OBSCORE model estimates 10-year risk across 18 obesity-related outcomes

What happened

Researchers developed OBSCORE to estimate the 10-year risk of 18 outcomes associated with overweight and obesity. The model uses 20 shared clinical features, including health history, symptoms, smoking status, common blood tests, and body measurements. Its aim is to distinguish risk among people who can have a similar BMI but different health profiles.

The development analysis included 197,264 UK Biobank participants with a BMI of at least 27. Their median age at enrollment was 58, and 9.3% had type 2 diabetes at baseline. The team began with more than 2,300 possible predictors, removed highly correlated measures, and used machine-learning methods to select a smaller clinical set that could be applied across the 18 outcomes.

The outcomes included cardiovascular death, heart attack, stroke, type 2 diabetes, chronic kidney disease, sleep apnea, fatty liver disease, gout, hypertension, gallstones, joint disease, and several other cardiovascular and mechanical complications. In the cardiovascular-mortality analysis, 10-year incidence was 0.1% for the bottom risk quintile and 5.7% for the top quintile.

The investigators externally evaluated versions of the model in EPIC-Norfolk and Genes & Health. Eighteen of the 20 features were available for 14 outcomes in 2,112 EPIC-Norfolk participants. A 15-feature version was tested for incident type 2 diabetes in 1,740 British Pakistani and British Bangladeshi participants in Genes & Health. The paper reports retained discrimination, although not every outcome or feature could be evaluated in both cohorts.

A separate analysis applied 12 available features to SURMOUNT-1 trial data. Predicted risks decreased after 72 weeks in the tirzepatide groups. That analysis measured changes in model-generated risk scores, not observed prevention of 18 clinical outcomes during the trial.

What it means

OBSCORE illustrates why BMI alone can miss meaningful differences in future health risk. In the study, people within the same BMI category had a wide spread of predicted risks, and some participants with overweight appeared in the highest-risk groups.

A risk model could eventually help clinicians and health systems identify people who may benefit most from earlier or more intensive intervention. The authors describe possible use in referral and decision support, including medication, nutrition, behavioral care, or surgery. That potential still depends on further validation, calibration, implementation work, and evidence that using the score improves decisions and outcomes.

What it does not mean

OBSCORE is not a diagnosis and does not predict an individual's future with certainty. It does not replace an evaluation of current disease, symptoms, treatment preferences, contraindications, or surgical candidacy. A high estimate does not prove that a specific medicine or operation will prevent the predicted event.

UK Biobank participants are not fully representative of the general population. External testing used fewer than 20 features and covered only the outcomes available in each cohort. The paper also notes that the model overestimated risk for some outcomes before recalibration.

The SURMOUNT-1 exercise does not validate OBSCORE as a treatment-response calculator. It found changes in predicted risk measures after treatment, while the trial was not long enough to observe 10-year complication rates. The score should not be used as an insurance rule or a reason to start, stop, or rank treatments without clinical review.