Table 3.

Performance metrics for all classifiers

	Train	Test	SVM results
	AUC	AUC	F1	BAC	Acc	Sens	Spec	PPV	NPV	PSI	PLR	NLR	DOR
Multi	78.86 (2.81)	67.74 (13.86)	67.15	63.05	65.72	57.45	68.65	41.64	81.57	23.21	1.83	0.62	2.96
Clinical	73.59 (1.64)	62.81 (11.14)	64.30	60.10	62.62	54.73	65.44	37.04	79.79	16.83	1.58	0.69	2.29
Bio	63.12 (1.04)	57.09 (11.47)	55.80	51.47	53.64	46.90	56.03	27.44	74.78	2.22	1.07	0.95	1.13
sMRI	64.53 (1.74)	56.75 (10.46)	56.94	52.03	55.31	45.00	59.06	29.16	74.96	4.12	1.10	0.93	1.18
Cardio	61.44 (1.48)	56.03 (13.12)	56.15	54.22	53.65	55.55	52.90	29.62	77.18	6.80	1.18	0.84	1.40
Serum	60.70 (0.75)	54.43 (8.31)	51.81	50.00	63.97	20.00	80.00	5.41	58.57	−36.02	1.00	1.00	1.00
PGRS	59.72 (1.26)	50.52 (13.62)	53.32	50.55	50.85	49.91	51.18	27.88	73.30	1.18	1.02	0.98	1.04

All classifiers used a Linear Support Vector Machine with Platt scaling, only predictor modalities varied across models. Mean (SD) scores from the outer 10-fold cross-validation loops are presented. Model abbreviations: Multi our multimodal model (all biomarker modalities, clinical, and demographic variables), Clinical clinical and demographic predictors only, Bio model with all biomarker modalities (no clinical or demographic data), sMRI structural imaging predictor model only, Cardio electrocardiography and bioelectrical impedance analysis predictor model only, Serum blood biomarkers only, PGRS PGRS model only. Metric abbreviations: AUC area-under-the-curve, F1 Harmonic mean of Sens + Spec, BAC balanced accuracy, Acc accuracy, Sens sensitivity, Spec specificity, PPV positive predicted value, NPV negative predicted value, PSI prognostic summary index, PLR positive likelihood ratio, NLR negative likelihood ratio, DOR diagnostic odds ratio