(Caution – this piece is wonkish.)

An interesting aspect of the Oregon Health Study, which I wrote about a little while ago, is that it uses the “intent to treat” format of a randomized controlled trial. In this kind of study, the control group and treatment group are still randomized. The people in the treatment group, however, don’t necessarily receive the treatment; they just get the opportunity (or have the “intent”) to receive it. This has deep implications for the study’s results.

People who were randomized into the “intent to treat” group got the chance to enroll in Oregon’s Medicaid program. Not all of them signed up. If there were any uniform differences between the people who signed up and those who didn’t, it would have affected the study’s results. I think there probably were, and they did.

Who were the people most likely to sign up for Medicaid, given the chance? I’d guess they fell into two groups: 1) people who cared a lot about their health and seized the opportunity to get insurance, and 2) people who were really sick and knew they needed health care. In other words, the healthiest and sickest people may have been the ones who signed up.

A few months after the lottery, the researchers checked to see whether people with Medicaid coverage had better scores in cholesterol, blood pressure, and other health tests. They didn’t find much of a difference in several areas.

I don’t think this is surprising, if you believe my story about self-selection and adverse selection in the previous paragraph. The healthiest people were already doing everything to be healthy, and the sickest may have been too sick to improve in such a short period of time. The people who might have shown a change even over a few months may have been the least likely to sign up for Medicaid, even when they had the chance.

The “intent to treat” format does not alleviate selection problems within the “intent to treat” group. As the researchers’ appendix explains, it simply magnifies the effect measured for the selected subset of the group. Until they reveal more about the initial health differences between the two subsets of the “intent to treat” group, I’ll have a hard time believing their results.

Image courtesy of Shutterstock