Evaluation of Marginal Healthcare Expenditures and Health-related Quality of Life in Multiple Sclerosis with a Bayesian Quantile Machine Learning Approach
Author(s)
Xi Lu, PhD, Jieni Li, MPH, PhD, Rajender Aparasu, PharmD, PhD.
Pharmaceutical Health Outcomes and Policy, University of Houston College of Pharmacy, HOUSTON, TX, USA.
Pharmaceutical Health Outcomes and Policy, University of Houston College of Pharmacy, HOUSTON, TX, USA.
OBJECTIVES: Traditional regression models may lead to biased estimations due to the inherent distribution and heterogeneity of health outcome measures. Therefore, this study evaluated the marginal healthcare expenditure and health-related quality of life (HRQoL) in multiple sclerosis (MS) with the Bayesian quantile machine learning (BQR) approach and compared the model performance with other models.
METHODS: This study includes adult participants (≥18 years) using the 2017-2022 Medical Expenditure Panel Survey data. BQR approaches were applied to examine the marginal healthcare expenditure and HRQoL between MS adults and non-MS populations under a variety of quantile levels. Markov Chain Monte Carlo (MCMC) with Gibbs sampling was used to estimate the posterior distribution of model parameters. Alternative models, including Bayesian least absolute shrinkage and selection operator (LASSO) and the multivariate generalized linear models (GLM), were also considered.
RESULTS: The study sample included 310 (unweighted) MS patients with a mean age of 50.03±18.28. The results of the BQR model found that MS patients had $8,873.26 more expenditures than non-MS adults under the quantile level of 0.25. Furthermore, MS is associated with -1.19 unit lower mental component and -8.37 units lower physical component of HRQoL under the quantile level 0.75. Meanwhile, BQR outperforms GLM with a low prediction error compared with the alternative methods. When estimating healthcare expenditures, the prediction error of BQR is as small as 1883.46, while that of Bayesian LASSO is 8165.52, and that of GLM (with Gaussian family) is 8173.11. The convergence of the MCMC chain was assessed to ensure the reliability and stability of the posterior estimates.
CONCLUSIONS: Overall, the study found that the BQR model shows better prediction accuracy and estimation stability with improved computational efficiency through MCMC compared to alternative models. Therefore, BQR provides a robust and efficient modeling option for heterogeneous or skewed data like expenditure and HRQoL.
METHODS: This study includes adult participants (≥18 years) using the 2017-2022 Medical Expenditure Panel Survey data. BQR approaches were applied to examine the marginal healthcare expenditure and HRQoL between MS adults and non-MS populations under a variety of quantile levels. Markov Chain Monte Carlo (MCMC) with Gibbs sampling was used to estimate the posterior distribution of model parameters. Alternative models, including Bayesian least absolute shrinkage and selection operator (LASSO) and the multivariate generalized linear models (GLM), were also considered.
RESULTS: The study sample included 310 (unweighted) MS patients with a mean age of 50.03±18.28. The results of the BQR model found that MS patients had $8,873.26 more expenditures than non-MS adults under the quantile level of 0.25. Furthermore, MS is associated with -1.19 unit lower mental component and -8.37 units lower physical component of HRQoL under the quantile level 0.75. Meanwhile, BQR outperforms GLM with a low prediction error compared with the alternative methods. When estimating healthcare expenditures, the prediction error of BQR is as small as 1883.46, while that of Bayesian LASSO is 8165.52, and that of GLM (with Gaussian family) is 8173.11. The convergence of the MCMC chain was assessed to ensure the reliability and stability of the posterior estimates.
CONCLUSIONS: Overall, the study found that the BQR model shows better prediction accuracy and estimation stability with improved computational efficiency through MCMC compared to alternative models. Therefore, BQR provides a robust and efficient modeling option for heterogeneous or skewed data like expenditure and HRQoL.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR80
Topic
Methodological & Statistical Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Neurological Disorders