Weighted quantile regression for analyzing health care cost data with missing covariates

Area of Science:

• Biostatistics
• Health Economics
• Econometrics

Background:

• Health care cost data analysis is challenging due to skewness, heteroscedasticity, and missing data.
• Existing research often focuses on modeling the conditional mean, neglecting other distributional aspects.

Purpose of the Study:

• To develop a robust method for estimating conditional quantiles of health care costs with missing covariates.
• To propose a variable selection technique for quantile regression in the presence of missing data.

Main Methods:

• Weighted quantile regression for estimating conditional quantiles.
• A modified Bayesian Information Criterion (BIC) for variable selection with missing covariates.
• Semiparametric approach not requiring likelihood specification for errors or covariates.

Main Results:

• The weighted quantile regression estimator is consistent and asymptotically normal, outperforming naive estimators.
• The proposed modified BIC facilitates effective variable selection in quantile regression models with missing data.
• Simulations demonstrate the procedure's validity and efficiency.

Conclusions:

• Weighted quantile regression offers a more complete understanding of covariate effects on health care costs, accommodating data skewness and heterogeneity.
• The method provides a flexible and powerful tool for analyzing complex health care cost data, even with missing information.