Fig. 2. Performance of the baseline, 12-, 24- and 36-month probability classification models.

A Area under the receiver operating characteristic curve (AUROC). B Brier score. C Decision curve. The results indicate that models with longer longitudinal information performed better. D The average feature importance rank was calculated across 100 repetitions of the autoML procedures, where a higher value indicates greater importance of a feature. Dx diagnosis, EIS early intervention service, DUP duration of untreated psychosis, SA suicide attempts, NSSI non-suicidal self-injury, OCD obsessive-compulsive disorder, SOFAS Social and Occupational Functioning Assessment Scale, A&E accident & emergency, OPD out patient departments, DDD daily defined dose, ECT electroconvulsive therapy.