Iatrogenic Inpatient Hypoglycemia: Risk Factors, Treatment, and Prevention: Analysis of Current Practice at an Academic Medical Center With Implications for Improvement Efforts Free

Our medical center (University of California, San Diego Medical Center[UCSD MC]) is an academic teaching institution in a major metropolitan center with an average daily census of 540 adult patients. The study population was derived from inpatients on general medical and surgical units with POC or serum glucose values obtained in the course of their usual care, and all were on inpatient glucose-lowering agents. Patients were identified from a daily report generated from the clinical UCSD MC database between December 2007 and February 2008. Approval from the Internal Review Board and a waiver of individual informed consent was obtained.

This was a matched case-control design. A case series of consecutive patients suffering from an iatrogenic hypoglycemic day were compared to controls that did not have hypoglycemia on that same day. Cases were identified via a computer-generated daily report that captured all patients with a POC or serum glucose value ≤ 60 mg/dl.

The inclusion criteria for cases included 1) adults ≥ 18 years of age; 2) a patient on UCSD MC medical, surgical, or orthopedic service (noncritical care); 3) POC or serum glucose value ≤ 60 mg/dl; and 4) an event occurring while on a glucose-lowering agent,and the timing of the event was consistent with an iatrogenic etiology. For patients with more than one hypoglycemic event in a day, the first event recorded was the index event.

All patients who did not have a POC or serum glucose value ≤ 60 mg/dl that same day were also identified in a daily report. Eligible controls selected for the study were randomly chosen from this pool of patients using a Microsoft Excel random number generator “plug-in” utility, and all controls were also on glucose-lowering agents.

Controls were matched by similar lengths of stay to cases, to avoid bias regarding the frequency of prior hypoglycemic events. Patients were excluded if they had a hypoglycemia admission diagnosis or a hypoglycemic event within 24 hours after admission and were also excluded if their hypoglycemia was not induced by insulin or an oral anti-hyperglycemic agent prescribed in the hospital. Intensive care, psychiatry, senior behavioral health, and obstetrics units were also excluded to limit the study population to those services that used the subcutaneous insulin protocol.

An electronic and chart audit of patient demographics and possible hypoglycemia risk factors (Table 1) using medical charts and hospital databases was performed 24–48 hours after the hypoglycemic event. Direct nursing staff queries were also used only to clarify chart documentation if needed. Demographic data included sex, date of birth, admit date and time, hospital site, unit,service, attending name, blood glucose value and source (POC or chemistry),result date and time (generating the hospital date and time the event occurred), and admission diagnosis.

Information was then extracted from the administrative database, chart review, or Siemens Pharmacy systems on both cases and controls; this included subjects' height, weight, BMI, and race; disease states that may affect glycemic control (as defined below); and nutritional intake and medication regimen information.

Diagnoses of type 1 or type 2 diabetes were derived from admission and emergency department documentation. Liver disease was defined as a documented diagnosis of cirrhosis, end-stage liver disease, portal hypertension, or attendant complications (such as ascites or encephalopathy). Kidney disease was defined as a glomerular filtration rate < 30 ml/min or active dialysis,and congestive heart failure was defined by an ejection fraction < 40% or a documented diagnosis in the admission history and physical.

Medications (outpatient and inpatient) with the potential to affect glucose control were oral hypoglycemic agents, insulin, steroids, and β-blockers. Nutritional status (type of nutrition/diet patients were on at the time of event) data were collected along with the antihyperglycemic regimen. We also collected data on the presence or absence of a prior hypoglycemic day in the current hospitalization and on the presence or absence of potential nutritional interruption or discordance of nutrition with the antihyperglycemic regimen. We placed nutritional interruption/discordance into one of three categories: new NPO status, new documented nausea/emesis or anorexia, or other causes, such as the administration of nutritional insulin out of sync with nutrition or giving an incorrect or extra dose of insulin. These nutritional interruption/discordance parameters had to be documented in the record and needed to be present within the 12-hour time period proximal to the index hypoglycemic event.

For every case with a hypoglycemic event, additional information was collected on management of the event and adherence to hypoglycemia treatment protocols: documentation of the hypoglycemic event other than the value itself; any temporary or permanent harm induced by the event; minutes until the next glucose value was checked after event; minutes until resolution(defined as a documented glucose value ≥ 80 mg/dl); amount of carbohydrate given; changes in regimen (i.e., increase in dextrose or nutrition or decrease in hypoglycemic agents); and description of actions taken.

