Why the Geographic Variation in Health Care Spending Can’t Tell Us Much about the Efficiency or Quality of our Health Care System
By Louise Sheiner
Brookings Institution, September 2014
Abstract
This paper examines the geographic variation in Medicare and non-Medicare health spending and finds little support for the view that most of the variation is likely attributable to differences in practice styles. Instead, I find that socioeconomic factors that affect the need for medical care, as well as interactions between the Medicare system and other parts of the health system, can account, in an econometric sense, for most of the variation in Medicare health spending.
The paper also explores the econometric differences between controlling for health attributes at the state level (the method used in this paper) and controlling for them at the individual level (the approach used by the Dartmouth group.) I show that a state-level approach can explain more of the state-level variation associated with omitted health attributes than the individual-level approach, and argue that this econometric difference likely explains much of the difference between my results and those of the Dartmouth group.
More broadly, the paper shows that the geographic variation in health spending does not provide a useful way to examine the inefficiencies of our health system. States where Medicare spending is high are very different in multiple dimensions from states where Medicare spending is low, and thus it is difficult to isolate the effects of differences in health spending intensity from the effects of the differences in the underlying state characteristics. I show, for example, that previous findings about the relationships between health spending, the share of physicians who are general practitioners, and quality, are likely the result of omitted factors rather than the result of causal relationships.
I. Introduction
It is well known that Medicare spending per beneficiary varies widely across geographic areas. The conventional wisdom from the leaders in this research area, the Dartmouth group, is that little of this variation is accounted for by variation in income, prices, demographics, and health status, and, instead, most of the variation represents differences in “practice styles.” Further, the Dartmouth research suggests that the additional health spending of the high-spending areas does not improve the quality of health care, and, indeed, might even diminish it.
One of the implications of the Dartmouth work is that health care spending can be reduced without significant effects on health outcomes. For example, Sutherland, Fisher, and Skinner (2009) argue “Evidence regarding regional variations in spending and growth points to a more hopeful alternative: we should be able to reorganize and improve care to eliminate wasteful and unnecessary service.” This view was promoted by the Obama Administration as part of the effort to reform health care. In a Wall Street Journal op-ed, then OMB-director Peter Orszag, referring to the Dartmouth work, noted “If we can move our nation toward the proven and successful practices adopted by lower-cost areas and hospitals, some economists believe health-care costs could be reduced by 30% — or about $700 billion a year — without compromising the quality of care.
The Dartmouth group has also argued that this geographic variation holds the key to reducing excess cost growth in health care. According to Fisher, Bynum, and Skinner (2009), “By learning from regions that have attained sustainable growth rates and building on successful models of delivery-system and payment system reform, we might… manage to “bend the cost curve.” ……. Reducing annual growth in per capita spending from 3.5% (the national average) to 2.4% (the rate in San Francisco) would leave Medicare with a healthy estimated balance of $758 billion, a cumulative savings of $1.42 trillion.”
In this paper, I reexamine the geographic variation in health spending at the state level and find little support for the Dartmouth views. I find that most of the geographic variation in Medicare spending is explainable, at least in an econometric sense, by differences in socioeconomic factors that affect the need for medical care and the resources available in the nonelderly population to finance it. Although it is not possible to rule out the Dartmouth view that the differences in spending reflect differences in practice styles, I show that there are other explanations for the variation in spending that seem to be better supported by the data. Furthermore, I show that the relationships between health spending (both Medicare and non-Medicare), physician composition, and quality are likely the result of omitted factors rather than the result of causal relationships.
More broadly, the paper shows that the geographic variation in health spending does not provide a useful way to examine the inefficiencies of our health system. States where Medicare spending is high are very different from states where Medicare spending is low, and thus it is difficult to isolate the effects of differences in health spending intensity from the effects of the differences in the underlying state characteristics. Insights into the relationship between health spending and outcomes are more likely to be provided by natural experiments such as that analyzed by Doyle (2007), who showed that among visitors to Florida who had heart attacks, outcomes were better at hospitals with higher spending, the true experiment run in Oregon in which a group of uninsured low-income adults was selected by lottery to be given the chance to apply for Medicaid (Finkelstein et al, 2011), or the recent paper by Finkelstein et al which focuses on Medicare beneficiaries who move (Finkelstein, 2013).
It is important to note at the outset that nothing in this paper suggests that improvements in our health system are unattainable. Rather, the paper suggests that comparisons of spending between high cost states and low costs states are unlikely to provide a measure of how much we can hope to gain by efforts to improve health system efficiency.
