ML flags overtreatment in older T2D
A new BMJ Open analysis used machine learning to estimate the prevalence of glucose‑lowering overtreatment among older people with type‑2 diabetes in long‑term care, linking predictors and cohort data to spotlight where de‑intensification is needed. The study provides a data‑driven framework for identifying patients at risk of hypoglycemia from overly aggressive glycemic targets. (x.com)
The paper, led by Greg Carney and coauthors, was published in BMJ Open (received 17 June 2025, accepted 18 February 2026) under DOI e106707. (bmjopen.bmj.com) Researchers used province‑wide linked administrative health claims from British Columbia covering 2016–2023 and restricted the cohort to long‑term care residents aged >65 and community‑dwelling people aged >75 with an A1C ≤7.0%. (bmjopen.bmj.com) The analytic cohort included 133,773 patients; 38,074 individuals (28.5%) met the study’s prespecified criteria for potential overtreatment, with a mean age of 79.6 years, 47% female and a median A1C of 6.4%. (bmjopen.bmj.com) Overtreatment was defined a priori as overlapping prescriptions for ≥2 glucose‑lowering medications or any insulin or sulfonylurea dispensing within 90 days after the index A1C test. (bmjopen.bmj.com) A gradient‑boosting model that combined expert‑selected and data‑driven variables performed best, achieving an AUC of 0.87, sensitivity of 0.81 and a negative predictive value of 0.89 when tested on a temporally distinct 2021–2023 holdout set. (bmjopen.bmj.com) Top contributors to predicted overtreatment were use of blood‑glucose test strips, A1C testing volume, polypharmacy, specialist involvement and measures of diabetes severity, and the authors state external validation is required before clinical implementation while noting these factors could guide future quality‑improvement efforts. (bmjopen.bmj.com)
