Thread by @Jabaluck

Does health insurance save lives? While there are dozens of studies with a wide range of estimates, the best studies consistently suggest that the answer is yes and the magnitude is considerable.

I recently received a referee report on a grant application suggesting that the RAND and Oregon experiments demonstrated that insurance doesn't matter for physical health, so it seems like a thread is needed.

In 1993, a JAMA study showed that conditional on gender, race, and baseline age, education, income, employment status, uninsured people had 25% higher mortality.
jamanetwork.com/journals/jama/…

But maybe some people are more conscientious. Conscientious people are healthier and more likely to be insured. Maybe controls for "Are you employed?" don't capture all the different complex ways in which employment status might impact both insurance and health.

The bottom-line, as usual, is that it's extremely hard to figure out the impact of insurance through "back door" methods of trying to control for confounds. Other studies using this approach find no effect of insurance: onlinelibrary.wiley.com/doi/full/10.11…

The fundamental problem is that after you control for stuff, you still need a story about why some people get insurance and others don't, and why that mechanism is otherwise unrelated to outcomes

https://twitter.com/Jabaluck/status/1209949896297635840

One of the first studies to do this was Card, Dobkin and Maestas in the 2009 QJE, showing that inpatient mortality for non-deferrable conditions drops at age 65, when people qualify for Medicare.
academic.oup.com/qje/article/12…

Chay, Kim and Swaminathan showed that these discontinuities in mortality at age 65 appear only after Medicare's introduction: hec.ca/iea/seminaires…

But even with this strategy, we don't *always* find effects at age 65. There are, for example, no differences in COVID mortality. Insurance doesn't fix *everything*, but it seems to help with some things. jamanetwork.com/journals/jama-…

Now to the RAND and Oregon experiments -- the claim that they showed that getting insurance has no effect on physical health is mistaken in several respects.

First, the RAND experiment does not compare the effect of being insured to being uninsured. It compares the effect of being in insurance plans with different levels of coverage, ranging from zero cost-sharing to catastrophic coverage.

The RAND experiment found effects on physical health for the poorest and sickest patients, including better control of hypertension, improved version, and a reduction in "serious symptoms"
rand.org/pubs/research_…

In the RAND experiment, 11 out of 1,294 people died in the plan that offered full coverage. 29 of 2,664 people died in plans offering more coverage. I get a 95% CI ranging from a 90% reduction to a 43% increase. Hardly definitive evidence that insurance doesn't impact mortality!

What about the Oregon Health Insurance experiment? They found a somewhat surprising mix of results. On the one hand, participants randomized to insurance were 25% more likely to say they were in good or excellent rather than fair or poor health.

On the other hand, they found no impact on blood pressure, cholesterol, hemoglobin, or a 10-year cardiovascular risk index combining these factors. Their effects were reasonably precise.
nber.org/programs-proje…

They found no statistically significant impact on mortality. But the point estimate was a 16% reduction with a CI from [-82%,+50%] (as Kate Baicker, a principal architect of the Oregon experiment with Amy Finkelstein, frequently highlights)
nejm.org/doi/full/10.10…

Ben Sommers, Kate Baicker and co-authors have conducted several diff-in-diff studies of the MA health reform and Medicaid expansions, consistently finding mortality impacts (well within the CI of RAND and Oregon)
pubmed.ncbi.nlm.nih.gov/24798521/
nejm.org/doi/full/10.10…

There remained some ambiguity as other diff-in-diff studies found no impact on health, although the outcome was typically self-reported health rather than administratively measured mortality.
nejm.org/doi/full/10.10…
ncbi.nlm.nih.gov/pmc/articles/P…

The two highest-quality studies to date have been published in the last two years. First, @smilleralert, Norman Johnson and @LaurawherryR looked at the impact of the Medicaid expansions using detailed administrative data on who qualifies for Medicaid.
academic.oup.com/qje/article-ab…

Using this vastly larger sample than previous studies as well as far more granular data on who is likely to be impacted, they find clear evidence of declines in mortality following Medicaid expansions, specifically among those who gained Medicaid eligibility.

The study conducts a number of compelling placebo checks showing that we see no effect in pre-ACA years or age > 65, as expected if they've isolated the effect of the Medicaid expansions. This evidence is extremely persuasive in my view that Medicaid reduces mortality.

Finally, @jacobsgoldin, Ithai Lurie and Janet McCubbin report results from an RCT actually powered to impact mortality -- the IRS sent letters to 3.9 million households informing them about the tax penalty from the ACA.
academic.oup.com/qje/article/13…

The bottom line is that 42,900 people gained insurance as a result, and mortality was lower in the treatment group, with the effect concentrate among those 45 years or older:

There are many subtleties in interpreting the magnitudes in both studies. But even the lower bounds are large -- ACA coverage for a year appears to reduce mortality by at least 22% while a year of Medicaid reduces mortality by at least 11.9%, possibly several times more.

Many questions remain about who benefits and why. I have truly marvelous comments on the mechanism by which insurance impacts mortality which the 25-tweet per thread limit is too narrow to contain.