Table 5.

Leave-one-out cross validation for TdP risk prediction at 1× C_max.

	Left-out drug	Category	P(0)	P(1)	P(2)	Prediction error
1× Cmax	Dofetilide	2	0	0.033 (0)	0.967 (1)	0.033 (0)
	Bepridil	2	0	0	1	0
	Sotalol	2	0	0.3475 (0)	0.6525 (1)	0.3475 (0)
	Quinidine	2	0	0	1	0
	Cisapride	1	0	1	0	0
	Terfenadine	1	0	0.5455 (0)	0.4545 (1)	0.4545 (1)
	Ondansetron	1	0	1	0	0
	Chlorpromazine	1	0.1575 (1)	0.8425 (0)	0	0.1575 (1)
	Verapamil	0	0.9995 (1)	0.0005 (0)	0	0.0005 (0)
	Ranolazine	0	0.9215 (1)	0.0785 (0)	0	0.0785 (0)
	Mexiletine	0	1	0	0	0
	Diltiazem	0	1	0	0	0
10× Cmax	Dofetilide	2	0.0373 (0)	0.0580 (0)	0.9047 (1)	0.1326 (0)
	Bepridil	2	0	0	1	0
	Sotalol	2	0	0.712 (0)	0.288 (1)	0.712 (0)
	Quinidine	2	0	0	1	0
	Cisapride	1	0	0.728 (1)	0.272 (0)	0.272 (0)
	Terfenadine	1	0	1	0	0
	Ondansetron	1	0	1	0	0
	Chlorpromazine	1	0.9945 (1)	0.0055 (0)	0	0.9945 (1)
	Verapamil	0	0.3075 (1)	0.6925 (0)	0	0.6925 (0)
	Ranolazine	0	1	0	0	0
	Mexiletine	0	1	0	0	0
	Diltiazem	0	1	0	0	0

The TdP risk levels were assigned category values of 2 (High), 1 (Intermediate), and 0 (Low). A classifier was trained on 11 of 12 drugs and then used to predict the category probabilities [P(x), where x is the category value] and to obtain an overall prediction error for the remaining drug (see section Cross Validation). Uncertainty model simulations were used for training and prediction. For comparison, probabilities, and prediction errors from Dutta et al. (2017) are shown in parentheses when they differed from uncertainty results.