2051-LB: Predictive Modeling for Presymptomatic Type 1 Diabetes Detection Using Open Claims Data

Introduction and Objective: Type 1 Diabetes mellitus (T1D) progresses through distinct stages before clinical diagnosis, with early stages being asymptomatic. Integrating pre-symptomatic T1D detection can enable the proactive identification of patients at higher risk and facilitate earlier interventions. This study employed machine learning approaches on a U.S. open claims dataset to identify people at risk of presymptomatic T1D. Methods: A retrospective cohort analysis was conducted on 89,453 patients selected based on the modified Klompas algorithm with initial T1D diagnosis or insulin prescription between Jan 2016 and Dec 2023. The study utilized tree-based (XGBoost) and transformer-based (BERT) models. The performance was evaluated using Precision, Recall, Bayes factor, and Number Needed to Test (NNT). Feature importance was assessed using Shapley values. Results: BERT was found to have better performance than tree-based models featuring Precision of 0.09 vs 0.05; Recall of 0.8 vs 0.72; Bayes factor of 19.92 vs 10.75; NNT values of 11.6 vs 22.7, respectfully. In both models, elevated risk of T1D was most associated with lower age and male gender. Tree-based model showed that elevated occurrence of diagnoses related to digestive system and thyroid tests were also associated with presymptomatic T1D. Conclusion: Our findings suggest the potential for using claims records to identify presymptomatic T1D individuals. The models improved the detection rate from the standard prevalence of 1:200 to less than 1:23, significantly enhancing early detection efforts. Existing demographic, frequent digestive system diagnoses and routine endocrine tests could signal T1D risk. Utilizing the recency and frequency of medical events may enhance the predictability of early T1D onset. Such models could enable more targeted screening and access to timely interventions for improved clinical outcomes. Disclosure D. Einor: None. O. Buiko: None. K.C.S. Lee: Employee; Sanofi. L. Adamek: Employee; Sanofi. B. Rufino: Employee; Sanofi. I. Yarova: None. J. Josleyn: Employee; Sanofi.