Table 2.

Study characteristics and predictive performance of studies reporting on a prediction model developed on one or more ad-hoc datasets.

Study characteristics			ML and DL modelling			Performance						Validation
Authors	No. cases/controls	Database	Algorithm	Features	Prediction interval	Sensitivity	Specificity	PPV	NPV	AUC	Accuracy	Internal	External
Acharya et al.31	20 SCD/18 controls	SCDH; NSRD	DT, KNN, SVM∗	ECG (DWT decomposition, non-linear feature extraction: FD, entropy)	2 min	100.0	97.2	97.5	100.0	N/A	98.7	10-fold CV	N/A
Acharya et al.30	20 SCD/18 controls	SCDH; NSRD	KNN∗, PNN, SVM, DT	HRV (RQA, non-linear)	4 min	94.4	80.0	N/A	N/A	N/A	86.8	10-fold CV	N/A
Alfarhan et al.32	20 SCD/20 controls	SCDH; NSRD	KNN∗ and LDA	HRV (frequency-domain, time-domain and QRS complex features and VR features)	10 min	N/A	N/A	N/A	N/A	N/A	97.0	10-fold CV	N/A
Amezquita-Sanchez et al.33	23 SCD/18 controls	SCDH; NSRD	Enhanced probabilistic NN	WPT to decompose data in frequency bands, non-linear feature extraction (homogeneity index)	20 min	N/A	N/A	N/A	N/A	N/A	95.8	50% train, 20% validation, 30% test	N/A
Bayasi et al.34	16 VA/18 controls	NSRD; SCDH; CUDB; AHADB	LDA	Segments (PQ, PS, RT, QP, SP)	3 h	98.9	N/A	98.4	N/A	99.9	99.1	10-fold CV	N/A
Calderon et al.35	16 SCD/20 controls	SCDH and Fantasia database	DT, KNN, SVM, LR and ANN∗	Segments (PS,Q T, ST, PR and RR)	2 min	91.0	93.0	N/A	N/A	N/A	92.0	8-fold CV	N/A
Cappielo et al.36	32 VA/32 controls	CUDB; PTBDB	Hybrid prediction index	Phase-space portraits characteristics	354 ECG beats	96.9	100.0	100.0	97.0	N/A	98.4	LOOCV	N/A
Devi et al.37	18 SCD/18 controls/15 CHF	SCDH; BIDMC Congestive Heart Failure, NSRD	DT, KNN∗, SVM	HRV (CWT, time-domain, frequency-domain, time-frequency and non-linear)	10 min	75.0	87.5	75.0	75.0	N/A	83.33	75% train, 25% test	N/A
Ebrahimzadeh et al.39	35 SCD/35 normal)	SCDH; NSRD	MLP, KNN, SVM, ME classifier∗	HRV (time-domain, frequency-domain, time-frequency, non-linear)	13 min	82.2	85.7	83.3	85.3	N/A	82.9	train 70%, test 30%	N/A
Ebrahimzadeh et al.40	23 SCD/18 controls	SCDH; NSRD	MLP	HRV (time-domain, frequency-domain, time-frequency, non-linear)	12 min	82.7	85.1	84.7	83.1	N/A	83.9	LOOCV	N/A
Ebrahimzadeh et al.42	35 SCD/35 normal	SCDH; NSRD	MLP∗ and KNN	HRV (time-domain, frequency-domain, time-frequency and non-linear (DFA, Poincaré))	4 min	83.8	16.0	84.0	83.8	N/A	83.9	LOOCV	N/A
Ebrahimzadeh et al.41	35 SCD/35 controls	SCDH; NSRD	ANN	HRV (time-domain, frequency-domain, time-frequency-domain)	2 min	N/A	N/A	N/A	N/A	N/A	91.23	LOOCV	N/A
Fairooz et al.43	18 SCD/18 normal	SCDH; NSRD	SVM	CWT transformation, subsequent feature extraction (intervals, amplitudes, TWA)	30 min	100.0	100.0	100	100	N/A	100	Train 77.78%, test 22.22%	N/A
