Glycemic control in the inpatient setting has received increasing attention in recent years, with the demonstration that appropriate blood glucose (BG) control prevents adverse events in both intensive care unit (ICU) and non‐ICU settings.1 Recent recommendations set target blood glucose levels near euglycemia for most hospitalized patients.1 Unfortunately, the risk of hypoglycemia increases with tighter glycemic control,2 and hypoglycemia may result in catastrophic events.35 Although hyperglycemia is associated with postoperative infection,6 and effective management decreases wound infections,7 few reports have detailed the hypoglycemia rates among surgical patients.8 Hypoglycemia rates on medical services are as high as 28%,9, 10 and efforts to achieve more normal BG levels in hospitalized patients have been associated with more hypoglycemia.11

We undertook a study of hypoglycemia in all adult hospitalized patients receiving hypoglycemic therapy at our institution. The purpose of this study was to determine the incidence, natural history, associations, and consequences of hypoglycemia in this broad inpatient population in order to have a baseline prior to introducing any formal hospital strategies to achieve the newer targets for glycemic control.

Research Design and Methods

Thomas Jefferson University Hospital (TJUH) is a 675‐bed acute care teaching institution in center‐city Philadelphia with more than 30,000 patient admissions each year. We undertook a prospective, consecutive medical record review from August 16, 2004, to November 15, 2004, of hospitalized patients who had experienced at least 1 hypoglycemic episode, defined as at least one blood glucose (BG) 60 mg/dL within 48 hours of administration of an antihyperglycemic agent in the hospital. The definition of hypoglycemia was consistent with our hospital policies and a compromise between the BG 70 mg/dL proposed by the American Diabetes Association (ADA) hypoglycemia workgroup12 and the BG 40 mg/dL used by authors studying glycemic control in the ICU.13, 14

Hypoglycemic episodes were identified by a daily electronic search of the online medication administration record (MAR) where nurses document all point‐of‐care (POC) BG values. Two of the authors (P.V. and V.G.) reviewed the medical record for each episode and excluded pediatric (<18 years), emergency department, and maternity patients. Intensive care, step‐down, and medical/surgical unit patients were all included if the hypoglycemia had occurred within 48 hours of hospital administration of an antihyperglycemic agent. All medication orders at TJUH are placed through the computerized prescriber order entry (CPOE) system (Centricity Enterprise), which links all antihyperglycemic agents to a standardized hypoglycemia treatment protocol. The protocol includes instructions to administer glucose and/or glucagon and check the BG 15 minutes after a hypoglycemic episode. We established operational definitions prior to chart review (Appendix). A symptomatic hypoglycemia‐related adverse event was defined as any documented event occurring at the time of the hypoglycemic episode involving symptoms, change in care, temporary or permanent injury, or increased length of hospitalization. We did not include following our hypoglycemic protocol with the administration of 50% dextrose or glucagon as a change of care, as we considered this usual care, unless symptoms or signs also accompanied the hypoglycemic event.

We searched the University Health System Consortium Clinical Database (CDB) to quantify the number of patients at TJUH receiving any antihyperglycemic agent during the study period and to identify the specific agents received. The CDB receives all patient, physician, and pharmacy dose‐ specific information from the hospital clinical and billing information systems. We defined subgroups of patients taking insulin(s) only, taking oral agent(s) only, and taking a combination.

Differences between proportions were evaluated using the chi‐square statistic; differences between means were evaluated using the Student t test. Probabilities of the null hypothesis less than .05 were considered significant.

The project was approved by the Institutional Review Board for Human Subjects at Thomas Jefferson University.

RESULTS

Over the 2‐month study period 8140 patients were admitted, of whom 2174 (27%) received an antihyperglycemic agent. Five hundred and sixty‐eight hypoglycemic episodes (BG 60 mg/dL) occurred in 265 patients. We excluded 84 episodes among 59 patients who did not receive antihyperglycemic agents, resulting in 484 episodes of hypoglycemia occurring within 48 hours of hospital administration of an antihyperglycemic agent in 206 patients, an average of 5.26 episodes per day. Of the 2174 of patients receiving antihyperglycemic agents, 206 (9.5%) experienced 1 or more episodes of hypoglycemia.

Patient ages ranged from 20 to 93 years, with an average of 62 years. Fifty‐seven percent (118 of 206) of participants were female. About one‐fourth of all episodes (23.8%) occurred in the ICU setting. The distribution of patients by decade and their ICU status are presented in Figure 1. Of the 206 patients, 29% (59) had type 1 diabetes mellitus (DM), 49% (102) had type 2 DM, 1% (2) had new‐onset diabetes, and 21% (43) had no diagnosis of DM. Of the 484 hypoglycemic episodes, 37.8% occurred in patients with type 1 DM, 46.9% in patients with type 2 DM, and 0.6% in patients with new‐onset DM. The remaining 14.5% occurred in patients with no documented history of DM, although they were receiving antihyperglycemic agents. More than 1 episode was experienced by 44% of patients, and 12% experienced 5 or more episodes.

Figure 1

The BG was between 51 and 60 mg/dL in 282 of the episodes (58.2%), between 41 and 50 mg/dL in 149 episodes (30.8%) and 40 mg/dL or less in 53 episodes (11%). In 20 episodes (4.1% of episodes, representing fewer than 1% of all patients receiving an antihyperglycemic agent), a symptomatic hypoglycemia‐related adverse event was documented. All but 1 adverse event occurred outside the ICU. Ten of these events (2.1% of all hypoglycemic episodes) in 10 patients involved symptoms including headache, agitation, disorientation, and tremors. Of these patients 9 had type 1 DM, and 1 had type 2 DM. Six events (1.2% of hypoglycemic episodes) in 4 patients involved seizures. Two of these patients had type 1 DM, and 2 had type 2 DM. Four events (0.8% of hypoglycemic episodes) in 4 patients involved an unresponsive or unarousable state, including the sole ICU episode of symptomatic hypoglycemia. Three of these patients had type 1 DM, and 1 had type 2 DM. Patients with hypoglycemia‐related adverse events had a mean BG of 43.0 mg/dL, significantly lower (P = .01) than the mean BG of 50.9 mg/dL for hypoglycemic episodes without such events. However, 35% of these events occurred with a measured BG between 50 and 60 mg/dL. The distributions of BG values associated with symptomatic and asymptomatic events are shown in Figure 2. There is no useful threshold that separates symptomatic from asymptomatic hypoglycemia. No deaths or irreversible consequences were associated with hypoglycemia.

Figure 2

Approximately 40% (195 of 484) of the hypoglycemic episodes were related to decreased enteral intake (Table 1). In addition, 6.1% (30 of 484) of hypoglycemic episodes were related to insulin adjustment and 0.4% (2 of 484) to steroid withdrawal. In 43% (209 of 484) of the episodes the cause of the hypoglycemia was unclear. The remaining 10.4% of episodes were attributed to diverse causes.

One third of patients had a documented BG rechecked within 60 minutes, and fewer than half of the hypoglycemic patients had documented euglycemia within 2 hours of their low blood glucose measurement. The average time to documented resolution of a hypoglycemic episode was 4 hours, 3 minutes, with a median of 2 hours, 25 minutes.

Table 2 delineates the various combinations of antihyperglycemic agents that the 206 patients received in the 48 hours prior to a hypoglycemic episode. Of the 484 hypoglycemic episodes, 362 involved insulin. Of patients receiving insulin, 38 of 362 of episodes of hypoglycemia occurred in patients receiving sliding‐scale insulin (SSI) dosing as the only insulin order. In 163 hypoglycemic episodes, insulin was dosed with a combination of SSI and infusion or SSI with daily long‐acting insulin. The remaining 161 episodes involved administration of insulin to patients without an accompanying sliding‐scale order.

The prevalence of hypoglycemia did not significantly differ among patients treated with oral agents alone (9 of 85, 10.6%), patients treated with insulin alone (149 of 1497, 10%), and patients treated with both (47 of 592, 7.9%). However, there was a significant relationship between specific oral agent and probability of hypoglycemia. Glyburide was associated with a higher risk of hypoglycemia (19.1%, P < .01) than were other oral agents (Table 3).

DISCUSSION

Recently, many have called for substantive changes in the management of the hospitalized diabetic.15, 16 Most have recommended replacing sliding‐scale insulin with basal bolus insulin dosing and have challenged the historic tolerance of hyperglycemia during an acute hospital stay.17 However, as hospitals and physicians transform the management of inpatient hyperglycemia, they must assess the frequency of hypoglycemia and evaluate the risk/benefit ratio of strict glycemic control.18 Thus, one study found that eliminating sliding‐scale insulin markedly improved diabetes control but hypoglycemia (BG 60 mg/dL) was more frequent.10 The cost of euglycemia is hypoglycemia.2

We report a 9.5% rate of hypoglycemia among adult hospitalized patients being treated for hyperglycemia, including those in the ICU and those in non‐ICU settings. In widely publicized landmark trials, 5.2% of intensively treated surgical ICU patients14 and 18.7% of intensively treated medical ICU patients13 experienced hypoglycemia with no adverse events. Using those studies' definition of hypoglycemia (BG 40 mg/dL), only 2.4% (53 of 2174) of our patients experienced hypoglycemia. However, our survey included general medical and surgical patients as well as ICU patients treated at their physicians' discretion, reflecting the greater variability in care that exists outside a randomized, ICU trial.

We did not anticipate the duration to documented resolution of hypoglycemic episodes, nor did we anticipate the number of hypoglycemia‐related adverse events. We believe that hospitals will need to develop formal strategies to minimize the hypoglycemic risk from tight glycemic control. The frequency and duration of the time it took to recheck the glucose, coupled with the 4.1% symptomatic event rate, suggests that inpatient hypoglycemia deserves more attention. One potential focus is the interruption of nutrition, as medications may not be readjusted when patients' oral intake declines or when they travel for tests.19

More than 40% of our hypoglycemic patients experienced recurrent episodes. This may reflect a lack of adjustment of medications following hypoglycemia. However, recurrent hypoglycemia may also be explained by hypoglycemia‐associated autonomic failure and the desensitization to hypoglycemia that occurs once a patient has lower blood glucose.2 Thus, hypoglycemic patients are at high risk of repeat episodes and often require more frequent BG monitoring. Of note, patients with hypoglycemia unawareness may not have symptoms despite low BG, and thus unless they develop signs of hypoglycemia, they would not meet criteria for an adverse event in our study, despite a very low BG.

Medical error can precipitate hypoglycemia,4, 20 and the Institute for Safe Medication Practices21 and the Joint Commission on Accreditation of Healthcare Organizations22 consider insulin a high‐risk medication. We found no hypoglycemic episodes associated with a medication error. Our CPOE system eliminates ambiguity from poor penmanship, and hospital policy requires 2 nurses to check all administered insulin. However, despite the apparent lack of dispensing/administration medication errors, nearly 10% of patients receiving hypoglycemic therapy experienced iatrogenic hypoglycemia. Thus, strategies to reduce hypoglycemia must expand beyond the prevention of medication errors.

Contrary to our expectation, we found the prevalence of hypoglycemia was at least as high, if not higher, for patients taking only oral hypoglycemics than for patients taking either insulin alone or insulin in combination with oral antihyperglycemic agents. Glyburide appeared to carry the most risk both in our population and in previous studies,2325 perhaps because of a moderately active hepatic metabolite.26 The risk of hypoglycemia with different oral agents warrants further study.

The study stimulated several actions. First, we augmented the online nursing flow sheet to permit documentation of hypoglycemic episodes, including the administration of orange juice or food. Second, our CPOE system now prevents a physician from inadvertently deselecting the hypoglycemia protocol. Third, the CPOE system prompts the nurse to recheck the BG as specified in the hypoglycemia protocol. Finally, the CPOE system warns physicians to adjust antihyperglycemic agents when they institute nutritional changes. We propose that monitoring hypoglycemia rates must become a necessary component of inpatient diabetes care that is both effective and safe and plan to monitor these rates to determine the impact of interventions designed to reduce the frequency of hypoglycemia‐related adverse events.

Our study had several limitations. We only included episodes of hypoglycemia that were identified with a POC BG. This excluded patients treated for symptomatic hypoglycemia without a measured POC BG, potentially underestimating the event rate. Moreover, defining time to resolution of a hypoglycemic episode as that documented with a serum BG but not a POC BG may have resulted in overestimating the duration, and nurses may have documented POC BGs in the MAR after a substantial delay, also artificially lengthening the time to resolution. Capillary BG may underestimate the true degree of hypoglycemia,27 thus confounding the relationship between BG and adverse events. Our study was not designed to evaluate subtle suboptimal management of hyperglycemia as a cause of hypoglycemia, although expansion of the types and combinations of insulin has increased the possibility of prescribing errors. Nor were we able to assess the preventability of hypoglycemic episodes or the independent risk factors for hypoglycemia. Finally, this study originated from a single academic hospital and thus may reflect its unique idiosyncrasies.

We have reported a comprehensive survey of hypoglycemia in patients treated with antihyperglycemic agents at a single hospital. At the time the study took place, we had not instituted hospitalwide strategies to maintain BG near euglycemic targets, although such strategies have since begun. To detect untoward events that may follow from our efforts to better control hyperglycemia, we believe it is important to establish baseline measurements. Even without aiming for tighter glucose control, we identified the need to aim for and possible strategies to achieve, better prevention of hypoglycemia.

