Machine learning-predicted insulin resistance is a risk factor for 12 types of cancer

2026-02-16 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Health & Medical Research, Life Sciences & Biology, Mathematics & Computational Sciences · Depth: Advanced, extended

Summary

A machine learning model, artificial intelligence–derived insulin resistance (AI-IR), predicts insulin resistance using nine clinical parameters and identifies it as a significant risk factor for 12 types of cancer. Applied to the UK Biobank cohort, AI-IR demonstrated superior predictive performance for diabetes incidence (AUC 0.798) compared to traditional metrics like BMI (0.721), metabolic syndrome (0.748), TG/HDL ratio (0.702), and TyG index (0.703). The study found AI-IR to be significantly associated with an increased risk of six cancers (uterine, kidney, esophagus, pancreas, colon, and breast) and nominally associated with six others (renal pelvis, small intestine, stomach, liver and gallbladder, leukemia, and bronchial and lung). For composite cancers, the age- and sex-adjusted hazard ratio for AI-IR positive individuals without diabetes was 1.25 (95% CI, 1.20-1.31; P < 1 × 10⁻¹¹). AI-IR also improved cancer risk stratification, even for individuals with obesity.

Key takeaway

For AI Scientists developing predictive health models, this research highlights the value of machine learning in identifying complex disease risk factors like insulin resistance. Your models can integrate multiple clinical parameters to create more robust "digital biomarkers" that surpass traditional metrics. Consider how your predictive tools can capture both BMI-dependent and BMI-independent effects to enhance risk stratification for conditions like diabetes and cancer, leading to more targeted clinical interventions and improved patient outcomes.

Key insights

AI-IR, a machine learning model, robustly predicts insulin resistance and identifies it as a risk factor for 12 cancer types.

Principles

• AI-IR outperforms traditional metrics for diabetes prediction.
• Insulin resistance is a BMI-independent cancer risk factor.

Method

An XGBoost model was trained on nine clinical parameters to predict HOMA-IR > 2.5, achieving an AUC of 0.88. This AI-IR model was then applied to the UK Biobank to assess disease incidence.

In practice

• Utilize AI-IR for early identification of high-risk individuals.
• Implement focused screening for diabetes and specific cancers.

Topics

• Machine Learning Models
• Insulin Resistance
• Cancer Risk Prediction
• UK Biobank
• Diabetes Prediction

Best for: AI Scientist, AI Researcher, Data Scientist, Research Scientist

Related on AIssential

• Mapping neighbourhood-level drivers of type 2 diabetes for precision public health using predictive and causal machine learning — Neighbourhood-level factors, identified via ML and causal inference, predict Type 2 diabetes risk and inform targeted public health.
• This AI Model Predicts Breast Cancer Years Before Humans — Mirai uses AI to predict future breast cancer risk from mammograms, enabling personalized screening.
• Beyond the Blood Draw: Explainable Machine Learning for Non-Invasive Dysglycemia Risk Screening — Explainable ML models can non-invasively screen for dysglycemia risk using readily available data, outperforming existing clinical scores.
• 🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik — Improving cancer drug success requires advanced AI models for precise patient selection based on multimodal biological data.
• A deep learning–based digital biopsy for predicting early recurrence in gastric cancer — The RSA model combines deep learning with clinical data to predict gastric cancer recurrence.
• AI identifies early risk patterns for skin cancer — AI models can predict melanoma risk with high accuracy using routine healthcare data, enabling targeted screening.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.