Predicting long-term poor outcomes in individuals at clinical high risk for psychosis using real-world clinical data: the OASIS1000 prospective study

World Psychiatry. 2026 Jun;25(2):295-306. doi: 10.1002/wps.70063.

ABSTRACT

Most research into the clinical high risk for psychosis (CHR-P) state has focused on predicting transition to psychosis in the short term, and largely in research samples that are not representative of real-world preventive care. However, CHR-P individuals experience a broad range of poor long-term clinical outcomes in real-world settings, such as hospitalization for psychiatric reasons, initiation of a new antipsychotic treatment, or suicide. Improving the detection of individuals at greater risk of these long-term poor outcomes is critical to advance preventive care in real-world clinical settings. This study aims to develop and internally validate a personalized clinical prediction model for real-world, long-term poor outcomes beyond transition risk in CHR-P individuals. This RECORD-compliant, real-world, long-term prospective cohort study included electronic health records (EHR) data from all CHR-P individuals receiving preventive care from the Outreach and Support in South London (OASIS) service in South London and Maudsley National Health Service Foundation Trust in the UK from 2001 to 2024. The primary outcome was the long-term cumulative risk of first real-world poor outcome, operationalized by pragmatic parameters informing clinical practice: transition to psychosis, receiving a first antipsychotic treatment at a dosage necessary to treat first-episode psychosis, receiving a first voluntary or compulsory hospitalization for psychiatric reasons, or dying by suicide. A clinical prediction model (regularized Cox regression) was developed and validated using internal-external cross-validation to predict long-term poor outcomes, utilizing real-world predictors available in routine care. Model performance was indexed by discrimination (Harrell’s C), calibration (slope, intercept), overall performance (Brier score), and potential clinical utility (decision curve analysis). One thousand CHR-P OASIS patients were included (mean age: 22.51±4.99 years; 53.60% males, 44.73% White) and followed up to a maximum of 21 years. The cumulative risk of real-world poor outcome was 0.644 (95% CI: 0.547-0.742) at 14-18 years. The validated clinical prediction model showed statistically significant discrimination (C=0.69; 95% CI: 0.63-0.74), calibration (slope = 1.61, 95% CI: 0.74-2.48; intercept = -0.03, 95% CI: -0.62 to 0.55) and overall performance (Brier score = 0.18; 95% CI: 0.13-0.22). Decision curve analysis demonstrated that the model was associated with greater net benefit than the clinical alternatives at risk thresholds from 0% to 50%. These data suggest that most CHR-P individuals have long-term poor outcomes, beyond transition to psychosis, in real-world care, which should become the focus of a new generation of research. The clinical prediction model presented here can identify these individuals and facilitate the personalized provision of preventive care, thereby improving outcomes in this population. CHR-P services should extend their duration of care to address the substantial long-term clinical outcomes experienced by young individuals.

PMID:42136514 | DOI:10.1002/wps.70063