Knowing the cost of production is a basic business principle that often eludes residency training programs. A variable and complex aggregation of hospitals, clinics, laboratories, libraries, doctors' offices, service centers (sports medicine, cancer, rehabilitation, surgery), nursing homes, and community agencies provides the nexus whence well-trained physicians emerge, prepared to practice one of medicine's specialties. Even the most diligent residency directors and financial officers are thwarted in their pursuit of precise accounting of residency costs, in no small part because of the far-flung components of residencies and the irreducible ambiguity of what comprises an educational, service, or research expense. No absolute chart of accounts exists that captures, line by line, all the costs of residency training. There should be little wonder, then, that reported costs vary. Indeed, large variations in reported costs have been documented, lamented, and largely accepted as just the way things are.
Medicare is the largest source of funding for graduate medical education (GME) in the United States, and one critical variable that helps determine the amount of Medicare payments for GME that an individual institution receives is the institution's reported costs of training. Higher reports of costs of training result in higher Medicare payments for GME. In this study, we used public data sets to extend prior research to describe the actual direct costs of residency training upon which Medicare payments are based, both within and across states, and to assess the stability of these costs over two consecutive years. We also assessed how direct medical education funding related to a specific workforce policy issue: the number of primary care physicians in training and practice, state by state.
To understand the financial morass of GME funding that teaching institutions face, it is important to understand the legislation and mechanism for receiving GME funding from Medicare.
The House Ways and Means Committee, upon implementation of the Medicare program in 1965, proclaimed that the services provided to Medicare beneficiaries are enhanced by the educational activities related to their care at teaching hospitals, and that this increase in quality warranted the provision of support for some of the associated costs.1 That expectation of increased quality of care is still widely held, and there is documentation of such an effect.2 In 1999, the Medicare Payment Advisory Commission recommended that resultant clinical care enhancement be recognized in any formula for funding GME.
Medicare payments for GME are part of a complex web of financing for teaching hospitals. Many teaching hospitals function as safety-net facilities, particularly in providing services to underserved inner-city populations. Much of the care that they provide has gone uncompensated, and has recently become a source of serious administrative confusion because many uninsured and low-income patients are actually eligible to receive Medicaid or other benefits from programs in which they are no longer enrolled.3 The provision of free medical services to the poor has been possible, in part, because of disproportionate-share hospital (DSH) payments and because these hospitals are sustained by GME funding.4 But the demands on these institutions to compete in a rapidly changing marketplace strain their ability to provide charitable services, and they post difficulties for research programs at academic health centers. Although National Institutes of Health funding has increased, its funding is designated largely for research. An academic heath center's ability to shift revenues from one source to cover costs elsewhere in the academic enterprise has diminished. Faculty time, once devoted to teaching and research, is often diverted to clinical practice in the effort to generate additional income.5 GME payments have been especially important to the enterprises of academic health centers that are not research-intensive, because these institutions have not benefited substantially from the growth in appropriations for the National Institutes of Health.
The accounting structure for GME payments is administered by the Health Care Financing Administration (HCFA). The structure is rigid, is not clearly related to activities required to produce well-trained physicians, and does not reflect how the teaching hospitals actually use the funding. HCFA Medicare policies distinguish between the direct medical education (DME) costs and indirect medical education (IME) costs that teaching hospitals incur for medical education, and hospitals are required to differentiate between these two expense categories in their yearly cost reports submitted to HCFA. The amounts of DME and IME payments are determined by separate formulas.
From 1965 to 1983, HCFA provided DME funds on a cost-reimbursement basis to defray the costs of residents' and faculty's salaries, and office space, as well as other training program expenses. Since the adoption of the inpatient prospective payment system, DME payments have been based on a hospital's direct GME costs per resident for the year 1984, adjusted annually for inflation and multiplied by the number of full-time-equivalent residents and the proportion of total inpatient days allotted to Medicare patients. More recently, funding has been made somewhat dependent on the hospital's specialty mix among its residents. Payments that hospitals now receive for primary care residents' GME are about 6% higher than they are for residents in other specialties.6
The rationale for IME funding differs from that for DME funding. It is intended to reimburse teaching hospitals for the expensive and technically complex services they provide to their patients, who tend to be more severely ill than are those admitted to other facilities. Thus, IME constitutes a separate add-on to payments otherwise received for the hospital's reported diagnosis-related group case mix under the prospective payment system. It is a function of the resident-to-bed ratio and is adjusted for the local costs of both capital and labor. The larger of the two components of GME payments is IME (in 1997, $4.44 billion versus $2.09 billion for DME). DME costs, however, are more directly related to the teaching function, and these payments are made to hospitals at a rate per resident. Thus, DME was the major focus of our study.
