Abstract
Disclosure: B. Lalani: None. R. Penno: None. A. Lalani: Atul Lalani has an ownership stake in Endocrine Technologies, which developed the algorithm used in this study.
Background: Timely glycemic control is crucial in the critically ill but remains challenging due to interpatient variability in insulin sensitivity, counter-regulatory hormonal responses, comorbidities, and co-medications. A patient’s changing metabolism during their hospital stay can further complicate hyperglycemia management. Computerized insulin infusion protocols can allow for personalized titration by leveraging patient-level electronic health record (EHR) data. Previously, we evaluated Glucopilot (formerly LIIP) in critically ill COVID-19 patients and, separately, compared Glucopilot's performance against a commercially available algorithm in patients receiving IV insulin. Herein, we present four-year longitudinal performance data on Glucopilot, an adaptive insulin dosing tool. Methods: We conducted a retrospective analysis of patients who received insulin therapy managed by Glucopilot across a six-hospital health system in the Phoenix metropolitan area from August 18, 2020, to March 6, 2024. All point-of-care blood glucose measurements taken hourly during insulin runs were extracted from the EHR and reported on a Microsoft Power BI dashboard. Primary endpoints included time to euglycemia (70 - 180 mg/dL) and percentage of time in euglycemia, hyperglycemia (>180 mg/dL), and hypoglycemia (<70 mg/dL). Secondary analyses were conducted in subgroups where attaining glycemic control was known to be historically challenging. Subgroups were defined by diabetic ketoacidosis/hyperosmolar hyperglycemic state, sepsis, and steroid use. Results: Among 3743 patients, the median time to euglycemia from baseline glucose was 181 minutes, with patients spending 89.15% in euglycemia, 10.63% in hyperglycemia, and 0.22% in hypoglycemia. In the DKA/HHS subgroup (n=843), patients had a median time to euglycemia of 259 minutes and spent 84.25% in euglycemia, 15.43% in hyperglycemia, and 0.31% in hypoglycemia. Similar results were found for sepsis patients (n=451; 277 minutes: 86.75% euglycemia, 13.01% hyperglycemia, 0.24% hypoglycemia) and patients on steroids (n=789; 255 minutes: 85.07% euglycemia, 14.74% hyperglycemia, 0.19% hypoglycemia). Conclusion: We implemented a personalized decision support tool that integrates EHR data and glucose trends over time to calculate patient-specific variables (e.g., insulin sensitivity, insulin dose), which continuously adjust throughout the hospital stay. Glucopilot led to acceptable time in range and time to control, with low hypoglycemia incidence. Notably, Glucopilot required no prior algorithm modifications for traditionally difficult-to-control subgroups (e.g., patients on parenteral or oral glucocorticoids).
Presentation: Sunday, July 13, 2025