A question for the health-economics statisticians, also an aside about the health care costs of smoking

Atul Gawande wrote an interesting article about health-care costs, focusing on McAllen, Texas, which he describes as “one of the most expensive health-care markets in the country. . . . In 2006, Medicare spent fifteen thousand dollars per enrollee here, almost twice the national average.” In some ways, Gawande’s article is like a case-control regression analysis without the numbers: he compares McAllen to the national average and to various other places in the United States, and looks the similarities and differences to find systematic patterns. He concludes that the key problem is “untenably fragmented, quantity-driven systems of health care,” in which doctors are motivated to do more and more, with no apparent beneficial effects on the patients.

What do the experts say?

I imagine this is an area where health-economics statisticians have done some research. I’d be interested to hear the comments of Sharon-Lise Normand, Alan Zaslavsky, or some other expert in this field. They very well may have run some regression analyses to try to understand the factors that explain variation in health care costs at the regional, state, and local levels.


As a minor point, I was puzzled by an offhand comment that Gawande made:

An unhealthy population couldn’t possibly be the reason that McAllen’s health-care costs are so high. (Or the reason that America’s are. We may be more obese than any other industrialized nation, but we have among the lowest rates of smoking and alcoholism, and we are in the middle of the range for cardiovascular disease and diabetes.)

I don’t know how things go with alcoholism, but my impression of smoking was that it caused a net decrease in health care costs: smokers tend to die younger, and to die quickly once they get seriously ill, thus sparing the health care system some of the big-ticket end-of-life costs. For example, from this 1997 article by Barendregt et al. on the health care costs of smoking:

Health care costs for smokers at a given age are as much as 40 percent higher than those for nonsmokers, but in a population in which no one smoked the costs would be 7 percent higher among men and 4 percent higher among women than the costs in the current mixed population of smokers and nonsmokers. . . . If people stopped smoking, there would be a savings in health care costs, but only in the short term. Eventually, smoking cessation would lead to increased health care costs.

7 thoughts on “A question for the health-economics statisticians, also an aside about the health care costs of smoking

  1. Just reading the article now. Regarding the second point, on smoking: the health care costs described in the article are specifically per person (Medicare spending per recipient). So the first case in the last paragraph (of higher costs during individuals' lifetimes) is what's relevant to Gawande's point. It's not inconsistent with a "net decrease" in costs at all. Were smoking rates higher, we'd probably see an overall decrease in health care costs, but we'd also see a decrease in survival rates, so the per person costs would go up.

  2. In a similar vein, K. Ullrich of Wharton has a working paper speculating that doing more bicycling than car driving actually INCREASES lifetime fossil fuel use — all that exercise makes you live longer, giving you a longer time to use fossil fuels.

  3. Andrew,

    You can find an excellent semi-recent overview of the literature from the CBO here:


    Their finding is that much of the regional variation in healtcare spending remains unexplained. This may be why Gawande struggled to find a cause for the El Paso / Mcallen disparity. It also suggests the true cause is probably more complicated that Gawande's hypothesis.

  4. Atul Gawande's article in the New Yorker is an excellent review of some of the issues we have been struggling with in health policy research. While a lot of energy has gone into looking at the impact of various incentive schemes (public reporting of quality measures, "pay for performance") on quality of healthcare, it has been difficult to address the kinds of issues of organizational culture described in the article. The differences in provider structure and culture that are key to success or failure in providing high-quality, efficient care are not that readily brought into analysis — the variables are just not measured and available. So instead we have very convincing case studies — the McAllen market at one extreme and the integrated Kaiser, Geisinger etc systems at the other.

    One problem is that many of the analyses take place at the level of the health (insurance) plan, but health plans in most cases are not in the business of providing care, they are in the business of buying it. (Even some of the original staff-model HMOs went through the transition to being insurance companies, like HIP in New York.) The variables that are routinely available for analysis at the health plan level are very crude proxies for the underlying organizational structures and cultures. For example, some economists have told me that they are perplexed by findings that not-for-profit plans provide better-quality care than for-profits, since both types of firms should be subject to similar incentives. What this leaves out is the distinct histories of some of the leading not-for-profits, and consequent differences in organizational cultures.

    I do highly recommend the work of the Dartmouth group (including Jon Skinner and Elliott Fisher, mentioned in the article) on area variations to interested readers. However, it does suffer from some of the same limitations — the variations can be found and clearly show that more is not always better, but it is hard to say what actually drives the differences or what can be done to implant the cultural and organizational features that would make more areas look like the best areas.

    While I am a proponent of a universal system of health care, I don't think that will be enough to solve our problems without some fundamental changes in the incentives and structures under which care providers operate. Paradoxically, "rationalization" has meant squeezing out not the most wasteful aspects of care but some of the unprofitable but essential services that could make care more effective.

  5. The real question is whether Medicare is a good proxy for overall spending.

    At present, it appears that it is not. If so, the Dartmouth Atlas work isn't particularly useful. Cooper's 4 December 2008 paper in Health Affairs, web version, makes the interesting point that unlike Medicare spending, total spending is correlated with quality. This makes intuitive sense. It is also supported by Doyles work on mortality for out-of-state visitors to Florida ERs in high and low spending areas. In high spending areas, mortality is lower which is what one would ordinarly expect.

    The "high spending is mostly waste" argument fails to make intuitive sense and does not appear to be well supported when all costs are taken into account.

Comments are closed.