Glycemic control in the inpatient setting has received increasing attention in recent years, with the demonstration that appropriate blood glucose (BG) control prevents adverse events in both intensive care unit (ICU) and non‐ICU settings.1 Recent recommendations set target blood glucose levels near euglycemia for most hospitalized patients.1 Unfortunately, the risk of hypoglycemia increases with tighter glycemic control,2 and hypoglycemia may result in catastrophic events.35 Although hyperglycemia is associated with postoperative infection,6 and effective management decreases wound infections,7 few reports have detailed the hypoglycemia rates among surgical patients.8 Hypoglycemia rates on medical services are as high as 28%,9, 10 and efforts to achieve more normal BG levels in hospitalized patients have been associated with more hypoglycemia.11

We undertook a study of hypoglycemia in all adult hospitalized patients receiving hypoglycemic therapy at our institution. The purpose of this study was to determine the incidence, natural history, associations, and consequences of hypoglycemia in this broad inpatient population in order to have a baseline prior to introducing any formal hospital strategies to achieve the newer targets for glycemic control.

Research Design and Methods

Thomas Jefferson University Hospital (TJUH) is a 675‐bed acute care teaching institution in center‐city Philadelphia with more than 30,000 patient admissions each year. We undertook a prospective, consecutive medical record review from August 16, 2004, to November 15, 2004, of hospitalized patients who had experienced at least 1 hypoglycemic episode, defined as at least one blood glucose (BG) 60 mg/dL within 48 hours of administration of an antihyperglycemic agent in the hospital. The definition of hypoglycemia was consistent with our hospital policies and a compromise between the BG 70 mg/dL proposed by the American Diabetes Association (ADA) hypoglycemia workgroup12 and the BG 40 mg/dL used by authors studying glycemic control in the ICU.13, 14

Hypoglycemic episodes were identified by a daily electronic search of the online medication administration record (MAR) where nurses document all point‐of‐care (POC) BG values. Two of the authors (P.V. and V.G.) reviewed the medical record for each episode and excluded pediatric (<18 years), emergency department, and maternity patients. Intensive care, step‐down, and medical/surgical unit patients were all included if the hypoglycemia had occurred within 48 hours of hospital administration of an antihyperglycemic agent. All medication orders at TJUH are placed through the computerized prescriber order entry (CPOE) system (Centricity Enterprise), which links all antihyperglycemic agents to a standardized hypoglycemia treatment protocol. The protocol includes instructions to administer glucose and/or glucagon and check the BG 15 minutes after a hypoglycemic episode. We established operational definitions prior to chart review (Appendix). A symptomatic hypoglycemia‐related adverse event was defined as any documented event occurring at the time of the hypoglycemic episode involving symptoms, change in care, temporary or permanent injury, or increased length of hospitalization. We did not include following our hypoglycemic protocol with the administration of 50% dextrose or glucagon as a change of care, as we considered this usual care, unless symptoms or signs also accompanied the hypoglycemic event.

We searched the University Health System Consortium Clinical Database (CDB) to quantify the number of patients at TJUH receiving any antihyperglycemic agent during the study period and to identify the specific agents received. The CDB receives all patient, physician, and pharmacy dose‐ specific information from the hospital clinical and billing information systems. We defined subgroups of patients taking insulin(s) only, taking oral agent(s) only, and taking a combination.

Differences between proportions were evaluated using the chi‐square statistic; differences between means were evaluated using the Student t test. Probabilities of the null hypothesis less than .05 were considered significant.

The project was approved by the Institutional Review Board for Human Subjects at Thomas Jefferson University.

RESULTS

Over the 2‐month study period 8140 patients were admitted, of whom 2174 (27%) received an antihyperglycemic agent. Five hundred and sixty‐eight hypoglycemic episodes (BG 60 mg/dL) occurred in 265 patients. We excluded 84 episodes among 59 patients who did not receive antihyperglycemic agents, resulting in 484 episodes of hypoglycemia occurring within 48 hours of hospital administration of an antihyperglycemic agent in 206 patients, an average of 5.26 episodes per day. Of the 2174 of patients receiving antihyperglycemic agents, 206 (9.5%) experienced 1 or more episodes of hypoglycemia.

Patient ages ranged from 20 to 93 years, with an average of 62 years. Fifty‐seven percent (118 of 206) of participants were female. About one‐fourth of all episodes (23.8%) occurred in the ICU setting. The distribution of patients by decade and their ICU status are presented in Figure 1. Of the 206 patients, 29% (59) had type 1 diabetes mellitus (DM), 49% (102) had type 2 DM, 1% (2) had new‐onset diabetes, and 21% (43) had no diagnosis of DM. Of the 484 hypoglycemic episodes, 37.8% occurred in patients with type 1 DM, 46.9% in patients with type 2 DM, and 0.6% in patients with new‐onset DM. The remaining 14.5% occurred in patients with no documented history of DM, although they were receiving antihyperglycemic agents. More than 1 episode was experienced by 44% of patients, and 12% experienced 5 or more episodes.

Figure 1

The BG was between 51 and 60 mg/dL in 282 of the episodes (58.2%), between 41 and 50 mg/dL in 149 episodes (30.8%) and 40 mg/dL or less in 53 episodes (11%). In 20 episodes (4.1% of episodes, representing fewer than 1% of all patients receiving an antihyperglycemic agent), a symptomatic hypoglycemia‐related adverse event was documented. All but 1 adverse event occurred outside the ICU. Ten of these events (2.1% of all hypoglycemic episodes) in 10 patients involved symptoms including headache, agitation, disorientation, and tremors. Of these patients 9 had type 1 DM, and 1 had type 2 DM. Six events (1.2% of hypoglycemic episodes) in 4 patients involved seizures. Two of these patients had type 1 DM, and 2 had type 2 DM. Four events (0.8% of hypoglycemic episodes) in 4 patients involved an unresponsive or unarousable state, including the sole ICU episode of symptomatic hypoglycemia. Three of these patients had type 1 DM, and 1 had type 2 DM. Patients with hypoglycemia‐related adverse events had a mean BG of 43.0 mg/dL, significantly lower (P = .01) than the mean BG of 50.9 mg/dL for hypoglycemic episodes without such events. However, 35% of these events occurred with a measured BG between 50 and 60 mg/dL. The distributions of BG values associated with symptomatic and asymptomatic events are shown in Figure 2. There is no useful threshold that separates symptomatic from asymptomatic hypoglycemia. No deaths or irreversible consequences were associated with hypoglycemia.

Figure 2

Approximately 40% (195 of 484) of the hypoglycemic episodes were related to decreased enteral intake (Table 1). In addition, 6.1% (30 of 484) of hypoglycemic episodes were related to insulin adjustment and 0.4% (2 of 484) to steroid withdrawal. In 43% (209 of 484) of the episodes the cause of the hypoglycemia was unclear. The remaining 10.4% of episodes were attributed to diverse causes.

One third of patients had a documented BG rechecked within 60 minutes, and fewer than half of the hypoglycemic patients had documented euglycemia within 2 hours of their low blood glucose measurement. The average time to documented resolution of a hypoglycemic episode was 4 hours, 3 minutes, with a median of 2 hours, 25 minutes.

Table 2 delineates the various combinations of antihyperglycemic agents that the 206 patients received in the 48 hours prior to a hypoglycemic episode. Of the 484 hypoglycemic episodes, 362 involved insulin. Of patients receiving insulin, 38 of 362 of episodes of hypoglycemia occurred in patients receiving sliding‐scale insulin (SSI) dosing as the only insulin order. In 163 hypoglycemic episodes, insulin was dosed with a combination of SSI and infusion or SSI with daily long‐acting insulin. The remaining 161 episodes involved administration of insulin to patients without an accompanying sliding‐scale order.

The prevalence of hypoglycemia did not significantly differ among patients treated with oral agents alone (9 of 85, 10.6%), patients treated with insulin alone (149 of 1497, 10%), and patients treated with both (47 of 592, 7.9%). However, there was a significant relationship between specific oral agent and probability of hypoglycemia. Glyburide was associated with a higher risk of hypoglycemia (19.1%, P < .01) than were other oral agents (Table 3).

DISCUSSION

Recently, many have called for substantive changes in the management of the hospitalized diabetic.15, 16 Most have recommended replacing sliding‐scale insulin with basal bolus insulin dosing and have challenged the historic tolerance of hyperglycemia during an acute hospital stay.17 However, as hospitals and physicians transform the management of inpatient hyperglycemia, they must assess the frequency of hypoglycemia and evaluate the risk/benefit ratio of strict glycemic control.18 Thus, one study found that eliminating sliding‐scale insulin markedly improved diabetes control but hypoglycemia (BG 60 mg/dL) was more frequent.10 The cost of euglycemia is hypoglycemia.2

We report a 9.5% rate of hypoglycemia among adult hospitalized patients being treated for hyperglycemia, including those in the ICU and those in non‐ICU settings. In widely publicized landmark trials, 5.2% of intensively treated surgical ICU patients14 and 18.7% of intensively treated medical ICU patients13 experienced hypoglycemia with no adverse events. Using those studies' definition of hypoglycemia (BG 40 mg/dL), only 2.4% (53 of 2174) of our patients experienced hypoglycemia. However, our survey included general medical and surgical patients as well as ICU patients treated at their physicians' discretion, reflecting the greater variability in care that exists outside a randomized, ICU trial.

We did not anticipate the duration to documented resolution of hypoglycemic episodes, nor did we anticipate the number of hypoglycemia‐related adverse events. We believe that hospitals will need to develop formal strategies to minimize the hypoglycemic risk from tight glycemic control. The frequency and duration of the time it took to recheck the glucose, coupled with the 4.1% symptomatic event rate, suggests that inpatient hypoglycemia deserves more attention. One potential focus is the interruption of nutrition, as medications may not be readjusted when patients' oral intake declines or when they travel for tests.19

More than 40% of our hypoglycemic patients experienced recurrent episodes. This may reflect a lack of adjustment of medications following hypoglycemia. However, recurrent hypoglycemia may also be explained by hypoglycemia‐associated autonomic failure and the desensitization to hypoglycemia that occurs once a patient has lower blood glucose.2 Thus, hypoglycemic patients are at high risk of repeat episodes and often require more frequent BG monitoring. Of note, patients with hypoglycemia unawareness may not have symptoms despite low BG, and thus unless they develop signs of hypoglycemia, they would not meet criteria for an adverse event in our study, despite a very low BG.

Medical error can precipitate hypoglycemia,4, 20 and the Institute for Safe Medication Practices21 and the Joint Commission on Accreditation of Healthcare Organizations22 consider insulin a high‐risk medication. We found no hypoglycemic episodes associated with a medication error. Our CPOE system eliminates ambiguity from poor penmanship, and hospital policy requires 2 nurses to check all administered insulin. However, despite the apparent lack of dispensing/administration medication errors, nearly 10% of patients receiving hypoglycemic therapy experienced iatrogenic hypoglycemia. Thus, strategies to reduce hypoglycemia must expand beyond the prevention of medication errors.

Contrary to our expectation, we found the prevalence of hypoglycemia was at least as high, if not higher, for patients taking only oral hypoglycemics than for patients taking either insulin alone or insulin in combination with oral antihyperglycemic agents. Glyburide appeared to carry the most risk both in our population and in previous studies,2325 perhaps because of a moderately active hepatic metabolite.26 The risk of hypoglycemia with different oral agents warrants further study.

The study stimulated several actions. First, we augmented the online nursing flow sheet to permit documentation of hypoglycemic episodes, including the administration of orange juice or food. Second, our CPOE system now prevents a physician from inadvertently deselecting the hypoglycemia protocol. Third, the CPOE system prompts the nurse to recheck the BG as specified in the hypoglycemia protocol. Finally, the CPOE system warns physicians to adjust antihyperglycemic agents when they institute nutritional changes. We propose that monitoring hypoglycemia rates must become a necessary component of inpatient diabetes care that is both effective and safe and plan to monitor these rates to determine the impact of interventions designed to reduce the frequency of hypoglycemia‐related adverse events.

Our study had several limitations. We only included episodes of hypoglycemia that were identified with a POC BG. This excluded patients treated for symptomatic hypoglycemia without a measured POC BG, potentially underestimating the event rate. Moreover, defining time to resolution of a hypoglycemic episode as that documented with a serum BG but not a POC BG may have resulted in overestimating the duration, and nurses may have documented POC BGs in the MAR after a substantial delay, also artificially lengthening the time to resolution. Capillary BG may underestimate the true degree of hypoglycemia,27 thus confounding the relationship between BG and adverse events. Our study was not designed to evaluate subtle suboptimal management of hyperglycemia as a cause of hypoglycemia, although expansion of the types and combinations of insulin has increased the possibility of prescribing errors. Nor were we able to assess the preventability of hypoglycemic episodes or the independent risk factors for hypoglycemia. Finally, this study originated from a single academic hospital and thus may reflect its unique idiosyncrasies.

We have reported a comprehensive survey of hypoglycemia in patients treated with antihyperglycemic agents at a single hospital. At the time the study took place, we had not instituted hospitalwide strategies to maintain BG near euglycemic targets, although such strategies have since begun. To detect untoward events that may follow from our efforts to better control hyperglycemia, we believe it is important to establish baseline measurements. Even without aiming for tighter glucose control, we identified the need to aim for and possible strategies to achieve, better prevention of hypoglycemia.

