Using Bayesian Statistical Inference to Improve the Measurement of Adequacy of Mental Health Care Utilization in a Nationally Representative Sample

• Activity Number: 288
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Chih-Nan Chen*+ and Benjamin Cook and Margarita Alegria
• Companies: National Taipei University and Center for Multicultural Mental Health Research and Center for Multicultural Mental Health Research
• Keywords: Bayesian estimation ; Mental health ; Health care research

Abstract

In health care research, longitudinal analysis of health care episodes is preferred to cross-sectional analysis because it avoids arbitrary observation periods and provides more information on temporal changes in utilization behavior. Analysis of longitudinal claims must account for left and right censoring to construct valid episodes of care. Analyzing individuals with probable mental illness from the Medical Expenditure Panel Survey (MEPS), we apply two-part negative binomial models to account for a significant proportion of zero service use. We use the Gibbs sampler to simulate the posterior mean and standard deviations from the joint distribution of observations. We compare Bayesian predictions to three alternatives: a method naïve to censoring; a naïve method that drops censored cases; and a maximum likelihood estimator of the censored two-part negative binomial model. Bayesian and MLE estimators reported similar mean numbers of mental health care visits per treatment episode. Typically used naïve methods substantially under-estimated the number of visits per episode. We conclude that Bayesian approaches to censoring can greatly improve health care episodes analyses.