Validation of a Population-Based Algorithm to Detect Chronic Psychotic Illness

Abstract

OBJECTIVE: To validate algorithms to detect people with chronic psychotic illness in population-based health administrative databases. METHOD: We developed 8 algorithms to detect chronic psychotic illness using hospitalization and physician service claims data from administrative health databases in Ontario to identify cases of chronic psychotic illness between 2002 and 2007. Diagnostic data abstracted from the records of 281 randomly selected psychiatric patients from 2 hospitals in Toronto were linked to the administrative data cohort to test sensitivity, specificity, and positive predictive values (PPV) and negative predictive values. RESULTS: Using only hospitalization data to capture chronic psychotic illness yielded the highest specificity (range 69.9% to 84.7%) and the highest PPV (range 55.2% to 80.8%). Using physician service claims in addition to hospitalization data to capture cases increased sensitivity (range 90.1% to 98.8%) but decreased specificity (range 31.1% to 68.0%) and PPV (range 38.4% to 71.1%). CONCLUSION: Using health administrative data to study population-based outcomes for people with chronic psychotic illness is feasible and valid. Researchers can select case identification methods based on whether a more sensitive or more specific definition of chronic psychotic illness is desired.