Table 2.

Ten-fold cross-validation results from penalized logistic regression models using the first 50 PCA modes (PCA50) and from the reference baseline mass/volumes (MassVol) measurement models

Factor	Model	AUC	Sensitivity	Specificity	Precision	F1-score	npv	Cutoff
Cholesterol	MassVol	0.587	0.704	0.451	0.346	0.464	0.787	0.495
	PCA50*	0.751	0.792	0.579	0.437	0.564	0.871	0.485
Smoking	MassVol	0.594	0.856	0.287	0.334	0.480	0.826	0.490
	PCA50*	0.699	0.525	0.795	0.517	0.521	0.800	0.511
Diabetes	MassVol	0.652	0.740	0.554	0.342	0.468	0.872	0.494
	PCA50*	0.817	0.704	0.781	0.502	0.586	0.894	0.504
Previous MI	MassVol	0.733	0.699	0.663	0.235	0.351	0.937	0.490
	PCA50*	0.853	0.677	0.911	0.529	0.594	0.950	0.518
Angina	MassVol	0.671	0.700	0.637	0.196	0.307	0.944	0.497
	PCA50*	0.773	0.663	0.763	0.262	0.376	0.947	0.493
Obesity	MassVol	0.682	0.675	0.598	0.638	0.656	0.637	0.494
	PCA50*	0.839	0.760	0.763	0.771	0.766	0.752	0.500
Hypertension	MassVol	0.651	0.693	0.543	0.718	0.705	0.514	0.495
	PCA50*	0.770	0.752	0.663	0.789	0.770	0.615	0.494

Statistical tests were performed between MassVol and PCA50. Significant differences (p < 0.005) are represented with an asterisk

PCA principal component analysis, MI myocardial infarction