Researchers from the University of Tokyo developed a machine learning tool called AI-IR that predicts insulin resistance using nine standard clinical parameters (age, sex, race, BMI, fasting plasma glucose, HbA1c, triglycerides, total cholesterol, and HDL).
Using UK Biobank data from over 370,000 individuals without prior cancer history, the study found elevated cancer risk in those with metabolic dysfunction. For composite cancers (10 types common to both sexes), risk was 25 – 40% higher in individuals with insulin resistance or diabetes. Similar patterns emerged for uterine and breast cancer in females, with 23–26% elevated risk.
Insulin resistance has long been linked to diabetes and cardiovascular disease, but establishing its connection to cancer at a population scale has been difficult. The AI-IR tool addresses limitations of using BMI alone, which produces false positives (metabolically healthy obese individuals) and false negatives (normal-weight people with insulin resistance).
Since AI-IR uses parameters obtained through standard health checkups, it may help identify high-risk individuals for focused screening. The team is now investigating how genetic differences influence this risk.
Published in: Nature Communications (February 2026)
