Research documenting geographic variation in Medicare spending has attracted considerable attention, generating calls for payment reform and focusing attention on differences across markets.11,25
Our findings demonstrate that spending variation is present in the population insured by large commercial firms as well as the Medicare population. But the correlation between Medicare spending and spending by the commercially insured sample is weak to nonexistent.
Importantly, our work highlights the significance of understanding that variation in spending reflects variation in both price and utilization. Our analysis suggests a positive correlation in utilization for the elderly and nonelderly, but a small negative correlation in spending. This implies a negative correlation in price (though we did not observe prices directly), which may reflect differences in pricing mechanisms.
The degree of provider competition in local markets should affect prices in commercial markets but not Medicare, which uses administered pricing. For example, significant hospital capacity or competition may allow commercial insurers to bargain successfully relative to Medicare, whereas markets with little provider competition may result in commercial payers being charged more relative to Medicare. Past empirical evidence supports this view.26
Our descriptive statistics suggest price effects may be important as well.
Although our period of observation included passage of the Balanced Budget Act, which had the immediate effect of reducing Medicare spending but may have increased rates to commercial payers,27
our analysis does not necessarily indicate cost shifting. The pattern of results we observed, particularly the association with market structure, may merely reflect differential market power as opposed to a causal relationship between prices in different sectors.
Our analysis has several limitations. The weak correlation may reflect data issues (including noise due to small sample sizes in some HRRs) or population differences. The Thomson Reuters data reflect the experiences only of large firms. We make no claim that our results are generalizable to other sectors of the commercially insured market (small firms) or to other segments of the under-65 population (eg, Medicaid, the uninsured).
Greater standardization of the commercial population, their plan types, and benefit generosity, as well as better standardization for the presence of supplemental coverage for Medicare beneficiaries, could also influence our findings, but the impact of those factors and their variation across HRRs would need to be substantial to alter the conclusion that commercial and Medicare spending are not highly correlated.
In fact, unreported analysis of retirees from the Thomson Reuters (Medstat) MarketScan Database suggests a positive but modest correlation with the more representative Dartmouth data, suggesting that differences between individuals insured by large firms and the overall Medicare population may be important. Our message is simply that correlation in spending across areas will be sensitive to the populations studied.
The low correlation we found between Medicare and non-Medicare spending across areas is consistent with recent work by Rettenmaier and Saving, who used much broader measures of non-Medicare spending.28
However, our results suggest that the correlation in measures of utilization may be more positive than the correlation in spending, suggesting correlation in prices may be negative. To confirm this hypothesis, measures of utilization would be needed.
The restriction to hospital and physician services is another limitation. This focus recognizes that we did not have Medicare data on prescription drug spending and that spending on post–acute care services is much less important in commercial plans. There also were differences in the unit of observation. The inpatient data from the ARF are based on the county in which the hospital is located (for both Medicare and non-Medicare data), and the spending data are based on HRRs using the beneficiary’s residence (for both Medicare and non-Medicare data). Moreover, county and MSA-level data are assigned to HRRs using crosswalks that may introduce some error.
Finally, we did not observe the relationship between spending variation and quality. Research suggests spending and utilization are not highly correlated with quality, but we cannot address that issue in this work.4,9,29,30
Despite these issues, the story surrounding geographic variation in medical care spending appears more complex than what might be suggested by variation in Medicare-only spending patterns. Market structure likely affects Medicare and commercial spending differently because commercial insurers can better exploit competition and are correspondingly more vulnerable to market concentration. This highlights the importance of understanding variation in pricing as well as in utilization.
The potential susceptibility of private payers to provide market power has important implications when assessing the merits of private markets or public markets in setting prices. Administrative price systems have many flaws, which are fundamentally related to the difficulty in determining the appropriate price when costs are heterogeneous, are not known very precisely, are changing over time, and may reflect discretionary provider behavior. MedPAC, in its role of advising Congress about Medicare reimbursement, struggles constantly with this issue. Moreover, administered prices can be subject to political manipulation. For example, over a third of hospitals use geographic adjustment indices that are exceptions to the standard adjustment estimates.31
Some of these reclassifications may reflect legitimate concerns that adjustments do not appropriately reflect cost differences, which is a general problem in administered pricing systems. But others reflect ad hoc rules inserted simply to benefit particular providers or areas.
The concern is not limited to hospitals. There is widespread concern that the system for setting physician payment, which relies on the Relative Value Scale Update Committees, is biased against primary care services.32,33
Finally, payment rates for services such as graduate medical education have been set above the rates suggested by statistical analysis.34
These are just a subset of possible examples of where the political system could distort administrative prices.
yet despite all the concerns about administrative pricing, our analysis appears to suggest that administratively set prices seem to reduce purchaser vulnerability to provider market power. The challenge for policymakers interested in administered prices must be how to mitigate distortions in the price-setting process, although policymakers will never have enough information to establish perfect (economically efficient) prices (bundled or otherwise).
The analogous challenge for policymakers interested in market systems is how to avoid the pitfalls associated with provider market power. It is not clear whether concerns about market systems are more important or will be easier to mitigate than concerns about administered pricing. However, as the country moves forward with changing the healthcare system, these concerns will be paramount. Medicare, despite numerous inefficiencies in pricing, may be better able to avoid problems with market power in certain markets, suggesting that if private markets are to work better, strategies need to be developed to promote competition (or at least competitive pricing) for provider services. Descriptive analyses such as ours can only raise these issues. Analysis that does a better job of measuring prices and provides more detailed utilization patterns is needed to confirm these suspicions and inform potential policy solutions.