The current geographic practice cost index (GPCI) incorporates a geographic adjustment for the price of physician work. The Centers for Medicare and Medicaid Services (CMS) computes the work GPCI using the relative median hourly earnings in seven nonphysician occupations collected as part of the Bureau of Labor Statistics Occupational Employment Statistics data.
The adjustment is set to 25% of the relative wage differences for each area compared to the national average. Thus, for example, if wages in an area are 8% above the national average, the adjustment factor would be 1 + (25%) (8%) = 1.02. Although the partial adjustment with an inclusion factor of 25% is in law, the committee did not find an explicit scientific or policy basis for the choice of 25% as opposed to any other percentage value. This led the committee to consider normative principles and empirical analyses that might form the basis for the choice of an inclusion factor of 0%, 100%, or some other value. The committee also considered alternatives for the reference group on which the base index of wage differentials should be based.
The argument against any physician work adjustment is based on the view that physicians providing an equivalent service for a federal program should receive the same reimbursement regardless of where they are located: “work is work.” According to this view, Medicare’s work relative value unit (RVU) already takes into account physician work effort, and it takes no more or less effort to provide the same medical service in different areas (Goertz, 2011). Furthermore, self-employed physicians are more like suppliers than employees and should be paid equivalently for the commodity (health care) that they supply.
A counterargument to this position is that wage rates in the private sector, including the health care industry, vary across labor markets. Federal wage rates for a variety of occupations ranging from census workers to highly skilled professionals and managers also vary geographically. Indeed, geographic variation in wages for nonphysician health care workers is reflected in the geographic adjustment of hospital and physician office labor expenses. Furthermore, a substantial and growing share of physicians (nearly 50% of new physicians in 2010), according to the Medical Group Management Association (MGMA) (2010) are
employees who must be paid at locally prevalent salary scales, and self-employed physicians should be paid at a rate that allows them to compensate themselves in line with salaries of their local employed colleagues.
Since the objective at this point is to assess the relative costs of equivalent physician labor to practices in different areas, an obvious solution would be to use current mean or median earnings of a group of physicians (or a standardized mix of specialties) to determine the ratios, thus making physicians their own reference group (corresponding to an inclusion factor of 100%). The committee rejected this solution, however, because of the same concerns about circularity that motivated the search for wider reference groups than hospital employees (in the hospital wage index) and physician office employees (in the practice expense GPCI). Because almost all physicians work in the health care industry, expanding the data source for physician earnings beyond the health care industry would not solve the circularity problem inherent in using physician wages for the work GPCI. Such an approach would incorporate local wage distortions into the wage rate, potentially making it possible for a large practice or group of practices to affect or even manipulate their physician work reimbursement rates within a market.
The committee therefore turned to economic theory for a rationale for a more indirect approach. The economic argument for varying physician compensation across areas is that, in general, compensation varies inversely with the affordability and desirability of an area as a place to live and work; thus, both a lower cost of living and greater availability of amenities (cultural attractions, low crime, and access to outdoor activities, for example) will tend to depress wages. (See the discussion of the theory of compensating wage differentials in Chapters 2 and 5.) Under this theory, wages will adjust so that the marginal physician choosing among locations will be indifferent among high-wage but less desirable options and lower-wage but more desirable options, while those with various preferences off the margin will sort into the locations in the quantities required to satisfy demand.
There is no way to directly assess the relative desirability of areas to physicians. For the reasons given above, the committee prefers not to rely on physicians as the reference group. However, it seems reasonable to assume that other reference groups with similar levels of education and income to physicians and similar degrees of professionalization might have similar location preferences, particularly with regard to the trade-off between income and amenities. A wage index calculated from such groups might then be used to estimate appropriate payment to physicians. The current GPCI adjustment starts with such an index, calculated from seven professional groups: architecture and engineering; computer, mathematical, and life and physical sciences; social science, community and social service, and legal; education, training and library; registered nurse; pharmacists; and art, design, entertainment, and sports and media (CMS, 2010).
A limitation of this approach is that different factors might affect wages for physicians and other professional occupations. For example, a physician’s skills are geographically nonspecific and highly portable—oncologists or pediatricians who practice in Nashville have much the same skills as their respective counterparts who practice in New York City. But lawyers who practice in New York City include a much higher proportion of employees of large corporations and investment banks, and differences between median incomes of lawyers in these cities reflect this difference in professional mix within the occupation as well as the amenities and cost of living differences between cities. Similarly, teachers’ wages are affected by factors such as local school funding policies and unionization, which are not relevant to physicians.
