BACKGROUND The temporal stability of a diagnosis is one measure of its predictive validity.
AIMS To measure diagnostic stability in first-episode psychosis using ICD-10 and DSM-III-R.
METHOD Between 1992 and 1994 we ascertained a cohort of persons with first-episode psychosis (n = 168), assigning to each a consensus diagnosis. At three-year follow-up, longitudinal consensus diagnoses, blind to onset diagnoses, were made. Stability was measured by the positive predictive values (PPVs) of onset diagnoses. For onset schizophrenia, we also calculated sensitivity, specificity and concordance (kappa).
RESULTS First-episode ICD-10 and DSM-III-R schizophrenia had a PPV of over 80% at three years. Over one-third of cases with ICD-10 F20 schizophrenia at three years had non-schizophrenia diagnoses at onset. Manic psychoses showed the highest PPV (91%). For onset schizophrenia, both systems had high specificity (ICD-10: 89; DSM-III-R: 93%), but low sensitivity (ICD-10: 64%; DSM-III-R: 51%) and moderate concordance (ICD-10: 0.54; DSM-III-R: 0.46).
CONCLUSIONS Bipolar disorders and schizophrenia showed the highest stability. DSM-III-R schizophrenia did not have greater stability than ICD-10 schizophrenia.