Presentation (CTI)

OBJECTIVES: Traditional frequentist approaches may yield unstable estimation when subgroup sample sizes are limited. To address this limitation, we apply Bayesian hierarchical models (BHM) to assess marginal healthcare expenditures of dementia across race and sex subgroups.
METHODS: This study includes older adults (≥65 years) using the 2021-2022 Medical Expenditure Panel Survey data. One-to-one propensity score matching on key sociodemographic characteristics and comorbidities is used to achieve balanced groups. BHM is applied to examine the marginal annual overall healthcare, medication, and inpatient expenditures between dementia patients and non-dementia populations, stratified across subgroups defined by race and sex. Markov Chain Monte Carlo (MCMC) with Gibbs sampling is used to estimate the posterior distribution of model parameters. The convergence of the MCMC chains is assessed using the potential scale reduction factor, trace plots, and effective sample sizes to ensure reliability and stability.
RESULTS: After matching, this study identified 198 matched patients with a mean age of 80.27± 5.47. Despite the limited sample size among females (50 non-dementia and 65 dementia adults), the BHM consistently identifies higher expenditures for dementia patients across all subgroups. For total healthcare expenditures, dementia is associated with an estimated increase of $6,851(CI: [$6693, $7010]) among males and $6,600(CI: [$6429, $6770]) among females per year. Across race subgroups, including those with extremely limited sample sizes (e.g., only eight individuals in the “Other” category), the BHM detects higher expenditures among dementia patients across all expenditure categories. For instance, for total healthcare expenditures, dementia patients spend an additional $33.59 (CI: [$23.28, $44.96]) among Hispanic, $32.36 (CI: [$22.15, $42.42]) among non-Hispanic White and $33.17 (CI: [$22.99, $43.87]) among non-Hispanic White.
CONCLUSIONS: The BHM yields reliable estimates, indicating a persistent and substantial economic burden among patients with dementia across sex and race subgroups. These findings support the consideration of BHM in healthcare analysis with limited subgroup sample sizes.