Lately, I’ve had the feeling that “progressive” journalists and commentators too often pull their punches in calling attention to social problems, by underestimating the magnitude of problem-related statistics such as the unemployment rate and the number of fatalities due to lack of health insurance in the United States. My theory about this is that “progressives” are being defensive in their approach and bending over backwards to give the right wing the benefit of the doubt by understating numbers out of an abundance of caution.
If this is right, then my reply is that underestimating problem-related numbers is just as bad as over-estimating them, and that what people ought to do is try to provide the best estimates they can and let the critical chips fall where they may. In this post, I want to raise a question about the accuracy of the numerical estimates of US deaths due to lack of health insurance often seen in articles and blogs authored by people, like myself, who favor eliminating insurance coverage shortfalls by passing Medicare for All.
The Woolhandler/Harvard Study and Its headline Finding
When progressives write about the number of deaths associated with lack of health insurance they, almost invariably, directly cite or project from the headline finding of an analysis of data, gathered by the Third National Health and Nutrition Examination Survey (NHANES III), done at the Harvard Medical School whose results were published in the American Journal of Public Health in December of 2009, with pre-publication prints appearing at least as early as September of 2009. The peer-reviewed article reporting the results of the study directed by Steffie Woolhandler, was co-authored by Andrew P. Wilper, MD, MPH, Steffie Woolhandler, MD, MPH, Karen E. Lasser, MD, MPH, but pre-publication Danny McCormick, MD, MPH, David H. Bor, MD,and David U. Himmelstein, MD.
The NHANES III survey data was gathered over the period 1988 – 1994, and followup interviews were then conducted in the year 2000. The Harvard researchers analyzed the data using a Cox Regression proportional hazards model, a method widely used in medical research and epidemiology. They calculated “hazard ratios” from the data using this modeling and analysis method and then went on to calculate propensity scores, and finally used those to estimate fatalities among the uninsured using Census data.
The article reporting their study was frequently cited in the run-up to passage of the Affordable Care Act (ACA) in the Fall of 2009 – 2010, and its headline conclusion has been cited ever since in articles and blog posts much too numerous to cite. The headline conclusion, as normally stated is that there are 44,789 or sometimes approximately 45,000 fatalities annually associated with, or more boldly, caused, by lack of health insurance.
In the Wilper et al article however, the headline finding stated is (p. 2292):
Replicating the methods of the IOM panel with updated census data24,25 and this hazard ratio, we calculated 27424 deaths among Americans aged 25 to 64 years in 2000 associated with lack of health insurance. Applying this hazard ratio to census data from 200526 and including all persons aged 18 to 64 years yields an estimated 35327 deaths annually among the nonelderly associated with lack of health insurance. When we repeated this approach without age stratification, (thought by investigators at the Urban Institute to be an overly conservative approach)23 we calculated approximately 44789 deaths among Americans aged 18 to 64 years in 2005 associated with lack of health insurance.
IOM in the above quote is the Institute of Medicine of the National Academy of Sciences, and the citations in the above passage are to Centers for Disease Control and Prevention, National Center for Health Statistics, Census Bureau, and Urban Institute Reports. Note that the number of deaths estimate applies to the population 18 – 64 years for the year 2005. There’s no claim in the above quote that the same estimate would apply to the years since that time or to the whole population under 65. But on p. 2294, in the conclusions section, the authors say:
Lack of health insurance is associated with as many as 44789 deaths per year in the United States, . . .
This is clearly an overdrawn conclusion since it’s dependent on the 1) estimated hazard ratio remaining constant, 2) estimated base populations of both insured and uninsured remaining the same when these are changing every year, and 3) extent to which the various limitations they mention on pp. 2293 – 2294 of their article lead to an underestimate of the hazard associated with being uninsured. The likelihood of these limitations leading to an underestimate of deaths in their study is very great, as the authors themselves say.
So, given this, why say: “. . . as many as 44789 deaths . . . “ above. Why not say “. . . . at least 44,789 deaths in 2005 and most likely substantially more than that . .. “? And why not also point out that since in the years after 2005 the number of uninsured has increased compared to 2005, and that the deaths have also increased, even assuming that the hazard ratio has remained the same, and hasn’t gotten worse?
A Question About the Study and Two Error Scenarios
But beyond this, the article by the Harvard researchers raises a question that, depending on the answer, suggests an error that, in turn, suggests they may have provided a substantial underestimate of the number of annual fatalities even based on the data. In the quote from p. 2292, the 44,789 deaths associated with lack of health insurance in 2005 are assigned to the 18 – 64 population, not the whole population under 65.
