Obtaining population-based estimates for survey data using Bayesian hierarchical models with poststratification

For large-scale surveys such as the National Health and Aging Trends Study (NHATS), investigators may wish to combine data from two (or more) cohorts in a single analysis to obtain larger sample sizes. Unfortunately, it is not possible to combine the 2011 and 2015 NHATS cohorts while retaining the sample weights. We applied Bayesian hierarchical models with poststratification as an alternative strategy for obtaining population-based estimates from NHATS. As proof of principle, we compared prevalence estimates of frailty obtained from our Bayesian approach with those obtained from the 2011 and 2015 cohorts using the NHATS sample weights. Once validated, we applied our strategy to combine the cohorts into a single analytical dataset without overlap of participants, and generated Bayesian estimates of frailty for the combined cohort. Estimates from the Bayesian model closely matched the weighted NHATS estimates. The ability to combine cohorts while generating population-based estimates will allow investigators to address questions that require larger sample sizes, thereby enhancing the value of NHATS to the scientific community.