Geographic Variation in Medicare Beneficiaries' Medical Costs Is Largely Explained by Disease Burden

Rethinking What Drives Geographic Differences in Medicare Spending

Published in May 2013 in Medical Care Research and Review, a study by James D. Reschovsky, Jack Hadley, and Patrick S. Romano challenged a widely held assumption in health policy: that geographic variation in Medicare spending is primarily driven by differences in how aggressively physicians practice medicine, rather than by differences in how sick patients actually are.

Using claims data from 1.6 million fee-for-service Medicare beneficiaries spread across 60 representative communities, the researchers found that population health -- the actual burden of disease in a given area -- likely accounts for more than 75 to 85 percent of the cost variation observed across geographic regions. That conclusion stood in sharp contrast to influential earlier work, most notably from the Dartmouth Atlas of Health Care, which had suggested that much of the geographic spending variation reflected overuse and inefficiency rather than genuine differences in patient needs.

The Problem: How You Adjust for Patient Illness Changes Everything

At the center of the debate was casemix adjustment -- the statistical methods researchers use to account for differences in how sick patient populations are across different parts of the country. If one region has sicker residents than another, you would naturally expect higher medical spending there. The question was how to measure and control for those differences in sickness accurately.

Different studies had used different casemix adjustment approaches, and the choice of method produced dramatically different estimates of how much geographic variation was "real" versus how much was explained by patient health status. The Reschovsky, Hadley, and Romano study took a hard look at this methodological disagreement and tested the assumptions behind the two most common approaches.

Two Competing Methods for Measuring Patient Sickness

The study evaluated two main casemix adjustment techniques that researchers had been using to study geographic variation in Medicare costs.

Diagnosis-Based Adjustment

The first approach controls for patient conditions by using the diagnoses recorded on Medicare claims. If a patient's claims show diagnoses for diabetes, heart failure, and chronic kidney disease, for example, those conditions are used to predict expected spending. The concern with this method, raised by critics including the Dartmouth researchers, was that physicians in high-spending areas might simply be more aggressive in coding diagnoses -- recording more conditions or more severe conditions on claims -- which would make their patients appear sicker than they truly were. If that were the case, the diagnosis-based adjustment would be biased, making it look like patient illness explained the spending differences when the real driver was physician behavior.

End-of-Life Spending Adjustment

The second method, favored by the Dartmouth Atlas researchers, took a different tack. Instead of relying on diagnoses from claims, it looked at spending on patients who were in their last year or two of life. The logic was that people close to death are presumably equally sick regardless of where they live, so any spending differences among this group would reflect practice style rather than patient health. If Miami spent twice as much as Minneapolis on patients in their final months, the argument went, that difference must be due to how doctors practice, not how sick the patients were.

What the Study Found

The researchers tested both methods and reached two conclusions that pushed back against the conventional wisdom.

No Evidence of Diagnosis Coding Bias Across Areas

First, the authors found no evidence that diagnosis-based adjustment was biased by area-level differences in physician coding patterns. The concern that doctors in high-spending regions were simply recording more diagnoses -- thereby inflating how sick their patients appeared -- was not supported by the data. When the researchers tested for this kind of coding bias, they came up empty. This was an important finding because it meant that using diagnoses from claims to measure patient illness was more reliable than critics had suggested.

The End-of-Life Assumption Does Not Hold Up

Second, and more damaging to the Dartmouth approach, the study found that the core assumption behind end-of-life spending comparisons was flawed. Patients close to death were not equally sick across all areas. The severity and mix of conditions among dying patients varied by region, meaning that spending differences among this group could not simply be attributed to physician practice style. Some areas had sicker dying patients than others, and that mattered for costs.

This finding undercut one of the most commonly cited pieces of evidence for the claim that physician behavior, not patient illness, was the primary driver of geographic spending variation.

Current-Year Diagnoses Outperform Prior-Year Diagnoses

The study also examined a technical question within the diagnosis-based approach: whether it was better to use diagnoses from the current year (the same year as the spending being measured) or from a prior year. Some researchers had used prior-year diagnoses to avoid the potential circularity of using same-year medical encounters to both measure illness and predict spending. The Reschovsky team found that current-year diagnoses produced more accurate and appropriate casemix adjustments. Using prior-year diagnoses missed newly developed conditions and changes in health status that directly affected current-year spending, leading to less reliable estimates of how much geographic variation was attributable to patient health versus other factors.

The Bottom Line: Disease Burden Explains Most of the Variation

When the researchers applied their preferred diagnosis-based adjustment using current-year data, they concluded that population health differences probably explained more than 75 to 85 percent of the cost variation observed across the 60 communities in their sample. That left only a relatively small share of geographic spending differences that could be attributed to factors like physician practice patterns, health system efficiency, or local medical culture.

This was a significant finding for health policy. If most of the geographic variation in Medicare spending reflected genuine differences in how sick people were in different parts of the country, then the policy response needed to focus more on addressing underlying health disparities and disease prevention rather than solely on changing how doctors practice or penalizing high-spending regions.

Policy Implications and the Broader Debate

The geographic variation debate carried enormous stakes for Medicare policy. Proposals to reduce Medicare spending by bringing high-cost areas in line with low-cost areas rested on the assumption that spending differences were largely driven by waste and inefficiency. If the Dartmouth view was correct, there was potentially hundreds of billions of dollars in savings available by getting physicians in high-cost regions to practice more like those in low-cost regions.

But the Reschovsky, Hadley, and Romano findings suggested a more sobering reality. If most of the variation reflected actual differences in disease burden, then the savings from standardizing practice patterns would be considerably smaller than some policymakers had hoped. The more productive path might involve investing in prevention, managing chronic diseases more effectively, and addressing the social and economic factors that made populations in some regions sicker than in others.

The study was published in Medical Care Research and Review and included a supplementary methodological appendix. It was conducted through the Center for Studying Health System Change (HSC), a nonpartisan policy research organization in Washington, D.C., affiliated with Mathematica Policy Research.

Sources and Further Reading

CMS -- Medicare -- Official Medicare program information and resources.

Kaiser Family Foundation -- Medicare -- Medicare policy analysis and data.

MedPAC -- Medicare Payment Advisory Commission -- Congressional advisory body on Medicare payment policy.

Health Affairs -- Peer-reviewed health policy research.

Commonwealth Fund -- Research on health care system performance.