Development of a hypoglycemia prediction tool for use in patients with diabetes mellitus admitted to Beaumont Hospital, Dearborn

Document Type

Conference Proceeding

Publication Date


Publication Title

Ohio Pharmacy Residency Conference


Purpose: Hypoglycemia during hospital stay is associated with longer lengths of stay and increased risk of mortality. Although the danger of hypoglycemia is known, there are few tools to guide the initial intensity of diabetes management in hospitalized patients. Thus we rely on clinical judgment and institutional resources when deciding insulin doses and glycemic targets. An objective and simple predictive model for hypoglycemia could be helpful in guiding the initial intensity of treatment. While studies published in the last decade indicate that there is some agreement on major risk factors, the results vary accordingly due to differences in region, methods, scale, and variables utilized limiting their external validity. In this study, we sought to derive and validate a hypoglycemia risk model utilizing variables readily available upon admission at Beaumont Hospital, Dearborn. Methods: Patients in the derivation cohort were identified by ICD-10 codes related to diabetes mellitus (both type 1 and type 2) noted during admission occurring between November 1, 2019 and April 30, 2020. Subjects were excluded if they were less than 18 years old at the time of noted admission. The prediction rule was derived from a stepwise logistic regression, with hypoglycemia during the first 72 hours of hospitalization as the primary outcome. Hypoglycemia was defined as a serum or blood glucose reading less than 70 mg/dL, while severe hypoglycemia was defined as less than 50 mg/dL. Risk scores were generated on the basis of the beta-coefficients of the model. Patients in the validation cohort were identified in a similar manner from records between November 1, 2020 and February 28, 2021. Demographic and clinical variables previously reported were obtained for comparison of baseline characteristics and validation of the prediction rule. Results: The hypoglycemia incidence was 12.61% in the derivation cohort. Characteristics independently associated with hypoglycemia included history of hypoglycemic episode within the past 6 months (OR 3.25, p) Conclusions: The hypoglycemia prediction rule showed fair accuracy in predicting hypoglycemia.

This document is currently not available here.