New genetic risk score predicts diabetes and obesity across ancestries - and who will end up on GLP-1 drugs

Body mass index is a blunt instrument. It captures one dimension of metabolic health - weight relative to height - and misses nearly everything else. Two people with identical BMIs can have vastly different risks for diabetes, cardiovascular disease, and stroke, depending on how their bodies handle fat distribution, insulin signaling, glucose regulation, and dozens of other metabolic processes.

A team at Mass General Brigham set out to build a genetic risk score that captures that complexity. Their metabolic polygenic risk score (PRS), published in Cell Metabolism, integrates genetic variants associated with 20 different metabolic traits - not just BMI, but fat distribution patterns, insulin sensitivity, glucose control, and other measures of metabolic function. The training data came from genome-wide association studies encompassing over 8.5 million participants globally.

Two scores, twenty traits

The researchers built two versions: one optimized for obesity prediction and another for type 2 diabetes. Both look beyond the usual suspects. Rather than relying primarily on BMI-associated variants, the scores incorporate genetic signals linked to where the body stores fat, how efficiently it processes glucose, how insulin secretion responds to meals, and other metabolic characteristics that BMI alone cannot capture.

This broader genetic view produced scores that outperformed existing PRS models in predicting both conditions. But the more striking finding was downstream: the scores predicted not just who would develop obesity or diabetes, but who would go on to experience cardiovascular disease and stroke as consequences.

Predicting treatment trajectories

Among individuals who were initially healthy, those in the high-PRS group were roughly twice as likely to later receive GLP-1 receptor agonist medications or undergo bariatric surgery compared to those with mid-range scores, during a median follow-up of 5.5 years. This is a meaningful finding because it suggests the score can identify people headed for aggressive intervention before they arrive there - potentially opening a window for earlier, less intensive prevention.

Co-senior author Akl Fahed, an interventional cardiologist at Massachusetts General Hospital, framed the ambition directly: early identification of people likely to have a worsening metabolic trajectory, before they develop conditions, can improve both prevention and clinical interventions.

Working across ancestries

A persistent problem with polygenic risk scores has been their poor transferability across populations. Scores developed primarily in European cohorts often lose predictive power in people of African, East Asian, or South Asian descent - the very populations that in many cases face the highest metabolic disease burden.

The Mass General Brigham team addressed this by deliberately incorporating multi-ancestry GWAS data, with particular attention to non-European populations. The resulting scores surpassed prior PRS models when tested in African, East Asian, and South Asian individuals. This improvement is not just statistical; it has direct equity implications. A risk score that works only in European-descent populations would entrench rather than reduce existing health disparities.

From population statistics to individual prediction

Polygenic risk scores remain probabilistic, not deterministic. A high score does not guarantee disease, and a low score does not prevent it. Environment, behavior, access to healthcare, and chance all play substantial roles. The score also cannot currently distinguish between genetic subtypes of type 2 diabetes or obesity - distinctions that may matter for treatment selection.

The researchers plan to refine the scores further, working toward genetic subtyping that could improve patient classification and stratification for clinical trials. If different genetic profiles respond differently to specific interventions - a plausible but unproven hypothesis - then a more granular score could guide treatment decisions, not just risk predictions.

Clinical implementation also faces practical barriers. Genomic data is not routinely collected in primary care, and integrating PRS results into clinical workflows requires infrastructure, education, and clear guidelines for how scores should influence decision-making. A score that identifies someone as high-risk is only useful if it triggers an action that changes their outcome.

But the underlying advance is real. By moving beyond BMI-centric genetics and incorporating the full spectrum of metabolic function, this PRS provides a more complete picture of inherited metabolic risk - one that works across the world's diverse populations rather than just those who have been most studied.