(This is cross-posted on Taking Note.)
Avik Roy is an creative and engaging blogger. So is Austin Frakt. In addition, though, Frakt is a highly-skilled empirical health services researcher. So it’s a bit odd that Roy chooses to mix things up with Austin regarding instrumental variables studies that explore the impact of Medicaid coverage on health outcomes.
It is odder still that Roy writes a piece called “Why Medicaid is a Humanitarian Catastrophe.” He cites a useful but readily-misinterpreted surgical study which finds notably higher adjusted mortality rates among Medicaid patients than among the privately insured. Estimated mortality rates among Medicaid patients were also higher than that found among Medicare patients and the uninsured. Roy concludes the following from these findings:
This is, simply put, the greatest scandal in America. Bigger than Madoff, bigger than the Wall Street bailout, bigger even than the plight of the uninsured.
For all of TARP’s flaws, the government actually made money—$22 billion—bailing out the banks. For Medicaid, in contrast, we spend half a trillion dollars every year to provide the poor with worse care than is gained by the uninsured. How is this even possible? …
After I wrote about the issue, I expected many left-of-center policy types to share my concern. But instead I met indifference, denial, resistance, and even an accusation: that I was using the UVa study as a pretext to abandon the poor. In fact, the truth was and is quite the opposite. I want the poor to have great health care. On the other hand, very few people on the Left were willing to take on the challenging task of exploring why Medicaid patients fare worse than the uninsured, and how we might make Medicaid better.
I am just baffled by Roy’s essay. The study he cites cannot show what Roy claims it shows. (Along the way, Roy gratuitously insults Jonathan Gruber’s exemplary two-decade record of Medicaid policy research. I’ll let Jonathan respond for himself if he chooses.)
Roy calls this a landmark study, suggesting that is shows that people fare worse on Medicaid than they can by going uninsured and getting what people can on their own. In fairness, the study is titled “Primary payer status affects mortality for major surgical operations.” That title should not have passed peer review.
This is a valuable descriptive study which demonstrates that adjusted mortality is high among Medicaid surgical patients, even after one sensibly corrects for the kinds of potential confounders readily available in large-scale clinical databases. Indeed Medicaid patients have higher adjusted mortality than the uninsured. (The authors’ reported statistical findings compare Medicare, Medicaid, and uninsured patients to the privately-insured rather than each other. The authors don’t seem to report the statistical significance of the rather modest differences between Medicaid patients and the uninsured.)
As Austin notes, Medicare patients also fare worse than the uninsured on many outcomes. This pattern–and others reported in the paper–suggests that we are seeing straightforward selection bias. Medicaid patients face more difficult personal risk-factors and circumstances that lead to worse outcomes. The original paper even includes a postscript in which experts identify many ways in which such selection biases could arise.
In this study sample, Medicaid patients receive more costly care and had longer lengths of stay, which is consistent with some selection bias account. Some patients are enrolled into Medicaid during their hospital stay. Hospital staff and social service agencies may have greater motivation to pursue such processes when patients are especially sick or needy. Medicaid patients likely face greater economic, family, and educational obstacles. They use different hospitals and live in different communities. Some receive Medicaid because they have specific disabilities. Privately insured patients with minor issues may have better access to surgery. Some of these factors can be controlled in the statistical analysis. Others are unobserved. (See Austin for more on why this is an inherently dicey study design to test the causal impact of insurance coverage.)
Don’t get me wrong. The fact that Medicaid patients exhibit high mortality rates is an important descriptive finding. If this is more than an artifact of the data, understanding why this is happening may point the way to better policies.
I share Roy’s dismay that we don’t see better health outcomes among Medicaid patients. We don’t actually spend $500 billion on Medicaid. We do spend $373 billion, which is no small amount. We can do better.
The Affordable Care Act includes some elements to improve Medicaid–namely large subsidies to states and measures to raise primary care reimbursement rates. I’d like to see much more. Addressing the programs flaws will require more money spent on Medicaid–precisely the opposite of what is now being contemplated in Washington and across the states.
Expanding public insurance is not the only or the most powerful way to improve population health, but it’s important. Roy is correct that the evidence we have to-date regarding the health impact of Medicaid could be much better. Still, misinterpreting observational studies to disparage the program doesn’t advance this debate.