The cumulative impact of health insurance on health status

This is an open access article under the terms of the http://creativecommons.org/licenses/by/4.0/ License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Associated Data

Abstract

Objective

To add to the evidence base on causal linkages between health insurance coverage and health status, controlling for sociodemographic factors, by analyzing longitudinal data.

Data Source

Secondary data from the Panel Study of Income Dynamics (PSID), 2009‐17, which is a longitudinal, multigenerational study covering a wide array of socioeconomic topics that began in 1968 but has only recently begun collecting useful information on individual health insurance.

Study Design

2017 data on self‐reported health status, work limitations, and death were analyzed as outcomes based upon the degree of exposure to health insurance in 2011‐17. All variables were collected biannually for four years beginning in 2011. Having health insurance at each point in time was, in turn, modeled as a function of several sociodemographic factors.

Data Extraction Methods

Data were downloaded using the crosswalk tool available at the PSID website. Because individual health insurance questions were only asked of heads and spouses in households beginning in 2011, we analyzed only these records.

Principal Findings

Among respondents who were not in fair or poor health in 2009, each additional 2 years of subsequent reported insurance coverage reduced the chance of reporting fair or poor health in 2017 by 10 percent; however, this effect was not present for black respondents.

Conclusions

Our results suggest that the effect of health insurance on health status may compound over time, although unevenly by race. Since people who report fair or poor health status represent the bulk of utilization and spending, our findings provide evidence in support of viewing coverage expansions as investments that will pay dividends in the form of lower utilization over time. More work is needed to produce detailed estimates of cost savings, which may in turn influence policy, as well as to understand and address the source of racial disparity.

What is already known

A large and growing literature finds a positive association between health status and self‐reported health, using relatively short study periods, population subgroups, and survey data at the county level.

What this study adds

Controlling for demographics including age, poverty, and education, we find evidence of a “dose/response” relationship between consistent health insurance coverage and self‐reported health status in our sample of all adults, although the relationship is not present for black respondents.

Each additional two years of health coverage is associated with about a 10 percent reduction in the chance that an individual will fall into fair or poor health.

This finding suggests that policies that support increased access to health insurance may help people maintain better health over time.

1. INTRODUCTION

While it is a compelling narrative that health insurance, by ensuring uninterrupted access to health care and in particular early access to preventive care, is likely to have a positive impact on health, 1 much of the available evidence is unable to directly answer this question. Commentary by Levy and Meltzer in 2008, 2 referring to their systematic review of the literature in 2004, 3 stated that causality has not been firmly established; they suggested at that time that large investments in health and social programs would be needed to demonstrate causality. Subsequent studies using the longitudinal Health and Retirement Survey have examined the effects of health insurance coverage on health. In one study, continuously and intermittently uninsured adults ages 50‐60 were found to be more likely to have a major decline in overall health (defined by self‐reported health status) and the development of new physical difficulties compared with their continuously insured counterparts. 4 Another study examined the effects of gaining Medicare at age 65 among people who were insured continuously prior to that time, compared with those who spent time uninsured before enrolling. Researchers found gains in health over multiple years among people who had been previously uninsured, compared with those who had always had coverage, especially for individuals with diabetes and cardiovascular disease. 5 It was also calculated that public spending to insure individuals prior to age 65 would be offset by about half from savings due to improved health. 6

In a systematic review of the causal effect of health insurance on outcomes in adults, Freeman et al conclude that health insurance improves health. 7 However, the authors note that studies they identified in their review assessed the effect of insurance among individuals who were already sick. Similarly, in the 1971‐1982 large‐scale randomized RAND Health Insurance Experiment, the poorest and sickest sample at the start of the experiment were found to have better health outcomes under the free plan for four out of the thirty conditions measured compared to their counterparts with cost‐sharing. However, for the average person, no difference was found in health outcomes. 8 Results from the more recent 2008 randomized control trial, the Oregon Health Insurance Experiment, support these findings in that no significant effect of gaining Medicaid coverage was found on clinical outcomes or mortality in the years following the lottery. Researchers did however find that the group assigned to health insurance coverage had better self‐reported physical and mental health compared with the control group, supporting results from longitudinal studies discussed above. 9 , 10 Recent results from a randomized pilot study in which the IRS sent informational letters to 3.9 million taxpayers who paid a tax penalty for lacking health insurance coverage under the Affordable Care Act found that the increase in coverage in the two years following reduced mortality among middle‐aged adults. 11 Additionally, new evidence on Medicaid expansion supports the conclusion that new access to Medicaid among low‐income adults is significantly associated with reduced mortality, improved coverage, access to care, and self‐reported health. The authors discuss that their findings support a “plausible causal chain,” but do not use causal language, as the secondary outcomes of the study were not based upon individual‐level data. (The percentages of persons with Medicaid and in “excellent” or “very good” health were compared with the percentages without any health insurance.) 12 Moreover, the study examines three Medicaid expansion states compared with three control states and focuses on the impact within the low‐income population, which may limit generalizability to the entire US population. The effect of health insurance coverage on health outcomes has been mixed, with most prior studies evaluating the relationship over a shorter time frame and for specific populations.

