Body mass index is, when you strip it back, a Victorian arithmetic formula. Weight divided by height squared. It was never designed to tell you whether you were going to develop type 2 diabetes or die early of cardiovascular disease, and yet here we are, a century and a half later, still reaching for the bathroom scales to make clinical decisions about metabolic health. A team of researchers at Mass General Brigham reckons it can do considerably better than that.
The alternative they’ve developed is called MetPRS, a genetic risk calculator built not from one disease, not from one measurement, but from 20 different metabolic traits simultaneously, drawing on data from over 8.5 million people worldwide. It outperforms existing prediction models across six different ancestral populations. And it can apparently tell, years in advance, who will end up needing bariatric surgery or GLP-1 drugs like Ozempic.
The basic idea behind polygenic risk scores is perhaps familiar by now: your genome contains thousands of tiny variants, each nudging your disease risk slightly up or slightly down, and if you add them all together you get something more predictive than any single variant alone. Most existing scores for obesity and type 2 diabetes do this for one condition at a time. What the Mass General team realised, and what turns out to matter quite a lot, is that obesity and diabetes share substantial overlapping biology. Fat distribution, insulin sensitivity, glucose regulation: these things are entangled with each other in ways that single-disease scores simply miss.
Not a small oversight, as it happens.
So Min Seo Kim and colleagues built a score that integrates genetic data from conditions ranging from BMI (the old faithful) to more granular measurements: waist circumference adjusted for overall body mass, hip circumference, visceral adipose tissue deposits, subcutaneous fat, insulin sensitivity index, and others. Twenty traits in total, each illuminating a slightly different corner of metabolic function. Feed all of that into a model trained on the UK Biobank (one of the largest repositories of linked genetic and clinical data in existence) and you get a score that is, in a meaningful sense, biologically enriched. It has absorbed more of the underlying complexity of what metabolic dysfunction actually is.
“In the future, this genomic approach could complement established clinical risk factors to inform patient care and preventative strategies,” said Kim, a physician-scientist and co-first author on the paper, which was published in Cell Metabolism.
The validation work is where things get genuinely interesting. The team tested their score not just in the UK Biobank population (which skews heavily European) but in three additional multi-ethnic cohorts totalling up to 300,000 people. In African, East Asian, and South Asian individuals, MetPRS outperformed every previous polygenic risk score for these conditions, which reflects a real shift: earlier models were built predominantly on European genetic datasets and tended to transfer poorly across ancestries. By using multi-ancestry genome-wide association data from the outset, the team has made something that’s more portable. Whether it’s portable enough for routine clinical use across diverse global populations is a different question, and one the researchers are careful not to overclaim.
The most striking clinical finding involves what happens at the extreme end of the score distribution. Individuals in the top decile (the 10% with highest MetPRS) who were still metabolically healthy at baseline were roughly twice as likely, compared with people in the middle quintile, to go on to start GLP-1 receptor agonist medications or receive bariatric surgery over the following five-and-a-half years or so. This isn’t just a statistical curiosity. GLP-1 drugs are expensive, surgery is serious, and knowing who is on a trajectory toward needing them, potentially years before the clinical signs emerge, could meaningfully change how and when preventative action is taken.
BMI wouldn’t have flagged most of those people. That’s the point.
Akl Fahed, an interventional cardiologist at Massachusetts General Hospital and co-senior author, put the ambition plainly: “Early identification of people who are likely to have a worse trajectory of poor metabolic health, before they even develop these conditions, can help us improve prevention and clinical interventions. That is how we can cure disease, and that is the bold mission that we are after.”
There are limits, of course. Polygenic risk scores remain expensive and logistically complex to run at population scale. They don’t explain why an individual will develop disease, only that the odds are somewhat stacked. And a high score is not destiny: lifestyle, environment, and chance still matter enormously. The researchers are also candid that the score does not yet replace BMI in clinical practice; the hope is that it eventually adds another layer of information on top.
What the study perhaps changes most is the conceptual frame. Metabolic disease has long been thought of as a cluster of separate conditions with some shared features. MetPRS treats it as something more unified: a single underlying susceptibility, expressed differently depending on your particular genetic architecture and life circumstances. If that framing holds up as the score is refined and validated further, it could shift how clinicians think about which patients to watch closely, when to intervene, and whether a number on a bathroom scale was ever really the right thing to be looking at.
DOI / Source: https://doi.org/10.1016/j.cmet.2026.02.009
Frequently Asked Questions
A polygenic risk score adds up the effects of thousands or even millions of common genetic variants across your genome, each of which has a tiny individual effect on disease risk. This is different from a test for a single high-impact mutation (such as BRCA1 for breast cancer), which identifies a rare variant with a large effect. Most common diseases like type 2 diabetes and obesity are polygenic, meaning they’re shaped by many genes working together, and a polygenic risk score is designed to capture that complexity rather than look for any single culprit.
Most earlier polygenic risk scores were trained predominantly on datasets of European ancestry, which meant they worked less well when applied to people from African, East Asian, or South Asian backgrounds. Genetic variants and their frequencies differ between populations, so a score calibrated on one group can give systematically different, and often less accurate, estimates in others. By incorporating multi-ancestry genome-wide association data from the outset, MetPRS is better calibrated across populations, though validating performance across every ancestry remains ongoing work.
Not yet routinely, though the trajectory is in that direction. Running a polygenic risk score requires genomic data, which most people don’t have on file with their GP, and the infrastructure for integrating these scores into clinical workflows is still being built. The Mass General team envisions MetPRS eventually sitting alongside standard measures like blood pressure, cholesterol, and yes, BMI, adding a layer of genetic context. The immediate near-term application is more likely to be in research settings, clinical trials (helping enrol patients who are most likely to benefit), and specialist preventive care programs.
GLP-1 receptor agonists are a class of drugs originally developed for type 2 diabetes that have become widely used for weight management. They include semaglutide (marketed as Ozempic and Wegovy) and related compounds. They work by mimicking a gut hormone that regulates appetite and blood sugar. They’re effective but expensive, and demand has significantly outstripped supply globally. Knowing years ahead of time who is on a metabolic trajectory likely to require these medications could allow for earlier, cheaper preventive interventions, or ensure the people most likely to benefit get access sooner.
No. A high polygenic risk score increases your statistical likelihood of developing these conditions, but it is not a deterministic prediction. Environment, diet, physical activity, and chance all shape outcomes, sometimes dramatically. The roughly twofold increased risk seen in the top decile is a population-level statistical finding: of a large group of people with high scores, more will develop disease than from a group with average scores, but many high-scorers will not. The score is most useful as a tool for prioritising who gets closer monitoring or earlier preventive support, not as a verdict on any individual’s future.
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.
Key Takeaways
- Body mass index (BMI) lacks precision for predicting metabolic health; researchers at Mass General Brigham developed MetPRS to improve predictions.
- MetPRS uses genetic data from 20 metabolic traits, outperforming traditional models in diverse ancestral populations.
- This tool helps identify individuals at risk for conditions requiring interventions, such as bariatric surgery or GLP-1 drugs, years in advance.
- Unlike BMI, MetPRS offers a more integrated view of metabolic health, recognizing the shared biology of obesity and diabetes.
- Challenges remain, including cost and complexity of implementation, but MetPRS may redefine clinical approaches to metabolic disease.