Fig. 4: Effect of machine learning-predicted insulin resistance on the incidence of composite cancers, and its BMI-dependent and -independent effect on cancer incidence.

a Kaplan–Meier plots of the cumulative incidence of the composite cancers in the DM(-); AI-IR (-), DM(-); AI-IR (+), and DM (+) groups. In the graph, the lines represent the estimated cumulative incidence, and the shaded error bands represent 95% confidence intervals. The HRs for the incidence of the composite cancers adjusted for age and sex or adjusted for age, sex, and BMI were also shown. Cox proportional hazards model was used with a two-sided Wald test. b Kaplan–Meier plots of the cumulative incidence of the uterine and breast cancer in females from the DM(-); AI-IR (-), DM(-); AI-IR (+), and DM (+) groups. In the graph, the lines represent the estimated cumulative incidence, and the shaded error bands represent 95% confidence intervals. The HRs for the incidence of these cancers adjusted for age or adjusted for age and BMI were also shown. Cox proportional hazards model was used with a two-sided Wald test. c For 13 cancer types whose incidences were significantly or nominally affected by AI-IR, the effects of AI-IR after adjustment for BMI were shown. Cox proportional hazards model was used with a two-sided Wald test. In the graph, points represent the estimated HR, and error bars represent 95% confidence intervals. HRs are adjusted for age and sex. The Bonferroni-corrected P value for significance was 0.05/43 = 1.163 × 10⁻³. Source data are provided as a Source data file.