Poster Paper: Measurement of Healthcare Disparities- A Difference-in-difference-in-differences (DIDID) approach to Evaluate the Maryland Multi-Payor Patient Centered Medical Home Program

Thursday, November 2, 2017
Regency Ballroom (Hyatt Regency Chicago)

*Names in bold indicate Presenter

Lanlan Xu1, Eberechukwu Onukwugha2, Ilene Harris1, Zippora Kiptanui1, Christine Franey2, Jill Marsteller3 and Ben Steffer4, (1)IMPAQ International, LLC, (2)University of Maryland, (3)Johns Hopkins University, (4)Maryland Health Care Commission

Key Words: Patient Centered Medical Home Program, Measurement of Health Disparity, DIDID

Research Objective:

Despite improvements in overall health of the American population, disparities persist. Delivery system reforms such as the patient centered medical home (PCMH) could have important implications for disparities. However, there is little empirical evidence on whether and how PCMHs can reduce healthcare disparities. We add to this evidence by evaluating the Maryland Multi-Payor Patient Centered Medical Home Program (MMPP)—a three-year pilot launched in April 2011. We employ a difference-in-difference-in-differences (DIDID) approach to measure and compute healthcare disparities.

Study Design:

52 primary care practices with over 300 providers participated in the MMPP. We used propensity score matching to select 104 comparison practices that were not participating in the MMPP. A DIDID estimate was calculated for all quality and utilization measures comparing the difference in the MMPP and comparison sites between 2010 and 2013 for each disparity. The disparities included were (1) race comparing non-white to white, (2) sex comparing females to males, (3) income proxy comparing Medicaid insured to privately insured, and (4) practice location comparing non-large metropolitan to large metropolitan locations.

For count measures, a log link and Poisson distribution was used, and for proportion measures a logit link and binomial distribution was used. For each model, a robust variance was estimated to account for the multiple records per practice. All models were adjusted for case-mix for the patients in the practice meeting the measure’s criteria using the Johns Hopkins ACG® software. Additionally, due to the multiple comparisons performed, the p-values were adjusted using the False Discovery Rate (FDR) method developed by Benjamini and Hochberg (Benjamini 1995).

Population Studied

Patients attributed to MMPP practices or comparison practices in 2010 or 2013 were included in the analysis. The sample size was 104,695 patients in 2010 and 92,765 patients in 2013 attributed to MMPP practices, and 83,164 in 2010 and 82,321 in 2013 attributed to comparison practices.

Principal Findings

The DIDID analysis detected only a few statistically significant differences between the baseline year (2010) and 2013 between the MMPP and comparison practices for the disparities included. Out of the 40 measures investigated, two were statistically significant for racial disparities, one for sex disparity, zero for income proxy, and eight for practice location when using the FDR p-value. While the two measures for racial disparities and the one measure for sex disparity showed significant reductions, 7 out of the 8 significant practice location measures showed increases in disparities.


Findings suggest that the MMPP did not achieve much success in reducing healthcare disparities in race, sex, income or practice location using the measures specified, highlighting the need to consider healthcare disparities in the overall impact evaluation of the program.

Implications for Policy or Practice

This evaluation offers an opportunity to understand how practice transformation models or initiatives such as PCMHs influence healthcare disparities. These findings help policymakers consider how to best identify successful practice level interventions and create initiatives to drive reduction in disparities.