Glycemic control in the inpatient setting has received increasing attention in recent years, with the demonstration that appropriate blood glucose (BG) control prevents adverse events in both intensive care unit (ICU) and non‐ICU settings.1 Recent recommendations set target blood glucose levels near euglycemia for most hospitalized patients.1 Unfortunately, the risk of hypoglycemia increases with tighter glycemic control,2 and hypoglycemia may result in catastrophic events.35 Although hyperglycemia is associated with postoperative infection,6 and effective management decreases wound infections,7 few reports have detailed the hypoglycemia rates among surgical patients.8 Hypoglycemia rates on medical services are as high as 28%,9, 10 and efforts to achieve more normal BG levels in hospitalized patients have been associated with more hypoglycemia.11

We undertook a study of hypoglycemia in all adult hospitalized patients receiving hypoglycemic therapy at our institution. The purpose of this study was to determine the incidence, natural history, associations, and consequences of hypoglycemia in this broad inpatient population in order to have a baseline prior to introducing any formal hospital strategies to achieve the newer targets for glycemic control.

Research Design and Methods

Thomas Jefferson University Hospital (TJUH) is a 675‐bed acute care teaching institution in center‐city Philadelphia with more than 30,000 patient admissions each year. We undertook a prospective, consecutive medical record review from August 16, 2004, to November 15, 2004, of hospitalized patients who had experienced at least 1 hypoglycemic episode, defined as at least one blood glucose (BG) 60 mg/dL within 48 hours of administration of an antihyperglycemic agent in the hospital. The definition of hypoglycemia was consistent with our hospital policies and a compromise between the BG 70 mg/dL proposed by the American Diabetes Association (ADA) hypoglycemia workgroup12 and the BG 40 mg/dL used by authors studying glycemic control in the ICU.13, 14

Hypoglycemic episodes were identified by a daily electronic search of the online medication administration record (MAR) where nurses document all point‐of‐care (POC) BG values. Two of the authors (P.V. and V.G.) reviewed the medical record for each episode and excluded pediatric (<18 years), emergency department, and maternity patients. Intensive care, step‐down, and medical/surgical unit patients were all included if the hypoglycemia had occurred within 48 hours of hospital administration of an antihyperglycemic agent. All medication orders at TJUH are placed through the computerized prescriber order entry (CPOE) system (Centricity Enterprise), which links all antihyperglycemic agents to a standardized hypoglycemia treatment protocol. The protocol includes instructions to administer glucose and/or glucagon and check the BG 15 minutes after a hypoglycemic episode. We established operational definitions prior to chart review (Appendix). A symptomatic hypoglycemia‐related adverse event was defined as any documented event occurring at the time of the hypoglycemic episode involving symptoms, change in care, temporary or permanent injury, or increased length of hospitalization. We did not include following our hypoglycemic protocol with the administration of 50% dextrose or glucagon as a change of care, as we considered this usual care, unless symptoms or signs also accompanied the hypoglycemic event.

We searched the University Health System Consortium Clinical Database (CDB) to quantify the number of patients at TJUH receiving any antihyperglycemic agent during the study period and to identify the specific agents received. The CDB receives all patient, physician, and pharmacy dose‐ specific information from the hospital clinical and billing information systems. We defined subgroups of patients taking insulin(s) only, taking oral agent(s) only, and taking a combination.

Differences between proportions were evaluated using the chi‐square statistic; differences between means were evaluated using the Student t test. Probabilities of the null hypothesis less than .05 were considered significant.

The project was approved by the Institutional Review Board for Human Subjects at Thomas Jefferson University.

RESULTS

Over the 2‐month study period 8140 patients were admitted, of whom 2174 (27%) received an antihyperglycemic agent. Five hundred and sixty‐eight hypoglycemic episodes (BG 60 mg/dL) occurred in 265 patients. We excluded 84 episodes among 59 patients who did not receive antihyperglycemic agents, resulting in 484 episodes of hypoglycemia occurring within 48 hours of hospital administration of an antihyperglycemic agent in 206 patients, an average of 5.26 episodes per day. Of the 2174 of patients receiving antihyperglycemic agents, 206 (9.5%) experienced 1 or more episodes of hypoglycemia.

Patient ages ranged from 20 to 93 years, with an average of 62 years. Fifty‐seven percent (118 of 206) of participants were female. About one‐fourth of all episodes (23.8%) occurred in the ICU setting. The distribution of patients by decade and their ICU status are presented in Figure 1. Of the 206 patients, 29% (59) had type 1 diabetes mellitus (DM), 49% (102) had type 2 DM, 1% (2) had new‐onset diabetes, and 21% (43) had no diagnosis of DM. Of the 484 hypoglycemic episodes, 37.8% occurred in patients with type 1 DM, 46.9% in patients with type 2 DM, and 0.6% in patients with new‐onset DM. The remaining 14.5% occurred in patients with no documented history of DM, although they were receiving antihyperglycemic agents. More than 1 episode was experienced by 44% of patients, and 12% experienced 5 or more episodes.

Figure 1

The BG was between 51 and 60 mg/dL in 282 of the episodes (58.2%), between 41 and 50 mg/dL in 149 episodes (30.8%) and 40 mg/dL or less in 53 episodes (11%). In 20 episodes (4.1% of episodes, representing fewer than 1% of all patients receiving an antihyperglycemic agent), a symptomatic hypoglycemia‐related adverse event was documented. All but 1 adverse event occurred outside the ICU. Ten of these events (2.1% of all hypoglycemic episodes) in 10 patients involved symptoms including headache, agitation, disorientation, and tremors. Of these patients 9 had type 1 DM, and 1 had type 2 DM. Six events (1.2% of hypoglycemic episodes) in 4 patients involved seizures. Two of these patients had type 1 DM, and 2 had type 2 DM. Four events (0.8% of hypoglycemic episodes) in 4 patients involved an unresponsive or unarousable state, including the sole ICU episode of symptomatic hypoglycemia. Three of these patients had type 1 DM, and 1 had type 2 DM. Patients with hypoglycemia‐related adverse events had a mean BG of 43.0 mg/dL, significantly lower (P = .01) than the mean BG of 50.9 mg/dL for hypoglycemic episodes without such events. However, 35% of these events occurred with a measured BG between 50 and 60 mg/dL. The distributions of BG values associated with symptomatic and asymptomatic events are shown in Figure 2. There is no useful threshold that separates symptomatic from asymptomatic hypoglycemia. No deaths or irreversible consequences were associated with hypoglycemia.

Figure 2

Approximately 40% (195 of 484) of the hypoglycemic episodes were related to decreased enteral intake (Table 1). In addition, 6.1% (30 of 484) of hypoglycemic episodes were related to insulin adjustment and 0.4% (2 of 484) to steroid withdrawal. In 43% (209 of 484) of the episodes the cause of the hypoglycemia was unclear. The remaining 10.4% of episodes were attributed to diverse causes.

One third of patients had a documented BG rechecked within 60 minutes, and fewer than half of the hypoglycemic patients had documented euglycemia within 2 hours of their low blood glucose measurement. The average time to documented resolution of a hypoglycemic episode was 4 hours, 3 minutes, with a median of 2 hours, 25 minutes.

Table 2 delineates the various combinations of antihyperglycemic agents that the 206 patients received in the 48 hours prior to a hypoglycemic episode. Of the 484 hypoglycemic episodes, 362 involved insulin. Of patients receiving insulin, 38 of 362 of episodes of hypoglycemia occurred in patients receiving sliding‐scale insulin (SSI) dosing as the only insulin order. In 163 hypoglycemic episodes, insulin was dosed with a combination of SSI and infusion or SSI with daily long‐acting insulin. The remaining 161 episodes involved administration of insulin to patients without an accompanying sliding‐scale order.

The prevalence of hypoglycemia did not significantly differ among patients treated with oral agents alone (9 of 85, 10.6%), patients treated with insulin alone (149 of 1497, 10%), and patients treated with both (47 of 592, 7.9%). However, there was a significant relationship between specific oral agent and probability of hypoglycemia. Glyburide was associated with a higher risk of hypoglycemia (19.1%, P < .01) than were other oral agents (Table 3).

DISCUSSION

Recently, many have called for substantive changes in the management of the hospitalized diabetic.15, 16 Most have recommended replacing sliding‐scale insulin with basal bolus insulin dosing and have challenged the historic tolerance of hyperglycemia during an acute hospital stay.17 However, as hospitals and physicians transform the management of inpatient hyperglycemia, they must assess the frequency of hypoglycemia and evaluate the risk/benefit ratio of strict glycemic control.18 Thus, one study found that eliminating sliding‐scale insulin markedly improved diabetes control but hypoglycemia (BG 60 mg/dL) was more frequent.10 The cost of euglycemia is hypoglycemia.2

We report a 9.5% rate of hypoglycemia among adult hospitalized patients being treated for hyperglycemia, including those in the ICU and those in non‐ICU settings. In widely publicized landmark trials, 5.2% of intensively treated surgical ICU patients14 and 18.7% of intensively treated medical ICU patients13 experienced hypoglycemia with no adverse events. Using those studies' definition of hypoglycemia (BG 40 mg/dL), only 2.4% (53 of 2174) of our patients experienced hypoglycemia. However, our survey included general medical and surgical patients as well as ICU patients treated at their physicians' discretion, reflecting the greater variability in care that exists outside a randomized, ICU trial.

We did not anticipate the duration to documented resolution of hypoglycemic episodes, nor did we anticipate the number of hypoglycemia‐related adverse events. We believe that hospitals will need to develop formal strategies to minimize the hypoglycemic risk from tight glycemic control. The frequency and duration of the time it took to recheck the glucose, coupled with the 4.1% symptomatic event rate, suggests that inpatient hypoglycemia deserves more attention. One potential focus is the interruption of nutrition, as medications may not be readjusted when patients' oral intake declines or when they travel for tests.19

More than 40% of our hypoglycemic patients experienced recurrent episodes. This may reflect a lack of adjustment of medications following hypoglycemia. However, recurrent hypoglycemia may also be explained by hypoglycemia‐associated autonomic failure and the desensitization to hypoglycemia that occurs once a patient has lower blood glucose.2 Thus, hypoglycemic patients are at high risk of repeat episodes and often require more frequent BG monitoring. Of note, patients with hypoglycemia unawareness may not have symptoms despite low BG, and thus unless they develop signs of hypoglycemia, they would not meet criteria for an adverse event in our study, despite a very low BG.

Medical error can precipitate hypoglycemia,4, 20 and the Institute for Safe Medication Practices21 and the Joint Commission on Accreditation of Healthcare Organizations22 consider insulin a high‐risk medication. We found no hypoglycemic episodes associated with a medication error. Our CPOE system eliminates ambiguity from poor penmanship, and hospital policy requires 2 nurses to check all administered insulin. However, despite the apparent lack of dispensing/administration medication errors, nearly 10% of patients receiving hypoglycemic therapy experienced iatrogenic hypoglycemia. Thus, strategies to reduce hypoglycemia must expand beyond the prevention of medication errors.

Contrary to our expectation, we found the prevalence of hypoglycemia was at least as high, if not higher, for patients taking only oral hypoglycemics than for patients taking either insulin alone or insulin in combination with oral antihyperglycemic agents. Glyburide appeared to carry the most risk both in our population and in previous studies,2325 perhaps because of a moderately active hepatic metabolite.26 The risk of hypoglycemia with different oral agents warrants further study.

The study stimulated several actions. First, we augmented the online nursing flow sheet to permit documentation of hypoglycemic episodes, including the administration of orange juice or food. Second, our CPOE system now prevents a physician from inadvertently deselecting the hypoglycemia protocol. Third, the CPOE system prompts the nurse to recheck the BG as specified in the hypoglycemia protocol. Finally, the CPOE system warns physicians to adjust antihyperglycemic agents when they institute nutritional changes. We propose that monitoring hypoglycemia rates must become a necessary component of inpatient diabetes care that is both effective and safe and plan to monitor these rates to determine the impact of interventions designed to reduce the frequency of hypoglycemia‐related adverse events.

Our study had several limitations. We only included episodes of hypoglycemia that were identified with a POC BG. This excluded patients treated for symptomatic hypoglycemia without a measured POC BG, potentially underestimating the event rate. Moreover, defining time to resolution of a hypoglycemic episode as that documented with a serum BG but not a POC BG may have resulted in overestimating the duration, and nurses may have documented POC BGs in the MAR after a substantial delay, also artificially lengthening the time to resolution. Capillary BG may underestimate the true degree of hypoglycemia,27 thus confounding the relationship between BG and adverse events. Our study was not designed to evaluate subtle suboptimal management of hyperglycemia as a cause of hypoglycemia, although expansion of the types and combinations of insulin has increased the possibility of prescribing errors. Nor were we able to assess the preventability of hypoglycemic episodes or the independent risk factors for hypoglycemia. Finally, this study originated from a single academic hospital and thus may reflect its unique idiosyncrasies.

We have reported a comprehensive survey of hypoglycemia in patients treated with antihyperglycemic agents at a single hospital. At the time the study took place, we had not instituted hospitalwide strategies to maintain BG near euglycemic targets, although such strategies have since begun. To detect untoward events that may follow from our efforts to better control hyperglycemia, we believe it is important to establish baseline measurements. Even without aiming for tighter glucose control, we identified the need to aim for and possible strategies to achieve, better prevention of hypoglycemia.

In November 1999, the Institute of Medicine released its landmark report entitled To Err Is Human: Building A Safer Health System.1 The report claimed that more than 1 million people in the United States suffer from preventable medical injuries each year and that as many as 98,000 people die annually in hospitals from medical errors. Although evidence‐based methods are available to prevent adverse events, there is concern that the current lack of standardization among hospitals implementing such safe practices has the potential to both diffuse and dilute efforts to improve patient safety.

