Table 3.

Mean absolute error (MAE) and accuracy (ACC) ± standard deviation measured on the replication dataset when blanking specific waves of a median heartbeat.

Variable	Metric	Median ECG	Blanking P-wave	Blanking QRS complex	Blanking T-wave
QT interval (ms)	MAE ↓	3.13 ± 0.21	3.20 ± 0.22	31.76 ± 6.34	47.35 ± 12.38
PR interval (ms)	MAE ↓	2.73 ± 0.06	32.95 ± 10.31	40.55 ± 5.88	3.84 ± 0.71
QRS duration (ms)	MAE ↓	2.58 ± 0.08	3.99 ± 0.48	40.55 ± 5.88	3.50 ± 0.10
Heart Rate (bpm)	MAE ↓	1.57 ± 0.07	2.92 ± 0.19	3.62 ± 1.89	4.79 ± 1.10
J-point elevation (µV)	MAE ↓	5.77 ± 0.18	6.43 ± 0.56	23.07 ± 3.29	8.62 ± 0.42
T-wave amplitude (µV)	MAE ↓	5.80 ± 1.29	6.13 ± 1.45	8.70 ± 1.49	339.04 ± 8.0
R-wave amplitude (µV)	MAE ↓	8.35 ± 1.12	8.64 ± 1.11	927.38 ± 16.0	10.48 ± 3.56
Sex classification (%)	ACC ↑	88.80 ± 0.7	87.50 ± 1.0	62.40 ± 6.6	79.80 ± 2.5

Prediction errors increased dramatically when the feature in question is blanked out. Prediction errors also often increased slightly when other parts of the ECG are blanked. In the metric column, the arrow signifies whether higher or lower values are better, i.e., an arrow pointing downwards means that lower values are preferable.