2058-LB: Identification of Earlier Stage Autoimmune Type 1 Diabetes Using Machine Learning Algorithms

Abstract

Introduction and Objective: Autoimmune type 1 diabetes (T1D) often goes undiagnosed until a major clinical event triggers disease recognition. Identifying individuals in early T1D stages remains a clinical challenge given inefficient screening thus limiting opportunities for early intervention. This study aimed to develop a predictive machine learning model that identified individuals before the onset of stage 3 T1D. Methods: This was a retrospective cohort study that utilized medical claims data and lab test results from the US Managed Markets Insight & Technology (MMIT) dataset to develop two age specific AI/ML Models (0-24 years and 25+ years) for identifying individuals with presumed early stage T1D at least one year from first observed T1D diagnosis. Confirmed stage 3 T1D cases, used to train and validate the model, were required to have ≥2 claims for T1D, a ratio of T1D : type 2 diabetes claims of ≥0.5, ≥1 claim for insulin or continuous glucose monitoring, and claims activity of at least 1 medical and 1 pharmacy claim in each year for two years before first observed T1D diagnosis or treatment (index). The model was trained on patient data >12 months prior to index to identify patients at least one year before the appearance of a T1D diagnosis or treatment. Variables included T1D and non-T1D associated clinical variables, autoimmune markers, comorbidities, demographic factors, and sequential medical events. Results: Both models were able to detect diagnosed T1D patients (~80% sensitivity in the 0-24 model; ~92% in the 25+model) at ~8% precision in the 0-24 model (~14k true positives in ~167k predicted positives) and ~10% in the 25+ model (~16k in ~169k). Conclusion: The study demonstrates the potential clinical utility of machine learning models for the early detection of type 1 diabetes. This may enable earlier diagnosis through increased screening efficiency and yield, allowing for timely intervention and better management of T1D, ultimately improving patient outcomes. Disclosure F. Lam: Consultant; Sanofi. V. Tiwari: Consultant; Sanofi. G. Mion: Consultant; Sanofi. D. Khedekar: Consultant; Sanofi. P. Mukherjee: Consultant; Sanofi. S. Pandey: Consultant; Sanofi. D. Desai: Consultant; Sanofi. L. Wilson: Employee; Sanofi-Aventis U.S. Stock/Shareholder; Sanofi-Aventis U.S. J. Dunne: Employee; Sanofi, Novo Nordisk. L. Hao: Employee; Sanofi. M. Wieloch: Employee; Sanofi. Stock/Shareholder; Sanofi. J.H. Zaccai: Employee; Sanofi. R.B. McQueen: Speaker's Bureau; Sanofi. Other Relationship; Sanofi. K.M. Simmons: Consultant; Sanofi. Research Support; Sanofi. Advisory Panel; Sanofi, Shoreline Biosciences. E.K. Sims: Consultant; Sanofi. Speaker's Bureau; Med Learning Group. Other Relationship; American Diabetes Association. Funding This study was funded by Sanofi.