Analysis of Health-Related Quality of Life in Multiple Sclerosis: A Bayesian Quantile LASSO Approach

Background/Objectives: Multiple Sclerosis (MS) is a complex, chronic autoimmune neuroinflammatory disorder that significantly impacts patients’ health-related quality of life (HRQoL) and increases the burden of healthcare costs. However, evidence that quantifies the covariate-adjusted differences between the MS and non-MS populations is limited due to the presence of outliers, which frequentist approaches may not adequately address. Therefore, this study aims to examine healthcare expenditure and HRQoL in patients with MS compared to the non-MS population using a robust Bayesian approach. Methods: This retrospective cross-sectional study includes adults (≥18 years) with MS and those without MS using the 2017–2022 Medical Expenditure Panel Survey (MEPS) data. The Bayesian quantile LASSO (BQL) is applied to examine the association between MS and different response variables under three quantile levels. Markov Chain Monte Carlo (MCMC) with Gibbs sampling was used to estimate the coefficients from the posterior distribution of model parameters. The convergence of the MCMC chain has also been assessed to ensure the reliability and stability of the posterior estimates. Alternative methods, including Bayesian LASSO and the multivariate Generalized Linear Models (GLMs), are also applied for comparison in both prediction and estimation. Results: The results of the BQL show that for the healthcare expenditures, the estimated total healthcare expenditure in patients diagnosed with MS is $29,860.11 (95% credible interval $27,826.96 to $31,825.63) more compared to those without MS under the quantile level 0.5. With the coefficient −3.16 (95% credible interval −4.31 to −2.07), MS is negatively related to the mental component of VR-12 under the median quantile. Compared with individuals without MS, patients with MS have a 13.40-point lower score on the physical component of the VR-12 (95% credible interval −14.56 to −12.19) at the 0.5 quantile. At the median quantile, BQL achieves prediction errors of 6723.86 for healthcare expenditure, 6.57 and 6.41 for the mental and physical components of the VR-12, respectively. Conclusions: BQL with a quantile level of 0.5 shows the lowest in-sample prediction error when examining the healthcare costs and HRQoL in MS.