Long-term Performance of a Dynamic Clinical Decision Support System for Intravenous Insulin Therapy: A Retrospective Cohort Study in Patients With and Without Diabetes.
OBJECTIVE To report and analyze the long-term performance of Glucopilot, a clinical decision support system for intravenous (IV) insulin therapy. METHODS This retrospective cohort study included adult patients across 9 hospitals who received IV insulin guided by Glucopilot between August 18, 2020 and May 30, 2025. Admissions with fewer than 10 hours of IV insulin or insufficient glucose monitoring were excluded. Using point-of-care glucose data from the electronic health record, we evaluated time to euglycemia, time in glycemic ranges, and severe dysglycemic events, with subgroup analyses (eg, diabetes type, diabetic ketoacidosis, sepsis, and steroid use) and multivariable models to assess glycemic performance and factors associated with sustained euglycemia (ie, 100% time in euglycemia after initial attainment) and severe hyperglycemia (ie, ≥300 mg/dL). RESULTS There were 1239 admissions included. All patients presenting with hyperglycemia achieved euglycemia, with a median time to euglycemia of 174 minutes. Thereafter, median time in euglycemia was 96%, with 4% median time in hyperglycemia, and 0% median time in hypoglycemia. Sustained euglycemia was observed in 43.4% of admissions, and steroid use was not associated with reduced likelihood of achieving sustained euglycemia. Baseline glucose was strongly associated with time to euglycemia but not with subsequent time in euglycemia after attainment. Glycemic control was consistent across subgroups. CONCLUSIONS Use of Glucopilot, an electronic medical record-integrated adaptive insulin dosing software, was associated with timely attainment of euglycemia and sustained glycemic control with a median of 0% time in hypoglycemia in real-world practice.