Medium-term prediction of clinically relevant outcomes in first-episode schizophrenia patients

Background Predicting long-term outcomes in first-episode schizophrenia (FES) remains difficult, despite being especially important early in the illness, when timely intervention is most critical. It also remains unclear how much data from the initial phase of illness is required to improve prognostic accuracy. Methods We analysed 68 FES patients assessed at baseline (V1; mean 0.5 years post-onset), one-year follow-up (V2; mean 1.2 years), and outcome (V3; mean 4.9 years). Elastic-net models were trained to predict three V3 outcomes — negative symptoms (PANSS Negative factor; Wallwork/Fortgang), global functioning (GAF), and quality of life (WHOQOL-BREF psychological domain) — using either V1 predictors alone (23 variables) or V1 + V2 combined (43 variables). Performance was evaluated using nested cross-validation on held-out data. Results With V1 + V2 predictors, all three outcomes were predicted at statistically significant levels: PANSS Negative R 2 = 0.22 (driven by log(DUP), PANSS Negative at V1/V2, and PANSS Disorganised at V2); WHOQOL-BREF Psychological Health R 2 = 0.22 (driven by WHOQOL Psychological Health and GAF at V2); and GAF R 2 = 0.14 (driven by GAF, PANSS Positive, WHOQOL Psychological Health at V2, and hospitalisation burden). With V1 predictors alone, only PANSS Negative showed meaningful predictive power (R 2 = 0.15); GAF and WHOQOL-BREF did not outperform the intercept-only baseline. Conclusion Long-term functioning and quality of life in FES cannot be predicted from first-episode data alone; at least one year of follow-up is required, suggesting post-onset changes shape these outcomes. Negative symptoms are an exception: comparatively stable after initial treatment and predictable from baseline, with past symptomatology along with DUP selected as predictors – indicating stronger persistency and predictability than in the other two investigated outcomes.