Researchers led by the University of Tokyo applied a machine‑learning model (AI‑IR) to predict insulin resistance from routine clinical data across 500,000 UK Biobank participants and found insulin resistance was a population‑level risk factor for 12 cancer types. The study, published in Nature Communications, used predicted insulin resistance scores to associate metabolic dysfunction with cancer incidence. AI‑IR estimates insulin resistance using nine standard clinical measures, enabling large‑scale epidemiologic assessment where direct insulin‑resistance testing is impractical. The authors reported significant associations across multiple cancer types after adjusting for covariates, suggesting metabolic dysregulation may play a broader oncologic role than previously quantified. The work highlights the utility of ML‑derived biomarkers for population studies, but authors note that causality and potential clinical translation require further mechanistic and interventional research.