Classifying coronary heart disease risk from NHANES survey data (2011-2018), with a full leakage audit and calibration check [P]

A GitHub repository details the classification of coronary heart disease (CHD) risk using NHANES survey data from 2011-2018. The project achieved ROC-AUC scores of 0.875 and PR-AUC of 0.239 with logistic regression, random forest, and gradient boosting models. Age was a significant predictor, yielding an AUC of 0.83, with blood pressure, cholesterol, and body size explaining additional variance. The positive predictive value (PPV) was 0.13, reflecting the rarity of CHD in the dataset, a point explicitly addressed in the report.

Unlike many predictive models that omit limitations, this report explicitly details a full leakage audit and calibration check, directly addressing the low positive predictive value of 0.13.