Extended Data Fig. 2: Evaluating the calibration and predictive value of feature categories for the meta-prediction model in the UK Biobank.

This figure evaluates the performance of the meta-prediction model compared to existing clinical risk scores within the test set of the incident CAD cohort (n = 33,419), highlighting both the model’s calibration and the predictive contribution of distinct feature categories. a. Calibration plot showing the predicted vs. observed 10-year CAD risk by decile. Brier’s scores are provided as a summary calibration measure. Conventional risk scores were re-calibrated using the same approach within the UK Biobank training cohort. b. Predictive performance of individual feature categories and their combinations in the meta-prediction model. The left panel illustrates the Area Under the Receiver Operating Characteristic (AUROC) for various models. The right panel shows the Area Under the Precision-Recall Curve (AUPRC). The models compared include the full meta-prediction model, partial models restricted to a single feature class and feature class combination, including 15 meta-features; 22 polygenic risk scores (PRSs) and 13 modifiable risk factors (MFs); 13 MFs; sex and age with 12 PRSs; 12 PRSs; and sex and age alone. Abbreviation: MF: Modifiable factors.