Overview
- This model, published Thursday in Nature Medicine, estimates a person’s 10-year risk of 18 conditions using 20 routine measures such as blood tests, blood pressure and medical history.
- Researchers used interpretable machine learning on data from about 200,000 UK Biobank participants with a BMI of 27 or higher, narrowing more than 2,000 variables to the strongest predictors.
- The tool, called Obscore, was externally validated in the Genes & Health and EPIC-Norfolk cohorts, and the authors call for wider testing and cost-effectiveness trials before clinical use.
- A sizable share of those flagged as highest risk were classed as overweight rather than obese by BMI, pointing to earlier identification and possible use to prioritize GLP-1 treatments like semaglutide or tirzepatide.
- Experts caution that some inputs are not routinely collected in NHS care, UK Biobank volunteers are healthier than average, and real-world rollout would need checks on data availability, equity and practicality.