Table 1. Characteristics of the included studies.

Study	True positive	False positive	False negative	True negative	ECG data	Machine learning model	Dataset	Definition of heart failure	Sample size	Number of subjects	Origin
CHF: congestive heart failure; cML: classical machine learning; DL: deep learning; ECG: electrocardiogram; HRV: heart rate variability; LVSD: left ventricular systolic dysfunction.
Pecchia, et al.[20]	26	0	3	54	HRV	cML	Public	CHF	< 1000	83	America
Chen, et al.[22]	36	0	1	22	HRV	cML	Public	CHF	< 1000	116	America
Acharya, et al.[21]	779	15	24	840	HRV	cML	Public	CHF	> 1000	73	America
Acharya, et al.[23]	29,660	794	340	79,206	ECG signals	DL	Public	CHF	> 1000	73	America
Attia, et al.[28]	3567	6977	564	41,762	Raw ECG	DL	Patient-level	LVSD	> 1000	97,829	America
Hussain, et al.[24]	39	5	3	69	HRV	cML	Public	CHF	< 1000	116	America
Lih, et al.[25]	29,811	213	189	120,055	ECG signals	DL	Public	CHF	> 1000	262	America & Europe
Adedinsewo, et al.[29]	121	183	43	1259	Raw ECG	DL	Patient-level	LVSD	> 1000	1606	America
Yang, et al.[27]	7348	642	152	11,858	ECG signals	DL	Public	CHF	> 1000	40	America & Europe
Attia, et al.[30]	7	111	19	4140	Raw ECG	DL	Patient-level	LVSD	> 1000	4277	America
Jahmunah, et al.[26]	29,790	241	210	120,027	ECG signals	DL	Public	CHF	> 1000	262	America & Europe