Jonathan Skinner

Department of Economics, Dartmouth College, Hanover, NH, USA

Contents

1 Introduction

2 An Economic Model of Regional Variations in Health Care

2.1 The Demand Side

2.2 The Supply Side

2.3 A Typology of Health Care Services

3 Empirical Evidence on Geographic Variations in Expenditures and Utilization

3.1 Units of Measurement and Spatial Correlations

3.2 Health Care Expenditures

3.2.1 Adjusting for Prices

3.2.2 Adjusting for Differences in Health Status

3.2.3 Adjusting for Income

3.2.4 Regional Variation in Non-Medicare Expenditures

3.3 Effective Care (Category I)

3.4 Preference-sensitive Treatments with Heterogeneous Benefits (Category II)

3.5 Supply-sensitive (Category III) Treatments with Unknown or Marginal Benefits

4 Estimating the Consequences of Regional Variation: Geography As An Instrument

5 Inefficiency and the Policy Implications of Regional Variations

6 Regional Variations in Health Outcomes

7 Discussion and Conclusion

References

Abstract

There are widespread differences in health care spending and utilization across regions of the US as well as in other countries. Are these variations caused by demand-side factors such as patient preferences, health status, income, or access? Or are they caused by supply-side factors such as provider financial incentives, beliefs, ability, or practice norms? In this chapter, I first consider regional health care differences in the context of a simple demand and supply model, and then focus on the empirical evidence documenting causes of variations. While demand factors are important—health in particular—there remains strong evidence for supply-driven differences in utilization. I then consider evidence on the causal impact of spending on outcomes, and conclude that it is less important how much money is spent, and far more important how the money is spent—whether for highly effective treatments such as beta blockers or anti-retroviral treatments for AIDS patients, or ineffective treatments such as feeding tubes for advanced dementia patients.

Keywords

• health economics • health care productivity • spatial models • regional variations • small-area analysis

JEL Codes

• I100 • I110 • I120 • I180 • R120

1 Introduction

A recently published Atlas documented dramatic differences in the utilization of an important health input. Relative to the rates observed in the Boston area, utilization was 74 percent higher in New Haven and more than 200 percent higher in San Francisco.2 One might explain differences in utilization by variations in income, health status, or prices, yet these factors do not appear to explain away the wide variations we observe. So why have these patterns—in per capita consumption of meat and poultry, ranging from 31 pounds in the Boston region to 113 pounds per capita in the San Francisco region—not received more attention from health experts or health economists?

In this chapter, I attempt to distinguish between the admittedly puzzling geographic variation in meat and poultry consumption and geographic variation in health care utilization. There are certainly many reasons why regional variations in utilization can be justified by underlying health status, preferences and income, or productivity differences among providers—surgical rates should be higher in regions where surgeons get better results (Chandra and Staiger, 2007). But there are also reasons why such variations might not be efficient. We know that specific components of the health care industry, such as the widespread use of insurance with modest co-payments, leads to the twin problems of moral hazard and adverse selection. And beginning with Arrow (1963), economists have argued that many of the unique features and institutions associated with medical care—ranging from not-for-profit ownership to licensure of providers to the structure of insurance—are best understood as the result of “uncertainty in the incidence of disease and in the efficacy of treatment” (Arrow, 1963, p. 941).

Health care is not the only sector of the economy with these characteristics—car repair, home construction, and management consultancies exhibit uncertainty about the efficacy of the fix, even if they are not typically insured to the same degree. While regional variations surely exist in the quality and cost of automotive repairs, it is less likely that the government would propose accountable car repair organizations, for example.

Regional variations in health care are different, for at least two reasons. First, assuming that the variation we observe is “unwarranted” in the sense that it cannot be explained by legitimate causes such as health status, the magnitude of variation is so large that the potential gain from erasing such inefficiencies—3 percent of GDP or more in the United States—is worth pursuing. And second, these inefficiencies are unlikely to be shaken out by normal competitive forces, given the patchwork of providers, consumers, and third-party payers each of which faces inadequate incentives to improve quality or lower costs (Fuchs and Milstein, 2011).

Following on the earlier survey by Charles Phelps in the Handbook (Phelps, 2000), this review considers five general questions in reflecting on the economics of regional variations in health care. First, what are the theoretical causes of such differences? Traditionally, the presence of regional variation in medical care utilization has been viewed through the lens of “supplier-induced demand,” the idea that regional variations can be explained by the utility-maximizing behavior of health care providers responding to a fee-for-service environment and relative scarcity (or abundance) of providers (McGuire, 2011). The problem arises when individual physicians in two seemingly similar regions—with identical insurance mechanisms and similar patients—end up providing much different quantities of health care. That is, standard supplier-induced demand models may argue that physicians do more for their patients than is optimal, but does not typically explain why physicians in McAllen, TX, do so much more for their Medicare patients than those in El Paso (Gawande, 2009). To address these issues, I adopt a model based on Chandra and Skinner (2011) and Wennberg et al. (2002) to parse out both supply and demand factors that might be expected to explain regional variation in specific types of treatment, ranging from highly effective care that clearly saves lives at minimal cost (such as beta blockers for heart attack patients) to very expensive treatment without known benefits for patients (like proton beam therapy for prostate cancer).

Second, I use this basic framework to consider the empirical evidence on causes of geographic variation in health care utilization and expenditures. This is the key section to assess the evidence on whether supply-side variations really do exist. If all of the regional variation in observed utilization rates can be explained by other factors such as patient preferences, relative prices, income, and health, then the puzzle of regional variations is not even a puzzle any more. I focus in this chapter largely on US regional variations, but document also a growing literature reflecting international variations both within and across countries.3

Third, what are the consequences of higher health care spending? Does more spending yield better outcomes—or is how the money spent more important for health? A key focus of this section is to understand how geography might be used as a statistical instrument in health economics to help estimate whether greater intensity of care is associated with better health outcomes. Starting with the early work by Glover (1938) and Wennberg and Gittelsohn (1973), and continuing through the more formal analysis using instrumental variables, there has been a long-standing tradition of using geography as an instrument to make inferences about the health care “production function”(Fisher et al., 2003b; McClellan et al., 1994). Some studies have suggested a negative association between spending and outcomes, while others have found a positive association, but what is most striking is how much variability there is in outcomes across providers or regions, and how poorly such variability is associated with factor inputs.

Fourth, what are the policy implications of observing variations in health care utilization? If there are enormous variations in the productivity of concrete, a seven-digit SIC code output with readily apparent quality measures (Syverson, 2004), is it any surprise that there are even greater disparities in the productivity of health care across regions or hospitals where outputs are difficult to measure and rarely made public? The...