Unlocking treatment success: predicting atypical antipsychotic continuation in youth with mania

Table 2 Performance measures for assessing the predictive accuracy of mania treatment continuation prediction models

Reference model^c	AUC	P value^d	Sensitivity	Specificity	PPV	NPV	F1 Score	Intercept	Slope	Recall	Precision
Generalized Linear Model (Reference)	0.82 (0.79–0.85)	NA	0.71	0.73	0.72	0.82	0.73	-1.24	1.00	0.73	0.75
Support Vector Machines with Linear Kernel	0.82 (0·79–0·85)	0.73	0.72	0.73	0.72	0.76	0.78	-1.23	1.14	0.77	0.76
Generalized Additive Model using LOESS	0.82 (0·78–0·86)	0.58	0.73	0.74	0.75	0.88	0.74	-1.25	1.12	0.73	0.75
Stochastic Gradient Boosting	0.81 (0·78–0·84)	0.49	0.70	0.76	0.73	0.89	0.74	-1.26	1.19	0.72	0.75
Random Forest	0.78 (0·75–0·81)	0.02	0.70	0.88	0.74	0.83	0.71	-0.29	1.78	0.74	0.75
Super Learner	0.82 (0·79–0·85)	0.84	0.74	0.72	0.74	0.89	0.74	-1.24	1.01	0.74	0.74

^c As the reference model, we employed a generalized linear model
^d Using the DeLong’s test, we compared the area under the curve of each machine-learning-based prediction model with the reference model
P values less than 0.05 are bold

ISSN: 1472-6947