A new study indicates that insulin resistance may be a risk factor for 12 different types of cancer. The research, which analyzed data from approximately half a million individuals, utilized an artificial intelligence (AI) model to assess this risk on a population-wide scale. Understanding cancer risk factors is crucial for developing preventative strategies and improving patient outcomes.
Insulin resistance occurs when the body no longer responds effectively to insulin, the hormone that regulates blood glucose levels. It is a fundamental cause of diabetes and is too associated with cardiovascular, kidney, and liver diseases. While its connection to obesity is well-known, directly evaluating insulin resistance in clinical practice can be challenging.
For the first time, an international team of researchers, including a team from the University of Tokyo in Japan, applied a machine learning-based prediction model to data from roughly 500,000 participants in the UK Biobank.
The results, published on Monday, February 16, 2026, in the journal Nature Communications, demonstrate an association between insulin resistance and an increased risk for 12 cancer types.
The team developed a tool called AI-IR, which estimates insulin resistance based on nine routinely collected medical parameters. The model was trained on independent cohorts from the United States and Taiwan, using anonymized medical data from over half a million people.
Testing revealed that AI-IR has a superior predictive ability compared to body mass index (BMI), which is frequently used to approximate the risk of metabolic imbalances associated with insulin resistance. This improved accuracy could lead to more targeted risk assessments.
BMI can produce inaccurate results. Some individuals with obesity may have a relatively healthy metabolic profile, while others with a normal weight may exhibit insulin resistance. By combining the nine parameters into a single indicator, AI-IR can identify insulin resistance that is not solely explained by BMI.
Compared to direct measurement of insulin resistance in validation sets, the model demonstrated high predictive performance. Direct measurement of insulin resistance is difficult to implement on a large scale and is typically performed in specialized diabetes clinics. The authors believe AI-IR offers a population-level applicable alternative for evaluating this risk.
Researchers are currently analyzing how genetic differences between individuals may influence the risk associated with insulin resistance, with the goal of correlating extensive data with molecular biology studies and developing more effective strategies for managing this condition.
