Teaching Hospital Costs

Policy Perspectives Teaching Hospital Costs Implications for Academic Missions in a Competitive Market Robert Mechanic, MBA; Kevin Coleman; Allen Dobson, PhD Context. As the managed care environment demands lower prices and a greater focus on primary care, the high cost of teaching hospitals may adversely affect their ability to carry out academic missions. Objective. To develop a national estimate of total inpatient hospital costs related to graduate medical education (GME). Design. Using Medicare cost report data for fiscal year 1993, we developed a series of regression models to analyze the relationship between inpatient hospital costs per case and explanatory variables, such as case mix, wage levels, local market characteristics, and teaching intensity (the ratio of interns and residents to beds). Setting and Participants. A total of 4764 nonfederal, general acute care hospitals, including 1014 teaching hospitals. Major Outcome Measures. Actual direct GME hospital costs and estimated indirect GME-related hospital costs based on the statistical relationship between teaching intensity and inpatient costs per case. Results. In 1993, academic medical center (AMC) costs per case were 82.9% higher than those for urban nonteaching hospitals (actual cost per case, $9901 vs $5412, respectively). Non-AMC teaching hospital costs per case were 22.5% higher than those for nonteaching hospitals (actual cost per differences in case, $6630 vs $5412, respectively). After adjustment for case mix, wage levels, and direct GME costs, AMCs were 44% more expensive and other teaching hospitals were 14% more costly than nonteaching hospitals. The majority of this difference is explained by teaching intensity. Total estimated US direct and indirect GME-related costs were between $18.1 billion and $22.8 billion in 1997. These estimates include some indirect costs, not directly educational in nature, related to clinical research activities and specialized service capacity. Conclusions. The cost of teaching hospitals relative to their nonteaching counterparts justifies concern about the potential financial impact of competitive markets on academic missions. The 1997 GME-related cost estimates provide a starting point as public funding mechanisms for academic missions are debated. The efficiency of residency programs, their consistency with national health workforce needs, financial benefits provided to teaching hospitals, and ability of AMCs to maintain higher payment rates are also important considerations in determining future levels of public financial support. JAMA. 1998;280:1015-1019 From The Lewin Group, Fairfax, Va. Mr Mechanic is now with the Massachusetts Hospital Association, Burlington. Reprints: Robert Mechanic, MBA, Massachusetts Hospital Association, 5 New England Executive Park, Burlington, MA 01803 (e-mail: rmechanic@mhalink.org). THE GROWTH of managed care has created an environment in which hospitals face demands for lower prices and a greater focus on primary care. 1 Teaching hospitals and academic medical centers (AMCs) may be especially vulnerable under these conditions because their patient care costs are generally higher than those of nonteaching community hospitals, in large part because of their roles in graduate medical education (GME) and clinical research. Teaching hospital costs are also affected by their subspecialty orientation and highly specialized service capacity. Teaching hospital costs historically have been financed through higher charges to insured patients and by special payments, such as Medicare s indirect medical education (IME) and direct GME adjustments. Many teaching hospitals command high payment rates to support their cost structure because of their reputations for providing high-quality health care and their specialized health service capacity. However, in markets where price is an increasingly important determinant of which providers get health insurance contracts (and patients), teaching hospitals may finditmoredifficulttocompeteeffectivelywhilemaintainingtheir current commitments to education and research. Federal and state policymakers continue to debate new dedicated funding sources to support teaching hospital missions. The US Congress has considered proposals for GME trust funds during the past 3 legislative sessions, and states such as Minnesota, California, and Utah are actively exploring new GME-funding mechanisms. 2 New York State recently established a $1.4 billion GME pool, financed in part by assessments on private health insurance plans. 3 As federal policymakers consider support for teaching hospital activities, 4 major questions must be addressed. Should the current Medicare-only approach to paying for GME and related activities be changed to recognize costs incurred on behalf of all third-party payers? What level of public funding is appropriate to support academic activities? How should these activities be financed? How should funds be allocated across institutions and programs? Determining the appropriate level of funding for academic missions requires a clear understanding of hospital costs and the factors that cause them to vary. Prior analyses of the indirect costs of teaching hospitals have focused specifically on Medicare. 4-6 However, as broader funding mechanisms for financing academic missions such as all-payer trust funds are debated, costs incurred on behalf of the entire patient population must be considered. This article applies a series of hospital cost models to analyze GME-related costs for all payers and all patients. 7 The models estimate the effect of such factors as case mix, wage levels, and teaching intensity on hospital costs. We first analyzed the differences in inpatient cost per case in teaching and nonteaching hospitals to assess potential price pressures JAMA, September 16, 1998 Vol 280, No. 11 Policy Perspectives 1015

