Using machine learning to develop a clinical prediction model for SSRI-associated bleeding: a feasibility study

Table 3 Cohort size, number of bleeding events, and best model performance metrics for each drug cohorts

	Total number of patients, N	Number of bleeding events, n (%)	Best models based on largest AUC score
	Total number of patients, N	Number of bleeding events, n (%)	ML model	AUC	YI optimized sensitivity	YI optimized specificity
Clopidogrel	2,159	234 (10.8)	LR	0.638	64.4%	59.5%
Warfarin	1,855	293 (15.8)	XGBoost	0.682	69.0%	61.0%
Citalopram	3,151	286 (9.1)	RF	0.698	67.8%	66.7%
Escitalopram	2,597	156 (6.0)	RF	0.656	67.3%	59.1%
Fluoxetine	2,719	226 (8.3)	DT	0.664	36.8%	85.4%
Fluvoxamine^a	117	≤ 20	-	-	-	-
Paroxetine	1,100	97 (8.8)	RF	0.632	58.9%	63.2%
Sertraline	4,052	336 (8.3)	RF	0.665	66.8%	61.9%
Vortioxetine^a	149	≤ 20	-	-	-	-
Combined SSRI	10,362	996 (9.6)	XGBoost	0.688	57.9%	70.6%

Abbreviations: AUC area under the receiver operating characteristic curve statistic, DT decision tree, LR logistic regression, ML machine learning, RF random forest, XGBoost extreme gradient boosting, YI Youden’s index
Results were reported in compliance with the AoU Data and Statistics Dissemination Policy prohibiting the display of participant counts ranging 1 to 20
^aThe models for fluvoxamine and vortioxetine were excluded due to the small number of participants in the cohorts relative to other drugs

ISSN: 1472-6947