The objective of this work was to add to the evidence base on causal linkages between health insurance coverage and health status, controlling for sociodemographic factors, by analyzing individual‐level longitudinal data that are representative of the entire US population over a time horizon sufficient to detect a benefit if one exists. In particular, while it may be argued that there is a selection effect in that people choose to purchase—or not—health insurance coverage, the reality is that for most Americans, the choice is severely limited by one's education (making one eligible for the type of job that is likely to provide employer sponsored coverage) and one's income (as private market plans are not affordable for many higher‐income individuals, and a very low income qualifies a person for Medicaid coverage). Therefore, controlling for education and income, in particular, should be helpful in correcting for the selection effect. The longitudinal nature of the data helps us separate out the competing explanations for an association between health insurance and health: especially prior to the ACA, a person might become sick and then become uninsured either due to becoming unemployed or being denied coverage. The PSID helps identify such a chronology while also providing a cumulative health insurance variable that essentially allows us to capture a dose/response effect.

Evidence from the 2017 Medical Expenditure Panel Survey shows that self‐reported health status is highly correlated with total medical spending. Expenses incurred by those in excellent or very good health in 2017 averaged $3029; by those in good health averaged $6,545; and by those in fair or poor health averaged $16,670. A decline from “good” to “fair” doubled the average spending from $6545 to $13,918, 13 so any policy that reduces the chance that an individual's health becomes “fair” or “poor” is likely to be associated with significantly less utilization and cost.

2. METHODS

The study was observational, using secondary data on self‐reported health status, work limitations, and death. These variables were analyzed as outcomes in a set of logistic models in which the degree of exposure to health insurance in 2011‐17 is the independent variable of interest. The models controlled for additional sociodemographic variables, including sex, race (white, black, and other race), educational status (high school less, some college, college degree or more), and poverty level. All variables were collected biannually for four years beginning in 2011. Having health insurance at each point in time was, in turn, modeled as a function of several sociodemographic factors. The data are from the Panel Study of Income Dynamics (PSID), 2009‐17, which is a longitudinal, multigenerational study covering a wide array of socioeconomic topics that began in 1968 but has only recently begun collecting useful information on individual health insurance. Data were downloaded using the crosswalk tool available at the PSID website. 14 Because individual health insurance questions were only asked of heads and spouses in households beginning in 2011, we analyzed only these records, creating a total sample size of 9744. The specific question asked was “Do you currently have health insurance?” In each model, the sample was restricted to those with nonmissing data for the outcome variable and regressors.

Certain variable translations were made. Because health care utilization and expenditures increase significantly for individuals reporting “fair” and “poor” health, 15 we combined these categories to create an outcome variable for being in either fair or poor health, vs. responding that one's health was “good,” “very good,” or “excellent.” For the insurance and poverty status variables, an adjustment was necessary due to some members of the sample dying during the sampling period. Therefore, we created a “percent of time insured” variable that equals 100 if the individual was insured over all 4 observation years—or if they were insured for all of the years in which they were alive. If they were insured for half of the years that they were alive, the variable equaled 50. If death occurred after 3 years, and the individual was insured for 1 of 3 years, then the variable equaled 33. A similar variable was constructed based upon the amount of time, in years, that an individual's annual income was above the federal poverty level (FPL). Someone alive all 4 years, with income below 100 percent FPL in 2 of the years, would have a value of 50. An individual who died after 3 years but lived in poverty for none of them would have a value of 100. These variables are constructed such that higher values correspond to better circumstances and are therefore hypothesized to reduce the chances of an individual falling into fair or poor health.

We used SAS PROC SURVEYLOGISTIC (SAS Enterprise Guide version 7.15) and applied appropriate sampling weights (which correct for oversampling and attrition) to estimate a set of logistic models to determine whether the degree of exposure to health insurance between 2011 and 2017 affected the chance that a person would rate their health in 2017 as fair or poor. We ran unconditional and conditional models; in the latter, we subset our sample to include only individuals who were not in fair or poor health in 2009. Similar models were run for the outcome variables of death (by 2017) and reporting of health limitations in ability to perform work in 2017.

3. RESULTS

Table 1 displays characteristics of the 9064 PSID respondents with nonmissing data in 2009 and who either had nonmissing data in 2011‐17 or who died in one of those years. Of these, 53 percent were females, while 82 percent were white, 13 percent were black, and 5 percent identified as another race. About 43 percent had a high school education or less, while 33 percent had a college degree or more education. In 2009, more than half (52 percent) reported being in excellent or very good health, while only 17 percent reported fair or poor health. A sizeable majority (79 percent) were insured for all four measurement years, with 19 percent experiencing partial coverage over time and only 2 percent being continuously uninsured in all years. Similarly, a large majority (81 percent) had incomes above the federal poverty level in all four years, while 16 percent had incomes that fluctuated above and below, and only 2 percent were consistently in poverty in all years. Table 2 further breaks down the sample by displaying the distribution of self‐reported health status categories for each demographic group. It shows that females, blacks, and those with less education and more consistent low incomes are somewhat more likely to have been in fair or poor health in 2009, which highlights the importance of comparing unconditional models with those conditioned on not being in fair or poor health in 2009.

Table 1

Characteristics of PSID sample population, weighted a