Criticism of this country's GME payment system has been rekindled by the debate leading to enactment of the Balanced Budget Amendment of 1997 and subsequent refinements to that legislation. Inequities in the compensation of teaching hospitals for residency training have received much of the attention.1,7,8 The revelation of similarly large and unexplained variations in other areas of the health care field (i.e., consumption of services and related expenditures) earlier in the decade precipitated a period of serious reassessment of national policy.9 Documented variation in GME payments, in spite of strictly defined methods for their calculation, has not yet received similar attention, even though it has eroded confidence in our system of funding for residency training. The need for its reevaluation has been further stimulated by the renewed interest in revising GME policy to effect significant changes in the U.S. physician workforce.1,8
Focus of Analysis
In this study, we comprehensively examined both inter- and intrastate variances in the reimbursement rate structure, and in corresponding GME payments by Medicare to teaching hospitals, in an effort to better understand the relationships that exist between policy objectives and the current payment system. We emphasized primary care residents in our analyses because increasing the proportion of primary care physicians in the United States and geographically distributing them to meet the needs of all Americans, especially underserved and otherwise disadvantaged populations, are major policy objectives.10 We specifically determined intrastate variations in DME costs upon which HCFA payments to teaching hospitals were based in 1996 and 1997, and correlated those costs with the numbers of primary care residents being trained in each facility receiving Medicare funding, primary care health personnel shortage areas (PCHPSAs), and primary care physician-to-population ratios by state. We also examined the relationships of an institution's proportion of inpatient days for Medicare beneficiaries, another component of DME payment, with the numbers of both specialty and primary care residents in training.
Data for this study were taken from Public Use Files, obtained from HCFA, and the 1999 Area Resource File, maintained under contract for the Bureau of Health Professions. Because both are public-domain data sets, the analyses described in this paper can be replicated or conducted for other years.
The primary Public Use File we used in this study was the latest available, and it contained financial and other information from the Medicare Hospital Cost Report submitted by the fiscal intermediaries for certified hospitals for fiscal year 1997 (a one-year period beginning on or after 10/1/96 and before 10/1/97). Another file, for FY96, was also examined to evaluate the consistency of GME payment policy over time. Not all of the hospital cost reports in the Public Use Files had been finally settled by HCFA, and they were, therefore, officially subject to revision. Later changes could also conceivably occur as a result of a report's being reopened.
We conducted most analyses using data for the 648 hospitals that had records in both the FY96 and the FY97 Public Use Files, and indications of at least moderate GME volume in FY97. This restriction was used to reduce outliers and to limit comparisons to those with significant financial and training implications. We included training facilities that had reported at least ten primary care resident full-time equivalents and $100,000 in total DME payments for Part A and Part B Medicare combined. Because of the extremely small Medicare proportion of inpatient days at children's hospitals and the importance of that variable in GME payment formulas, those institutions were also omitted from examination. Training facilities in Puerto Rico and several states were eliminated from certain comparisons. Alaska, Mississippi, Idaho, Vermont, and Wyoming had fewer than two hospitals that met the inclusion criteria, and therefore they were not used for intrastate contrasts. Thus, Table 1 contains data for 642 facilities located in 45 states. It should be noted that four hospitals, all in different states, with major detectable data errors in key payment and resident full-time equivalent fields were excluded without applying the study inclusion criteria.
The February 1999 release of the Area Resource File was maintained under contract for the Office of Research and Planning of the Bureau of Health Professions. It contained the 1997 federal designation status of each U.S. county pertaining to a PCHPSA status. A ratio of population-to-primary care physicians higher than 3,500:1 was the principal criterion used for designation. The actual calculations of the ratio used in one correlation with states' average DME costs was conducted with 1998 data because the Area Resource File did not contain osteopathic generalist physician totals for 1997.