To address this issue, the National Quality Forum (NQF) in 2003 released an evidence‐based consensus report that presented 30 safe practices for better health care with a recommendation that all be universally adopted.2 The purpose of this study is to use information collected from a voluntary patient safety program in Georgia3 and an Agency for Healthcare Research and Quality (AHRQ) reporting demonstration study4 to (1) describe the current statewide adoption rates for NQF medication safe practices and safety culture (Table 1), and (2) examine if hospital adoption varies by hospital size, ownership, and rural or urban location.

METHODS

Mapping Survey Questions to Safe Practices

A subset (n = 57) of the SMU survey questions directly related to safe medication processes (ie, prescribing, transcribing, dispensing, administration, and monitoring) were selected for inclusion in the study (Fig. 1). A nonoverlapping subset of Leadership (n = 35) and SMU (n = 10) survey questions related to safety culture were also identified. Clinical members of the project team independently reviewed and mapped medication process survey questions to 1 of 9 NQF indicators of safe medication practices. Assignment was based on face validity and best fit with the intent of the NQF indicator. Social science team members mapped culture‐related survey questions to the NQF indicator create a health care culture of safety using the 5 domains of safety culture.5

Figure 1

RESULTS

DISCUSSION

This is the first study to use existing data sources to characterize the current progress and barriers to further adoption of NQF safe practices and safety culture related to medication use in a statewide sample of hospitals. Several findings are notable. First, most of the hospitals surveyed had adopted 7 of 9 medication‐related NQF safe practices by 2004. Similar to findings from the earlier ISMP Safety Self‐Assessment for Hospitals, hospitals scored most highly on practices related to drug storage, packaging, and labeling and lowest on CPOE implementation.12 Results from the 2003 Leapfrog Group Quality and Safety Survey also found that only 3.7% of participating hospitals had fully implemented a CPOE system.13 Medication safe practices that directly affect physicians, such as verbal orders, standardized abbreviations, and access to relevant clinical information when prescribing had only intermediate adoption rates.

Most hospitals have developed policies around nonpunitive safety cultures, but fewer have adopted proactive error reduction systems. Safety culture is more difficult to measure than safe medication processes. A previous survey of Iowa hospitals assessed only whether hospitals reported progress toward creating a culture of safety.14 In this study we attempted to break down the broad concept of safety culture into specific actionable components. Three widely recognized components of a safe hospital culture are creating a nonpunitive environment for staff, using data to identify and analyze errors and system causes, and safe staffing levels.8 Most but not all surveyed hospitals had adopted these safety culture strategies. Other more resource‐intensive practices, such as unit visits by senior management and FMEA, were less likely to have been adopted. The adoption rates reported here for 20032004 are in most cases higher than those found in the 2000 ISMP survey, which may be explained by the more recent survey reported here and variations in question wording as well as response scoring.

Variations in adoption rates of NQF‐recommended safe practices generally were not explained by hospital characteristics such as ownership, size, or geographic location. Instead, barriers appear to be related to resource constraints as well as the ability of hospitals to directly control the specific safe practice. The ISMP survey found that hospital demographic factors explained only 3% of the variation in adoption, which is similar to our finding of few differences in adoption of safe practices by hospital type. Cost and health care culture may explain why certain safe practices remain less than fully adopted.12 Resource constraints may explain the lower adoption rate of several practices: CPOE, pharmacist consultation, and physical environment improvements. Other safe practices with lower adoption rates require active physician participation, for example, minimizing verbal orders, standardizing abbreviations, and ensuring accurate information transfers. Hospital‐based physicians can play a key role in advocating for effective processes to promote these practices.

Another general factor that distinguished highly adopted practices from less adopted practices was the extent to which reactive as opposed to proactive actions were required. Hospitals were more likely to report reactive policies such as reading back verbal orders than proactive policies to minimize verbal orders. Pharmacists were generally available for telephone consultation but in only half the hospitals were they available on the hospital units for consultation. A similar pattern was seen for culture of safety practices; systems were generally in place for nonpunitive error reporting, but a minority of hospitals had senior leadership making unit rounds or multidisciplinary teams proactively testing new systems to identify potential errors before they occur. Again, there is a leadership role that hospital‐based physicians can play as effective team builders for safety culture and as clinical leaders for improvement of medication processes. Much research has demonstrated the impact that a culture of safety can have on error reduction.811 As physicians who spend most of their clinical time directly on patient care units, hospital‐based physicians are uniquely positioned to promote positive changes in culture. Research on the impact of hospitalists on hospital costs and patient outcomes should be broadened to include an assessment of their impact on safety culture and error reduction.

This study had several limitations, the first being that it was based on voluntarily provided self‐assessment data. The surveys used in this project have been refined and administered over 3 years in a nonpunitive process improvement program with a consistently high participation rate. The hospital‐reported survey results have not been independently verified for accuracy, similar to most of the prior research in this area. The surveys measure management's perception of safety culture and do not assess actual employee perceptions of the safety culture on their particular units. Thus, although management may believe they are implementing policies to create a nonpunitive environment, actual assessments of employees' views are needed to confirm this. Because the study was based on previously collected data, several steps were used to map the existing questions to NQF safe practices. Given the broad nature of the NQF topics, at least 1 relevant survey question was identified for each of the medication‐related safe practices. When more than 1 question was judged to be relevant, the responses were averaged. The survey was limited to adult acute care hospitals in Georgia, which may not be nationally representative, and federal and Veterans Administration hospitals were not included. However, Georgia has a relatively high proportion of smaller rural hospitals and offers interesting baseline data on similar rates of adoption of safe practices in rural and smaller hospitals compared with that in urban hospitals. Because we were using previously collected surveys, we could only look at the adoption of selected safe practices. Further work is needed to look at the adoption of other safe practices.

In summary, it is encouraging that the most studied NQF‐recommended safe practices have already been adopted by a wide range of hospitals, including rural and small hospitals. Resource constraints as well as health care culture and structure remain barriers to broader diffusion. Some barriers may be addressed by technology and improvements in physical environments, but others relate to culture and may be more challenging to address. Active physician participation in medication‐related patient safety initiatives will be key to promoting further adoption of safe practices.

In November 1999, the Institute of Medicine released its landmark report entitled To Err Is Human: Building A Safer Health System.1 The report claimed that more than 1 million people in the United States suffer from preventable medical injuries each year and that as many as 98,000 people die annually in hospitals from medical errors. Although evidence‐based methods are available to prevent adverse events, there is concern that the current lack of standardization among hospitals implementing such safe practices has the potential to both diffuse and dilute efforts to improve patient safety.

To address this issue, the National Quality Forum (NQF) in 2003 released an evidence‐based consensus report that presented 30 safe practices for better health care with a recommendation that all be universally adopted.2 The purpose of this study is to use information collected from a voluntary patient safety program in Georgia3 and an Agency for Healthcare Research and Quality (AHRQ) reporting demonstration study4 to (1) describe the current statewide adoption rates for NQF medication safe practices and safety culture (Table 1), and (2) examine if hospital adoption varies by hospital size, ownership, and rural or urban location.

METHODS

Mapping Survey Questions to Safe Practices

A subset (n = 57) of the SMU survey questions directly related to safe medication processes (ie, prescribing, transcribing, dispensing, administration, and monitoring) were selected for inclusion in the study (Fig. 1). A nonoverlapping subset of Leadership (n = 35) and SMU (n = 10) survey questions related to safety culture were also identified. Clinical members of the project team independently reviewed and mapped medication process survey questions to 1 of 9 NQF indicators of safe medication practices. Assignment was based on face validity and best fit with the intent of the NQF indicator. Social science team members mapped culture‐related survey questions to the NQF indicator create a health care culture of safety using the 5 domains of safety culture.5

Figure 1

RESULTS

DISCUSSION

This is the first study to use existing data sources to characterize the current progress and barriers to further adoption of NQF safe practices and safety culture related to medication use in a statewide sample of hospitals. Several findings are notable. First, most of the hospitals surveyed had adopted 7 of 9 medication‐related NQF safe practices by 2004. Similar to findings from the earlier ISMP Safety Self‐Assessment for Hospitals, hospitals scored most highly on practices related to drug storage, packaging, and labeling and lowest on CPOE implementation.12 Results from the 2003 Leapfrog Group Quality and Safety Survey also found that only 3.7% of participating hospitals had fully implemented a CPOE system.13 Medication safe practices that directly affect physicians, such as verbal orders, standardized abbreviations, and access to relevant clinical information when prescribing had only intermediate adoption rates.

Most hospitals have developed policies around nonpunitive safety cultures, but fewer have adopted proactive error reduction systems. Safety culture is more difficult to measure than safe medication processes. A previous survey of Iowa hospitals assessed only whether hospitals reported progress toward creating a culture of safety.14 In this study we attempted to break down the broad concept of safety culture into specific actionable components. Three widely recognized components of a safe hospital culture are creating a nonpunitive environment for staff, using data to identify and analyze errors and system causes, and safe staffing levels.8 Most but not all surveyed hospitals had adopted these safety culture strategies. Other more resource‐intensive practices, such as unit visits by senior management and FMEA, were less likely to have been adopted. The adoption rates reported here for 20032004 are in most cases higher than those found in the 2000 ISMP survey, which may be explained by the more recent survey reported here and variations in question wording as well as response scoring.

Variations in adoption rates of NQF‐recommended safe practices generally were not explained by hospital characteristics such as ownership, size, or geographic location. Instead, barriers appear to be related to resource constraints as well as the ability of hospitals to directly control the specific safe practice. The ISMP survey found that hospital demographic factors explained only 3% of the variation in adoption, which is similar to our finding of few differences in adoption of safe practices by hospital type. Cost and health care culture may explain why certain safe practices remain less than fully adopted.12 Resource constraints may explain the lower adoption rate of several practices: CPOE, pharmacist consultation, and physical environment improvements. Other safe practices with lower adoption rates require active physician participation, for example, minimizing verbal orders, standardizing abbreviations, and ensuring accurate information transfers. Hospital‐based physicians can play a key role in advocating for effective processes to promote these practices.

Another general factor that distinguished highly adopted practices from less adopted practices was the extent to which reactive as opposed to proactive actions were required. Hospitals were more likely to report reactive policies such as reading back verbal orders than proactive policies to minimize verbal orders. Pharmacists were generally available for telephone consultation but in only half the hospitals were they available on the hospital units for consultation. A similar pattern was seen for culture of safety practices; systems were generally in place for nonpunitive error reporting, but a minority of hospitals had senior leadership making unit rounds or multidisciplinary teams proactively testing new systems to identify potential errors before they occur. Again, there is a leadership role that hospital‐based physicians can play as effective team builders for safety culture and as clinical leaders for improvement of medication processes. Much research has demonstrated the impact that a culture of safety can have on error reduction.811 As physicians who spend most of their clinical time directly on patient care units, hospital‐based physicians are uniquely positioned to promote positive changes in culture. Research on the impact of hospitalists on hospital costs and patient outcomes should be broadened to include an assessment of their impact on safety culture and error reduction.

This study had several limitations, the first being that it was based on voluntarily provided self‐assessment data. The surveys used in this project have been refined and administered over 3 years in a nonpunitive process improvement program with a consistently high participation rate. The hospital‐reported survey results have not been independently verified for accuracy, similar to most of the prior research in this area. The surveys measure management's perception of safety culture and do not assess actual employee perceptions of the safety culture on their particular units. Thus, although management may believe they are implementing policies to create a nonpunitive environment, actual assessments of employees' views are needed to confirm this. Because the study was based on previously collected data, several steps were used to map the existing questions to NQF safe practices. Given the broad nature of the NQF topics, at least 1 relevant survey question was identified for each of the medication‐related safe practices. When more than 1 question was judged to be relevant, the responses were averaged. The survey was limited to adult acute care hospitals in Georgia, which may not be nationally representative, and federal and Veterans Administration hospitals were not included. However, Georgia has a relatively high proportion of smaller rural hospitals and offers interesting baseline data on similar rates of adoption of safe practices in rural and smaller hospitals compared with that in urban hospitals. Because we were using previously collected surveys, we could only look at the adoption of selected safe practices. Further work is needed to look at the adoption of other safe practices.

In summary, it is encouraging that the most studied NQF‐recommended safe practices have already been adopted by a wide range of hospitals, including rural and small hospitals. Resource constraints as well as health care culture and structure remain barriers to broader diffusion. Some barriers may be addressed by technology and improvements in physical environments, but others relate to culture and may be more challenging to address. Active physician participation in medication‐related patient safety initiatives will be key to promoting further adoption of safe practices.

In November 1999, the Institute of Medicine released its landmark report entitled To Err Is Human: Building A Safer Health System.1 The report claimed that more than 1 million people in the United States suffer from preventable medical injuries each year and that as many as 98,000 people die annually in hospitals from medical errors. Although evidence‐based methods are available to prevent adverse events, there is concern that the current lack of standardization among hospitals implementing such safe practices has the potential to both diffuse and dilute efforts to improve patient safety.

To address this issue, the National Quality Forum (NQF) in 2003 released an evidence‐based consensus report that presented 30 safe practices for better health care with a recommendation that all be universally adopted.2 The purpose of this study is to use information collected from a voluntary patient safety program in Georgia3 and an Agency for Healthcare Research and Quality (AHRQ) reporting demonstration study4 to (1) describe the current statewide adoption rates for NQF medication safe practices and safety culture (Table 1), and (2) examine if hospital adoption varies by hospital size, ownership, and rural or urban location.

