The strengths of this study include its prospective data collection methods, which included rigorous inclusion criteria and collection of detailed clinical data.

Our study findings suggest several approaches to improve care in the future. To combat resistance to change, the American Association of Clinical Endocrinologists strongly recommends that each institution ensure that all its clinicians involved agree about general philosophies of diabetes management.19 A more expansive, hospital‐wide educational and promotional plan may increase the initial acceptance of the protocol. Interviews with residents also indicated there was unfamiliarity with diabetes management and significant concerns about the harmful affects of tight glucose control (ie, risk of hypoglycemia), especially in certain patient subgroups. These results confirmed the need for more practical individualized training and sparked the implementation of small‐group, case‐based educational sessions on inpatient diabetes management for all house officers, with a particular focus on patients with multiple comorbidities, on steroid tapers, and/or with renal failure.

The lack of nutritional insulin orders, delays in ordering basal insulin, and use of inadequate doses of insulin may be counteracted by the use of an order set, in our case built into our computer physician order entry (CPOE) system. The use of CPOE also allows reminders to be automatically sent to clinicians if eligible patients are not started on these orders. Clinical inertia (eg, failure to adjust the insulin doses of specific patients despite hyperglycemia) is more difficult to combat but may be addressed through better organization of clinical data, individualized, case‐based education, and CPOE reminders and eventually through culture change.

As a result of our pilot study, additional revisions were made to the protocol in hopes of increasing protocol adherence. For example, for patients eating discrete meals who are not taking insulin at home, the pilot protocol had suggested a starting insulin dose range for basal and nutritional insulin that required 2 separate calculations. The revised protocol was simplified to recommend a total daily insulin dose to be split evenly between basal and nutritional insulin. The daily adjustment instructions were also simplified. The pilot protocol had included a complicated table of adjustment recommendations based on bedside glucose trends. The revised protocol recommends adjusting the new daily dose by adding the total units of insulin given the previous day (including supplemental doses), making minor adjustments for hyper‐ or hypoglycemia and other clinical factors (like renal failure), and splitting this dose evenly between scheduled basal and nutritional insulin. In addition, 3 order sets were built into our computerized physician order entry system to facilitate early and appropriate insulin orders for patients with different diets (discrete meals, continuous tube feeds, and nothing by mouth); 3 different insulin sliding scales were created for patients with different degrees of insulin resistance; a diabetes management page for our electronic medication administration record is being developed to better organize clinical data; and hospital‐wide education and individualized training are ongoing.

In conclusion, the adherence to an inpatient glycemic management protocol that focused on increasing use of basal insulin and performing daily insulin adjustments was only fair. Barriers to successful implementation included clinical inertia regarding individual patients, unfamiliarity with inpatient diabetes management strategies, fear of hypoglycemia, and resistance to changing the current standard of practice. Targeted education, standard order sets, better organization of clinical data, protocol simplification, and institutional culture changes may be necessary for successful protocol implementation and improved inpatient glucose control.

Diabetes mellitus is a common comorbid condition in hospitalized patients. In 2003, diabetes was listed as a diagnosis in 17.2% of hospital discharges in the United States.1 Because these diagnosis codes do not account for undiagnosed diabetes or hospital‐related hyperglycemia, the true prevalence of diabetes or hyperglycemia in hospitalized patients is likely higher and has been estimated to be as great as 38%.2 Hyperglycemia has been associated with adverse outcomes among hospitalized patients, including infectious complications, increased length of stay, and increased mortality.27 However, because hyperglycemia is not usually the primary reason patients with diabetes are hospitalized, its management is often not a focus in the inpatient setting. Sliding‐scale insulin alone continues to be commonly prescribed despite clinical evidence showing it to be ineffective in achieving glycemic control.8, 9

Recent randomized controlled trials have demonstrated that aggressive treatment of inpatient hyperglycemia improves outcomes in surgical and medical intensive care units10, 11 and in patients admitted for myocardial infarction.12, 13 Based on this clinical evidence and strong observational data linking hyperglycemia to poor patient outcomes in the non‐ICU setting,27 the American Diabetes Association (ADA) now advocates good metabolic control, defined as preprandial glucose levels of 90‐130 mg/dL and peak postprandial glucose levels < 180 mg/dL in hospitalized non‐ICU patients with hyperglycemia14 (note that these targets are less aggressive than those for ICU patients, for whom randomized controlled trials showed the benefits of reduced mortality provided by tight glucose control).11 To reach these targets, the ADA and American College of Endocrinology suggest that multidisciplinary teams develop and implement hyperglycemia management guidelines and protocols.15 Protocols should promote the use of continuous intravenous insulin or scheduled subcutaneous insulin as opposed to the use of sliding‐scale insulin alone. Subcutaneous insulin protocols should include target glucose levels; basal, nutritional, and supplemental insulin; and daily adjustments based on previous glucose levels, insulin sensitivity, nutritional intake, illness, and medications.6, 15 To date, few published protocols or algorithms for inpatient subcutaneous insulin have been shown to be effective.16, 17 It is therefore not known how best to design and implement an inpatient diabetes management protocol that is effective, efficient, and self‐perpetuating. The aims of our pilot study were to develop and implement a subcutaneous insulin protocol on a general medicine service, to identify barriers to implementation, and to determine the effect of this protocol on glycemic control.