Table 1 shows variations within and among states in their reporting of DME costs provided by their hospitals. Medicare paid a considerable share of these amounts, the proportion of Medicare beneficiaries' inpatient days to total inpatient days. DME average costs per primary care resident ranged from $43,707 in North Dakota to $92,072 in New York, but intrastate differences were much more remarkable. The range (highest to lowest updated amount per primary care resident) within each of 11 states was greater than $100,000. In these and in nine additional states, the range exceeded the average hospital DME cost. It was more than $156,000 in California, where one teaching hospital reported costs more than 12 times that of another.
Stability of the Variation
Of the teaching hospitals that met the inclusion criteria for analysis in FY97, 588 met the same criteria for FY96, enabling year-to-year correlation. There were strong associations between FY96 and FY97 data for the numbers of primary care residents being trained at a facility (r = .93, p < .001) and for the updated per-resident amounts for payment a facility received (r = .96, p < .001). Data for FY97 differed little from those for FY96, indicating stability over time. For the 588 hospitals, reported DME costs for FY97 averaged $73,823, versus $73,542 for the preceding year, with standard deviations of $26,272 and $26,281, respectively.
Primary Care Residents
The average number of primary care residents in training among the 588 teaching hospitals increased slightly, from 58.0 in FY96 to 58.7 in FY97. Changes in primary care full-time equivalents were inversely related to the numbers of primary care residents already at a hospital in FY96, and to the addition of non—primary-care residents from one year to the next (means = 54.2 in FY96 and 56.0 in FY97). There was no association between changes in hospitals' primary care resident full-time equivalents and their per-resident DME costs.
Other Determinants of Medicare GME Payments
The proportion of Medicare payments relative to total inpatient days for Medicare recipients is another important component in determining the DME payments made to teaching hospitals. Hospitals exhibited a wide range in the percentages of Medicare beneficiaries' inpatient days, from 3.8% to 73.7%, with a mean of 42.4%. We observed an inverse relationship between this percentage and the number of primary care residents being trained (r = −.41, p < .001). Table 2 depicts this association. It might be assumed that a higher concentration of older inpatients, indicated by higher concentrations of Medicare beneficiaries, would require more specialists' services. For the 471 of 648 hospitals in this analysis with at least ten non—primary-care residents, however, a similar negative correlation was found between specialist resident full-time equivalents and the percentage of Medicare beneficiaries' inpatient days by Medicare enrollees (r = −.37, p < .001). The relationship between the primary care resident DME costs paid in part by Medicare and the number of primary care residents claimed by the hospitals was weak (r = .07, p = .08).
Direct GME Payments and Public Access to Primary Care Physicians
According to the Area Resource File, almost one fourth (756, 24.5%) of all U.S. counties were totally designated as PCHPSAs in 1997. Partial designations of PCHPSAs typically indicate that the needs of special populations, e.g., the medically indigent or ethnic minorities, are unmet. By contrast, whole designations are more a function of the ratio of primary care physicians to the general population of the county, a problem that can be solved simply by increasing the number of primary care doctors in the area. The federal government invests in the National Health Service Corps, Area Health Education Centers, and other interventions in an effort to improve accessibility to primary health care in PCHPSAs. Yet, among the 47 states with at least five counties, a negative association was found between the average hospital DME cost and the percentage of counties totally designated as PCHPSAs (r = −.39, p < .01). For example, 15 counties in North Dakota (28.3%) versus one in New York (1.6%) were PCHPSAs in 1997, while New York recorded the highest DME payment rate approved by HCFA during that year and North Dakota recorded the lowest DME payment rate (see Table 1).
A relationship of similar strength was found between states' average hospital DME costs and the numbers of primary care physicians per 1,000 people in 1998 (r = .38, p = .01). New York had both the highest average DME cost and the highest number of primary care doctors per 1,000 people of any state. Mississippi and Idaho had the least favorable primary care physician-to-population ratios among the states, but they were excluded from the reported correlation because they did not have at least two teaching hospitals, required for comparisons and study inclusion.
Earlier studies have revealed general variation in DME payment rates.7,8 Intrastate differences have also been examined.1 The results of our analyses are consistent with prior findings, but we document more completely the magnitude of variations within individual states and throughout the country. This pattern, particularly its consistency in smaller states and the District of Columbia, supports the conclusion that this phenomenon is not attributable to differences in labor and capital costs across market areas. The congruence of these findings with those from other research, and validating the profile of DME costs for FY97 with those for FY96, suggests these results are not simply a single-year statistical aberration. Rather, they indicate a need to review the other factors we used in the formula (e.g., the 1984 calculations that remain the annually adjusted basis for Medicare payments to teaching hospitals). Such a review is timely because the Medical Payment Advisory Committee is pondering the effect of national standardization in its attempt to reduce disparities. While the range of payment rates may be diminished incrementally by subtle revisions of the present policy, any strategy that preserves application of this questionable base is unlikely to promote real equity throughout the system or provide incentives for the training of residents in states that are most in need of additional primary medical care resources.
