Life expectancy and other health outcomes in the United States are associated with income and socioeconomic status. Low-income populations have poorer health outcomes and, unlike high-income populations, outcomes also vary by geography. While these relationships have been described for decades in prior literature, the underlying mechanisms of these connections have not been identified. Understanding the determinants of such health disparities is critical to improving health outcomes and preventing disease. The impact of health insurance on health outcomes in low-income populations is unknown, as it has not been studied empirically on a national scale. The proposed research will be the first ever study to employ randomization of health plans in Medicaid, the government insurance program for individuals who lack resources to pay for health care. We will establish the role of insurance coverage on health outcomes in low-income populations. This will involve a novel data source containing enrollment and health care utilization information for the universe of low-income households in Massachusetts from 2006 to 2015. We are uniquely positioned to study the effect of Medicaid and type of Medicaid coverage as our data contains enrollees who were randomly assigned to plans and those who were not. These data are linked to important health outcomes, including cause of death, birth weight, and APGAR scores. We will additionally exploit our partial random assignment to examine effect modification by quality of care. Disparities exist in quality of care along income lines, and low-income populations tend to have lower quality scores. To accomplish these goals, we propose the creation of a nonparametric machine learning framework for generalizing results from experimental and quasi-experimental studies. This flexible robust methodology will be developed and tailored for our study of health disparities in order to generalize results from within-state studies to i) the full population of low-income individuals within a state and ii) populations of low-income individuals in other states. With average monthly Medicaid enrollment of over 55 million individuals across the country, this project will have significant broad impacts on low-income populations. Our findings and new analytic tools will provide foundational understanding of the role of insurance coverage on health outcomes among low-income populations, including any effect modification by quality of care. Unraveling the complicated relationships between insurance and health outcomes in low-income populations is an essential component toward improving these outcomes. The lack of randomization in most public health studies on disparities has contributed to diverging results and confusion around interpretation for health policy. The development of our machine learning framework for the generalizability of population health studies is innovative as it will revolutionize the ability of researchers to assess the effects of health exposures and interventions.

Public Health Relevance

Health outcomes and quality of care in low-income populations lag behind other groups, and the impact of health insurance on these disparities among low-income individuals is currently unknown. The goal of this proposal is to examine the role of insurance coverage on health outcomes in low-income populations with rigorous new tools in partially randomized data. This will be achieved by developing a novel machine learning framework for the generalizability of experimental and quasiexperimental studies, providing population health scientists with robust methodology to assess the effects of health interventions and exposures.

National Institute of Health (NIH)
National Institute on Minority Health and Health Disparities (NIMHD)
NIH Director’s New Innovator Awards (DP2)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1)
Program Officer
Berzon, Richard
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Stanford University
United States
Zip Code