METHODS

RESULTS

After screening all 785 admissions to the 2 medical teams during the study period, we prospectively identified 109 patients (14%) for the pilot study. Twenty patients were subsequently excluded: 7 patients who were discharged the same day they were identified, 4 who did not have a fasting blood glucose value greater than 140 mg/dL, 4 patients who had type 1 diabetes, 2 patients who were admitted with diabetic ketoacidosis, and 3 patients whose data could not be accessed because of repeated unavailability of the medical record. Characteristics of the remaining 89 study subjects are shown in Table 2 and are compared to 91 baseline subjects. The mean age of the study subjects was 68.7 years; 45% were men. Five patients (6%) did not have a previous diagnosis of diabetes, and 51% were taking insulin prior to admission; the median A1C was 6.8%.

The medical residents agreed, at least in theory, to follow the subcutaneous insulin protocol for 50 patients (56%), partially accepted it for 8 (9%), and declined for 31 (35%). Reasons for declining the protocol included fear of hypoglycemia, severity of patient's other disease states or overall poor health of patient, concern for the effects of renal insufficiency on insulin clearance, concern for the effect of steroid tapers on glucose levels, desire to titrate oral medications, and anticipation of patient's imminent discharge. Other reasons such as the glucose levels are not that bad and let's watch the glucose levels for one more day suggest that some residents did not view hyperglycemia as an acute problem requiring immediate attention.

Regarding insulin‐ordering practices (Table 3), basal insulin was prescribed for 57 patients (64%) in the pilot group compared to 45 patients (49%) in the baseline group (P = .05). Nutritional insulin was prescribed to 12 patients (13%) in the pilot group compared to no patients in the baseline group (P < .001). Oral hypoglycemic agents were prescribed less often in the pilot study than at baseline (20% vs. 38%, P = .01). The use of a standard default sliding scale from the hospital computer order set was high and was not significantly different in the pilot study compared with that at baseline (93% vs. 90%, P = .78). Twenty‐four of the 83 patients in the pilot group (29%) received sliding‐scale insulin without ever receiving basal or nutritional insulin during hospitalization compared to 45 of 91 patients in the baseline group (49%; P = .01 for comparison). Among patients started on basal insulin, 42% (24 of 57) were started after the first full hospital day. The initial basal insulin dose was appropriate according to the protocol (within 20%) in 38 of 57 patients (67%). Only 20 of 61 patients (33%) who had any hypo‐ or hyperglycemia had any change to their insulin regimen made during days 2 through 7 of their hospitalization on GMS, similar to the rate noted at baseline (36%).

Regarding glucose control (Table 3), the mean percentage of glucose readings per patient greater than 180 mg/dL was not significantly different in the pilot study compared to baseline (31.6% vs. 33.3%, P = .85). Despite implementation of the protocol and increased use of basal and nutritional insulin, 76% of patients had at least 1 routine glucose reading greater than 180 mg/dL, and 37% of patients had at least 40% of their routine glucose readings greater than 180 mg/dL, comparable to baseline (74% and 38%, respectively, P = NS for both comparisons). At least 1 hypoglycemic event (glucose reading below 60 mg/dL) occurred in 7% of patients at baseline and 13% during the pilot study (P = .20). Eleven hypoglycemic events in the pilot study were between 50 and 59 mg/dL (55%), 6 were between 40 and 49 mg/dL (30%), 3 were between 30 and 39 mg/dL (15%), and none were less than 30 mg/dL. Nine occurred before breakfast (45%), 5 before dinner (25%), 3 before lunch (15%), and 3 at bedtime (15%).

During the pilot study, the use of basal insulin did improve over the first 5 days of hospitalization (Fig. 1), in both the percentage of patients prescribed any basal insulin and the percentage of each patient's total insulin dose (basal, nutritional, and supplemental) given as basal (both P < .001 for trend). Hyperglycemia rates also improved during hospitalization (Fig. 1), decreasing from 48% on hospital day 1 to 34% on hospital day 5 (P = .004 for trend). These trends were not observed in the baseline group, with hyperglycemia rates of 37% on hospital day 1 and 34% on hospital day 5 (P = .16 for trend).

