Impact of classification methods when using Medicaid claims for mental health services research

Objective: Medicaid claims data are increasingly used for planning purposes to estimate treated prevalence of mental disorders in public sector mental health systems and to identify cohorts for mental health services research. This study investigates how different methods of estimation and identification affect case classification, prevalence rates and service cost estimates when individuals have multiple diagnoses over a study period. Methods: A cross-sectional, observational analysis of 224,654 adult Medicaid recipients in a managed care behavioral health plan was employed to identify treated individuals from 2005 claims records. ICD-9 diagnostic codes were used to classify individuals. Four classification methods were used to assign individuals to diagnostic categories of schizophrenia, major depressive disorder, bipolar disorder and other mental health disorders. Treated prevalence rates for all diagnostic categories were compared as were inpatient and emergency room costs. Results: Diagnostic group classification and cost estimates varied depending on the classification method employed. Annual treated prevalence for a mental disorder ranged from 14% to 22% of the Medicaid population. Identification of individuals with schizophrenia ranged from 2.6% to 4% depending on the method. Inpatient and emergency room costs were also affected by the method used. Conclusions: Methods chosen for identifying cohorts for mental health services research from Medicaid claims data will affect overall and diagnosis-specific treated prevalence rates and cost estimates. Researchers should be aware of the ways in which different methods of selecting cohorts may affect their findings and should make their classification procedure explicit to enhance the correct interpretation of results.