Who’s killing whom?
Writing in The New York Times the other day, Paul Krugman had this to say:
And while supposed ObamaCare horror stories keep on turning out to be false, it’s already quite easy to find examples of people who died because their states refused to expand Medicaid. According to one recent study, the death toll from Medicaid rejection is likely to run between 7,000 and 17,000 Americans each year.
Really? 17,000 deaths per year? This outrageous claim flies in the face of years of careful study by real economists — who have concluded that there is almost no relationship between health insurance and mortality in the general population. As we previously reported:
In independent empirical papers, Richard Kronick and David Card and his colleagues find little evidence that health insurance coverage significantly reduces mortality. Former Director of the Congressional Budget Office June O’Neill and her husband Dave also conclude that lack of insurance has little or no impact on mortality. See the discussion at this blog here, here and here.
So where does Krugman’s claim come from? Not from economists, it turns out, but from a Health Affairs blog post by Sam Dickman, David Himmelstein, Danny McCormick and Steffie Woolhandler. And notably, they do not do what serious scholarly papers do — acknowledge the work of other scholars who have addressed this same topic.
Their numbers rely on surveys known to overestimate the uninsured and on results from papers with methodological problems that are both serious and widely known. Their claim that failing to expand Medicaid in conformance with ObamaCare dictates will kill somewhere between 7,115 and 17,104 people a year confuses ideological posturing with scholarship.
Knock, knock, knockin’ on heaven’s door
The older studies concluding that being uninsured increases mortality tend to be of lower quality than those that find it has no discernable effect. Dickman et al. continue this tradition in selecting the research results underlying their methodology for calculating the upper and lower bounds for their death estimates. The upper bound estimates depend upon the results in a 2012 paper by Sommers et al. which neglects the effects of HIV infection when comparing all-cause mortality in New York with that of other states. The lower bound estimates depend upon the results in a 2009 paper by Wilper et al. It assumed that people who were uninsured at the time they were interviewed remained uninsured a decade later and found that health insurance was so important that having it reduced an individual’s chance of dying by 40 percent.
The Sommers paper concludes that Medicaid expansion reduced unadjusted nonelderly all-cause mortality by 19.6 deaths per 100,000 from a baseline of 320 deaths per 100,000. It compared mortality in three states that expanded Medicaid coverage in the early 2000s (New York, Maine and Arizona) with mortality in four neighboring states (Pennsylvania, New Hampshire and Nevada plus New Mexico) that did not. New York accounted for about 45 percent of the sample population in the paper. The authors write that their results were “largely driven by” New York.
The time periods used for the comparison were the five years immediately before the first full year in which a state’s Medicaid expansion took effect, and the first five years after the expansion.
Aside from the usual problems inherent in the sort of ecologic modeling the Sommers paper is attempting, its results were likely influenced by the higher than normal 2001 mortality rate caused by the World Trade Center attack, and by New York’s extremely high rate of HIV infection.
It is easy to see the World Trade Center mortality bump in the red line of the graph on the right, which comes from page 1,029 of the Sommers paper. Year 0 is the year in which each state’s Medicaid expansion was first implemented. It is 2002 for New York and Arizona, and 2003 for Maine. Maine’s population makes up about 5 percent of the sample population and deaths each year.
The expansion sample death rate rose sharply in the expansion states in year -1 to 0, or 2001, falling back to previous levels in year 0 to 1, or 2002. While it is true that mortality immediately declined the year the two populous states expanded Medicaid, the mortality rate looked good in 2002 only because terrorism had made it look so bad in 2001.
Using data from CDC Wonder and New York State vital statistics to estimate all-cause mortality for the nonelderly suggests that excluding the World Trade Center deaths would lower 2001 mortality for the expansion states from 326 per 100,000 to about 314 per 100,000. This would erase the bump in the graph. It would also erase part of the reduction in mortality seen after the Medicaid expansion in the “expansion states.”
The marked decline in the crude mortality rate in the expansion states also appears to reflect the fact that at a time when HIV infection was a leading cause of death for nonelderly adults, New York’s HIV death rate was one of the highest in the country. The introduction of antiretroviral therapies in 1995 reduced the HIV mortality rate from very high rates in the late 1990s to high, but steadily declining, rates in the Medicaid expansion period.
As the second graph shows, New York’s HIV death rate continued to decline after Medicaid was expanded, a decline that lowered the overall mortality rate in the expansion states. The reduced death rates were likely caused by antiretroviral therapies rather than Medicaid expansion. In Pennsylvania, a state that did not expand Medicaid and which was used as the “control state” for New York, death rates from HIV fell from 1997 to 1998, but were relatively constant thereafter and likely had little effect on the overall mortality rate simply because Pennsylvania did not have as many HIV deaths as New York.
Unless further research takes these differences into account, the Sommers paper will remain unconvincing in its claim that Medicaid expansion in New York, Arizona and Maine reduced all-cause mortality rates. As a result, the upper bound deaths from failing to expand Medicaid that are asserted in the Dicks et al. blog post are not credible.
In addition, here is what Chris Conover has to say about the Sommers paper:
I can state with great confidence that the authors have grossly overestimated any mortality gains to be had from Medicaid expansion. The evidence that Medicaid even has a positive effect on adult mortality risk is far more thin than the Harvard/CUNY team has led you to believe.
The Wilper paper uses the same strategy as the 1993 paper by Franks, Clancy and Gold to estimate that people who self-reported that they were uninsured in 1988-94 were 40 percent more likely to have died 6 to 12 years later in 2000. It applies this result to all states and claims that a lack of health insurance is associated with as many as 44,789 deaths a year in the United States.
This conclusion is based on self-reported insurance status for nonelderly adults who participated in the 1988-1994 NHANES III study. After excluding all the individuals who were in the survey and were covered by government health insurance, they were left with 12,112 people. Of those, 9,004 satisfactorily completed the interview and health examination and could be identified as living or dead in 2000. The error rate in the living or dead attribution is unknown. Earlier calibration studies estimated that 4 percent of NHANES respondents may be incorrectly classified as dead in follow-up surveys.
The problems with the Wilper methodology have been extensively discussed. Two examples may be found on this blog here and here. First, people who were uninsured at the time they were interviewed do not necessarily remain that way. Second, as Wilper et al. point out, other work suggests that between 7 and 11 percent of those who report that they are uninsured actually have insurance. Third, it is well known that the uninsured population differs from the privately insured population in important ways that are likely to have significant influence on health but are not controlled for in the Wilper study. Fourth, there is a literature showing that deaths are inversely related to income, and that higher incomes are associated with improved mental and physical health. Because Medicaid expansions and ObamaCare are likely to reduce incomes, their effects on health are highly unlikely to be all that their advocates claim.
Chris Conover at Forbes concludes that people will not die in states that don’t expand Medicaid.