A secure electronic database was maintained to store the patient data.

To determine power, we estimated a medium effect size of 30% difference in nutritional interruption/discordance between cases and controls. A sample size of 49 subjects in each group was needed to achieve power of 0.8 at an α< 0.05 using the χ² statistic.

We used χ² analysis to compare the cases and controls by the multiple factors. Initial univariate analyses were examined using a matched logistic model for predicting cases. Variables were than grouped by three categories: 1) disease (diabetes diagnosis, congestive heart failure,liver failure, chronic kidney disease), 2) medications, and 3) demographics, including age, sex, BMI, nutritional interruption/discordance, and prior hypoglycemic day. Based on these three separate analyses, variables that remained robust with a significant value of P < 0.10 were entered into a final multivariate model with adjustments for age and sex.

We also looked at adherence to the medical center hypoglycemia treatment protocol, treatment of the event, and changes in the regimen to prevent another occurrence. If the hypoglycemic event did not result from a temporary nutritional interruption/discordance, the patient was judged to remain at risk for future episodes unless the antihyperglycemic regimen was reduced or carbohydrate/nutritional delivery was increased.

There were a total of 65 sequential cases and 65 controls in the study. The demographic and baseline characteristics of both groups with unadjusted univariate comparisons between them are listed in Table 1.

There was a statistically insignificant trend for more males to be controls(P < 0.054) and no difference in age between the two groups. Cases had a significantly lower BMI than controls, with an average of 26 ± 17 and 31 ± 11 kg/m², respectively (P = 0.004), and had a mean weight 16 kg less than controls. Ethnicity was not significantly different in cases versus controls in the univariate analysis, except that cases were more likely to be Asian than controls (nine Asian cases vs. two Asian controls, P = 0.027).

Cases had a higher prevalence of chronic kidney disease (23 cases vs. 11 controls, P = 0.015) and a higher prevalence of congestive heart failure (24 cases vs. 10 controls, P = 0.015). There was no difference in liver disease prevalence between cases and controls. Seventy-five percent of the study population carried a prior diagnosis of diabetes. The diagnosis of type 1 diabetes did not differ significantly between cases and controls, but there were more cases (n = 52) than controls (n = 37) with type 2 diabetes (P < 0.004).

Before admission, 71 of 130 (55%) study participants were on scheduled insulin, and 33 (25%) were on oral antihyperglycemic agents. The inpatient medication regimens shifted toward more frequent use of scheduled insulin (86 of 130; 68%) and less use of oral antihyperglycemic regimens (15 of 130; 12%). Cases were more likely than controls to have scheduled insulin as an outpatient (48 cases vs. 23 controls, P < 0.001), and this same trend carried through in the inpatient setting, with 81% of the cases on scheduled insulin compared to 51% of the controls (P < 0.001).β-Blockers and steroids were both common in outpatient and inpatient settings, and inpatient β-blockers were more common in cases than controls (51% of cases vs. 32% of controls, P = 0.033). We did not detect discontinuation of steroids as an etiology for any hypoglycemic event.

New NPO order status was present in 13 cases and 12 controls. New nausea/emesis or anorexia was more common in cases than controls (18 cases vs. 8 controls, P = 0.05), as were “other” causes of nutritional/antihyperglycemic discordance (8 cases vs. 2 controls, P= 0.05). Nutritional interruption/discordance was present in some form in 49%of hypoglycemia cases vs. 26% of matched controls (P = 0.007). Twelve patients experienced more than one source of nutritional interruption/discordance, accounting for the discrepancy in totals of the component parts of this parameter with the composite totals.

Remarkably, more than half of hypoglycemic cases had experienced a prior hypoglycemic day during their inpatient stay, whereas only 8% of matched controls had experienced a prior hypoglycemic day during their inpatient stay(33 cases vs. 5 controls, P = 0.003).

Sub-analysis by the variables grouped within common sets of disease,medication, or demographics identified the variables entered into the final model. The age- and sex-adjusted final multivariate model is shown in Table 2. The presence of a prior hypoglycemic day during the admission was the dominant predictor of a hypoglycemic event, with nutritional interruption/discordance and insulin use as an outpatient also emerging as pertinent factors. Race, BMI, the disease states, and other factors did not add significant predictive ability to these factors.

