Regional Variation in Health Care: Physician Beliefs or Patient Preferences?

Health care analysts have long been puzzled by the existence of substantial regional variation in health care expenditures. In the Medicare population, for example, regional averages of price-adjusted per-patient expenditures range from under $7,000 to nearly $14,000, differences that cannot be explained by regional variation in patient illness or income.

In Physician Beliefs and Patient Preferences: A New Look at Regional Variation in Health Care Spending (NBER Working Paper 19320), researchers David Cutler, Jonathan Skinner, Ariel Dora Stern, and David Wennberg explore the causes of this phenomenon.

There are a number of possible explanations. On the demand side, patient preferences may play a role. Some patients facing a serious illness may prefer to try every possible treatment while others would prefer palliative care and comfort. If patients with similar preferences are grouped in the same geographic area, this could generate regional variation in spending. Differences in price and income, by contrast, are unlikely to be important in the Medicare context, where all patients essentially have access to the same, fairly generous insurance.

There are also possible supply-side explanations. Monetary incentives are one, as physicians may encourage patients to consume more health care if doing so raises physician income. But this alone cannot explain the observed regional variation in spending, since reimbursement rates do not vary much across areas. Alternatively, physicians may have differing beliefs about the efficacy of certain treatments. If physician beliefs are geographically correlated - for example, because many physicians in a given area received their training at the same medical school - this could create regional variation in spending.

The approach employed by the authors to distinguish between these competing hypotheses involves using "strategic surveys" that employ clinical vignettes to elicit information on physician beliefs and patient preferences. The authors analyze survey responses of over 1300 primary care physicians and cardiologists in 64 Hospital Referral Regions (HRRs). The vignettes present the physicians with hypothetical elderly patients suffering from conditions such as heart failure and ask the physicians how they would treat them. The authors characterize physicians as "cowboys" if they consistently and unambiguously recommend intensive care beyond clinical guidelines and as "comforters" if they consistently recommend palliative care for the severely ill.

On the patient side, the authors consider the preferences of over 1400 Medicare beneficiaries surveyed in the same set of HRRs. Respondents were asked questions relating to their preferences for unnecessary care and end-of-life care, such as whether they would like a test or cardiac referral even if their primary care physician did not think they needed one and whether they would want to be put on a respirator if it would extend their life for a short time.

There are a several key findings from these surveys. On the patient side, nearly three-quarters of patients would want an unnecessary test and over half would want an unneeded referral; in an end-of-life situation, about half would choose comfort over aggressive treatment. The physician survey revealed that nearly one-quarter of cardiologists would recommend more frequent follow-up visits than called for in current guidelines, while less than 1 percent would recommend fewer. By contrast, only 9 percent of primary care physicians would recommend more frequent follow-up visits than called for in practice guidelines and an equal number would recommend less frequent visits. There is substantial variation in all of these measures across HRRs, suggesting they could contribute to regional variation in spending.

To explore this hypothesis, the authors match their physician and patient survey data to Medicare expenditure data, focusing on expenditures in the last two years of life for beneficiaries with various fatal illnesses, and estimate models relating HRR-level expenditures to the survey measures. They find that average expenditures in an HRR rise with the percent of physicians that are cowboys - i.e. those that recommend more intensive care than clinical guidelines would suggest — and fall with the percent that are comforters - i.e. those who recommend more low-cost, palliative care for very sick patients. The estimated effects are large: a 10 percentage point increase in cowboys raises expenditures by 7.5 percent, while a 10 point increase in comforters lowers expenditures by 4.1 percent. A 10 percentage point increase in physicians recommending more frequent care than guidelines suggest is associated with an increase of 9.5 percent in end-of-life spending. However, the authors find a very modest relationship between regional patient preference measures and spending.

Next, the authors explore what factors affect physician beliefs. Older physicians are more likely to be cowboys and to recommend extra follow-up care, while men are less likely to be comforters. Board certification, a marker of quality, is associated with a reduced likelihood of being a cowboy or recommending more frequent follow-up care, consistent with earlier work showing that lower quality physicians spend more treating identical patients. Physicians in solo or 2-person practices are more likely to be cowboys, as are cardiologists who report that they accommodate the wishes of referring physicians. Yet the lion's share of differences in how doctors say they would treat patients are not explained by financial, organizational, or other factors, and likely results from differences in beliefs.

Finally, the authors use their results to estimate that, were physicians to follow professional guidelines, end-of-life Medicare expenditures would be 36 percent less, and overall expenditures would be 17 percent less. These results lead them to conclude "individual physician beliefs regarding treatment options can explain a substantial degree of regional variation in utilization among the U.S. Medicare population." While the authors note that economic incentives are not unimportant, the presence of large regional variation in environments where economic incentives are muted is consistent with a large role for physician beliefs. As yet, we know little about how physician beliefs arise and can be shaped, making this a productive area for future work.

The authors acknowledge funding from the National Institute on Aging (grants T32-AG000186-23 and P01-AG031098 to the NBER and P01-AG019783 to Dartmouth) and from the Laboratory for Economic Applications and Policy (LEAP) at Harvard University to Skinner.