The paper is organized as follows. I first give a brief overview of the literature on geographic variation. Then I present the basic results from the Medicare regressions, and show that the cross-state variation in average Medicare spending is well explained by differences in population characteristics across states. I compare my results to those of the Dartmouth group and suggest a number of reasons why my results differ. I show that, econometrically, there is a difference between controlling for attributes at the individual level (the Dartmouth approach) and controlling for them at the state level (the approach used here), and that this difference is likely to be empirically important when it comes to health care. I argue that my state-level approach better controls for the variation in health and other socioeconomic variables that affect health demand. In addition, to the extent that there are area differences in practice styles, I show that these too likely reflect systemic differences across states, and thus would likely be difficult to alter.
I then explore the relationships between Medicare and non-Medicare spending across the states, and show that the two appear to be somewhat negatively correlated. This correlation is quite important in thinking about the relationship between provider workforce characteristics, quality, and health spending. In particular, I show that taking into consideration some of the demographics and health insurance variables by state changes the conclusions one gets from previous studies. Finally, I show that the growth rates of Medicare spending are negatively related to the level of health spending—that is, low spending states tend to have higher growth rates than high-spending states. The conclusion assesses the implications of this work for Medicare policy.
VIII. Conclusions
The evidence presented in this paper shows that most of the variation in Medicare spending across states is attributable to factors that affect health and health behaviors, rather than to random variation in practice styles. Isolating the exact channels through which differences in health affect Medicare spending is difficult, however, because both the need for health spending and provider practice styles will likely be affected by variations in population health and variables that are correlated with it.
But the paper has several findings that suggest that the variation in Medicare spending does not represent wasteful spending that could be easily eliminated without significant effects on the health system. First, population characteristics have more explanatory power for Medicare spending than measures of social capital, indicating that the variation in patient characteristics is more important than variation in provider characteristics. Second, health measures are significantly more correlated at the state level than at the individual level, making it likely that state level regressions do a better job of controlling for unobserved variation in population health. Third, there does not seem to be a significant relationship between the use of “preference- sensitive procedures” and the level Medicare spending. Fourth, states with high levels of Medicare spending tend to have lower levels of non-Medicare spending. Providers in these states may face greater financial difficulties, and may “volume shift” to Medicare patients in order to cover costs.
The paper also shows that conclusions about the relationships between health spending, physician composition, and quality are sensitive to the inclusion of variables like the share of the population uninsured, black, or diabetic. What this sensitivity demonstrates is the difficulty of using the geographic variation in spending for hypothesis testing. It is not surprising that states in the South spend more on Medicare and have worse outcomes. These states perform significantly worse in numerous areas, including high school graduation rates, test scores, unemployment, violent crime, and teenage pregnancy. There are many ways that such differences can affect health utilization and outcomes, including differences in underlying health, social supports and social stressors, patient self-care and advocacy, ease of access to services, capabilities of hospital and physician nurses and technicians, and cultural differences in attitudes toward care. A comparison of health spending in Mississippi with health spending in Minnesota is not likely to provide a useful metric of the “inefficiencies” of the health system in isolation; rather, the difference in spending likely mirrors broader societal problems unrelated to the health system per se.
Finally, the evidence also suggests that low-cost states are not low-growth states. Thus, the geographic variation in Medicare spending is probably not the key to finding ways to slow spending growth while continuing to improve quality over time.
http://www.brookings.edu/~/media/projects/bpea/fall%202014/fall2014bpea_sheiner.pdf
****
Comment:
By Don McCanne, MD
Did this paper really say what it seems like it said? Wow! It is important because it seems to be a highly credible challenge to the principle that much of the waste in health care spending is due to variation in practice styles, as allegedly demonstrated by the Dartmouth group.
According to this Brookings paper by Louise Sheiner, “The evidence presented in this paper shows that most of the variation in Medicare spending across states is attributable to factors that affect health and health behaviors, rather than to random variation in practice styles.” Further, “But the paper has several findings that suggest that the variation in Medicare spending does not represent wasteful spending that could be easily eliminated without significant effects on the health system.”
Double wow! This suggests that the efforts of reducing waste through accountable care organizations (ACO) that are designed based on the Dartmouth studies of waste are mostly for naught. It also explains why to date the experiments with ACO models have had very little impact on either efficiency or quality.
Of particular concern is the finding that areas in the South with high Medicare spending have worse outcomes, and they also “perform significantly worse in numerous areas, including high school graduation rates, test scores, unemployment, violent crime, and teenage pregnancy.” You cannot help but think that more resources need to be directed toward these societal ills. It is not just health care that needs our attention.
For us, the important take-home point of this paper is that we should turn our attention away from puff programs such as ACOs that hold little promise of delivering on higher quality and lower costs, and turn instead to policies that have been proven in other nations to be effective. Of course we’re referring to a single payer national health program. We already know that it works.