Numbers do not add to totals because of rounding.
2 The sample sizes were probably not enough to yield reliable estimates for 10-year age bands. The study by Franks and colleagues used data from the National Health and Nutrition Examination Survey, with a cohort that included 1,287 respondents age 25–34, 1,035 respon- dents age 35–44, and so on. These cohorts were not large enough to esti- mate the impact of insurance status after controlling for gender, race, education, income, employment status, self-rated health, morbidity, exercise levels, smoking status, alcohol consumption, and obe- sity. The much larger CPS sample studied by Sorlie and colleagues was not large enough to yield valid estimates for blacks, much less for the smaller number of adults within various 10-year age bands.
3 This happens because younger adults are more likely to be uninsured, while older adults have higher mortality rates. In the studies on which IOM relied, the higher mortality rates for older adults and the higher uninsurance rates for younger adults were, in effect, distributed across the entire age range. That same effect arises when the IOM equations are applied evenly throughout the entire population of adults age 19–64, as this paper recommends, based on the unin- surance rate and the mortality rate of the entire group. It does not arise when the equations are applied separately to every 10-year age cohort, each with its own uninsurance and mortality rates.
The following hypothetical illustrates this peculiar result. Consider an imagi- nary group of 50,000 younger adults and 50,000 older adults with the following characteristics:
50 percent of the younger adults— or 25,000—are uninsured. 10 percent of the older adults—or 5,000—are uninsured. Altogether, 30,000 of the 100,000 adults—or 30 percent—are uninsured.
1 percent of the younger adults—or 500 people—die each year. 10 per- cent of the older adults—or 5,000—die each year. Altogether, 5,500 of the 100,000 adults—or 5.5 percent—die each year.
To condense the algebraic explana- tion in the text, the IOM’s methodology determines the number of excess deaths due to uninsurance based on the following equation: ED = DT – (DT/ (PI + (PU*1.25))), where: ED = the number of excess deaths
due to uninsurance DT = total deaths
PI = percentage of insured individuals PU = percentage of uninsured individuals
In our hypothetical, this equation can be applied either to the entire group at once or to each age cohort separately.
A single group-wide application. Applying this equation to the entire group of all adults, with a single consolidated death rate and a single consolidated rate of uninsurance, yields the estimate that 384 adults died because of uninsurance. For the group as a whole, ED = 5,500– (5,500/(70%+(30%*1.25))) = 5,500– (5,500/ (70%+37.5%)) = 5,500– (5,500/1.075) = 5,5 00–5,116.3 = 383.7.
Multiple cohort-specific applications. Applying this equation separately to younger adults and older adults, each group with its own death rate and unin- surance rate, yields the quite different conclusion that approximately 178 adults died because of uninsurance, including 56 younger adults and 122 older adults:
Among younger adults, ED = 500– (500/ (50%+(50%*1.25))) = 500– (500/ (50%+62.5%) = 500– (500/1.125) = 500–444.4 =55.6
Among older adults, ED = 5,000– (5,000/(90%+(10%*1.25))) = 5,000– (5,000/(90%+12.5%) = 5,000– (5,000/1.025) = 5000–4,878.0 = 122.0
Analysis. To yield the same number of deaths as the single groupwide appli- cation, the cohort-specific applications would need to apply either (a) to both cohorts, a common estimate that unin- surance increases risk of death by 57.5 percent, or (b) for each cohort, a different estimate for the impact of uninsurance on risk of death (e.g., 75 percent increase for older adults and 15 percent increase for younger adults). What does not yield the same number of deaths is applying to each age cohort the groupwide estimate for the impact of uninsurance on risk of death.
The longitudinal studies on which IOM relied derived a single, groupwide esti- mate that uninsurance increased the risk of death among all working-age adults by 25 percent. To apply such groupwide findings consistently with those studies requires groupwide, rather than cohort- specific, calculations.
In theory, cohort-specific calculations make more sense because they take into account that the adults least likely to be uninsured have the highest mortality rates. Accordingly, such calculations will be pre- ferred when the research matures to the point of providing consistent documenta- tion of a full range of age-cohort-specific estimates of the impact of uninsurance on mortality. However, since current research documents the impact of uninsurance on mortality across the full set of work- ing-age adults, the calculations of excess deaths likewise need to take place across that full set if they are to remain optimally
grounded in the research.
The Urban Institute is a nonprofit, nonpartisan policy research and educational organization that examines the social, economic, and governance problems facing the nation.
This research was funded by the Robert Wood Johnson Foundation. The author appreciates the helpful advice and assistance of Ms. Fiona Blackshaw, Ms. Allison Cook, Dr. Jack Hadley, Dr. John Holahan, Mr. Joel Ruhter, Mr. Bogdan Tereshchenko, and Dr. Stephen Zuckerman of the Urban Institute; Dr. Wilhelmine Miller of George Washington University (formerly of the Institute of Medicine); and Dianne Wolman of the Institute of Medicine.
About the author: Stan Dorn is a senior research associate at the Health Policy Center of the Urban Institute.
The views expressed are those of the author and should not be attributed to the individuals listed above, to the Robert Wood Johnson Foundation, or to the Urban Institute, its trustees, or its funders.
Uninsured and Dying Because of It: Updating the Institute of Medicine Analysis on the Impact of Uninsurance on Mortality