Table 1. Regression Model Variables* Total Hospital Inpatient Cost per Case Minus Direct Medical Education Costs Teaching intensity Ratio of interns and residents (net facility) to facility beds Payment Explanatory Case mix Medicare case-mix index Ratio of Medicare outlier payments to total Medicare inpatient payments Special care days (intensive care unit, coronary care unit, and other special care days) as a percentage Newborn days (excluding neonatal intensive and intermediate care days) as a percentage of total days Does the hospital provide long-term care services Long-term care days as a percentage Patient mix Medicare days as a percentage Medicaid days as a percentage Input prices Is hospital located in a metropolitan statistical area Medicare geographic hospital wage index Hospital characteristics Occupancy rate Ratio of average No. of employees (less net residents) to facility beds Total facility beds Demographic characteristics 1990 County population Families with incomes of $10 000 Families with incomes of $50 000 Persons aged 18 y Persons aged 65 y Market characteristics Health maintenance organization penetration in metropolitan statistical area No. of general hospital beds per 1000 persons, 1993, county State dummy variable in competitive markets. We then estimated the total GMErelated costs for all US teaching hospitals in 1993 and 1997 as a starting point for considering new funding mechanisms. We conclude by discussing issues for consideration as GME funding proposals are developed. METHODS Defining Direct and Indirect GME-Related Costs The presence of GME and related academic activities creates both direct and indirect costs. Direct GME costs include resident salaries, fringe benefits, the cost of faculty supervision, and associated overhead expenses, which are measured by using cost-accounting techniques. This analysis uses direct GME costs reported in hospitals 1993-1994 Medicare cost reports. Indirect GME-related costs are embedded in patient care and cannot be measured directly. They result from factors such as the higher severity of illnesses of patients and the use of more complex procedures in teaching hospitals, as well as reductions in clinical productivity associated with academic activities, such as increased use of diagnostic testing by residents. 8 Estimated with staffing State models only with staffing State models only *Model uses log form of selected variables. Data are from Interstudy Competitive Edge, Medicare Hospital Cost Report Information System (fiscal year 1993), American Hospital Association Hospital Annual Survey, and Area Resource File. indirect costs may also reflect activities that are not directly educational in nature, such as clinical research and maintenance of 24-hour standby capacity for services such as trauma care, where volume is difficult to predict. The Medicare prospective payment system and some state Medicaid programs recognize the higher patient care costs of teaching hospitals through an IME adjustment. However, its purpose is much broader than education. As noted by the House Ways and Means Committee: This adjustment is provided in light of doubts...about the ability of the DRG [diagnosis related group]-case classification system to account fully for factors such as severity of illness of patients requiring the specialized services and treatment programs provided by teaching institutions and the additional costs associated with the teaching of residents...the adjustment for indirect medical education is only a proxy to account for a number of factors which may legitimately increase costs in teaching hospitals. 9 This article focuses on measuring the inpatient costs of hospital-based academic activities including GME. The principal data used in this analysis are from prospective payment system year 10 (fiscal year 1993) Medicare cost reports. These data are supplemented with utilization measures from the American Hospital Association 1993 annual survey and county demographic and economic data from the Public Health Service area resource file. Our analytic database contains 4764 nonfederal, general acute care hospitals including 1014 teaching hospitals (reporting at least 1 resident). A double-log, weighted least squares regression model was used to analyze the cost of GME-related activities based on observed relationships between teaching intensity (measured by the intern- and resident-to-bed [IRB] ratio) and inpatient costs per case after controlling for other explanatory variables. 