Regional variation in US healthcare use: evidence from patient migration

There is considerable geographical variation in the use of healthcare by beneficiaries of Medicare, the US federal health insurance program for people who are 65 or older. This research explores the extent to which regional disparities are driven by the providers, whose use of expensive tests or procedures might vary across different places, or by the patients, who might have different healthcare needs and preferences. Analyzing data on Medicare beneficiaries who have migrated from one part of the country to another, the study finds that patients and providers account for roughly equal shares of the differences in regional spending. The results provide a better understanding of the components of medical costs, adding nuance to the debate about possible inefficiencies in US healthcare spending.

The use of healthcare varies widely across the United States. Adjusting for regional differences in age, sex, and race, healthcare spending for the average person enrolled in Medicare in Miami was $14,423 in 2010, but just $7,819 for the average enrollee in Minneapolis. Even within the same state, there can be wide differences: for example, the average enrollee in McAllen, Texas, spent $13,648, compared with $8,714 in nearby and demographically similar El Paso.

There are similar geographical variations in the frequency of specific treatments and in measures of total healthcare use that adjust for regional variation in administratively set prices. Higher area-level use is not generally correlated with better patient outcomes.

Understanding what drives geographical variation in healthcare use has important implications for policy. If high-use areas like McAllen and Miami are different mainly because their doctors’ incentives or beliefs lead them to order excessive treatments with low return, policies that change those incentives or beliefs could result in savings on the order of several percentage points of GDP.

If, on the other hand, patients in high-use areas are simply sicker and therefore more in need of healthcare, or if they prefer more intensive care, such policies could be ineffective or counterproductive.

Patients versus providers as drivers of regional variation

Our research makes use of patient migration to separate variation due to the ‘demand-side’ factors of patients’ characteristics, such as their health or healthcare preferences, from variation due to place-specific variables or ‘supply-side’ factors, such as doctors’ incentives and beliefs, endowments of physical or human capital, and hospital market structure.

To see the intuition for our approach, imagine a patient who moves from high-use Miami to low-use Minneapolis. If all of the difference in healthcare use between these cities arises from supply-side differences such as doctors’ incentives or beliefs, we would expect the migrant’s use to drop immediately following the move, to a level similar to other patients of the low-use doctors in Minneapolis.

But if all of the difference reflects the demand-side reality that residents of Miami are sicker, we would expect the migrant’s use to remain constant after the move, at a level similar to the typical person in Miami. Where the observed change falls between these two extremes identifies the relative importance of demand- and supply-side factors.

The healthcare use of Medicare beneficiaries

Our research analyzes the healthcare use of a sample of 2.5 million Medicare patients over the period from 1998 to 2008. Medicare is the US federal health insurance program for people who are 65 or older, a population that accounts for about a third of the country’s total annual healthcare spending.

We focus on a subgroup of enrollees who did not move during this time, as well as 500,000 Medicare enrollees who moved from one market – or ‘Hospital Referral Region’ (HRR) – to another. (The HRR concept was first developed by the Dartmouth Atlas of Health Care, a research project housed at Dartmouth College.)

By examining the healthcare use of the same people as they live in different places, we develop a natural experiment for addressing the question of whether it is providers or patients that have the most influence on the use of healthcare.

We find that nearly 50% of the spending differences across geographical areas stems from the characteristics of patients, meaning both their basic health and their varying preferences for the intensiveness of care. The rest of the spending differences derive from place-specific factors, potentially due to disparities in provider practices and incentives.

We also find that demand-side factors matter more for outcomes such as emergency room visits. A relatively larger share of the regional spending variation in emergency visits – 71% – is attributable to patients, who are likely to make more of the decisions about whether or not to seek care in those situations.

Demand-side factors matter less for outcomes such as diagnostic and imaging tests, where the physician is the main decision-maker: patients account for only 9% of the regional variation in diagnostic tests and 14% of the variation in imaging tests. In these cases, the variation by geographical region may be due to differing provider practices, with healthcare institutions in some places consistently spending more money on testing than other providers do elsewhere.

We also find that when a Medicare enrollee moves and receives a different level of healthcare spending, most of that change occurs in the first year after the move. The new level of use stays relatively constant in subsequent years.

Analysis of the mechanisms underlying our results suggests that a substantial share of the patient component can be attributed to differences across areas in average health status. The remainder is likely to reflect a combination of both unmeasured health and preferences.

We also find intriguing correlations with our estimated place effects. Among other things, they suggest that areas with higher place components of healthcare use also have physicians who believe in a more aggressive practice style, fewer non-profit hospitals, and a sicker population.

Implications for healthcare efficiency

In recent years, public discussion of geographical variation in healthcare use has largely highlighted providers, with the implication that reducing apparently excessive treatments could trim overall medical costs. While our results suggest that demand-side factors play a larger role than conventional wisdom might suggest, this does not translate immediately into conclusions about efficiency.

The correlation of healthcare use with demand-side factors (and with patient health in particular) may reflect differences in the marginal impact of treatment or the marginal utility from a given impact – and thus be consistent with efficiency. But it could also reflect differences in other drivers of demand, such as patients’ information or beliefs. A more careful examination of the efficiency implications of geographical variation is an important direction for further work.

The fact that a large part of demand-side geographical variation reflects variation in patients’ health may also point to limits to the effectiveness of demand-side policies aimed at changing patients’ beliefs or preferences. At the same time, the sharp adjustment we observe around moves suggests policies that affect the supply-side can have immediate impacts.

While we have taken a first step towards understanding the origins of the patient component we measure, it remains for future work to develop a better understanding of the mechanisms behind the place component. Particularly interesting questions concern the role of physicians’ training and practice patterns, and the role of healthcare organizations.

This article summarizes ‘Sources of Geographic Variation in Healthcare: Evidence from Patient Migration’ by Amy Finkelstein, Matthew Gentzkow and Heidi Williams, published in the Quarterly Journal of Economics 131(4): 1681-1726, 2016.