Real-World Data Predicts Long-Term Poor Outcomes in Clinical High-Risk Psychosis

Navigating Long-Term Risk in Early Psychosis Care

While early intervention for individuals at clinical high risk for psychosis (CHR-P) is a key goal, much of the existing research focuses narrowly on predicting the short-term transition to psychosis [1]. This focus often relies on highly selected research cohorts that do not reflect the complex realities of routine clinical practice [2]. In real-world settings, CHR-P individuals face a much broader spectrum of poor long-term outcomes, including psychiatric hospitalizations and suicide, which are not always captured in traditional study designs [3, 4, 5, 6]. A recent long-term prospective study addresses this gap, using two decades of electronic health records from a real-world clinical service to develop a personalized model for predicting a wider range of adverse outcomes [7].

Beyond Psychosis Transition: Defining Long-Term Outcomes

The historical focus on psychosis transition has overshadowed other significant adverse events that affect CHR-P individuals in community care. To create a more clinically relevant framework, a recent investigation expanded the definition of risk beyond psychosis onset. The researchers defined a composite primary outcome using pragmatic parameters that directly inform clinical management. This included not only transition to psychosis but also receiving a first antipsychotic treatment at a dosage sufficient for first-episode psychosis, a first voluntary or compulsory hospitalization for psychiatric reasons, or dying by suicide. By encompassing this range of events, the study aimed to develop and validate a personalized prediction model that could identify individuals at risk for these diverse and consequential long-term outcomes, reflecting the challenges clinicians manage daily.

A Real-World Cohort and Data-Driven Approach

To build a model grounded in clinical reality, the researchers conducted a long-term prospective cohort study adhering to the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement. The study drew upon a substantial dataset of electronic health records from all individuals receiving care at the Outreach and Support in South London (OASIS) service between 2001 and 2024. This provided a rich, longitudinal view with a follow-up period of up to 21 years. The final cohort included 1,000 CHR-P patients with a mean age of 22.51 years (standard deviation 4.99), of whom 53.60% were male and 44.73% were of White ethnicity. The use of this extensive, real-world dataset was fundamental to the study's goal of creating a tool applicable to everyday preventive care settings.

Developing and Validating the Prediction Model

The investigators developed a clinical prediction model using regularized Cox regression, a statistical method well-suited for analyzing time-to-event data from complex electronic health records. This technique predicts the risk of an outcome over time while preventing the model from becoming overly complex and tailored only to the initial dataset. To ensure the model's reliability and generalizability, it was assessed with internal-external cross-validation, a rigorous process where the model is repeatedly trained on one part of the data and tested on another. The model was built using only predictors readily available in routine clinical care. Its performance was evaluated using several standard metrics: discrimination was measured with Harrell's C, which assesses the model's ability to correctly distinguish between patients who will and will not experience a poor outcome. Calibration, measured by slope and intercept, evaluated how well the model's predicted risks align with the actual observed outcomes. The Brier score provided a measure of the model's overall predictive accuracy. Finally, its practical value was assessed through decision curve analysis, a method that determines if using the model to guide treatment decisions provides more net benefit than alternative strategies, such as treating all patients or none.

Key Findings: Quantifying Long-Term Risk and Model Performance

The study revealed the profound long-term vulnerability of the CHR-P population. Over a 14 to 18 year follow-up, the cumulative risk of experiencing a first real-world poor outcome was 0.644 (95% CI: 0.547-0.742), indicating that nearly two-thirds of individuals in this real-world cohort faced a significant adverse event. The clinical prediction model demonstrated moderate and statistically significant predictive ability. Its capacity to discriminate between higher and lower risk individuals was confirmed with a Harrell's C-statistic of 0.69 (95% CI: 0.63-0.74), suggesting a 69% probability of correctly ranking two randomly selected patients by their actual risk. The model also showed good calibration, with a slope of 1.61 (95% CI: 0.74-2.48) and an intercept of -0.03 (95% CI: -0.62 to 0.55), indicating its risk estimates are reliable. The model's overall accuracy was reflected in a Brier score of 0.18 (95% CI: 0.13-0.22), where lower scores signify better performance. Critically for clinical practice, decision curve analysis showed that using the model offered a greater net benefit than the default clinical strategies across risk thresholds from 0% to 50%.

Clinical Implications and Future Directions

These findings confirm that for most individuals identified as CHR-P, the clinical journey involves significant long-term challenges that extend well beyond the risk of developing psychosis. The high incidence of psychiatric hospitalization, need for antipsychotic treatment, and suicide risk underscores that these broader poor outcomes should become a primary focus for research and clinical care. The prediction model developed in this study represents a concrete step toward addressing this need. By identifying individuals at highest risk for this composite of adverse events, the model can help clinicians personalize the provision of preventive care, potentially improving long-term trajectories for this vulnerable population. The authors suggest that to adequately manage the substantial and persistent risks demonstrated in this cohort, CHR-P services should consider extending their duration of care beyond the short-term windows typical of many research protocols.

References

1. Camacho J, Olivares JDO, Hernández OR, et al. Quality of life in individuals at clinical high risk for psychosis: A systematic review and meta-analysis. European psychiatry. 2026. doi:10.1192/j.eurpsy.2026.10170

2. Kennedy‐Martin T, Hulin-Curtis SL, Faries DE, Robinson S, Johnston JA. A literature review on the representativeness of randomized controlled trial samples and implications for the external validity of trial results. Trials. 2015. doi:10.1186/s13063-015-1023-4

3. Norman R, Byambaa M, De R, Butchart A, Scott JG, Vos T. The Long-Term Health Consequences of Child Physical Abuse, Emotional Abuse, and Neglect: A Systematic Review and Meta-Analysis. PLoS Medicine. 2012. doi:10.1371/journal.pmed.1001349

4. Wang J, Mann F, Lloyd‐Evans B, Ma R, Johnson S. Associations between loneliness and perceived social support and outcomes of mental health problems: a systematic review. BMC Psychiatry. 2018. doi:10.1186/s12888-018-1736-5

5. Saha S, Chant D, Welham J, McGrath JJ. A Systematic Review of the Prevalence of Schizophrenia. PLoS Medicine. 2005. doi:10.1371/journal.pmed.0020141

6. Vancampfort D, Stubbs B, Mitchell AJ, et al. Risk of metabolic syndrome and its components in people with schizophrenia and related psychotic disorders, bipolar disorder and major depressive disorder: a systematic review and meta‐analysis. World Psychiatry. 2015. doi:10.1002/wps.20252

7. Logeswaran Y, Micheli AD, Toniol I, et al. Predicting long-term poor outcomes in individuals at clinical high risk for psychosis using real-world clinical data: the OASIS1000 prospective study.. World psychiatry : official journal of the World Psychiatric Association (WPA). 2026. doi:10.1002/wps.70063