Table 1.
Prevalence of outcomes stratified by sex and risk groups.
Fig 1.
AUC and average calibration error plots in the hold-out.
Comparison of AUC and average calibration error (average absolute observed vs. predicted difference over 10 deciles of risk) in the hold-out dataset for all methods tested for this study for each of the outcomes/strata combinations. Number is red represents outcome prevalence. Abbreviations: ANN, artificial neural network; AUC, area under the curve; CAD, coronary artery disease; GBM, gradient boosted machines; LR, logistic regression; RF, random forest; SVM, support vector machine.
Table 2.
Performance metrics in the holdout dataset for prediction models created using the artificial neural network strategy in each risk/sex strata.
Fig 2.
Calibration metrics in the hold-out dataset for the prediction models.
Calibration metrics in the holdout dataset for the prediction models created using the artificial neural network strategy in each risk/sex strata. The observed prevalence of outcomes in each decile is depicted with their 95% confidence intervals in red bar while the predicted prevalence is depicted as a black line/circle. Abbreviation: CAD, coronary artery disease.
Fig 3.
Heatmap of patient risk profiles for CAD stratified by risk strata and outcomes.
Heatmap (with hierarchical clustering) of patient risk factor profile for coronary artery disease stratified by risk strata and outcomes. Only factors that are either differentially expressed between groups and outcomes or those that substantially contributed to the final prediction models are reported, some highly correlated risk factors are not shown to improve readability. Abbreviations: ACE, angiotensin-converting enzyme; ADRB2, adrenoreceptor beta 2; AF, affected; ALT, alanine transaminase; AST aspartate aminotransferase; CEP, centrosomal protein; CLCNKA, chloride channel protein CLC-Ka; HR, high-risk; LR, low-risk; Meds, medication; N, negative; P positive; SNP, single nucleotide polymorphism; WWOX, WW domain containing oxidoreductase. (Bottom legend: P = positive, N = negative, F = female, M = male, LR = low risk, HR = high risk, AF = affected).
Fig 4.
Heatmap of patient risk profiles for HF stratified by risk strata and outcomes.
Heatmap (with hierarchical clustering) of patient risk factor profile for heart failure stratified by risk strata and outcomes. Only factors that are either differentially expressed between groups and outcomes or those that substantially contributed to the final prediction models are reported, some highly correlated risk factors are not shown to improve readability. Abbreviations: ACE, angiotensin-converting enzyme; ADRB2, adrenoreceptor beta 2; AF, affected; ALT, alanine transaminase; AST aspartate aminotransferase; CEP, centrosomal protein; CLCNKA, chloride channel protein CLC-Ka; HR, high-risk; LR, low-risk; Meds, medication; N, negative; P positive; SNP, single nucleotide polymorphism; WWOX, WW domain containing oxidoreductase. (Bottom legend: P = positive, N = negative, F = female, M = male, LR = low risk, HR = high risk, AF = affected).
Fig 5.
Feature importance plot averaged over all prediction models from ANN modeling strategy.
Feature importance averaged over all prediction models created using the artificial neural network strategy. Only features with an importance index above the 90th percentile are reported. Features with an asterisk are relevant only to a single sex. Abbreviations: GERD, gastroesophageal reflux disease; GGT, gamma-glutamyl transferase.