Séminaire MODAL'X : Antoine Moll (SCOR)

Résumé : Undiagnosed diabetes is a significant source of risk misclassifications in life and health insurance underwriting, affecting mortality, disability, and critical illness assessments. Because many applicants are unaware of their condition, insurers relying solely on self reported information may underestimate true medical risk. This study examines how complementary screening strategies can improve the detection of undiagnosed diabetes while controlling medical testing costs.
Using data from the National Health and Nutrition Examination Survey (NHANES) linked to mortality follow-up, several screening strategies are evaluated, ranging from simple questionnaires to a model without constraints. Logistic regression and gradient boosting models are used to estimate the probability of undiagnosed diabetes, while mortality is modeled using Cox proportional hazards. Economic performance is evaluated through a loss‑ratio framework that incorporates underwriting costs.
A screening strategy combining a medical examination with a urine albumin‑to‑creatinine test markedly improves detection and mortality segmentation while remaining cost‑efficient. Testing about 30-40% of the highest‑risk applicants maximizes underwriting gains, with diminishing returns beyond this threshold. These findings offer practical guidance for insurers seeking to refine diabetes screening policies.