7 Double-log models have been used by the Medicare Payment Assessment Commission (formerly ProPAC), the Health Care Financing Administration, and others since the early 1980s to evaluate and establish GME payment policy for Medicare. 10 This body of work provides a standard for validating our all-payer estimates. Other econometric techniques have been used to evaluate hospital costs. 11-14 However, these alternative approaches were considered less suitable for estimating GME-related costs. Our models estimate only inpatient GME costs because there are no data to measure teaching intensity in outpatient settings. Our final cost estimates were adjusted to account for nonfederal, general acute care teaching hospitals that were not in the database. Our dependent variable was 1993 hospital inpatient costs per case minus reported direct GME costs. 4-6 The model controlled for statistically significant independent variables including casemix and severity of illness measures; area wage levels; institutional characteristics, such as number of beds, staffing ratios, and occupancy rates; local demographic characteristics, such as age distribution and per capita income; and local market factors, such as health maintenance organization penetration (Table 1). Proxy measures for case mix and severity are particularly important because no national all-payer case-mix data exist. To supplement the Medicare case-mix index, we included the percentage of Medicare payments for discharges designated as cost or length-of-stay outliers. Variables were also added to reflect non-medicare case mix including the ratios of (1) intensive care, coronary care, and other special care days to total inpatient days; (2) newborn days to non-medicare days; and (3) long-term care days to total inpatient days. 1016 JAMA, September 16, 1998 Vol 280, No. 11 Policy Perspectives

Conceptual Approach Traditionally, 2 principal approaches have been used to estimateindirectteachinghospitalcostsformedicare. 7 Explanatory models attempt to include all observable statistically significant variables associated with differences in hospital costs per case. Explanatory models reflect a narrow view of GMErelated costs and have lower estimated teaching coefficients than models with fewer explanatory variables, therefore producing lower GME cost estimates. Explanatory models are often favored by those who believe that a narrower view of teaching costs is appropriate for determining public funding levels and that payment for residual costs should be left to private markets. Payment models, in contrast, include only factors that are explicitly included in hospital reimbursement systems and, therefore, assign a higher value to teaching intensity than do explanatory models. Payment models result in higher cost estimates and are generally preferred by those who favor broader public support for GME. There is no consensus about which approach is more appropriate. Use of payment models may reflect a belief that reimbursement systems should lean toward leveling the playing field. Use of explanatory models may reflect a desire to encourage teaching hospitals to be more efficient. Traditionally, these models have been estimated on a national basis, but we have also developed an approach that controls for state differences. National models assume that total costs per case are affected by hospital and market characteristics but not by state location. Under this approach, national coefficients are applied to the characteristics of individual teaching institutions to estimate GME costs. State models use the national database but differ from national models in 2 ways. First, state models include dummy variables that allow costs per case to vary across states. Second, by interacting state dummy variables with the IRB ratio, the share of total costs associated with differences in teaching intensity can vary for hospitals located in different states. Payment systems based on state models would distribute hospital paymentsmuchdifferentlyfromthosebasedonnationalmodels. Some of our models also include staffing variables. Teaching hospitals have more full-time staff per bed than do community hospitals. Explicitly accounting for staffing differences in the model results in lower estimated GME-related costs. Policymakers who believe that teaching hospital staffing patterns reflect operating inefficiencies might choose to include staffing in the model to remove its effect from GME cost estimates. In contrast, if higher staffing ratios are considered intrinsic to academic missions, policymakers may want to exclude staffing ratios from the model. RESULTS We estimated 6 different models for this analysis. Teaching intensity, case-mix, and wage coefficients are all significant at the 5% level and explain virtually all of the predicted cost differences between teaching and nonteaching hospitals. The adjusted R 2 statistics indicate that explanatory models explain more of the variation in costs per case than payment models and that state models explain more than national models (Table 2). Cost Differences in Teaching and Nonteaching Hospitals To