The GME payment arrangements addressed in this analysis relate only to the Medicare program's involvement as administered through HCFA. Subsidies for GME support from Medicaid have been substantial in some states but very modest in others,11 and there has been little commitment to medical training from private insurers. More recently, with the enrollment of Medicare beneficiaries in capitated arrangements, the base for GME payments has eroded due to managed care plans' diversion of patients to less expensive non-teaching hospitals.12 While Medicare has long shouldered the vast majority of the burden for funding GME in this country, the time appears opportune to consider again an all-payer system of the kind endorsed by the Institute of Medicine, the Council on Graduate Medical Education, the Association of American Medical Colleges, the American Medical Association, and the Commonwealth Fund. Establishing explicit funding streams for residents' training that are supported by other payers as well as Medicare might permit rationalizing payments to teaching institutions and linking them to policies that ensure a well-prepared and balanced medical workforce that is appropriate to the needs of the American people.
Prior to enactment of the 1997 Balanced Budget Amendment, there were powerful financial incentives for teaching hospitals to expand the GME programs they sponsored.13 However, in this study, institutional behavior was found not to be strongly correlated with DME costs. Change in the number of primary care residents from one year to the next was not significantly influenced by larger Medicare payment rates, and the number of residents being trained was not mediated by the proportion of inpatient days attributable to Medicare beneficiaries, although this is a substantial factor in determining both DME and IME payments to facilities. Nor was there a major preference among teaching hospitals for primary care versus other residents, even though training the former earns somewhat greater funding. Certainly, this is not to argue that hospitals do not seek financial advantage in their GME arrangements with HCFA. In fact, other ways exist by which teaching institutions can reduce costs and increase revenues associated with their residency programs.13 Many factors, such as accreditation standards and local opportunities and constraints, may influence their decisions, but it is to their credit that teaching hospitals do not seem driven directly to optimize revenues.
The failure of teaching hospitals to respond to increases in public policy directives concerning the primary care compositions of their residents' specialty mixes raises questions how best to create incentives for producing needed physicians. There is broad consensus that the U.S. physician workforce needs reform. The ratio of primary care to specialty doctors in this country is lower than it is in virtually any other developed nation.14 One can reasonably wonder what it would take to shift the ratio of primary care to specialist residents. Would it be prudent to fund the training of primary care residents at a rate 60% rather than 6% greater than that funded for other residents in order to achieve a clearly important policy objective? Such a large differential may be necessary for the present situation, in which much training continues in care settings where specialty residents may be inherently more valuable to the sponsoring organization than are primary care faculty and residents in terms of services and associated revenues.
The role of teaching hospitals in attaining health care policy objectives is important, but it is one of a constellation of factors. For example, developing an appropriately-sized primary care physician workforce in the United States must begin early in the medical education pipeline. We know that strong, creative primary care curricula and initiatives are necessary to further cultivate medical students' enthusiasm. Schools' admission policies that are sensitive to the primary care interests of applicants can also have an effect.
In addition to imbalances in medical specialties, the geographic maldistribution of doctors remains a problem in this country. Our analyses demonstrated that DME funding rates do not appear to provide an incentive for training residents in states where inaccessibility to primary health care is widespread. Future funding for the National Health Service Corps and Title VII programs, which address medical underservice, may be in jeopardy. Yet, at the same time that these successful but relatively small initiatives struggle for support, large DME expenditures are being made with little regard for their impact on long-standing unresolved problems of access to care. Our findings suggest a contradiction in policy. Despite the well-recognized tendency of physicians to establish practices in the states in which they are trained, we found that there is actually a financial incentive to train residents in areas in which there is relatively less need for their services.
Skepticism has developed concerning our ability to affect the physician labor force through GME policy using current approaches. A system characterized by irrational variation and irregular participation does not inspire confidence as a medium for change. Teaching hospitals and residencies might be more responsive to policies that seem fair and promote meaningful payment variations for obvious and compelling purposes than to those that appear arbitrary or reward program inefficiencies from decades long gone.