Fujita et al.44	20 SCD/18 normal	SCDH; NSRD	SVM∗, DT and KNN	HRV (DWT, non-linear features: Renyi entropy, fuzzy entropy, Hjorths parameters and Tsallis entropy)	4 min	95.0	94.4	95.0	94.4	N/A	94.7	10-fold CV	N/A
Houshyarifar et al.47	23 SCD/36 normal	SCDH; NSRD	KNN, SVM∗	HRV (non-linear, spectrum HOS features and time-domain)	4 min	N/A	N/A	N/A	N/A	N/A	94.5	10-fold CV	N/A
Houshyarifar et al.46	23 SCD/36 normal	SCDH; NSRD	KNN, SVM∗	HRV (non-linear recurrence and Poincaré plot)	4 min	84.25	96.8	N/A	N/A	N/A	93.3	10-fold CV	N/A
Jeong et al.48	58 VF/60 controls	CUDB, MVTDB, PAFDB and NSRDB	ANN	HRV (time-domain and non-linear Poincare)	80 s	N/A	N/A	N/A	N/A	N/A	88.18	10-fold CV	N/A
Joo et al.49	78 ICD patients	MVTB	ANN (VF)	HRV (time-domain, frequency-domain and non-linear Poincaré)	5 min	88.9	92.9	72.7	97.5	N/A	92.9	66% train 33% test	N/A
Khazaei et al.50	23 SCD/18 controls	SCDH; NSRD	DT∗, KNN, NB and SVM	HRV (non-linear: RQA and increment entropy)	6 min	95.0	95.0	N/A	N/A	N/A	95.0	10-fold CV	N/A
Lai et al.52	18 SCD/18 controls	SCDH; NSRD	KNN∗, DT, NB	Ventricular repolarisation features	60 min	99.5	98.3	98.3	N/A	N/A	98.9	5-fold CV	N/A,
Lai et al.51	28 SCD/18 controls	AHADB; SCDH; NSRD	KNN, DT, NB, SVM and RF∗	Ventricular repolarisation features∗	30 min	99.8	99.0	99.4	99.6	N/A	99.5	5-fold CV	N/A
Lopez-Caracheo et al.54	9 SCD/9 controls	SCDH; NSRD	HFD, BD, and KFD algorithms	HRV (Non-linear: Katz, Higuchi and Box Dimension)	14 min	N/A	N/A	N/A	N/A	N/A	91.4	50% train, 50% test	N/A,
Mandala et al.55	22 VA/18 controls	NSRD; VFDB	SVM, NB∗, DT	HRV (time-domain) and QRS complex features	25 min	93.3	86.7	N/A	N/A	N/A	N/A	5-fold CV	N/A
Mirhoseini et al.57	19 SCD/18 controls	SCDH; NSRD	SVM∗, DT	HRV (time-domain, frequency-domain, time-frequency, non-linear)	1 min	N/A	89.5	87.5	81.0	N/A	83.2	10-fold CV	N/A
Murugappan et al.58	18 SCD/18 controls	SCDH; NSRD	SVM,∗ subtractive fuzzy clustering, and neuro-fuzzy classifier	HRV (Non-linear features: Largest Lyapunov Exponent/approximate entropy/Sample entropy/Hurst exponent)	5 min	97.1	97.1	100.0	97.6	N/A	100.0	10-fold CV	N/A
Murugappan et al.59	20 SCD/18 controls	SCDH; NSRD (40 vs 36 holter)	KNN∗ and fuzzy classifier	HRV (time-domain)	5 min	92.2	95.3	95.4	N/A	N/A	93.71	10-fold CV	N/A
Murukesan et al.60	23 SCD/18 controls	SCDH; NSRD	SVM∗, PNN	DWT and HRV feature extraction (time-domain, frequency-domain, time-frequency, non-linear)	2 min	93.3	100.0	N/A	N/A	N/A	96.4	train 70% test 30%	N/A
Parsi et al.61	78 ICD carriers	MVTB (135 pre VT/126 controls)	SVM, RF and KNN∗	HRV (time and frequency-domain, HOS features, non-linear Poincaré)	5 min	88.8	94.2	N/A	N/A	N/A	91.5	LOOCV	N/A
Riasi et al.63	40 VT/40 controls	SCDH; NSRD and CUDB	SVM	Morphological features (area under ascending/descending/total T-wave and R-wave, beat to beat correlations, intervals)	20 s	88.0	100.0	N/A	N/A	N/A	94.0	75% train 25% test	N/A