Figure 1

Patients for whom the resident accepted or partially accepted the protocol had higher use of basal insulin (91% vs. 13%, P < .0001), higher use of nutritional insulin (21% vs. 0%, P = .01), and more frequent dose adjustments (47% vs. 7%, P = .01) compared with patients for whom the resident declined the protocol. However, the rate of hyperglycemia was higher in patients for whom the protocol was accepted or partially accepted than in patients for whom the protocol was declined (37% vs. 20%, P = .02).

DISCUSSION

Our subcutaneous insulin protocol focused on increasing the use of basal and nutritional insulin, avoiding the use of sliding‐scale insulin by itself, and performing daily insulin adjustments in response to the hypo‐ or hyperglycemia of general medical inpatients with diabetes or hyperglycemia.

The most notable finding of our pilot study was that residents were resistant to using the protocol, both in general and in its specific recommendations. Despite receiving education about inpatient diabetes control and protocol recommendations from the team pharmacist, and despite being on a hospitalist‐run medical service, the residents accepted use of the protocol for only half the eligible patients. Patients who were started on basal insulin were often underdosed or started after the first day of hospitalization, and daily dose adjustments were not consistently made despite persistent hypo‐ or hyperglycemia. Although the use of nutritional insulin was greater compared with that in the baseline group, it was still only prescribed for 13% of patients. Use of a standard sliding scale from the hospital computer order set was common in the pilot study and similar to that in the baseline group. These results suggest significant resistance to changing the current standard of practice.

Despite this lack of adherence to the protocol, some modest improvements in processes of care were seen. Basal insulin was ordered more often during the pilot study than at baseline, especially over the course of a hospital stay. Nutritional insulin was also ordered more often during the pilot study than at baseline, but was still infrequent. Oral antihyperglycemic agents were ordered less often during the pilot study than at baseline. This demonstrates that use of the protocol may be able to improve process outcomes. However, the modest improvements in process outcomes could have simply been a result of increased awareness and education, not the protocol itself.

Regarding patient outcomes, the overall hyperglycemia rate did not improve in the pilot study relative to that at baseline. Importantly, hypoglycemia rates did not increase significantly compared with those at baseline. However, because of the small number of hypoglycemia events, the sample size may not have been sufficient to detect a true difference between groups.

The most likely reason that the protocol did not show an effect on glycemic control was that its recommendations were not adhered to. In turn, this may have been a result of incomplete education, training, and implementation measures and/or inherent problems with the protocol that made its recommendations difficult to follow. Another possibility is that the protocol itself may not have been capable of improving glucose control, even when properly used. However, we do know that resident agreement to use the protocol did lead to higher rates of recommended best practices being carried out, such as basal insulin use and daily insulin dose adjustments, and that use of the protocol was associated with improvements in glucose control over the hospital stay. A larger study with a higher degree of protocol adherence would be better able to evaluate the merits of the protocol itself, as would a randomized controlled trial using instrumental variables to measure treatment efficacy. Another possibility explanation for the lack of effect is that glucose control on admission happened to be worse in the pilot group than in the control group: rates of hyperglycemia on day 1 were 48% in the pilot group compared with 37% in the baseline group (Fig. 1). Also, the decreased use of oral agents in the pilot group, a purposeful change to decrease the risk of hypoglycemia, may have counteracted the beneficial effects of more appropriate insulin use. Finally, there were few patients with poorly controlled diabetes at baseline (18 patients with A1C 8.0 in the baseline group and 12 such patients in the pilot group), arguably those most likely to benefit.

There is a pressing need to identify protocols that can improve glucose control in the non‐ICU inpatient setting and successfully implement these protocols with a minimum of resources and effort. To date, most studies that have improved glucose control in the non‐ICU setting have relied on frequent input from diabetologists or nurse‐practitioners.14, 15

The results of this study should be viewed in light of its limitations, including its relatively small sample size (thus limiting our ability to detect possible significant differences between groups) and that it was conducted at a single institution (thus limiting its generalizability). Patients were enrolled on weekdays, so patients admitted and discharged over a weekend or on a holiday may have been missed. Also, because of the nonrandomized design of the study, we cannot exclude the possibility that the improvements noted in the pilot study were a result of the increased education provided or of increased awareness and general improvement in diabetes management over the course of the study. Finally, implementation of the protocol was somewhat labor intensive and required staff support that could be difficult to replicate in other institutions. However, most of the study staff's effort was necessary either to implement the protocol in the absence of an order set or to evaluate barriers to implementation. Widespread implementation of a protocol with an order set, education, and the use of highly reliable tools should be possible with much less effort and resources. The strengths of this study include its prospective data collection methods, which included rigorous inclusion criteria and collection of detailed clinical data.