METHODS

Mapping Survey Questions to Safe Practices

A subset (n = 57) of the SMU survey questions directly related to safe medication processes (ie, prescribing, transcribing, dispensing, administration, and monitoring) were selected for inclusion in the study (Fig. 1). A nonoverlapping subset of Leadership (n = 35) and SMU (n = 10) survey questions related to safety culture were also identified. Clinical members of the project team independently reviewed and mapped medication process survey questions to 1 of 9 NQF indicators of safe medication practices. Assignment was based on face validity and best fit with the intent of the NQF indicator. Social science team members mapped culture‐related survey questions to the NQF indicator create a health care culture of safety using the 5 domains of safety culture.5

Figure 1

RESULTS

DISCUSSION

This is the first study to use existing data sources to characterize the current progress and barriers to further adoption of NQF safe practices and safety culture related to medication use in a statewide sample of hospitals. Several findings are notable. First, most of the hospitals surveyed had adopted 7 of 9 medication‐related NQF safe practices by 2004. Similar to findings from the earlier ISMP Safety Self‐Assessment for Hospitals, hospitals scored most highly on practices related to drug storage, packaging, and labeling and lowest on CPOE implementation.12 Results from the 2003 Leapfrog Group Quality and Safety Survey also found that only 3.7% of participating hospitals had fully implemented a CPOE system.13 Medication safe practices that directly affect physicians, such as verbal orders, standardized abbreviations, and access to relevant clinical information when prescribing had only intermediate adoption rates.

Most hospitals have developed policies around nonpunitive safety cultures, but fewer have adopted proactive error reduction systems. Safety culture is more difficult to measure than safe medication processes. A previous survey of Iowa hospitals assessed only whether hospitals reported progress toward creating a culture of safety.14 In this study we attempted to break down the broad concept of safety culture into specific actionable components. Three widely recognized components of a safe hospital culture are creating a nonpunitive environment for staff, using data to identify and analyze errors and system causes, and safe staffing levels.8 Most but not all surveyed hospitals had adopted these safety culture strategies. Other more resource‐intensive practices, such as unit visits by senior management and FMEA, were less likely to have been adopted. The adoption rates reported here for 20032004 are in most cases higher than those found in the 2000 ISMP survey, which may be explained by the more recent survey reported here and variations in question wording as well as response scoring.

Variations in adoption rates of NQF‐recommended safe practices generally were not explained by hospital characteristics such as ownership, size, or geographic location. Instead, barriers appear to be related to resource constraints as well as the ability of hospitals to directly control the specific safe practice. The ISMP survey found that hospital demographic factors explained only 3% of the variation in adoption, which is similar to our finding of few differences in adoption of safe practices by hospital type. Cost and health care culture may explain why certain safe practices remain less than fully adopted.12 Resource constraints may explain the lower adoption rate of several practices: CPOE, pharmacist consultation, and physical environment improvements. Other safe practices with lower adoption rates require active physician participation, for example, minimizing verbal orders, standardizing abbreviations, and ensuring accurate information transfers. Hospital‐based physicians can play a key role in advocating for effective processes to promote these practices.

Another general factor that distinguished highly adopted practices from less adopted practices was the extent to which reactive as opposed to proactive actions were required. Hospitals were more likely to report reactive policies such as reading back verbal orders than proactive policies to minimize verbal orders. Pharmacists were generally available for telephone consultation but in only half the hospitals were they available on the hospital units for consultation. A similar pattern was seen for culture of safety practices; systems were generally in place for nonpunitive error reporting, but a minority of hospitals had senior leadership making unit rounds or multidisciplinary teams proactively testing new systems to identify potential errors before they occur. Again, there is a leadership role that hospital‐based physicians can play as effective team builders for safety culture and as clinical leaders for improvement of medication processes. Much research has demonstrated the impact that a culture of safety can have on error reduction.811 As physicians who spend most of their clinical time directly on patient care units, hospital‐based physicians are uniquely positioned to promote positive changes in culture. Research on the impact of hospitalists on hospital costs and patient outcomes should be broadened to include an assessment of their impact on safety culture and error reduction.

This study had several limitations, the first being that it was based on voluntarily provided self‐assessment data. The surveys used in this project have been refined and administered over 3 years in a nonpunitive process improvement program with a consistently high participation rate. The hospital‐reported survey results have not been independently verified for accuracy, similar to most of the prior research in this area. The surveys measure management's perception of safety culture and do not assess actual employee perceptions of the safety culture on their particular units. Thus, although management may believe they are implementing policies to create a nonpunitive environment, actual assessments of employees' views are needed to confirm this. Because the study was based on previously collected data, several steps were used to map the existing questions to NQF safe practices. Given the broad nature of the NQF topics, at least 1 relevant survey question was identified for each of the medication‐related safe practices. When more than 1 question was judged to be relevant, the responses were averaged. The survey was limited to adult acute care hospitals in Georgia, which may not be nationally representative, and federal and Veterans Administration hospitals were not included. However, Georgia has a relatively high proportion of smaller rural hospitals and offers interesting baseline data on similar rates of adoption of safe practices in rural and smaller hospitals compared with that in urban hospitals. Because we were using previously collected surveys, we could only look at the adoption of selected safe practices. Further work is needed to look at the adoption of other safe practices.

In summary, it is encouraging that the most studied NQF‐recommended safe practices have already been adopted by a wide range of hospitals, including rural and small hospitals. Resource constraints as well as health care culture and structure remain barriers to broader diffusion. Some barriers may be addressed by technology and improvements in physical environments, but others relate to culture and may be more challenging to address. Active physician participation in medication‐related patient safety initiatives will be key to promoting further adoption of safe practices.

Because the incidence of hip fracture increases dramatically with age and the elderly are the fastest‐growing portion of the United States population, the number of hip fractures is expected to triple by 2040.1 With the associated increase in postoperative morbidity and mortality, the costs will likely exceed $16‐$20 billion annually.15 Already by 2002, the number of patients with hip fractures exceeded 340,000 in this country, resulting in $8.6 billion in health care expenditures from in‐hospital and posthospital costs.68 This makes hip fracture a serious public health concern and triggers a need to devise an efficient means of caring for these patients. We previously reported that a hospitalist service can decrease time to surgery and shorten length of stay without affecting the number of inpatient deaths or 30‐day readmissions of patients undergoing hip fracture surgery.9 However, one concern with reducing length of stay and time to surgery in the high‐risk hip fracture patient population is the effect on long‐term mortality because the death rate following hip fracture repair may be as high as 43% after 1 year.10 To evaluate this important issue, we assessed mortality over a 1‐year period in the same cohort of patients previously described.9 We also identified predictors associated with mortality. We hypothesized that the expedited surgical treatment and decreased length of stay of a hospitalist‐managed group would not have an adverse effect on 1‐year mortality.

METHODS

RESULTS

There were 236 patients with hip fractures (50.6%) admitted to the standard care service, and 230 patients (49.4%) admitted to the hospitalist service. As shown in Table 1, the baseline characteristics of the patients admitted to the 2 services did not differ significantly except that a greater proportion of patients with hypoxia were admitted to the hospitalist service (11.3% vs. 5.5%; P = .02). However, time to surgery, postsurgery stay, and overall length of hospitalization of the hospitalist‐treated patients were all significantly shorter.

Patients were followed for a median of 4.0 years (range 5 days to 5.6 years), and 192 patients were still alive at the end of follow‐up (April 2006). As illustrated in Figure 1, survival did not differ between the 2 treatment groups (P = .36). Overall survival at 1 year was 70.6% (95% confidence interval [CI]: 66.5%, 74.9%). Survival at 1 year in the standard care group was 70.6% (95% CI: 64.9%, 76.8%), whereas in the hospitalist group, it was 70.5% (95% CI: 64.8%, 76.7%). As delineated in Table 2, cardiovascular causes accounted for 34 deaths (25.6%), with 14 of these in the standard care group and 20 in the hospitalist group; 29 deaths (21.8%) had respiratory causes, 20 in the standard care group and 9 in the hospitalist group; and 17 (12.8%) were due to cancer, with 7 and 10 in the standard care and hospitalist groups, respectively. Unknown causes accounted for 21 cases, or 15.8% of total deaths.

Figure 1

In the univariate analysis, we found 29 variables that were significant predictors of survival (Table 3). A hospitalist model of care was not significantly associated with patient survival, despite the shorter length of stay (8.4 days vs. 10.6 days; P < .001) or expedited time to surgery (25 vs. 38 hours; P < .001), when compared with the standard care group, as previously reported by Phy et al.9 In the multivariable analysis (Table 4), however, the independent predictors of mortality were ASA class III or IV versus class II (hazard ratio [HR] 4.20; 95% CI: 2.21, 7.99), admission from a nursing home versus from home or assisted living (HR 2.24; 95% CI: 1.73, 2.90), and inpatient complications, which included patients requiring admission to the intensive care unit (ICU) and those who had a myocardial infarction or acute renal failure as an inpatient (HR 1.85; 95% CI: 1.45, 2.35). Even after adjusting for these factors, survival following hip fracture did not differ significantly between the hospitalist care patients and the standard care patients (HR 1.16; 95% CI: 0.91, 1.48).

DISCUSSION

In our previous study, length of stay and time to surgery were significantly lower in a hospitalist care model.9 The present study shows that neither the reduced length of stay nor the shortened time to surgery of patients managed by the hospitalist group was associated with a difference in mortality compared with a standard care group, despite significantly improved efficiency and processes of care. Thus, our results refute initial concerns of increased mortality in a hospitalist model of care.

Delivery of perioperative medical care to hip fracture patients by hospitalists is associated with significant decreases in time to surgery and length of stay compared with standard care, with no differences in short‐term mortality.9, 18 Although there have been conflicting reports on the impact of length of stay and time to surgery on long‐term outcomes, our findings support previous results that decreased time to surgery was not associated with an observable effect on mortality.1923 A recent study by Orosz et al. that evaluated 1178 patients showed that earlier hip fracture surgery (performed less than 24 hours after admission) was not associated with reduced mortality, although it was associated with shorter length of stay.19 Our study also corroborates the results of an examination of 8383 hip fracture patients by Grimes et al., who found that time to surgery between 24 and 48 hours after admission had no effect on either 30‐day or long‐term mortality compared with that of those who underwent surgery between 48 and 72 hours, between 72 and 96 hours, or more than 96 hours after admission.20 However, both these results and our own are contrary to those of Gdalevich, whose study of 651 patients found that 1‐year mortality was 1.6‐fold higher for those whose hip fracture repair was postponed more than 48 hours.21 However, time to surgery in both the standard care and hospitalist model in our study was well below the 48‐hour cutoff, suggesting that operating anywhere within the normally accepted 48‐hour time frame may not influence long‐term mortality.

Because of the small number of events in both groups, we were unable to specifically compare whether a hospitalist model of care has any specific impact on long‐term cause of death. Although causes of death of patients with hip fracture were consistent with those of previous studies,10, 24 our death rate at 1 year, 29.4%, was higher than that seen among similar population groups at tertiary referral centers.19, 20, 2429 This is most likely a result of the cohort having a high proportion of nursing home patients (22%)19, 24, 26 transferred for evaluation to St. Mary's Hospital, which serves most of Olmsted County, Minnesota. This hospital also has some characteristics of a community‐based hospital, as it is where greater than 95% of all county patients receive care for surgical repair of hip fracture. Mortality rates are often higher at these types of hospitals.30 Previous studies using patients from Olmsted County indicate results can also be extrapolated to a large part of the U.S. population.31 In Pitto et al.'s study, the risk of death was 31% lower in those admitted from home than for those admitted from a nursing home.32 The latter patients normally have a higher number of comorbid conditions and tend to be less ambulatory than those in a community home‐dwelling setting. Our study also demonstrated that admission from a nursing home was a strong predictor of mortality for up to 1 year in the geriatric population. This may reflect the inherent decreased survival in this patient group, which is in agreement with the findings of other studies that showed inactivity and decreased ambulation prior to fracture were associated with increased mortality.3335

Multiple comorbidities, commonly seen in a geriatric population, translate into a higher ASA class and an increased risk of significant in‐hospital complications. Our study confirmed the findings of previous studies that a higher ASA class is a strong predictor of mortality,21, 26, 30, 3537 independent of decreased time to surgery.38 We also noted that significant in‐hospital complications, including renal failure, respiratory failure, and myocardial infarction, are documented predictors of mortality after hip fracture.27 Although mortality may vary depending on fracture type (femoral neck vs. intertrochanteric),3941 these differences were not observed in our study, in line with the results of previous published studies.37, 42 Controlling for age and comorbidities may be why an association was not found between fracture type and mortality. Finally, in a model containing comorbidity, ASA class, and nursing home residence prior to fracture, age was not a significant predictor of mortality.

Our study had a number of limitations. First, this was a retrospective cohort study based on chart review, so some data may have been subject to recording bias, and this might have differed between the serial models. Because of the retrospective nature of the study and referral of some of the patients from outside the community, our 1‐year follow‐up was not complete, but approached a respectable 93%. Other studies have described the benefits derived by a hospitalist practice only following the first year of its implementation, likely because of the hospitalist learning curve.43, 44 This may be why there was no difference in mortality between the standard care and hospitalist groups, as the latter was only in its first year of existence. Additional longitudinal study is required to find out if mortality differences emerge between the treatment groups. Furthermore, although in‐hospital care may influence short‐term outcomes, its effect on long‐term mortality has been unclear. Our data demonstrate that even though a hospitalist service can shorten length of stay and time to surgery, there were no appreciable intermediate differences in mortality at 1 year. Further prospective studies are needed to determine whether this medical‐surgical partnership in caring for these patients provides more favorable outcomes of reducing mortality and intercurrent complications.

