Predicting schizophrenia spectrum disorders in pediatric outpatients: a prospective validation of the child psychosis-risk screening system

This study validated the prognostic ability of the Child Psychosis-risk Screening System (CPSS), an algorithm derived from the widely used Child Behavior Checklist (CBCL), and benchmarked its performance against a comprehensive machine learning model.

This prospective cohort study enrolled 491 pediatric psychiatric outpatients aged 6-18 years who visited clinics for developmental, psychosomatic, or psychiatric disorders. Baseline assessments (Visit 1) included the CBCL and extensive clinical data. Participants were followed-up for up to 5 years (n = 350; Visits 2+), including annual assessments for schizophrenia spectrum disorders (SSD). The predictive accuracy of the CPSS risk score was evaluated using receiver operating characteristic analysis and compared with that of a post-hoc machine learning benchmark model (Light Gradient Boosting Machine) incorporating 150 baseline variables.

The CPSS risk score demonstrated good predictive performance for final SSD status (area under the curve [AUC] = 0.906), significantly exceeding its discriminative ability for SSD at Visit 1 (AUC = 0.841). This discrepancy indicates that the CPSS sensitively captures prodromal vulnerabilities that do not yet meet diagnostic criteria. Although the complex Light Gradient Boosting Machine model achieved a marginally higher AUC (0.959), the CPSS yielded comparably robust accuracy using only eight CBCL subscales. Variable-importance analysis further showed that the CPSS risk score was a stronger predictor than the CBCL "Thought Problems" scale, a traditional marker of psychosis risk.

The CPSS is a parsimonious and effective screening tool that translates behavioral observations into objective risk stratification. Its use in primary care settings could streamline referral pathways and facilitate timely early intervention for children at high risk of psychosis who present with ambiguous somatic or behavioral symptoms.