Tuesday, November 17, 2009

Does Lack of Health Insurance Cause 45K Deaths Per Year in the US?

This paper has been cited by a lot of people recently, including Nicholas Kristof in his November 12th op-ed (I'll have more to say about Kristof later as he has been spreading health care falsehoods under the guise of countering health care myths).

The research has been interpreted as demonstrating that approximately 45K people die every year in the US because of lack of health insurance.

This is almost certainly wrong.

I note the following points to keep in mind about the paper:

1) the principal author is PNHP (Physicians for a National Health Plan) founder Dr. David Himmelstein;

2) in the paper itself the authors do not claim to have shown a causal relationship between lack of health insurance and higher mortality; they only claim to have shown that there is a positive correlation.

PNHP has an extreme agenda. The group was formed to promote socialized medicine in the US. In addition, PNHP recognizes that its goal of equal access to health care is undermined if a private market for medical services is allowed to develop in parallel with the government run network (such a 2-tier system exists in the UK). Therefore, PNHP wants essentially to ban private medical practice. Himmelstein et al want to make it impossible for someone to pay to see a doctor.

The fact that Himmelstein is the principal author of the paper does not necessarily mean that the paper is wrong (although one should probably be more skeptical than usual). Even if the paper is substantially correct, we should keep in mind that Himmelstein's political agenda may interfere in a number of ways. For example, data that undermines his political agenda may have been suppressed or at least not released publicly. Also, he may spin, or allow others to spin, the paper's results in a way which is not justified by the data.

I think the latter concern is clearly at play right now, and primarily with respect to the rather widespread confusion about causality vs correlation. That is, when sticking to the science, the paper only makes a claim about there being a positive correlation between lack of insurance and mortality rates (1.4x higher mortality rates for the uninsured vs the insured population when adjusted for a variety of factors). But correlation is not causation. Although the uninsured have a higher mortality rate, it may be due to some factor other than the lack of insurance. This is important because it might be that even if these people were given insurance for free, their mortality rates would remain high. I suspect this to be the case, at least in part.

Although not specifically mentioned in the Himmelstein paper (a surprising and telling omission I think), a reasonable hypothesis is that there is self-selection involved among people who are uninsured. People who are responsible about their health, who care about their health, and who therefore will go to the doctor when they are sick and then stick to a prescribed course of treatment, those people are more likely to care enough to go out and make sure they have insurance. It is the lazy, irresponsible people who will be skewed towards being uninsured.

I think that this effect is real, even in a system where most people get insurance through their employer. Remember, most honest economists would agree that health insurance is paid for out of employee wages, not out of company profits. All other things being equal, jobs which do not provide health insurance will pay higher wages than jobs that do. People who do not value health insurance will choose those jobs that do not provide health insurance because they will earn commensurately higher wages.

The authors of the paper try to normalize for various factors, including income, education, ethnicity, and current health status. But they cannot control for personal responsibility. In fact, their attempt to normalize for health status might exacerbate the effect they have captured (i.e. it makes the correlation between lack of insurance and mortality rates higher) because people who are already sick are probably focusing more on getting health insurance than people who are healthy. In other words, those who have good health status and are short-sighted about their health are more likely to be uninsured. So those without insurance are probably in a healthier pool after controlling for income, education, and ethnicity.

[To be fair, I should mention that the authors think that the effect of controlling for health status can go the other way because those who have poor health status at the start of the study might have already suffered from lack of health insurance.]

In any case, the paper only claims correlation, not causation. The researchers did not have access to cause of death data, and, accordingly, there was no way to know if lack of insurance played any role at all in the excess death rate for the "normalized" pool of uninsured.

Perhaps by design, however, the authors of the paper have created a perception in the media that causation has been proved and that 45K deaths per year could be prevented if we extended health insurance coverage to every US resident. I think that if you cornered Himmelstein himself, he would back off of that claim very quickly.

This paper finds that there is actually little correlation between excess mortality for the uninsured when mortality is restricted to "amenable" causes of death. Additionally, it finds that there is no change in the relative mortality rates for the uninsured pool of people once they become eligible for Medicare coverage at age 65. That is, giving the uninsured insurance does not improve their survival rates relative to the insured, at least once they've reached age 65.

Of course, as I've expressed before, even if the lack of health insurance causes excess mortality rates, we have to judge that effect relative to other wealth or income based factors that cause excess mortality. Eating cheap, starchy food, living in crime-ridden neighborhoods, driving unsafe cars, and working in dangerous jobs all cause excess deaths among the poor and are clearly direct consequences of being poor. If the money required to give out free health insurance could get better results in these other areas, then we shouldn't waste our limited resources on universal health insurance.