A Novel Casual Difference-in-differences Method to Study the Medical Home Effects

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): A Novel Causal Difference-in-differences Approach to Estimate the Effect of Medical Homes on Health Care Costs Among Children PI: Bing Han, co-PI: Hao Yu Other key personnel: Bonnie Ghosh-Dastidar, Mark Friedberg During the past decade, the medical home concept has received national attention as a mechanism for ensuring quality health care for all children and for adults with chronic conditions. Numerous studies have reported on the positive effects a medical home has on children health care access, utilization, and quality. However, mixed results have been reported about its effect on health care costs. Evaluations of the medical home effect on health care costs are often based on the difference-in-differences (DID) method. The classic DID method employs two core assumptions: the parallel trajectory and discrete treatment assumptions, both of which are likely violated in the medical home studies. Many medical home studies are based on observational data or quasi-experiments, which cannot safely guarantee the parallel trajectory assumption. Medical home statuses (known as medical homeness (MH) in the literature) are a continuous dosage, which has to be discretized to satisfy the discrete treatment assumption. These difficulties result in th loss of both accuracy and precision in the estimates. Thus, it is unclear whether the mixed results on health care cost reflect the true impact of medical home or the results of inefficient analytic methods. To address this issue, we propose an innovative causal inference method for the DID setting under specific aim 1. We propose a set of novel concepts-including the potential trajectory, the (dynamic) treatment scheme, and the longitudinal generalized propensity score (LGPS). These new concepts are natural extensions of the existing generalized propensity score and potential outcome (i.e., counterfactual) for dichotomous treatment statuses, but which can simultaneously relax both core assumptions in the classic DID method. We also propose to develop efficient nonparametric methods to estimate the LGPS and to fit the LGPS-adjusted estimate for various treatment effects. Under specific aim 2, we will examine the detailed impact of MH on health care costs using the new method, with the goal of exploring an accurate and precise estimate to fill in the gap of the existing literature.