Performance of a foundation model to generate thousands of predicted outcomes across multiple use cases

PhenOM(R) completes thousands of predictive tasks across clinical domains, with 80 percent reaching AUROC ≥ 0.70 and 30 percent reaching AUROC ≥ 0.80.

• Why it matters: Healthcare needs systematic, rapid generation of multi-condition predictions from one foundation-model architecture, but this remains uncommon across diagnoses, events, and outcomes.
• What they did: PhenOM(R) was evaluated on data excluded from training, using AUROC across four outcome categories (Future Risk, Screening, Care Opportunity, Treatment Effect) spanning multiple clinical domains.
• The result: Performance varied by outcome and clinical domain, and the model supports screening for undiagnosed conditions, forecasting future conditions and events, treatment effects, and care opportunities at patient and population levels.

AI summary of the abstract · may contain errors · not medical advice · read the paper.