Tables 3 and 4 summarize hypoglycemia therapy and regimen adjustment for the 65 patients who had either a POC or chemistry glucose value ≤ 60 mg/dl. Of those 65 patients, 11 (17%) had severe hypoglycemia, classified as a blood glucose ≤ 40 mg/dl. Only 2 of the 65 patients suffered any documented harm. One patient had lethargy and another was tremulous for a brief time beyond glucose value correction,prompting a brief transfer to an intermediate level of care for observation.

The UCSD MC hypoglycemia treatment protocol states that blood glucose has to be checked every 15 minutes after a hypoglycemic event until two consecutive values ≥ 80 mg/dl have been attained. The median time until the next glucose value was checked was 60 minutes, with a range of 8–600 minutes. The median time until documented resolution (with a glucose value of≥ 80 mg/dl) was 180 minutes with a range of 10–1,260 minutes. Nineteen of the 65 cases (29.2%) of hypoglycemic events were not documented anywhere in the chart or progress notes except for the recording of the low value itself.

As noted above, 32 of the 65 hypoglycemic cases had some form of nutritional interruption or antihyperglycemic/nutritional discordance, leaving 33 who had no such temporary reason for their hypoglycemic event. Of these 33 hypoglycemic cases without temporary nutritional interruption/discordance as a contributing factor, 17 had a reduction in their scheduled insulin dose, 2 had an oral agent discontinued, 4 had their carbohydrate consumption increased,and 10 (30%) had no adjustment made.

We sought to evaluate risk factors for iatrogenic hypoglycemia, compare the frequency of these risk factors in a matched control group, and examine the response to iatrogenic hypoglycemic events. Our study has several findings of interest with practical implications for multidisciplinary improvement teams.

Finding: The documentation and treatment of iatrogenic hypoglycemia is poor, and adherence to existing hypoglycemia protocols was suboptimal. There was a very long delay in documented blood glucose monitoring after an event, which clearly deviates from protocol. In 7 of the 65 hypoglycemic cases, the nurse reported that they did not receive proper notification of the hypoglycemic chemistry lab value, leading to long delays in the treatment and subsequent retesting.

Documentation of the hypoglycemic event varied from a detailed description of what the patient's current condition was leading up to the event and exactly how much carbohydrate was given as treatment to merely mentioning a patient's hypoglycemic glucose value. Documentation of the event and treatment was also difficult to retrieve because of inconsistencies in where the documentation was placed (nurses' progress notes vs. with POC glucose tests vs. in the medication administration record). Physician performance was also suboptimal, in that appropriate adjustments to avert repeated episodes of iatrogenic hypoglycemia often did not occur. This poor performance confirms the experience from other studies.^11,12,14 

Implications: The mere existence of a complete institutional hypoglycemia protocol provides no assurance that proper care of iatrogenic hypoglycemia is taking place. Monitoring and improving adherence to a hypoglycemia protocol should be a top priority for glycemic control steering committees. Simplification of hypoglycemia protocols, with a focus on treatment of the event, assessment of the etiology, notification of the treating clinician, and documentation in a standardized format should be stressed. Improvement efforts must include ongoing assessment of adherence to the hypoglycemia protocol.

Finding: Unexpected nutritional interruption/discordance is one of the most important risk factors for iatrogenic hypoglycemia. Our univariate analysis confirmed several risk factors for iatrogenic hypoglycemia(low BMI, congestive heart failure, chronic kidney disease, andβ-blockade). The failure to confirm other reported risk factors (advanced age, type 1 diabetes, liver disease) is likely the result of a limited sample size and multiple comparisons, and improvement teams should still consider these conditions as signals to use lower initial doses of insulin.

More importantly, nutritional interruption/discordance and a prior hypoglycemic event were strong predictors of subsequent hypoglycemia in both univariate and adjusted multivariate analyses. There were three categories making up the composite nutritional interruption/discordance parameter: new NPO order status, new documented nausea/emesis or anorexia, and other causes,such as the administration of nutritional insulin out of sync with nutrition or giving an incorrect or extra dose of insulin. The NPO order status was not a risk factor for hypoglycemia in our institution, which incorporates basal/nutritional/correction dose insulin terminology into orders and the medication administration record and standing orders in the care of NPO patients.^17,18 However, the other two categories of nutritional interruption/discordance were a risk factor at our institution.