Because the incidence of hip fracture increases dramatically with age and the elderly are the fastest‐growing portion of the United States population, the number of hip fractures is expected to triple by 2040.1 With the associated increase in postoperative morbidity and mortality, the costs will likely exceed $16‐$20 billion annually.15 Already by 2002, the number of patients with hip fractures exceeded 340,000 in this country, resulting in $8.6 billion in health care expenditures from in‐hospital and posthospital costs.68 This makes hip fracture a serious public health concern and triggers a need to devise an efficient means of caring for these patients. We previously reported that a hospitalist service can decrease time to surgery and shorten length of stay without affecting the number of inpatient deaths or 30‐day readmissions of patients undergoing hip fracture surgery.9 However, one concern with reducing length of stay and time to surgery in the high‐risk hip fracture patient population is the effect on long‐term mortality because the death rate following hip fracture repair may be as high as 43% after 1 year.10 To evaluate this important issue, we assessed mortality over a 1‐year period in the same cohort of patients previously described.9 We also identified predictors associated with mortality. We hypothesized that the expedited surgical treatment and decreased length of stay of a hospitalist‐managed group would not have an adverse effect on 1‐year mortality.

METHODS

RESULTS

There were 236 patients with hip fractures (50.6%) admitted to the standard care service, and 230 patients (49.4%) admitted to the hospitalist service. As shown in Table 1, the baseline characteristics of the patients admitted to the 2 services did not differ significantly except that a greater proportion of patients with hypoxia were admitted to the hospitalist service (11.3% vs. 5.5%; P = .02). However, time to surgery, postsurgery stay, and overall length of hospitalization of the hospitalist‐treated patients were all significantly shorter.

Patients were followed for a median of 4.0 years (range 5 days to 5.6 years), and 192 patients were still alive at the end of follow‐up (April 2006). As illustrated in Figure 1, survival did not differ between the 2 treatment groups (P = .36). Overall survival at 1 year was 70.6% (95% confidence interval [CI]: 66.5%, 74.9%). Survival at 1 year in the standard care group was 70.6% (95% CI: 64.9%, 76.8%), whereas in the hospitalist group, it was 70.5% (95% CI: 64.8%, 76.7%). As delineated in Table 2, cardiovascular causes accounted for 34 deaths (25.6%), with 14 of these in the standard care group and 20 in the hospitalist group; 29 deaths (21.8%) had respiratory causes, 20 in the standard care group and 9 in the hospitalist group; and 17 (12.8%) were due to cancer, with 7 and 10 in the standard care and hospitalist groups, respectively. Unknown causes accounted for 21 cases, or 15.8% of total deaths.

Figure 1

In the univariate analysis, we found 29 variables that were significant predictors of survival (Table 3). A hospitalist model of care was not significantly associated with patient survival, despite the shorter length of stay (8.4 days vs. 10.6 days; P < .001) or expedited time to surgery (25 vs. 38 hours; P < .001), when compared with the standard care group, as previously reported by Phy et al.9 In the multivariable analysis (Table 4), however, the independent predictors of mortality were ASA class III or IV versus class II (hazard ratio [HR] 4.20; 95% CI: 2.21, 7.99), admission from a nursing home versus from home or assisted living (HR 2.24; 95% CI: 1.73, 2.90), and inpatient complications, which included patients requiring admission to the intensive care unit (ICU) and those who had a myocardial infarction or acute renal failure as an inpatient (HR 1.85; 95% CI: 1.45, 2.35). Even after adjusting for these factors, survival following hip fracture did not differ significantly between the hospitalist care patients and the standard care patients (HR 1.16; 95% CI: 0.91, 1.48).

DISCUSSION

In our previous study, length of stay and time to surgery were significantly lower in a hospitalist care model.9 The present study shows that neither the reduced length of stay nor the shortened time to surgery of patients managed by the hospitalist group was associated with a difference in mortality compared with a standard care group, despite significantly improved efficiency and processes of care. Thus, our results refute initial concerns of increased mortality in a hospitalist model of care.

Delivery of perioperative medical care to hip fracture patients by hospitalists is associated with significant decreases in time to surgery and length of stay compared with standard care, with no differences in short‐term mortality.9, 18 Although there have been conflicting reports on the impact of length of stay and time to surgery on long‐term outcomes, our findings support previous results that decreased time to surgery was not associated with an observable effect on mortality.1923 A recent study by Orosz et al. that evaluated 1178 patients showed that earlier hip fracture surgery (performed less than 24 hours after admission) was not associated with reduced mortality, although it was associated with shorter length of stay.19 Our study also corroborates the results of an examination of 8383 hip fracture patients by Grimes et al., who found that time to surgery between 24 and 48 hours after admission had no effect on either 30‐day or long‐term mortality compared with that of those who underwent surgery between 48 and 72 hours, between 72 and 96 hours, or more than 96 hours after admission.20 However, both these results and our own are contrary to those of Gdalevich, whose study of 651 patients found that 1‐year mortality was 1.6‐fold higher for those whose hip fracture repair was postponed more than 48 hours.21 However, time to surgery in both the standard care and hospitalist model in our study was well below the 48‐hour cutoff, suggesting that operating anywhere within the normally accepted 48‐hour time frame may not influence long‐term mortality.

Because of the small number of events in both groups, we were unable to specifically compare whether a hospitalist model of care has any specific impact on long‐term cause of death. Although causes of death of patients with hip fracture were consistent with those of previous studies,10, 24 our death rate at 1 year, 29.4%, was higher than that seen among similar population groups at tertiary referral centers.19, 20, 2429 This is most likely a result of the cohort having a high proportion of nursing home patients (22%)19, 24, 26 transferred for evaluation to St. Mary's Hospital, which serves most of Olmsted County, Minnesota. This hospital also has some characteristics of a community‐based hospital, as it is where greater than 95% of all county patients receive care for surgical repair of hip fracture. Mortality rates are often higher at these types of hospitals.30 Previous studies using patients from Olmsted County indicate results can also be extrapolated to a large part of the U.S. population.31 In Pitto et al.'s study, the risk of death was 31% lower in those admitted from home than for those admitted from a nursing home.32 The latter patients normally have a higher number of comorbid conditions and tend to be less ambulatory than those in a community home‐dwelling setting. Our study also demonstrated that admission from a nursing home was a strong predictor of mortality for up to 1 year in the geriatric population. This may reflect the inherent decreased survival in this patient group, which is in agreement with the findings of other studies that showed inactivity and decreased ambulation prior to fracture were associated with increased mortality.3335

Multiple comorbidities, commonly seen in a geriatric population, translate into a higher ASA class and an increased risk of significant in‐hospital complications. Our study confirmed the findings of previous studies that a higher ASA class is a strong predictor of mortality,21, 26, 30, 3537 independent of decreased time to surgery.38 We also noted that significant in‐hospital complications, including renal failure, respiratory failure, and myocardial infarction, are documented predictors of mortality after hip fracture.27 Although mortality may vary depending on fracture type (femoral neck vs. intertrochanteric),3941 these differences were not observed in our study, in line with the results of previous published studies.37, 42 Controlling for age and comorbidities may be why an association was not found between fracture type and mortality. Finally, in a model containing comorbidity, ASA class, and nursing home residence prior to fracture, age was not a significant predictor of mortality.

Our study had a number of limitations. First, this was a retrospective cohort study based on chart review, so some data may have been subject to recording bias, and this might have differed between the serial models. Because of the retrospective nature of the study and referral of some of the patients from outside the community, our 1‐year follow‐up was not complete, but approached a respectable 93%. Other studies have described the benefits derived by a hospitalist practice only following the first year of its implementation, likely because of the hospitalist learning curve.43, 44 This may be why there was no difference in mortality between the standard care and hospitalist groups, as the latter was only in its first year of existence. Additional longitudinal study is required to find out if mortality differences emerge between the treatment groups. Furthermore, although in‐hospital care may influence short‐term outcomes, its effect on long‐term mortality has been unclear. Our data demonstrate that even though a hospitalist service can shorten length of stay and time to surgery, there were no appreciable intermediate differences in mortality at 1 year. Further prospective studies are needed to determine whether this medical‐surgical partnership in caring for these patients provides more favorable outcomes of reducing mortality and intercurrent complications.

Because the incidence of hip fracture increases dramatically with age and the elderly are the fastest‐growing portion of the United States population, the number of hip fractures is expected to triple by 2040.1 With the associated increase in postoperative morbidity and mortality, the costs will likely exceed $16‐$20 billion annually.15 Already by 2002, the number of patients with hip fractures exceeded 340,000 in this country, resulting in $8.6 billion in health care expenditures from in‐hospital and posthospital costs.68 This makes hip fracture a serious public health concern and triggers a need to devise an efficient means of caring for these patients. We previously reported that a hospitalist service can decrease time to surgery and shorten length of stay without affecting the number of inpatient deaths or 30‐day readmissions of patients undergoing hip fracture surgery.9 However, one concern with reducing length of stay and time to surgery in the high‐risk hip fracture patient population is the effect on long‐term mortality because the death rate following hip fracture repair may be as high as 43% after 1 year.10 To evaluate this important issue, we assessed mortality over a 1‐year period in the same cohort of patients previously described.9 We also identified predictors associated with mortality. We hypothesized that the expedited surgical treatment and decreased length of stay of a hospitalist‐managed group would not have an adverse effect on 1‐year mortality.

METHODS

RESULTS

There were 236 patients with hip fractures (50.6%) admitted to the standard care service, and 230 patients (49.4%) admitted to the hospitalist service. As shown in Table 1, the baseline characteristics of the patients admitted to the 2 services did not differ significantly except that a greater proportion of patients with hypoxia were admitted to the hospitalist service (11.3% vs. 5.5%; P = .02). However, time to surgery, postsurgery stay, and overall length of hospitalization of the hospitalist‐treated patients were all significantly shorter.

Patients were followed for a median of 4.0 years (range 5 days to 5.6 years), and 192 patients were still alive at the end of follow‐up (April 2006). As illustrated in Figure 1, survival did not differ between the 2 treatment groups (P = .36). Overall survival at 1 year was 70.6% (95% confidence interval [CI]: 66.5%, 74.9%). Survival at 1 year in the standard care group was 70.6% (95% CI: 64.9%, 76.8%), whereas in the hospitalist group, it was 70.5% (95% CI: 64.8%, 76.7%). As delineated in Table 2, cardiovascular causes accounted for 34 deaths (25.6%), with 14 of these in the standard care group and 20 in the hospitalist group; 29 deaths (21.8%) had respiratory causes, 20 in the standard care group and 9 in the hospitalist group; and 17 (12.8%) were due to cancer, with 7 and 10 in the standard care and hospitalist groups, respectively. Unknown causes accounted for 21 cases, or 15.8% of total deaths.

Figure 1

In the univariate analysis, we found 29 variables that were significant predictors of survival (Table 3). A hospitalist model of care was not significantly associated with patient survival, despite the shorter length of stay (8.4 days vs. 10.6 days; P < .001) or expedited time to surgery (25 vs. 38 hours; P < .001), when compared with the standard care group, as previously reported by Phy et al.9 In the multivariable analysis (Table 4), however, the independent predictors of mortality were ASA class III or IV versus class II (hazard ratio [HR] 4.20; 95% CI: 2.21, 7.99), admission from a nursing home versus from home or assisted living (HR 2.24; 95% CI: 1.73, 2.90), and inpatient complications, which included patients requiring admission to the intensive care unit (ICU) and those who had a myocardial infarction or acute renal failure as an inpatient (HR 1.85; 95% CI: 1.45, 2.35). Even after adjusting for these factors, survival following hip fracture did not differ significantly between the hospitalist care patients and the standard care patients (HR 1.16; 95% CI: 0.91, 1.48).

DISCUSSION

In our previous study, length of stay and time to surgery were significantly lower in a hospitalist care model.9 The present study shows that neither the reduced length of stay nor the shortened time to surgery of patients managed by the hospitalist group was associated with a difference in mortality compared with a standard care group, despite significantly improved efficiency and processes of care. Thus, our results refute initial concerns of increased mortality in a hospitalist model of care.

Delivery of perioperative medical care to hip fracture patients by hospitalists is associated with significant decreases in time to surgery and length of stay compared with standard care, with no differences in short‐term mortality.9, 18 Although there have been conflicting reports on the impact of length of stay and time to surgery on long‐term outcomes, our findings support previous results that decreased time to surgery was not associated with an observable effect on mortality.1923 A recent study by Orosz et al. that evaluated 1178 patients showed that earlier hip fracture surgery (performed less than 24 hours after admission) was not associated with reduced mortality, although it was associated with shorter length of stay.19 Our study also corroborates the results of an examination of 8383 hip fracture patients by Grimes et al., who found that time to surgery between 24 and 48 hours after admission had no effect on either 30‐day or long‐term mortality compared with that of those who underwent surgery between 48 and 72 hours, between 72 and 96 hours, or more than 96 hours after admission.20 However, both these results and our own are contrary to those of Gdalevich, whose study of 651 patients found that 1‐year mortality was 1.6‐fold higher for those whose hip fracture repair was postponed more than 48 hours.21 However, time to surgery in both the standard care and hospitalist model in our study was well below the 48‐hour cutoff, suggesting that operating anywhere within the normally accepted 48‐hour time frame may not influence long‐term mortality.