This statement may be an error of reference. The authors may have meant to say that this was the total for the under 65 population; or alternatively they may have stated their finding accurately. Their conclusion doesn’t shed light on this ambiguity, because it doesn’t mention the 18 – 64 age group. The New York Times article announcing release of the study quotes or paraphrases Dr. Steffie Woolhandler liberally, but uses only the “ . . . 45,000 deaths a year in the United States, . . .” without qualification by age group. Every other citation of the study I’ve seen since its release does the same, in plain contradiction with what the quote from p. 2292 states. So, which is it, an error of reference, or a more substantive error in not estimating deaths for the under 18 age group and adding them to the total?
Table One: Estimates of Population and Fatalities 2005 – 2012 Due to Lack of Insurance Under the Two Harvard Study Error Scenarios
Table One does a few things. First, it reports the most current Census Bureau population and uninsured population numbers for 2005 – 2012. Second, it provides estimates of deaths associated with lack of insurance from 2005 – 2012, assuming that the Harvard study made a misstatement in reporting that its deaths number of 44,789 applied to the 18 – 64 group, rather than to everyone under 65, as is implied by its conclusion. My estimates used the ratio of the 2005 death estimate to the uninsured estimate of 43,035,000 which is .0010407, and projected it to the years 2006 – 2012.
In effect, I’m assuming that the hazard ratio, on which the 2005 estimate is based, is constant over the period 2005 – 2012. It may not be, but I know of no data to the contrary. So, until there’s new data I think that assumption is the best one we can make.
Third, Table One also provides death estimates for the 2005 – 2012 period assuming that the deaths reported do apply to the 18 – 64 population and that this number needed to be supplemented by estimating 2005 deaths based on the population under 65 including children under 18 as well as the relatively small population over 65 that is uninsured (presumably people who have not enrolled in Medicare or any other insurance plan).
The Census Bureau report linked to above, estimates total uninsured in 2005 at 43,035,000, and uninsured children under 18 at 7,636,000. So, the total number of uninsured excluding the children is 35,399,000. It also turns out that there are 424,000 uninsured over 65. So, the uninsured population in the 18 – 64 group is estimated at 34,975,000, and this is the number of insured on which the estimate 44,789 is based, if the report of results as stated on p. 2292 is NOT in error.
The ratio of the death estimate to the uninsured population estimate in the 18 – 64 group is 44,789 to 34,975,000, or .0012806. Applying that ratio to the population of uninsured given by the census data for 2005 – 2012 gives the death estimates in the rightmost column of Table One. Again the underlying assumption is that the hazard ratio is constant over time.
So, which of the two errors was made in the Harvard Study has differing implications for estimates of deaths due to lack of insurance progressives that should be citing from the years 2005 – 2012. The second error scenario in Table One, if true, shows a much more serious underestimate of likely mortality than the first, but even the first error scenario involves a substantial underestimate of the deaths due to lack of insurance over the eight years from 2005 – 2012. Specifically, the assumption that deaths have averaged 45,000 per year and totaled 360,000 over the period, leads to a result that estimates 27,801 deaths less than can be reasonably projected. What reason is there for putting this kind of shortfall forward?
In the Harvard Study, the authors mention various limitations of the study. The limitations collectively suggest that the 44,789 estimate is itself a substantial underestimate of the reality. So what sense does it make to understate the problem of deaths due to lack of insurance further by continuing to cite the 2005 result without trying to update it for the increases in the number of uninsured that have occurred after 2005? It just gives away unnecessary ground to those who want to maintain the status quo and leave us with the fatalities their indifference to the problem is creating.
So, what am I saying? If you’re going to cite estimates of death due to lack of insurance, then please update them for changes in the size of the uninsured population before tossing them off.
And what if the second error scenario is true? Then the problem of under-estimation is still more serious, specifically the under-estimate under this scenario is 117,174, or an average of 14,647 per year, an under-estimate of the seriousness of the problem by more than 30%. If the second error really occurred in the Harvard study, then this degree of under-estimation is a lost opportunity to make the opponents of Medicare for All, including both those who want no Government in health insurance at all, and supporters of the ACA face the consequences of their actions.
Since 2010, those consequences have included anywhere from approximately 150,000 to 187,000 fatalities due to lack of health insurance. That’s unconscionable and immoral, especially because there was another choice, a road we failed to take. Let’s hope we can get on that road again rather than having to wait for it to be blindingly obvious that beyond all the spin, a neoliberal, market-placed based system of health insurance, can only lead to injustice, more death and misery for many Americans.
(Cross-posted from New Economic Perspectives)