Implications: New nausea/emesis or anorexia should trigger more frequent monitoring of glucose and a lower threshold for increasing carbohydrate delivery or reducing the intensity of antihyperglycemic therapy. Nutrition delivery, glucose monitoring, and insulin delivery should be synchronized as much as possible. Our findings suggest that interruption of nutrition is exceedingly common in inpatients and that staff often do not take action that could avert hypoglycemia in these situations. Although we have made attempts for years to improve the synchronization of nutrition,monitoring, and insulin delivery, many opportunities for improvement remain. Educational efforts should focus on these triggers for added attention and adjustment.

Finding: One of the strongest predictors of an iatrogenic hypoglycemic event is experiencing a prior hypoglycemic event, and adjustments to avert recurrent hypoglycemia are often absent or suboptimal. An impressive 51% of hypoglycemia cases had suffered a prior hypoglycemia event earlier in their hospitalization. Unfortunately, the opportunities to avert repeated episodes are often not realized, and suboptimal or no changes were made in many patients with hypoglycemia (even those without possible temporary mitigating circumstances). This finding has also been seen by others.^11,13–15 

Implications: Concurrent scrutiny and proactive adjustment of regimens in patients with iatrogenic hypoglycemia or near hypoglycemia could dramatically reduce iatrogenic hypoglycemia. As a result of this study,we are planning an interventional strategy for patients approaching or developing hypoglycemia. A daily report of all patients with glucose values in the danger zone (for example, any glucose < 80 mg/dl) will be forwarded to a nurse/hospitalist consultant team. This team will assess the glucose values and treatment regimen and intervene with focused interventions, such as templated consultation notes and phone calls, with recommendations for changes in either the antihyperglycemia regimen or nutrition plan. Although confirmatory research is needed, we believe this proactive approach looking at concurrent daily glucose values is much more likely to be successful in securing improvement than retrospective review of hypoglycemia cases.

Finding: Many disciplines influence iatrogenic hypoglycemia. Surgical, medical, pharmacy, laboratory, dietary,nutritional, and nursing personnel were all involved in the care of these patients, and opportunities for improvement were noted in all of them.

Implications: Multidisciplinary approaches are needed for multidisciplinary problems. We have already started educating our hospital staff, including nurses, physicians, pharmacists, dietary services staff, and laboratory staff, about the insulin management and hypoglycemia treatment protocols. Nursing in-service training sessions and skills workshops on proper documentation have been given and will continue throughout the medical center, and educational sessions are taking place for a variety of services. All disciplines will be involved as we measure the impact of revising and simplifying our current hypoglycemia protocol and other educational efforts.

Our study had several limitations. Sample size calculations, while sufficient to look at key variables, may not be sufficient to support a robust analysis of multiple factors. As mentioned previously, documentation of hypoglycemia treatment was often poor, so possible causes of the hypoglycemic event, as well as the exact treatment and changes made to regimens, were subject to the data collectors' search and interpretation. Also, we waited only 48 hours after an index event to collect information on treatments given or adjustments to regimens, which means we may have missed any changes made after 48 hours.

Our study also included patients who did not have a diagnosis of diabetes. There was a trend for having more cases with a diabetes diagnosis than controls. Although this trend did not reach statistical significance, controls were not on scheduled insulin as frequently in the outpatient or inpatient setting, potentially influencing other comparisons.

Still, we think our findings and implications will hold up despite these relative shortcomings because our study also demonstrates several strengths,including rigorous data collection techniques, the presence of a matched control group, and a logistic regression analysis that adjusts for the effect of multiple cofactors.

Iatrogenic hypoglycemia is pervasive and is the major barrier to achieving improved inpatient glycemic control. The treatment of iatrogenic hypoglycemia is often suboptimal, and improvement efforts should focus on simplifying hypoglycemia protocols and monitoring adherence to institutional standards of treatment, assessment, notification, and documentation of hypoglycemia episodes. Although traditional disease state–related risk factors for hypoglycemia remain important in calculating initial insulin dosages, the more dynamic risk factors of sudden nutritional interruption, mistiming of the antidiabetes regimen with nutritional intake, and prior hypoglycemic events are more important areas on which to focus improvement and prevention efforts.

Multidisciplinar y efforts to increase the frequency of monitoring,initiate appropriate adjustment of antihyperglycemic medication, and increase carbohydrate supply in the face of these risk factors should result in significant reductions in iatrogenic hypoglycemia without significant loss of glycemic control. Proactive adjustment of the monitoring and treatment regimens of patients with hypoglycemia or near-hypoglycemia is a strategy worthy of further study and improvement activities.