Because of the small number of events in both groups, we were unable to specifically compare whether a hospitalist model of care has any specific impact on long‐term cause of death. Although causes of death of patients with hip fracture were consistent with those of previous studies,10, 24 our death rate at 1 year, 29.4%, was higher than that seen among similar population groups at tertiary referral centers.19, 20, 2429 This is most likely a result of the cohort having a high proportion of nursing home patients (22%)19, 24, 26 transferred for evaluation to St. Mary's Hospital, which serves most of Olmsted County, Minnesota. This hospital also has some characteristics of a community‐based hospital, as it is where greater than 95% of all county patients receive care for surgical repair of hip fracture. Mortality rates are often higher at these types of hospitals.30 Previous studies using patients from Olmsted County indicate results can also be extrapolated to a large part of the U.S. population.31 In Pitto et al.'s study, the risk of death was 31% lower in those admitted from home than for those admitted from a nursing home.32 The latter patients normally have a higher number of comorbid conditions and tend to be less ambulatory than those in a community home‐dwelling setting. Our study also demonstrated that admission from a nursing home was a strong predictor of mortality for up to 1 year in the geriatric population. This may reflect the inherent decreased survival in this patient group, which is in agreement with the findings of other studies that showed inactivity and decreased ambulation prior to fracture were associated with increased mortality.3335

Multiple comorbidities, commonly seen in a geriatric population, translate into a higher ASA class and an increased risk of significant in‐hospital complications. Our study confirmed the findings of previous studies that a higher ASA class is a strong predictor of mortality,21, 26, 30, 3537 independent of decreased time to surgery.38 We also noted that significant in‐hospital complications, including renal failure, respiratory failure, and myocardial infarction, are documented predictors of mortality after hip fracture.27 Although mortality may vary depending on fracture type (femoral neck vs. intertrochanteric),3941 these differences were not observed in our study, in line with the results of previous published studies.37, 42 Controlling for age and comorbidities may be why an association was not found between fracture type and mortality. Finally, in a model containing comorbidity, ASA class, and nursing home residence prior to fracture, age was not a significant predictor of mortality.

Our study had a number of limitations. First, this was a retrospective cohort study based on chart review, so some data may have been subject to recording bias, and this might have differed between the serial models. Because of the retrospective nature of the study and referral of some of the patients from outside the community, our 1‐year follow‐up was not complete, but approached a respectable 93%. Other studies have described the benefits derived by a hospitalist practice only following the first year of its implementation, likely because of the hospitalist learning curve.43, 44 This may be why there was no difference in mortality between the standard care and hospitalist groups, as the latter was only in its first year of existence. Additional longitudinal study is required to find out if mortality differences emerge between the treatment groups. Furthermore, although in‐hospital care may influence short‐term outcomes, its effect on long‐term mortality has been unclear. Our data demonstrate that even though a hospitalist service can shorten length of stay and time to surgery, there were no appreciable intermediate differences in mortality at 1 year. Further prospective studies are needed to determine whether this medical‐surgical partnership in caring for these patients provides more favorable outcomes of reducing mortality and intercurrent complications.

My hospital's electronic medical record helpfully informs me after 1 week on service that there are 524 data available for my attention, a statistic that would be paralyzing without a cognitive framework for organizing and interpreting them in a manner that can be shared among my colleagues. Accurate information flow among clinicians was identified early on as an imperative of hospital medicine. Much attention has been focused on communication during transitions of care, such as that between inpatient and outpatient services and between inpatient teams, taking the form of the discharge summary and the sign‐out, respectively. But communication among physicians, consultants, and allied therapists must and inevitably does occur continuously day by day during even the most uneventful hospital stay. On academic services the need to keep multiple and ever‐rotating team members on the same page, so to speak, is particularly pressing.

The succinct and accurate problem list, formulated at the end of the history and physical examination and propagated through daily progress notes, is a powerful tool for promoting clear diagnostic and therapeutic planning and is ideally suited to meeting the need for continuous information flow among clinicians. Sadly, this inexpensive and potentially elegant device has fallen into disuse and disrepair and is in need of restoration.

In the 1960s, Dr. Lawrence Weed, the inventor of the SOAP note and a pioneer of medical informatics, wrote of the power of the problem list to impose order on the chaos of clinical information and to aid clear diagnostic thinking, in contrast with the simply chronological record popular in earlier years:

Weed proposed that the product of our diagnostic thinking and investigations should be a concise list of diagnoses, as precisely as we are able to identify them, or, in their absence, a clear understanding of the specific problems awaiting resolution and a clear appreciation of the interrelationships among these entities:

These principles were widely praised and adopted. An editorial in the New England Journal of Medicine proclaimed that his system is the essence of education itself,3 and it reigned throughout my own formal medical education.

In the decade that has seen our specialty flourish, with the attendant imperatives of clear thinking and communication, in teaching hospitals the problem list seems to have become an endangered species. The general pattern of its decline is that it is often supplanted by a list of organs, or worse, medical subspecialties, each followed by some assessment of its condition, whether diseased or not. The format resembles that used in critical care units for patients with multiple vital functions in jeopardy, on which survival depends from minute to minute, sometimes regardless of the original etiology of their failure. It is not clear how these notes began to spread from the ICU to the medical floor, where puzzles are solved and progress has goals more varied than mere survival. None of the residents I have queried over the years seem to know. The prevalence of this habit is also unknown, but it is widespread at both institutions at which I have been recently affiliated, and from the generation of notes in this format by trainees freshly graduated from medical schools across the land, I infer that it is no mere regional phenomenon. There may be an unspoken assumption that if this format is used for the sickest patients, it must be the superior format to use for all patients. Perhaps it reflects subspecialists teaching inpatient medicine, equipping trainees with vast technical knowledge of specific diseases and placing less emphasis on formulating coherent assessments. I believe its effects are pernicious and far‐reaching, affecting not only the quality of information flow among clinicians, but also the quality and rigor of diagnostic thinking of those in our training programs.

The history and physical examination properly culminate in the formulation of a problem list that establishes the framework for subsequent investigations and therapy. For each problem a narrative thread is initiated that can be followed in progress notes to resolution and succinctly reviewed in the discharge summary. It is now common to see diagnostic formulations arranged not by problem but by organ or subspecialty, for example, Endocrine: DKA. As everyone understands DKA to be an endocrine problem, the organ system preface adds nothing useful and only serves to bury the diagnosis in text. More tortured prose follows attempts to cram into the header all organs or specialties touched by the problem; hence pneumonia is often preceded by pulmonary/ID. A more egregious recent example was an esophageal variceal hemorrhage designated GI/Heme. And efforts to force an undifferentiated problem into an organ group can reach absurdity: Heme: Asymmetric leg swelling raised concern for DVT, but ultrasound was negative.

The organ preface at best merely adds clutter; the difficulty is compounded when the actual diagnosis or problem is omitted entirely in favor of mention of the organs, for example, for pneumonia: Pulm/ID: begin antibiotics. The reader may be left to guess exactly what is being treated, as with CV: begin heparin and beta‐blocker. The assessment and subsequent notes become even more unwieldy when the unifying diagnosis is approached circuitously on paper by way of its component elements, as with a recent patient with typical lobar pneumonia who was assessed by the house officer as having (1) ID: fever probably due to pneumonia; (2) Pulm: Hypoxia, sputum production and infiltrate on CXR consistent with pneumonia; and (3) Heme: leukocytosis likely due to pneumonia as well. Synthesis, the holy grail of the H&P, is thus replaced by analysis. Each tree is closely inspected, but we are lost in the forest. Weed wrote of such notes:

Often, however, as in the example above, the physician fully understands the unifying diagnosis but nonetheless insists on addressing involved systems separately. Each feature is then apt to be separately followed in isolation through the progress notes, sometimes without any further mention of pneumonia as such. Many progress notes thus omit stating what is actually thought to be wrong with the patient.

The failure to commit to a diagnosis on paper, even when having done so in practice, ultimately can make its way to the discharge summary, propagating confusion to the outpatient department and ricocheting it into future admissions. It also robs us of the satisfaction of declaring a puzzle solved. I was compelled to write this piece in part by the recent case of a young woman who presented with fever and dyspnea. Through an elegant series of imaging studies and serologic tests, a diagnosis of lupus pericarditis was established, and steroid therapy produced dramatic remission of her symptomsa diagnostic triumph by any measure. How disheartening then to read the resident's final diagnosis for posterity in the discharge summary: fever and dyspnea.

The disembodied organ list thus sows confusion and redundant, convoluted prose throughout the medical record. Perhaps even more destructive is its effect on diagnostic thinking when applied to undifferentiated symptoms or problems, the general internist's pice de rsistance. Language shapes thought, and premature assignment of symptoms to a single organ or subspecialty constrains the imagination needed to puzzle things out. Examples are everywhere. Fever of unknown origin may be peremptorily designated ID, by implication excluding inflammatory, neoplastic, and iatrogenic causes from consideration. The asymmetrically swollen legs cited earlier are not hematologic, but they are still swollen. Undiagnosed problems should be labeled as such, with comment as to the differential diagnosis as it stands at the time and the status of the investigation. When a diagnosis is established, it should replace the undifferentiated symptom or abnormal finding in the list, with cardinal manifestations addressed as such when necessary. Thus, for example, fever in an intravenous drug user becomes endocarditis, and anasarca becomes nephrotic syndrome becomes glomerulonephritis as the diagnosis is established and refined. Weed saw the promise of the well‐groomed, problem‐based record in teaching diagnostic thinking:

Hospitalists as generalist clinician‐educators have an opportunity to teach fundamental principles of medicine that span subspecialties. These principles must include clear organization and prioritization of complex medical information to enable coherent diagnostic and therapeutic planning and smooth continuity of care. The sign‐out and the all‐important discharge summary can be only as clear and as logical as the diagnoses that inform them. To these ends, let us maintain and reinvigorate the art of the problem list. As an exercise at morning report and attending rounds, we should emphasize the development of an accurate, comprehensive list of active problems before moving on to detailed discussion of any single issue, as Weed suggested nearly 40 years ago:

We should expect this list to be formulated at the end of the admission history and physical examination. We must ensure that trainees can correctly identify the level of resolution achieved for each item. They must learn to distinguish among undifferentiated symptoms, for example, passed out; undifferentiated problems, expressed by medical terms with precise meaning, such as syncope; and precise etiologic diagnoses, such as ventricular tachycardia. Daily progress notes and sign‐out documents must reflect the progressive refinement in classification of each item and give the current status of the diagnostic evaluation. When therapy has been established, daily notes must reflect its precise status relative to its end points; examples include place in the timeline for antibiotics or, for a bleeding patient, a tally of blood products and their impact. In the end, we must ensure that the discharge summary reflects the highest level of diagnostic resolution achieved for each problem we have identified. In so doing, we will help to ensure coherent and efficient care for our patients, save time and spare confusion for our colleagues, and teach our trainees to think and communicate clearly about our collective efforts.

My hospital's electronic medical record helpfully informs me after 1 week on service that there are 524 data available for my attention, a statistic that would be paralyzing without a cognitive framework for organizing and interpreting them in a manner that can be shared among my colleagues. Accurate information flow among clinicians was identified early on as an imperative of hospital medicine. Much attention has been focused on communication during transitions of care, such as that between inpatient and outpatient services and between inpatient teams, taking the form of the discharge summary and the sign‐out, respectively. But communication among physicians, consultants, and allied therapists must and inevitably does occur continuously day by day during even the most uneventful hospital stay. On academic services the need to keep multiple and ever‐rotating team members on the same page, so to speak, is particularly pressing.

The succinct and accurate problem list, formulated at the end of the history and physical examination and propagated through daily progress notes, is a powerful tool for promoting clear diagnostic and therapeutic planning and is ideally suited to meeting the need for continuous information flow among clinicians. Sadly, this inexpensive and potentially elegant device has fallen into disuse and disrepair and is in need of restoration.

In the 1960s, Dr. Lawrence Weed, the inventor of the SOAP note and a pioneer of medical informatics, wrote of the power of the problem list to impose order on the chaos of clinical information and to aid clear diagnostic thinking, in contrast with the simply chronological record popular in earlier years:

Weed proposed that the product of our diagnostic thinking and investigations should be a concise list of diagnoses, as precisely as we are able to identify them, or, in their absence, a clear understanding of the specific problems awaiting resolution and a clear appreciation of the interrelationships among these entities:

These principles were widely praised and adopted. An editorial in the New England Journal of Medicine proclaimed that his system is the essence of education itself,3 and it reigned throughout my own formal medical education.

In the decade that has seen our specialty flourish, with the attendant imperatives of clear thinking and communication, in teaching hospitals the problem list seems to have become an endangered species. The general pattern of its decline is that it is often supplanted by a list of organs, or worse, medical subspecialties, each followed by some assessment of its condition, whether diseased or not. The format resembles that used in critical care units for patients with multiple vital functions in jeopardy, on which survival depends from minute to minute, sometimes regardless of the original etiology of their failure. It is not clear how these notes began to spread from the ICU to the medical floor, where puzzles are solved and progress has goals more varied than mere survival. None of the residents I have queried over the years seem to know. The prevalence of this habit is also unknown, but it is widespread at both institutions at which I have been recently affiliated, and from the generation of notes in this format by trainees freshly graduated from medical schools across the land, I infer that it is no mere regional phenomenon. There may be an unspoken assumption that if this format is used for the sickest patients, it must be the superior format to use for all patients. Perhaps it reflects subspecialists teaching inpatient medicine, equipping trainees with vast technical knowledge of specific diseases and placing less emphasis on formulating coherent assessments. I believe its effects are pernicious and far‐reaching, affecting not only the quality of information flow among clinicians, but also the quality and rigor of diagnostic thinking of those in our training programs.

