In the 1970s, John Wennberg and Alan Gittelsohn noted something very peculiar about health care delivery in the United States: they found a great deal of variation within a sample New England towns in the rates of various surgical procedures. For example, the rates of hysterectomies and tonsillectomies in Vermont varied wildly across hospital service areas, with most intense areas sporting 3 and 10 times the number of procedures per capita as the least intense areas, respectively!
Since Wennberg and Gittelsohn's original paper, a great deal of work has been done on this phenomenon, coined "small area variation." This research has shown that (a) county/area level demographics, per capita income and other socioeconomic variables do not always explain away the cross-area variations and (b) variation in service delivery persists even today. (For a sense of (b), check out the Dartmouth Atlas of Health Care, a great resource and a fun way to spend a half hour or so.)
As you can probably imagine, the small-area variation literature created a huge health policy firestorm, whose echoes continue to reverberate in the present day. Basically, policymakers concluded the following things from this literature:
1) Wide variation in service delivery reflects uncertainty among physicians about what best practices are. This suggests that either physicians don't know as much as originally assumed, that information about best practices does not diffuse quickly among doctors or both.
2) In addition to (1), it may be that physicians in certain areas are more likely to get their patients to agree to certain procedures. The technical term for this is supplied-induced demand.
3) Areas that have higher rates of various procedures and, therefore, ostensibly spend more do not have better health outcomes. This suggests that the high intensity areas are wasteful. The empirical finding of no association between treatment rates/spending and outcomes lends even more support to (1) and (2).
Not surprisingly, small-area variation is widely seen as both a cause and symptom of very messy U.S. health care system, its existence viewed as indicative of pernicious processes that need to be curbed.
However, are there other explanations behind the empirical findings in the small area variations literature? Let's start with the question of why small area variation exists in the first place. Are there other theories besides supplied-induced demand and poor diffusion of information in the context of area-specific norms that can explain these empirical observations?
In a recent paper in the Journal of Political Economy, economists Amitabh Chandra and Douglas Staiger suggest the answer is "yes." In particular, the propose a theory where the key feature is that the benefit of a single procedure over another is contingent on the number of other patients receiving that procedure in the same area. How could this happen? First, since doctors learn from each other, a physician becomes more productive with a given procedure if he/she has more peers in the area doing the same thing. Second, support services in area hospitals that make a given procedure more productive might arise due to various historic factors. Third, doctors may self-select and practice in areas with different treatment intensities. Regardless of the relative importance of these factors, the key point is that these "productivity spillovers" are plausibly less icky than more "evil" stories like uninformed and tricky docs.
Chandra and Staiger test their theory in the data, looking at treatment patterns for heart attacks (intensive-surgical versus non-intensive-pharmacological) among a sample of Medicare patients, and find a very good fit. From a grad student perspective, the beauty of this paper is how the authors use their theory not only to explain the existing facts in the small-area variation literature, but also to generate further predictions that can be tested in the data. This is the key: what differentiates one theory from the other are the predictions it makes and, as a result, the only way to tease apart the underlying driving forces are to focus on predictions unique to each theory.
What about the consequences of small-area variation? Past research exploited this variation to conclude that increased treatment intensity and increased expenditure have no effect on health outcomes. Some recent research, however, suggests that this may be wrong. In a recent working paper (thanks to Brian Elbel for the reference), Joseph Doyle notes that past estimates of expenditure effects might be plagued by omitted variable bias: more money might be spent in areas where health outcomes are worse, thus biasing the relationship between the two towards zero. Here is how he gets around this:
The main innovation in this paper compares outcomes of patients who are exposed to different health care systems that were not designed for them: patients who are far from home when a health emergency strikes. The universe of emergencies in Florida from 1996-2003 is considered, and visitors who become ill in high-spending areas have significantly lower mortality rates compared to similar visitors in lower-spending areas. The results are robust across different types of patients and within groups of destinations that appear to be close demand substitutes.
Of course, the main threat to inference in this strategy is that richer tourists or visitors might self select to go into richer counties. Doyle has a variety of robustness checks to address this hypothesis. Check out the paper and see if you buy it.
I think this finding is interesting. Many people think that the U.S. is wasteful with its health care dollars and this might be true in some contexts. However, as the trajectory of the small-area variations literature illustrates, it is worth having a fresh look at our priors before making sweeping policy statements. Things may not be as bad as they seem.
No comments:
Post a Comment