Health Insurance and Mortality in US Adults Part 4

Limitations
Our study has several limitations.
NHANES III assessed health insurance at a single point in time and did not validate self-reported insurance status. We were unable to measure the effect of gaining or losing coverage after the interview. Point-in-time uninsurance is associated with subsequent uninsurance.6 Intermittent insurance coverage is common and accelerates the decline in health among middle-aged persons.33 Among the nearelderly, point-in-time uninsurance was associated with significant decline in overall health relative to those with private insurance.13 Earlier population-based surveys that did validate insurance status found that between 7% and 11% of those initially recorded as being uninsured were misclassified.13 If present, such misclassification might dilute the true effect of uninsurance in our sample. We excluded 29.5% of the sample because of missing data. These individuals were more likely to be uninsured and to die, which might also bias our estimate toward the null.

We have no information about duration of insurance coverage from this survey. Further, we have no data regarding cost sharing (out-of-pocket expenses) among the insured; cost sharing worsened blood pressure control among the poor in the RAND Health Insurance Experiment, and was associated with decreased use of essential medications, and increased rates of emergency department use and adverse events in a random sample of elderly and poor Canadians.37,38 Unmeasured characteristics (i.e., that individuals who place less value on health eschew both health insurance and healthy behaviors) might offer an alternative explanation for our findings.

Health Insurance and Mortality in US Adults Part 3

DISCUSSION
The uninsured are more likely to die than are the privately insured. We used a nationally representative data set to update the oft-cited study by Franks et al. and demonstrate the persistence of increased mortality attributable to uninsurance. Our findings are in accord with earlier research showing that lack of health insurance increases the likelihood of death in select illnesses and populations.5–7,13 Our estimate for annual deaths attributable to uninsurance among working-age Americans is more than 140% larger than the IOM’s earlier figure.23

By using methodologies similar to those used in the 1993 study, we found that being uninsured is associated with a similar hazard for mortality (1.40 for our study vs 1.25 for the 1993 study). Although the NHANES I study methodology and population were similar to those used in NHANES III, differences exist. The population analyzed in the original study was older on average than were participants in our sample (22.8% vs 55.6% aged 34 years or younger). The maximum length of follow-up was less (16 years vs 12 years), and the earlier analysis was limited to White and Black persons, whereas the present study also includes Mexican Americans.
The relative youthfulness and shorter follow-up in our study population would be expected to reduce our power to detect an elevated risk of death. In addition, if gaining Medicare reduces the effect of uninsurance on mortality, then the younger age and shorter length of follow-up in our study might strengthen the association between uninsurance and mortality compared with the earlier study. It is less clear how the differences in the racial and ethnic make-up of our study population would affect our ability to detect difference in risk of death. In fact, the increased likelihood of uninsurance among Mexican Americans who were nonetheless no more likely to die than non- Hispanic Whites might also be expected to reduce our power compared with the earlier study.

Health Insurance and Mortality in US Adults Part 2

We repeated the analysis of the relationship of insurance to mortality after forcing all covariates in the model. In this Cox proportional hazards analysis, we controlled for gender, age, race/ ethnicity (4 categories), income (poverty income ratio), education, current unemployment, smoking status (3 categories), regular alcohol use, self-rated health (4 categories), physicianrated health (4 categories), and BMI (4 categories). We tested for significant interactions between these variables and health insurance status (i.e., P<.05). We handled tied failure times by using the Efron method. We performed multiple sensitivity analyses to analyze the robustness of our results.


We developed a propensity score model and controlled for the variables in our previous models (with the exception of health insurance status), as well as marital status; household size; census region; number of overnight visits in hospital in past 12 months; number of visits to a physician in past 12 months; limitations in work or activities; job or housework changes or job cessation because of a disability or health problem; and number of self-reported chronic diseases, including emphysema, prior nonskin malignancy, stroke, congestive heart failure, hypertension, diabetes, or hypercholesterolemia. Next, we included the propensity score in the multivariable model with the indicator for insurance status. In addition, we tested for the effect of including those covered by Medicaid by using our original Cox model and the propensity score adjusted analysis.


In a subsidiary analysis, we excluded employment and self- and physician-rated health, as these covariates may be a result of limited access to health care because of uninsurance. To facilitate interpretation of our hazard ratio, we first replicated the calculation in the IOM report to estimate the number of US adults who die annually because of lack of health insurance. This approach applies the overall hazard ratio to 9-year age strata and sums these figures to arrive at an annual number of deaths attributable to lack of health insurance. We then recalculated this figure by using the slightly different approach utilized by the Urban Institute, which does not age stratify when calculating total mortality. We believe this approach to be more accurate than that used to produce the IOM estimate, as it calculates mortality from the entire age range that the hazard ratio was calculated from, as opposed to calculating mortality over 10-year age strata.23

RESULTS

Health Insurance and Mortality in US Adults Part 1

NHANES III staff interviewed respondents in their homes regarding demographics (including health insurance). Participants responded to questions about race, ethnicity, income, and household size. The sample design permits estimation for 3 racial/ethnic groups: non-Hispanic White, non-Hispanic Black, and

Mexican American. The NCHS created a variable that combined family income and the poverty threshold during the year of interview (the poverty income ratio), allowing income to be standardized for family size and compared across the 6 years of data collection.18

NHANES III interviewers also collected data on education, employment, tobacco use, alcohol use, and leisure exercise. We analyzed education dichotomously, comparing those with 12 years or more education to those with less than 12 years. We considered respondents to be unemployed if they were looking for work, laid off, or unemployed. All others, including the employed, students, homemakers, and retirees were considered ‘‘not unemployed.’’ We considered smokers in 3 categories: current smokers, former smokers (those who had smoked more than 200 cigarettes in their lifetime), and nonsmokers. We labeled those drinking more than 6 alcoholic beverages per week as regular drinkers. We analyzed exercise in 2 groups:those achieving greater than or equal to 100 metabolic equivalents (METs) per month, versus those achieving less than 100 METs per month.19,20

NHANES III

Health Insurance and Mortality in US Adults

The United States stands alone among industrialized nations in not providing health coverage to all of its citizens. Currently, 46 million Americans lack health coverage.1 Despite repeated attempts to expand health insurance, uninsurance remains commonplace among US adults.
Health insurance facilitates access to health care services and helps protect against the high costs of catastrophic illness. Relative to the uninsured, insured Americans are more likely to obtain recommended screening and care for chronic conditions2and are less likely to suffer undiagnosed chronic conditions3 or to receive substandard medical care.4 Numerous investigators have found an association between uninsurance and death.5–14

The Institute of Medicine (IOM) estimated that 18314 Americans aged between 25 and 64 years die annually because of lack of health insurance, comparable to deaths because of diabetes, stroke, or homicide in 2001 among persons aged 25 to 64 years.4 The IOM estimate was largely based on a single study by Franks et al.5 However, these data are nowmore than 20 years old; both medical therapeutics and the demography of the uninsured have changed in the interim.
We analyzed data from the Third National Health and Nutrition Examination Survey (NHANES III). NHANES III collected data on a representative sample of Americans, with vital status follow-up through 2000. Our objective was to evaluate the relationship between uninsurance and death.
METHODS
The National Center for Health Statistics (NCHS) conducted NHANES III between 1988 and 1994. The survey combined an interview, physical examination, and laboratory testing. NHANES III employed a complex sampling design to establish national estimates of disease prevalence among the noninstitutionalized civilian population in the United States.15 Staff performed interviews in English and Spanish.