Fig 1.
The data treatment and study design. a. clinical raw data treatment protocol; b. the design of prediction model development.
CHD: coronary heart disease; RF: random forest model; GBDT: gradient boosting decision tree; cat: model trained with all-categorical data; mix = model trained with mixture of numerical and categorical data; AUROC: area under receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value.
Table 1.
Typical baseline characteristics of patients with CHD and controls by mean value and standard deviation or percentage (in bracket). CHD: coronary heart disease.
Table 2.
The prediction performance from different models.
Fig 2.
The receiver-operating characteristics curves for CHD prediction models.
TLML: two-layer machine learning model; cat: model trained with all-categorical data; mix: model trained with mixture of numerical and categorical data.
Fig 3.
The Brier scores of the prediction from PCEs and TLML with different thresholds.
TLML: two-layer machine learning model, PCEs: pooled cohort equations.
Fig 4.
The receiver-operating characteristics curves for PCEs and RCM.
TLML: two-layer machine learning model; RCM: reduced complexity model.