Work

An extension of estimating equations to model longitudinal medical cost trajectory with Medicare claims data linked to SEER cancer registry

Year: 2023

Type: article

Abstract: Insurance claims data is an increasingly important health policy research resource, given its longitudinal assessment of cancer care clinical outcomes. Population-level information on medical cost trajectory from disease diagnosis to terminal events, such as death, specifically interests policy makers. Estimating the mean cost trajectory has statistical challenges. The shape of the trajectory is usually highly nonlinear with varying durations, depending on the diagnosis-to-death population time distribution. The terminal event may be right censored, resulting in missing subsequent costs. Medical costs often have skewed distributions with zero-inflation and heteroscedasticity, which may not fit well with the commonly used parametric family of distributions. In this paper, we propose a flexible semi-parametric model to address challenges without imposing a cost data distributional assumption. The estimation procedure is based on generalized estimating equations with censored covariates. The proposed model adopts a bivariate surface that quantifies the interrelationship between longitudinal medical costs and survival, and results in the nonlinear population mean cost trajectory conditional on the death time. We develop a novel generalized estimating equations algorithm to accommodate covariates subject to right-censoring, without fully specifying the joint distribution of the cost and survival data. We provide theoretical and simulation-based justification for the proposed approach, and apply the methods to estimate prostate cancer patient cost trajectories from the Surveillance, Epidemiology, and End Results (SEER)-Medicare linked database.

Source: The Annals of Applied Statistics

DOI: 10.1214/22-aoas1659

Authors: Shikun Wang, Jing Ning, Ying Xu, Ya-Chen Tina Shih, Yu Shen, Liang Li

Institutions: Columbia University, The University of Texas MD Anderson Cancer Center

Language: English

Topic: Statistical Methods and Bayesian Inference

Subfield: Statistics and Probability

Field: Mathematics

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: green

Funder: National Cancer Institute