BACKGROUND Outcome is important in the validation of psychiatric diagnosis, as most disorders lack clinicopathological correlates. We describe the predictive validity of four definitions of schizophrenia (DSM-III-R, ICD-10, ICD-9 and CATEGO S+), in a representative cohort of patients selected during their first episode of psychosis.

METHOD Each definition of schizophrenia was applied to 99 patients. Their respective ability to predict 13-year outcome (Global Assessment of Functioning scales) was assessed.

RESULTS DSM-III-R and ICD-10 diagnoses of schizophrenia have high predictive validity for long-term outcome, and both provide relatively stable diagnoses. ICD-9 is reasonably good at predicting disability, but not symptoms, and CATEGO S+ showed no predictive validity. Adding six-month duration criteria to ICD-10, ICD-9 and CATEGO S+ improved their predictive validity, and removing the six-month duration criterion from DSM-III-R commensurately reduced predictive validity.

CONCLUSIONS Modern diagnostic systems (DSM-III-R and ICD-10) have high predictive validity, and are superior to ICD-9. The six-month duration criterion of DSM-III-R schizophrenia accounts for its predictive validity and stability over 13 years, but restricts its use in first-episode studies. The one-month duration criterion of ICD-10 is less restrictive, without major compromises in predictive validity or stability.