Adrianna McIntyre is a PhD candidate in health policy at Harvard University. Her research focuses on barriers to health insurance coverage for low-income populations and the politics of health reform. 

For almost half a century, the conventional wisdom in health care has been that charging patients small copays will lead them to make better decisions about what care to get (or not get), and their health won’t be harmed. But a new study suggests giving patients too much “skin in the game” can actually lead to the greatest health harm of all: death.

Because death is such a rare outcome, it’s easy for mortality effects — even large ones — to hide in plain sight. Despite decades of related research, scholars have struggled to make any strong connections one way or the other between increased cost-sharing and mortality. But in a recent NBER working paper, Amitabh Chandra, Evan Flack and Ziad Obermeyer overcome conventional barriers to answering this question by exploiting a quirk of how cost-sharing for drugs is structured in Medicare: Drug benefits aren’t pro-rated, so people with birthdays later in the year are less likely to see a big jump in cost-sharing in their first year on Medicare. The authors combined this with machine learning to estimate how much people would spend on prescriptions without any cost-sharing. This allows them to compare how beneficiaries with similar prescription needs respond when their cost-sharing goes up, including whether those differences influence mortality.

Their findings are stark: Increasing cost-sharing by $10.40 per prescription drug fill causes a 23% drop in spending on prescription drugs and a 33% increase in mortality. Even more concerning is that the highest risk patients were most likely to cut back sharply on their medications, and a large portion of the unfilled prescriptions were high-value, life-saving drugs like medications for asthma, cholesterol and blood pressure.

This study stands out to me for two big reasons. One, it is really hard to get this kind of data connecting cost-sharing to mortality. Because death is such a rare outcome, you would have to have a huge study population to get these kinds of results, several times more than the 5,000-10,000 people included in landmark randomized health policy trials on health coverage and cost-sharing like the RAND health insurance experiment and Oregon Medicaid experiment.

Second, the findings accord with a small but growing literature documenting mortality improvements from insurance expansions. But this study goes further because all of the subjects have insurance; it’s the generosity of drug coverage that’s at issue, a nuance that is critical to understand for making effective policy. 

The authors are careful to note that their study looks at a very narrow population: a subset of mostly healthy 65 year olds newly enrolled in Medicare prescription drug plans. But cost-sharing is ubiquitous throughout the U.S. health care system, and there’s no reason to believe the response in this group is some kind of aberration. The authors conclude by saying that further rigorous study of the health consequences of cost-sharing — and policy solutions that address this evidence — are urgently needed.