89-OR: Effective Real-World Usage of Artificial Intelligence–Based Decision Support System for Diabetes Management

Abstract

Background: Artificial intelligence-based decision support system (AI-DSS) for frequent insulin dose adjustment was demonstrated to be effective in improving glycemic control in our previous clinical trials. We aimed to assess the effectiveness of AI-DSS in real world diabetes clinics. Methods: Data was collected and evaluated from routine AI-DSS usage for individuals with type 1 diabetes treated at 21 diabetes clinics (in the USA and 2 in Israel) . Two analyses were made: (1) Insulin dose recommendations were compared for rate of agreement/disagreement between those suggested by the AI-DSS and the approved recommendations by health care providers (HCP) . (2) Glycemic outcomes were compared between baseline and after 3 months of AI-DSS use. Included in this analysis were individuals who had mean baseline glucose level ≥ 182 mg/dL and at least one AI-DSS recommendation within the 3-months. Results: A total of 8 recommendations, provided to 370 individuals, were evaluated. Full agreement on the direction of insulin dose adjustments was observed in 87%, 87%, and 83% of the basal rate, carbs ratio (CR) , and correction factor (CF) pump settings parameters, respectively. Full disagreement on the direction of dose change was observed in only 0.9%, 0.5%, and 1.2% for the basal rate, CR and CF, respectively. Glycemic outcome analysis included 1 eligible individuals (mean of 1.6 recommendations within a 3-month interval) . Average sensor percentage of time in range [70-180 mg/dL] increased by 4% (p=0.003) : 43% increased time in range by more than 5% and 33% increased by 10% or more. Time in hyperglycemia [> 180 mg/dL] was reduced by 5% (p<0.01) while time in severe hypoglycemia [< 54 mg/dL] remained below 0.5%. Conclusion: A high rate of agreement with automated insulin adjustments was observed among HCPs at academic centers who used the AI-DSS in their workflow. Glycemic control was significantly improved for sub-optimally controlled individuals already after 3-months of follow up. Disclosure R.Nimri: Employee; DreaMed Diabetes, Ltd., Research Support; Abbott Diabetes, Dexcom, Inc., Insulet Corporation, Medtronic, Speaker's Bureau; Eli Lilly and Company, Novo Nordisk, Stock/Shareholder; DreaMed Diabetes, Ltd. A.Tirosh: Advisory Panel; Abbott Diagnostics, AstraZeneca, Boehringer Ingelheim International GmbH, Merck & Co., Inc., Novo Nordisk, Sanofi, Consultant; Bayer AG, DreaMed Diabetes, Ltd., Research Support; Medtronic, Speaker's Bureau; Eli Lilly and Company. I.Muller: Board Member; DreaMed Diabetes, Ltd. Y.Shtrit: Employee; DreaMed Diabetes, Ltd. M.Phillip: Advisory Panel; Dompé, Insulet Corporation, Medtronic, Pfizer Inc., Board Member; DreaMed Diabetes, Ltd., Consultant; QuLab Medical Ltd., Other Relationship; Dexcom, Inc., Dompé, DreaMed Diabetes, Ltd., Eli Lilly and Company, Medtronic, NG Solutions Ltd, Novo Nordisk, OPKO, Pfizer Inc., Sanofi, Stock/Shareholder; DreaMed Diabetes, Ltd., NG Solutions Ltd.