Shi et al.65	20 SCD/18 controls	SCDH; NSRD	KNN	HRV (EMD for entropy parameters, time-domain and frequency-domain)	14 min	97.5	94.4	N/A	N/A	N/A	96.1	10-fold CV	N/A
Shen et al.69	23 SCD/20 controls	SCDH and database	LSM∗, DBNN, BPNN	HRV (FFT and frequency-domain)	2 min	75.0	N/A	N/A	N/A	N/A	87.5	46% train, 56% test	N/A
Taye et al.66	78 ICD carriers	MVTDB (135 pre VT/126 controls)	1-D CNN	N/A	60 s	83.2	86.4	N/A	N/A	78.0	84.6	10-fold CV	N/A
Taye et al.67	27 VF/28 controls	CUDB, PAFDB, NSRDB	Fully connected ANN	HRV (time-domain, frequency-domain, non-linear Poincare), QRS complex features	30 s	98.4	99.0	N/A	N/A	99.0	98.6	10-fold CV	N/A
Tseng et al.68	81	CUDB	2D CNN, 2D-STFT	N/A	5 min	98.0	N/A	N/A	N/A	N/A	88.0	80% train and 20% validation	Two real cases as validation
Tsjui et al.70	20 SCD/20 controls	Not specified	R-LLGMn	HRV (time-domain)	5 min	N/A	N/A	N/A	N/A	90.0	82.5	LOOCV	N/A
Vargas-Lopez et al.71	23 SCD/18 controls	SCDH; NSRD	MLP	EMD, subsequent entropy and fractal dimension feature extraction	25 min	N/A	N/A	N/A	N/A	N/A	94.0	45% and 55% validation	N/A,

AHADB = AHA Database for Evaluation of Ventricular Arrhythmia Detectors, ANN = artificial neural network, AUC = area under the curve, BPNN = back-propagation neural network, CNN = convolutional neural network, CUDB=Creighton University ventricular tachyarrhythmia database, CV = cross validation, CWT=Continuous Wavelet Transform, DFA = detrended fluctuation analysis, DWT = Discrete wavelet transform, DBNN = decision-based neural network, DT = Decision Tree, ECG = electrocardiography, EMD = empirical mode decomposition, EMG = intracardiac electrogram, FD= Fractal Dimension, FFT = fast Fourier transform, HOS = higher order spectral, HRV = heart rate variability, KNN = k-nearest neighbour, LMS = least mean square, MVTDB = Spontaneous Ventricular Tachyarrhythmia Database, MLP = multi-layer perceptron, ME = maximum entropy, NSRBD = normal sinus rhythm database, PPV = positive predictive value, PAFDB = paroxysmal atrial fibrillation prediction challenge database, PNN = probabilistic neural network, LOOCV = leave-one-out cross validation, LDA = linear discriminant analysis, NB = naïve Bayes, NPV = negative predictive value, RF = random forest, R-LLGMn = recurrent log-linearised Gaussian mixture network, SCD = sudden cardiac death, SCDH = MIT-BIH SCD Holter database, SVM = support vector machine, RQA = recurrence quantification analysis, VFDB = Malignant Ventricular Arrhythmia Database, VF = ventricular fibrillation, VT = ventricular tachycardia, VR=Ventricular repolarisation, WPT = wavelet packet transform, 2D-STFT = two-dimensional short-time Fourier transform.