Patient Health Doesn't Explain Cost Differences
The Washington Post Style section recently declared that a new Brookings Institution report has "upended health-care research." The reality is more complex, and the new paper has not fundamentally changed anything.
The report, by Louise Sheiner, a senior fellow at Brookings, has drawn lots of attention because it brings into question decades of research by the Dartmouth Atlas of Health Care. The Dartmouth team's core findings have been that substantial variation exists in how health care is practiced in the U.S., and that, on average, higher cost does not correlate with higher quality. Together, these suggest it should be possible to reduce health-care costs without harming people's medical outcomes. (I have publicly embraced this perspective, and I continue to hold it.)
Enter Ms. Sheiner. She accepts that costs vary from state to state in the U.S., but finds "little support" for the notion that differences in medical practice are to blame. What does explain the variation, she claims, is the underlying health of patients: The states with higher costs simply have sicker people.
This claim is something the Dartmouth researchers themselves have tried to incorporate into their analyses. Sheiner, however, says they have done it wrong. Even after adjusting for a beneficiary's health in the manner embraced by the Dartmouth team, she shows, there is a positive correlation between a state's average cost for health care and its share of diabetics.
Now, in theory, the Dartmouth numbers should already account for the impact of diabetes on spending, so this finding is a puzzle. And kudos to Sheiner for highlighting it.
The question is, is the correlation a statistical artifact or an indication that the people in a state with higher average diabetes rates are actually sicker than they might initially appear? The presumption in Sheiner's argument -- hotly contested by the Dartmouth researchers -- is that a diabetic patient in a state with a relatively high average incidence of diabetes is less healthy than an ostensibly similar diabetic patient in a state with a lower average incidence of diabetes. (She actually claims it isn't even helpful to know whether an individual patient is diabetic, only whether the state the person lives in tends to have more diabetics.)
If you buy the Sheiner argument, then the variation in Medicare costs across states is explained mostly by the underlying health of the population, and not how differently doctors practice.
One promising way to evaluate the competing views is to examine what happens when people move from one state to another. (Ideally, we would randomly assign people to move to one location or another, but that's not feasible.)
Under the Dartmouth theory, because medical practices and costs are linked, when people move we should see a change in how much is spent on health care. Under the Sheiner theory, because the patient's health is the key determinant of spending, people's moving shouldn't affect it.
A recent study examined this question and concluded that spending changes noticeably when Medicare beneficiaries move from one place to another. The paper, by Amy Finkelstein and Heidi Williams of the Massachusetts Institute of Technology and Matthew Gentzkow of the University of Chicago, shows that spending changes almost immediately after a beneficiary moves, but not before. The change occurs, furthermore, regardless of whether someone moves from a low-spending area to a high-spending one (in which case the spending on the mover increases) or from a high-spending area to a low-spending one (the spending declines).
The conclusion? How ill people are explains, at most, half of the variation in Medicare costs across the U.S. That's only slightly more than the share the Dartmouth team itself attributes to health differences. To my reading, score one for the Dartmouth team.
Then, there's the fact that if you happen to be taken in an ambulance to one hospital rather than another, your total medical costs in the episode are substantially different, even though the overall quality of care isn't. There are the differences in the way doctors describe how they would treat hypothetical patients. And there are the noticeable improvements in value of care that places like Crystal Run Healthcare in New York have been able to achieve by reducing clinical variation within their practices.
All this evidence suggests that health-care spending varies significantly, that patients' health can't explain the majority of it, and that the potential therefore exists to restrain costs without harming outcomes.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
To contact the author on this story:
Peter R Orszag at email@example.com
To contact the editor on this story:
Mary Duenwald at firstname.lastname@example.org