The history and physical examination properly culminate in the formulation of a problem list that establishes the framework for subsequent investigations and therapy. For each problem a narrative thread is initiated that can be followed in progress notes to resolution and succinctly reviewed in the discharge summary. It is now common to see diagnostic formulations arranged not by problem but by organ or subspecialty, for example, Endocrine: DKA. As everyone understands DKA to be an endocrine problem, the organ system preface adds nothing useful and only serves to bury the diagnosis in text. More tortured prose follows attempts to cram into the header all organs or specialties touched by the problem; hence pneumonia is often preceded by pulmonary/ID. A more egregious recent example was an esophageal variceal hemorrhage designated GI/Heme. And efforts to force an undifferentiated problem into an organ group can reach absurdity: Heme: Asymmetric leg swelling raised concern for DVT, but ultrasound was negative.

The organ preface at best merely adds clutter; the difficulty is compounded when the actual diagnosis or problem is omitted entirely in favor of mention of the organs, for example, for pneumonia: Pulm/ID: begin antibiotics. The reader may be left to guess exactly what is being treated, as with CV: begin heparin and beta‐blocker. The assessment and subsequent notes become even more unwieldy when the unifying diagnosis is approached circuitously on paper by way of its component elements, as with a recent patient with typical lobar pneumonia who was assessed by the house officer as having (1) ID: fever probably due to pneumonia; (2) Pulm: Hypoxia, sputum production and infiltrate on CXR consistent with pneumonia; and (3) Heme: leukocytosis likely due to pneumonia as well. Synthesis, the holy grail of the H&P, is thus replaced by analysis. Each tree is closely inspected, but we are lost in the forest. Weed wrote of such notes:

Often, however, as in the example above, the physician fully understands the unifying diagnosis but nonetheless insists on addressing involved systems separately. Each feature is then apt to be separately followed in isolation through the progress notes, sometimes without any further mention of pneumonia as such. Many progress notes thus omit stating what is actually thought to be wrong with the patient.

The failure to commit to a diagnosis on paper, even when having done so in practice, ultimately can make its way to the discharge summary, propagating confusion to the outpatient department and ricocheting it into future admissions. It also robs us of the satisfaction of declaring a puzzle solved. I was compelled to write this piece in part by the recent case of a young woman who presented with fever and dyspnea. Through an elegant series of imaging studies and serologic tests, a diagnosis of lupus pericarditis was established, and steroid therapy produced dramatic remission of her symptomsa diagnostic triumph by any measure. How disheartening then to read the resident's final diagnosis for posterity in the discharge summary: fever and dyspnea.

The disembodied organ list thus sows confusion and redundant, convoluted prose throughout the medical record. Perhaps even more destructive is its effect on diagnostic thinking when applied to undifferentiated symptoms or problems, the general internist's pice de rsistance. Language shapes thought, and premature assignment of symptoms to a single organ or subspecialty constrains the imagination needed to puzzle things out. Examples are everywhere. Fever of unknown origin may be peremptorily designated ID, by implication excluding inflammatory, neoplastic, and iatrogenic causes from consideration. The asymmetrically swollen legs cited earlier are not hematologic, but they are still swollen. Undiagnosed problems should be labeled as such, with comment as to the differential diagnosis as it stands at the time and the status of the investigation. When a diagnosis is established, it should replace the undifferentiated symptom or abnormal finding in the list, with cardinal manifestations addressed as such when necessary. Thus, for example, fever in an intravenous drug user becomes endocarditis, and anasarca becomes nephrotic syndrome becomes glomerulonephritis as the diagnosis is established and refined. Weed saw the promise of the well‐groomed, problem‐based record in teaching diagnostic thinking:

Hospitalists as generalist clinician‐educators have an opportunity to teach fundamental principles of medicine that span subspecialties. These principles must include clear organization and prioritization of complex medical information to enable coherent diagnostic and therapeutic planning and smooth continuity of care. The sign‐out and the all‐important discharge summary can be only as clear and as logical as the diagnoses that inform them. To these ends, let us maintain and reinvigorate the art of the problem list. As an exercise at morning report and attending rounds, we should emphasize the development of an accurate, comprehensive list of active problems before moving on to detailed discussion of any single issue, as Weed suggested nearly 40 years ago:

We should expect this list to be formulated at the end of the admission history and physical examination. We must ensure that trainees can correctly identify the level of resolution achieved for each item. They must learn to distinguish among undifferentiated symptoms, for example, passed out; undifferentiated problems, expressed by medical terms with precise meaning, such as syncope; and precise etiologic diagnoses, such as ventricular tachycardia. Daily progress notes and sign‐out documents must reflect the progressive refinement in classification of each item and give the current status of the diagnostic evaluation. When therapy has been established, daily notes must reflect its precise status relative to its end points; examples include place in the timeline for antibiotics or, for a bleeding patient, a tally of blood products and their impact. In the end, we must ensure that the discharge summary reflects the highest level of diagnostic resolution achieved for each problem we have identified. In so doing, we will help to ensure coherent and efficient care for our patients, save time and spare confusion for our colleagues, and teach our trainees to think and communicate clearly about our collective efforts.

My hospital's electronic medical record helpfully informs me after 1 week on service that there are 524 data available for my attention, a statistic that would be paralyzing without a cognitive framework for organizing and interpreting them in a manner that can be shared among my colleagues. Accurate information flow among clinicians was identified early on as an imperative of hospital medicine. Much attention has been focused on communication during transitions of care, such as that between inpatient and outpatient services and between inpatient teams, taking the form of the discharge summary and the sign‐out, respectively. But communication among physicians, consultants, and allied therapists must and inevitably does occur continuously day by day during even the most uneventful hospital stay. On academic services the need to keep multiple and ever‐rotating team members on the same page, so to speak, is particularly pressing.

The succinct and accurate problem list, formulated at the end of the history and physical examination and propagated through daily progress notes, is a powerful tool for promoting clear diagnostic and therapeutic planning and is ideally suited to meeting the need for continuous information flow among clinicians. Sadly, this inexpensive and potentially elegant device has fallen into disuse and disrepair and is in need of restoration.

In the 1960s, Dr. Lawrence Weed, the inventor of the SOAP note and a pioneer of medical informatics, wrote of the power of the problem list to impose order on the chaos of clinical information and to aid clear diagnostic thinking, in contrast with the simply chronological record popular in earlier years:

Weed proposed that the product of our diagnostic thinking and investigations should be a concise list of diagnoses, as precisely as we are able to identify them, or, in their absence, a clear understanding of the specific problems awaiting resolution and a clear appreciation of the interrelationships among these entities:

These principles were widely praised and adopted. An editorial in the New England Journal of Medicine proclaimed that his system is the essence of education itself,3 and it reigned throughout my own formal medical education.

In the decade that has seen our specialty flourish, with the attendant imperatives of clear thinking and communication, in teaching hospitals the problem list seems to have become an endangered species. The general pattern of its decline is that it is often supplanted by a list of organs, or worse, medical subspecialties, each followed by some assessment of its condition, whether diseased or not. The format resembles that used in critical care units for patients with multiple vital functions in jeopardy, on which survival depends from minute to minute, sometimes regardless of the original etiology of their failure. It is not clear how these notes began to spread from the ICU to the medical floor, where puzzles are solved and progress has goals more varied than mere survival. None of the residents I have queried over the years seem to know. The prevalence of this habit is also unknown, but it is widespread at both institutions at which I have been recently affiliated, and from the generation of notes in this format by trainees freshly graduated from medical schools across the land, I infer that it is no mere regional phenomenon. There may be an unspoken assumption that if this format is used for the sickest patients, it must be the superior format to use for all patients. Perhaps it reflects subspecialists teaching inpatient medicine, equipping trainees with vast technical knowledge of specific diseases and placing less emphasis on formulating coherent assessments. I believe its effects are pernicious and far‐reaching, affecting not only the quality of information flow among clinicians, but also the quality and rigor of diagnostic thinking of those in our training programs.

The history and physical examination properly culminate in the formulation of a problem list that establishes the framework for subsequent investigations and therapy. For each problem a narrative thread is initiated that can be followed in progress notes to resolution and succinctly reviewed in the discharge summary. It is now common to see diagnostic formulations arranged not by problem but by organ or subspecialty, for example, Endocrine: DKA. As everyone understands DKA to be an endocrine problem, the organ system preface adds nothing useful and only serves to bury the diagnosis in text. More tortured prose follows attempts to cram into the header all organs or specialties touched by the problem; hence pneumonia is often preceded by pulmonary/ID. A more egregious recent example was an esophageal variceal hemorrhage designated GI/Heme. And efforts to force an undifferentiated problem into an organ group can reach absurdity: Heme: Asymmetric leg swelling raised concern for DVT, but ultrasound was negative.

The organ preface at best merely adds clutter; the difficulty is compounded when the actual diagnosis or problem is omitted entirely in favor of mention of the organs, for example, for pneumonia: Pulm/ID: begin antibiotics. The reader may be left to guess exactly what is being treated, as with CV: begin heparin and beta‐blocker. The assessment and subsequent notes become even more unwieldy when the unifying diagnosis is approached circuitously on paper by way of its component elements, as with a recent patient with typical lobar pneumonia who was assessed by the house officer as having (1) ID: fever probably due to pneumonia; (2) Pulm: Hypoxia, sputum production and infiltrate on CXR consistent with pneumonia; and (3) Heme: leukocytosis likely due to pneumonia as well. Synthesis, the holy grail of the H&P, is thus replaced by analysis. Each tree is closely inspected, but we are lost in the forest. Weed wrote of such notes:

Often, however, as in the example above, the physician fully understands the unifying diagnosis but nonetheless insists on addressing involved systems separately. Each feature is then apt to be separately followed in isolation through the progress notes, sometimes without any further mention of pneumonia as such. Many progress notes thus omit stating what is actually thought to be wrong with the patient.

The failure to commit to a diagnosis on paper, even when having done so in practice, ultimately can make its way to the discharge summary, propagating confusion to the outpatient department and ricocheting it into future admissions. It also robs us of the satisfaction of declaring a puzzle solved. I was compelled to write this piece in part by the recent case of a young woman who presented with fever and dyspnea. Through an elegant series of imaging studies and serologic tests, a diagnosis of lupus pericarditis was established, and steroid therapy produced dramatic remission of her symptomsa diagnostic triumph by any measure. How disheartening then to read the resident's final diagnosis for posterity in the discharge summary: fever and dyspnea.

The disembodied organ list thus sows confusion and redundant, convoluted prose throughout the medical record. Perhaps even more destructive is its effect on diagnostic thinking when applied to undifferentiated symptoms or problems, the general internist's pice de rsistance. Language shapes thought, and premature assignment of symptoms to a single organ or subspecialty constrains the imagination needed to puzzle things out. Examples are everywhere. Fever of unknown origin may be peremptorily designated ID, by implication excluding inflammatory, neoplastic, and iatrogenic causes from consideration. The asymmetrically swollen legs cited earlier are not hematologic, but they are still swollen. Undiagnosed problems should be labeled as such, with comment as to the differential diagnosis as it stands at the time and the status of the investigation. When a diagnosis is established, it should replace the undifferentiated symptom or abnormal finding in the list, with cardinal manifestations addressed as such when necessary. Thus, for example, fever in an intravenous drug user becomes endocarditis, and anasarca becomes nephrotic syndrome becomes glomerulonephritis as the diagnosis is established and refined. Weed saw the promise of the well‐groomed, problem‐based record in teaching diagnostic thinking:

Hospitalists as generalist clinician‐educators have an opportunity to teach fundamental principles of medicine that span subspecialties. These principles must include clear organization and prioritization of complex medical information to enable coherent diagnostic and therapeutic planning and smooth continuity of care. The sign‐out and the all‐important discharge summary can be only as clear and as logical as the diagnoses that inform them. To these ends, let us maintain and reinvigorate the art of the problem list. As an exercise at morning report and attending rounds, we should emphasize the development of an accurate, comprehensive list of active problems before moving on to detailed discussion of any single issue, as Weed suggested nearly 40 years ago:

We should expect this list to be formulated at the end of the admission history and physical examination. We must ensure that trainees can correctly identify the level of resolution achieved for each item. They must learn to distinguish among undifferentiated symptoms, for example, passed out; undifferentiated problems, expressed by medical terms with precise meaning, such as syncope; and precise etiologic diagnoses, such as ventricular tachycardia. Daily progress notes and sign‐out documents must reflect the progressive refinement in classification of each item and give the current status of the diagnostic evaluation. When therapy has been established, daily notes must reflect its precise status relative to its end points; examples include place in the timeline for antibiotics or, for a bleeding patient, a tally of blood products and their impact. In the end, we must ensure that the discharge summary reflects the highest level of diagnostic resolution achieved for each problem we have identified. In so doing, we will help to ensure coherent and efficient care for our patients, save time and spare confusion for our colleagues, and teach our trainees to think and communicate clearly about our collective efforts.