Our study findings suggest several approaches to improve care in the future. To combat resistance to change, the American Association of Clinical Endocrinologists strongly recommends that each institution ensure that all its clinicians involved agree about general philosophies of diabetes management.19 A more expansive, hospital‐wide educational and promotional plan may increase the initial acceptance of the protocol. Interviews with residents also indicated there was unfamiliarity with diabetes management and significant concerns about the harmful affects of tight glucose control (ie, risk of hypoglycemia), especially in certain patient subgroups. These results confirmed the need for more practical individualized training and sparked the implementation of small‐group, case‐based educational sessions on inpatient diabetes management for all house officers, with a particular focus on patients with multiple comorbidities, on steroid tapers, and/or with renal failure.

The lack of nutritional insulin orders, delays in ordering basal insulin, and use of inadequate doses of insulin may be counteracted by the use of an order set, in our case built into our computer physician order entry (CPOE) system. The use of CPOE also allows reminders to be automatically sent to clinicians if eligible patients are not started on these orders. Clinical inertia (eg, failure to adjust the insulin doses of specific patients despite hyperglycemia) is more difficult to combat but may be addressed through better organization of clinical data, individualized, case‐based education, and CPOE reminders and eventually through culture change.

As a result of our pilot study, additional revisions were made to the protocol in hopes of increasing protocol adherence. For example, for patients eating discrete meals who are not taking insulin at home, the pilot protocol had suggested a starting insulin dose range for basal and nutritional insulin that required 2 separate calculations. The revised protocol was simplified to recommend a total daily insulin dose to be split evenly between basal and nutritional insulin. The daily adjustment instructions were also simplified. The pilot protocol had included a complicated table of adjustment recommendations based on bedside glucose trends. The revised protocol recommends adjusting the new daily dose by adding the total units of insulin given the previous day (including supplemental doses), making minor adjustments for hyper‐ or hypoglycemia and other clinical factors (like renal failure), and splitting this dose evenly between scheduled basal and nutritional insulin. In addition, 3 order sets were built into our computerized physician order entry system to facilitate early and appropriate insulin orders for patients with different diets (discrete meals, continuous tube feeds, and nothing by mouth); 3 different insulin sliding scales were created for patients with different degrees of insulin resistance; a diabetes management page for our electronic medication administration record is being developed to better organize clinical data; and hospital‐wide education and individualized training are ongoing.

In conclusion, the adherence to an inpatient glycemic management protocol that focused on increasing use of basal insulin and performing daily insulin adjustments was only fair. Barriers to successful implementation included clinical inertia regarding individual patients, unfamiliarity with inpatient diabetes management strategies, fear of hypoglycemia, and resistance to changing the current standard of practice. Targeted education, standard order sets, better organization of clinical data, protocol simplification, and institutional culture changes may be necessary for successful protocol implementation and improved inpatient glucose control.

In a supplement to its inaugural issue, the Journal of Hospital Medicine published core competencies for hospitalists covering 3 areas: clinical conditions, systems in health care, and procedures.1 Completion of a traditional internal medicine residency may not provide hospitalists with the skills necessary to safely perform necessary procedures such as thoracentesis. A recent article reported that most internal medicine residents surveyed were uncomfortable performing common procedures, and their discomfort was higher for thoracentesis than for central line insertion, lumbar puncture, or paracentesis.2 This confirmed a previous report that family practice residents had low confidence in performing thoracenteses.3 Thoracentesis also carries the risk of the potentially life‐threatening complication of pneumothorax, which may be increased when performed by physicians‐in‐training.4

One method for improving training and assessment is the use of simulation technology. Simulation has been used to increase knowledge, provide opportunities for deliberate and safe practice, and shape the development of clinical skills.5, 6 Simulation has also been advocated for assessing competence in procedures including carotid angiography,7 emergency airway management,8 basic bronchoscopy,9 and advanced cardiac life support (ACLS).10, 11

Recently, we used simulation technology to help residents reach mastery learning standards for ACLS.11 Mastery learning,12 an extreme form of competency‐based education,13 implies that learners have acquired the clinical knowledge and skill measured against rigorous achievement standards. In mastery learning, educational results are equivalent, whereas educational practice time differs. To demonstrate mastery learning, we first documented a 38% improvement in skill after a simulation‐based educational intervention10 and used a multidisciplinary panel to determine mastery achievement standards for ACLS skills in 6 clinical scenarios.14 These standards were used in a study in which the amount of time needed to achieve skill mastery was allowed to vary while the skill outcomes of the residents were identical clinically.11

The present study had 4 aims. The first was to assess the baseline skill and knowledge of third‐year residents in thoracentesis. The second was to compare the thoracentesis‐related knowledge and skills of residents before and after an educational intervention. The third was to assess the correlation of medical knowledge and clinical experience with performance on a clinical skills examination after simulation training. The last was to document the feasibility of incorporating simulation‐based education into a training program.

METHODS

RESULTS

All residents consented to participate and completed the entire training protocol. Table 1 presents demographic data about the residents. Most had limited experience performing and supervising thoracenteses.