Wednesday, November 18, 2009

Another Media Failure Over A Questionable Study: Medicare Associated With Increased Trauma Deaths? No Better Than NOT Having Insurance?

Did you know that Americans with Medicare health insurance have a higher adjusted mortality rate from trauma compared to persons with commercial insurance?

The Disease Management Care Blog explains.

According to numerous news outlets such as ABC News, MSNBC, the Huffington Post and Reuters, a study's been published that shows that not being insured at the time of hospitalization for trauma leads to a higher death rate. Adding to the drumbeat of liberal dismay over lives lost for lack of a public option, this research by Rosen and colleagues, published in the prestigious Archives of Surgery, shows there are 'sky high' death rates (Democratic Underground) that are 'shocking' (Daily Kos) and could be the evil consquence of 'less enthusiasm on the part of providers.... once it was realized they would not be compensated' (Physicians for a National Health Program). Of course, the luster of this scientific report was only enhanced further by the inclusion as author of superstar surgeon Atul Gawande, that (in)famous author of the New Yorker article that used the shortcomings of single outlier McAllen Texas to call the entire U.S. healthcare system to task.

Like the media, should the DMCB be impressed by a published study in a prestigious journal by rock star authors? Just take their word for it? Actually no, because the purpose of peer reviewed literature is to allow readers to assess the research findings and decide for themselves. It's called trust, but verify.

So, unlike the attention-deficit disordered denizens of the oxygen-deprived mediasphere, the DMCB responded in a novel way. It decided to pull the study and actually take the time to read it.

Here's what it found. Data from 2.7 million U.S. hospitalized patients were mined, looking for correlations between death and insurance status while mathematically neutralizing the effect of age, gender, race, the type of trauma center as well as the type of trauma. Persistent and statistically significant correlations were found between dying and a) increasing age, b) race (being black plus young) c) the severity and mechanism of the trauma (for example penetrating injury is bad), d) an increased number of comorbid illnesses and e) insurance status.

In fact, two types of insurance status had statically significant associations with a higher death rate: 1) not having any health insurance and 2) being on Medicare. If you look at the graphic from the study, you'll see that compared to commercial insurance, Medicare had a statistically significant odds ratio of a higher death at about 1.5, while no insurance was also high at 1.8. Look carefully, because the lack of an asterix means the finding is statistically significant.

The DMCB thoughts about the implications:

1) What should the 'comparator' be? We don't know. One separate study used managed care as the baseline while this other study used Medicare. Since there is no generally accepted baseline for research like this, the author's choice of using commerical insurance as the gold standard in this article made not having insurance look bad. If Medicare had been the standard, persons without health insurance probably would not have turned out to be worse. Even more ironically, managed care (accounting for the majority of U.S. commercial insurance) would have turned out looking better. Interestingly, the authors neglected tell us why they chose this approach.

2) Are the authors (and the Archives' editors) politically motivated? There is one explanation for the use of the commerical comparator: it's a ringer that makes everyone else look bad. What's more, failing to mention that Medicare insurance status was also associated with lower survival rates is either a monumental lapse or an intentional attempt to understate a finding that is also important. Knowing that policy makers, media and many readers won't get past the title ('Downwardly Mobile. The Accidental Cost of Being Uninsured'), the abstract (no mention) and some juicy interviews, only half the story is being told. The full story is that when it comes to trauma care, giving everyone Medicare-style insurance is no better than no health insurance at all. This is a great example of framing, especially since the asterix seem to call the readers' attention away from Medicare's inconvenient......

3) Are some other potential explanations? While a lack of health insurance is associated with poor health outcomes, it is also known that having poor health leads to lack of insurance. Accordingly, the mathematical 'signal' from being nominally 'uninsured' in this study may really be due to the influence of unmeasured health or other issues that were not captured in the data base. That's called systematic bias and it went completely unmentioned as a potentional shortcoming in the authors' discussion of their results. Assuming the results are real, the DMCB believes post-hospitalization care (rehab and outpatient) is generally not well covered and coordinated under Medicare, which in turn leads to problems.

4) Association is not necessarily causality. Just because not having insurance is associated with death from trauma doesn't mean giving this poverty-prone population insurance will reverse things any more than the association between white hair and more heart attacks (due to age) can be fixed by dying everyone's hair black. Likewise unmentioned by the authors.

Once again, the media has demonstrated its shortcomings. What's more, the peer reviewed process has shown how necessary it is to look at the results for yourself: you can't always count on the authors or the editors to look at all sides of the data or rise above their prejudices. Lastly, there's an old joke out there that is sometimes told by us general internists: how do you create a double blind study? Get two or more surgeons involved.

This publication was double binded.

At the time of this posting, the DMCB had an email into the author asking for feedback. None yet

No comments: