Propensity Score Matching Unmasks Core Predictors of Psychosis Conversion in Clinical High-Risk Individuals

Predicting psychosis conversion in clinical high-risk for psychosis (CHR-P) individuals is limited by unaccounted cognitive and demographic confounding, which obscures the true prognostic value of clinical and functional markers.

A total of 920 CHR-P individuals from the Shanghai At Risk for Psychosis-extended program were followed for 3 years. Propensity score matching (PSM) balanced 5 demographic variables (age, gender, education, paternal/maternal education) and 6 core cognitive domains. Group comparisons (CHR converters [CHR-C] vs. non-converters [CHR-NC]) and multivariate logistic regression were conducted in the entire and PSM cohorts. Receiver Operating Characteristic curves quantified predictive accuracy.

After PSM, demographic and cognitive disparities were eliminated, while clinical distinctions (symptoms, Global Assessment of Functioning [GAF] drop rate) became prominent (all P < .05). The PSM-adjusted model identified bizarre thinking, lower unusual thought content, emotional experience disturbances, and GAF-related indicators as independent predictors, with higher discriminative ability (Area Under the ROC Curve [AUC] = 0.808) than the unadjusted model (AUC = 0.685). The GAF drop rate alone achieved 73% predictive accuracy post-PSM, versus weak performance in the entire cohort (AUC = 0.589).

Systematic control of cognitive and demographic confounding via PSM clarifies the true prognostic value of clinical and functional markers for CHR-P conversion. The findings provide actionable tools for risk stratification and highlight the critical need to account for cognitive function in early psychosis prediction research.