Bonnie C. Greenwood, PharmD

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

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

METHODS

RESULTS

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

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

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

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

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

Figure 1

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

DISCUSSION

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

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

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

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

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

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

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

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

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

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

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

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

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

METHODS

RESULTS

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

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

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

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

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

Figure 1

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

DISCUSSION

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

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

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

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

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

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

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

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

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

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

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

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

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

METHODS

RESULTS

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

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

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

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

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

Figure 1

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

DISCUSSION

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

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

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

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

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

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

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

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

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

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

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