On the other hand, amenities that might be attractive to some physicians, such as the opportunity to do research or teach in an academic medical center, are not relevant to other
occupations. Such considerations suggest a partial adjustment since the reference index would be partially but not perfectly indicative of appropriate wages for physicians; the current 25% adjustment might thus be justified in general principle, although there is little empirical basis for the choice of this specific number over any other value between 0% and 100%.
Empirically, we might expect that if the amenities and cost of living common to physicians and other professional occupations played a predominant role in determining compensation, the incomes of physicians and the reference occupations would be highly correlated across areas; this finding would support heavily weighting the reference-group incomes in determining a physician work adjustment. Conversely, a low correlation would suggest that the reference groups are poor proxies for factors affecting physicians, and thus relatively little weight should be given to their wage index.
This theoretical approach can be implemented through regression modeling. (The method described herein extends that of Gillis and colleagues  by estimating both the inclusion factor and the combination of occupational indices.) The data required for this model would be median physician wages (per RVU, to remove the effects of different work hours and specialty mixes) and median wages for the various reference occupations, each by metropolitan statistical area (MSA) or statewide non-MSA. (Data for employees in each group would be preferable, to exclude the entrepreneurial return obtained by the self-employed in their role as owners of a business.)
First, the MSA medians would be normalized for each occupation to obtain an index value by dividing each by the corresponding national mean of medians (weighted by physician population in the MSA). Then the raw physician index would be regressed on all of the reference occupation indices in a multivariate linear regression; the predictions under this model would become the new physician work adjustment factors. (By construction, this index would be 1 in an MSA in which all of the reference indexes are also 1, that is, an area with average wages for all occupations, and its weighted mean would also be 1.) This procedure would simultaneously form the combination of reference occupations that best predicts physician compensation (while excluding effects unique to physicians) and determine the weight to be given to this combination in determining the physician adjustment. Alternatively, the budget neutrality adjustment could be viewed as external to the model, in which case the statistical model would not be constrained to 1. This approach might improve the accuracy of the indexes; in this case, the budget neutrality adjustments would be performed afterwards.
The amount of variation in the predicted work adjustment in this model would implicitly take into account the observed amount of variation in physician compensation across payment areas (which might be different from that for the reference occupations) and also how well the reference population compensation predicts physician compensation, summarized by the correlation coefficient between the predictions and the raw physician index. Even with the best available choice of reference occupations, a low correlation such as 0.25, which is the same level as the adjustment currently used, would be an indication that the factors determining physician wages are too distinctive to be adequately captured by this methodology; in that case, a direct comparison of physician salary data from a variety of sources, such as MGMA or American College of Surgeons (ACS), might be the best available option. Another possibility might be to use an F-statistic (p < .05) to test the null hypothesis that the correlation between the geographic salary differentials for physicians and other occupations is 0, then using the model if the null hypothesis were rejected.
The relationship between the regression coefficients and the inclusion factor (now 25%) can
be clarified by a simple reparametrization of the regression equation. The present procedure is represented by formula of the form where W is the final physician work
index, C is the inclusion factor, and P is the reference (proxy) compensation index. Suppose the regression prediction is where Xk is the wage index for reference profession k.
This can be rewritten as is a weighted average of the proxy indexes, corresponding to P in the current method. Then a* is the multiplier C corresponding to the current 25%.
CMS (Centers for Medicare and Medicaid Services). 2010. Medicare program; payment policies under the Physician Fee Schedule and other revisions for Part B for CY 2010. Federal Register 75(228):73170–73860.
Gillis, K. D., R. J. Willke, and R. A. Reynolds. 1993. Assessing the validity of the geographic practice cost indexes. Inquiry 30:265–280. Center for Health Policy Research, American Medical Association.
Goertz, R. 2011. Testimony to the IOM Committee on Geographic Adjustment Factors in Medicare Payment on behalf of the American Academy of Family Physicians.
MGMA (Medical Group Management Association). 2010. MGMA physician placement report: 65 percent of established physicians placed in hospital-owned practices. Engelwood, CO: Medical Group Management Association. http://www.mgma.com/press/default.aspx?id=33777 (accessed March 8, 2011).