• Understanding geographic variation in health care

    By email Brad Flansbaum drew my attention to Geographic Variation in Health Care: Changing Policy Directions, by Jill Bernstein, James Reschovsky, and Chapin White from the Center for Studying Health System Change (ungated). It’s a very good summary of what is known and not known on this topic. It explains how and why some have misinterpreted the evidence on geographic variation in health spending. If you’ve been following work in this area, you’ll know that the story is more complex than most think. If you haven’t been, this paper is a great place to start.

    The paper includes the following conceptual model of the determinants of health care use. Because the diagram is circular (arrows flowing back from “practice patterns” to “health needs and references for health care” and “price of various health care services and supplies”) estimating causal effects is difficult. For this and other reasons, we do not yet have a good handle on how geographic variations in health care arise and how much of it is associates with wasteful care.

    However, we do know some things. As the authors write,

    The best available research does not provide a solid basis for drawing conclusions about how much of the variation in Medicare spending across localities reflects inappropriate or inefficient spending. […]

    Fee-for-service payment poses an obvious obstacle to reducing inappropriate or marginally effective care. […]

    Cutting reimbursement alone will not automatically make high-spending areas adopt the systems, culture, and experience of low-spending areas.

    In other words, not only are current FFS-based payments not helpful in eliminating wasteful health spending, the standard approaches to cutting spending — by cutting reimbursements — is not likely to work. The problem isn’t reimbursement levels but something about the (local) system they finance.

    The paper explains this with a model proposed by the Congressional Budget Office that considers two regions (see the figure below). Region 1 has low utilization (low spending) and high efficiency (i.e., very little low effectiveness care is provided). Region 2 has high utilization (high spending) and low efficiency (i.e., a lot of care that is not very effective is provided). The two regions have different health outcomes production functions. That is, the relationship between health care utilization (spending) and outcomes is different in the two regions.

    Suppose we observe health spending and outcomes for both regions at a point in time and find that spending and outcomes are represented by the black dots. Since spending is higher in region 2 than 1 but outcomes are worse, we might conclude that higher spending produces worse health outcomes. We would conclude that if we were not aware that the two regions had different production functions. However, if we believed that higher spending produced worse outcomes, then the policy solution is to reduce reimbursements in region 2.

    Suppose we reduced reimbursements so that region 2 had the same utilization level as region 1. That would shift region 2 to the empty circle in the figure. However, at that point, region 2’s outcomes are even worse than they were to begin with. In other words, cutting reimbursements does reduce utilization, but it does so in a way that is harmful to outcomes.

    The challenge is to reform policy to reduce spending while not harming outcomes (and, hopefully, improving them). A naive interpretation of geographic variations does not suggest the right reform. Simply spending less does is not the right approach.

    • I thought this was a great paper with lots of good points. As a practicing physician, I would concentrate on two of their points.

      First, if a physician (or hospital) converts to a high efficiency provider they suffer economically in a FFS system. That makes bottom up reform for these places very difficult.

      Secondly, it is still not clear what all of the factors are in determining utilization in different areas. My wife trained in one area and practiced, briefly, in two others. She was surprised at the frequency of tests in one area vs another. Some of this seems to be a kind of cultural norm that may or may not be directly related to the economics of care. This was most noticeable at the two academic centers where she worked. One place had a culture that demanded that residents order every conceivable test. In the other, testing was more focused.


    • Very good study.

      One of the most interesting things ever discovered in regional variation data was the Dartmouth Atlas’ look at the Mayo Clinic in Scottsdale, AZ.

      The Phoenix area is one of the areas of the country where high costs and relatively low outcome efficiency are a major issue. Mayo in its main campus in Rochester, MN, is a poster child for efficiency with low costs and excellent outcomes in the Dartmouth data — despite having relatively high actual charges on a per activity basis.

      Mayo in Scottsdale mimics the results — actually slightly improves on them — of the Mayo main center.

      This is by no means definitive, but strongly suggests that cost and outcome efficiency are almost purely related to institutional culture, which is tranplantable from region to region, and is not as strongly related to regional variation in patient characteristics as you might think.

      Mayo tends to be more efficient than regional norms in its other centers, but not as efficient as at Rochester and Scottsdale. Personally, I think that has to do with the fact that Scottsdale is an “elephant graveyard” for Mayo, with large numbers of staff and leadership there being experienced Rochester hands who have decided to move to the Sunbelt after many years in Rochester.

      Some other systems with centers in multiple regions display similar results, but the Mayo result is the most striking.

      The takeaway question is then how to move this type of cultural difference to other, less efficient cultures. I tend to think that is not likely to happen without a significant push from the government. However, as the Mayo model shows, the critical factor is not unit charges (again, Mayo is fairly high) but uitlization, so dropping unit charges is likely, as this study suggests, to have a negative effect on outcomes. I can see no way to effect that without specific management norms backed by refusal to pay for care that ignores the norms. That is what they do in other countries.

      The politics, however, is very very hard, with aggressive pushback by major stakeholders and by politicians who subscribe to the idea that price based rationing of access to care is superior to government involvement in management norms.

    • Good summary. As far as multiple production functions are concerned, this is pretty much what the Dartmouth people and others have been saying for years (before the CBO report). As far as how much of the variation reflects “inefficient” (or wasteful on the margin) spending, I’d say the case is pretty solid that it’s 20% or more. It’s difficult to pin this type of thing down, but the evidence that there’s a lot of wasteful spending is very strong by social science standards.

      As many have suggested, it might be best to frame geographic variation as an issue of misuse of care, both underuse and overuse. The result is not only inefficient spending, but also poorer quality.

    • Following up on Luke’s comments, interested readers may wish to consult the 2006 paper from which the CBO graph was adopted:

      Skinner, J.S., D. Staiger, E.S. Fisher, “Is Technological Change in Medicine Always Worth It? The Case of Acute Myocardial Infarction”, Health Affairs web exclusive (February 2006).

      For two more recent papers on the same topic:



    • Glad you found our paper useful. FYI, the link to the Bernstein, Reschovsky, and White paper is incorrect. Should be