illustrate potential financial pressures on teaching hospitals in competitive markets, we analyzed their actual costs Table 2. Teaching Intensity Coefficient and R 2 for Regression * IRB Ratio Adjusted R 2 National models Explanatory model (SE) 0.515 (0.0329) 0.658 Payment model with staffing (SE) 0.590 (0.0317) 0.642 Payment model without staffing (SE) 0.665 (0.0311) 0.635 State-specific models Explanatory model (SE) Range 0.725 Payment model with staffing (SE) Range 0.712 Payment model without staffing (SE) Range 0.705 *Range indicates that the intern- and resident-to-bed (IRB) coefficients are unique for each state. Data are from The Lewin Group estimates using prospective payment system year 10 Medicare cost report data. compared with those of nonteaching community hospitals. Teaching hospitals are designated as AMCs based on the Association of American Medical College s definition of an integrated AMC. 15 These 125 hospitals represent the nation s largest medical education and research programs. In 1993, actual AMC hospital costs per case were 83% higher than those of urban nonteaching hospitals with 100 or more beds (Table 3). Other teaching hospital costs per case were about 22.5% higher than nonteaching institutions. After adjusting for patient case mix and area wages, AMCs were still 63% more costly than nonteaching institutions, while other teaching hospitals were 20% more expensive than nonteaching hospitals. These percentages suggest the price premiums that teaching hospitals would need to command if no other subsidies were available. Direct GME accounted for about one third of the observed cost difference between teaching and nonteaching hospitals. Subtracting direct GME from costs per case reduces the difference to 44% for AMCs and 14% for other teaching hospitals. Relationship Between Teaching Intensity and Hospital Costs Our all-payer models indicate that each 0.1 increase in the IRB ratio (teaching intensity) is associated with increases in inpatient cost per case of between 5.2% and 6.7%. This is lower than Medicare s historical policy, which increased payments by about 7.7% for each 0.1 increase in a hospital s IRB ratio. The Balanced Budget Act of 1997 reduced Medicare s adjustment to 7.0% in fiscal year 1998 and 5.5% by fiscal year 2001. The Medicare Payment Assessment Commission s fiscal year 1995 estimate of the Medicare IME adjustment is 4.5%. 16 The commission s model is Medicare only and uses fixed case-mix and wage index coefficients, accounting for differences relative to our estimates. In the models that allow state variation, the majority of states have teaching coefficients between 3.0% and 10.0%. This variation partly reflects state cost differences. State-specific estimates are not presented because the model is unstable with small state sample sizes. Nevertheless, more than half of the nation s medical residents are concentrated in 7 states. 15 All but 1 of these states has a teaching coefficient above the national estimate. Estimated Total US GME-Related Costs in 1993 and 1997 Themodelwasusedtoestimateindirect1993inpatientGMErelated costs for nonfederal, general acute care hospitals. After adding reported 1993 direct GME expenses of $6.5 billion, total GME cost estimates ranged between $15.4 billion and $19.2 billion(table4). The1997estimatesassume2.8% annualgrowth in direct GME costs based on the projected consumer price JAMA, September 16, 1998 Vol 280, No. 11 Policy Perspectives 1017

Table 3. Comparison in Total Cost Per Case in Teaching and Nonteaching Hospitals, Fiscal Year 1993 (Urban Hospitals With More Than 100 Beds)* Nonteaching, Urban Community Hospitals (100 Beds) Table 4. Estimated US GME-Related Costs in 1993 and 1997 (Billions of Dollars)* Other Teaching Hospitals 1993 1997 Academic Medical Centers No. of hospitals in sample 1263 812 104 Actual cost per case, $ 5412 6630 9901 Percentage above community hospital average 0.0 22.5 82.9 Adjusted cost per case (with direct GME), $ 5393 6464 8769 Percentage above community hospital average 0.0 19.9 62.6 Adjusted cost per case (minus direct GME), $ 5393 6142 7754 Percentage above community hospital average 0.0 13.9 43.8 *GME indicates graduate medical education. Wage index and case-mix adjustments are based on our national explanatory model. Data are from The Lewin Group analysis of prospective payment system year 10 Medicare cost report data. Total, $ IME, $ Direct GME, $ Total, $ IME, $ Direct GME, $ National explanatory 15.4 8.9 6.5 18.1 10.9 7.2 National payment (with staffing) 16.5 10.0 6.5 19.5 12.3 7.2 National payment (no staffing) 17.7 11.2 6.5 20.9 13.7 7.2 State-specific explanatory 17.2 10.7 6.5 20.4 13.1 7.2 State-specific payment (with staffing) 18.1 11.6 6.5 21.5 14.2 7.2 State-specific payment (no staffing) 19.2 12.7 6.5 22.8 15.5 7.2 *GME indicates graduate medical education; IME, indirect medical education. Data are from Iglehart, 1 Salsberg, 2 Healthcare Association of New York, 3 Pettengill et al, 4 Lave, 5 and The Lewin Group estimates using prospective payment system year 10 Medicare cost report data. index and 5.2% annual inflation in IME-related costs based on actual and projected changes in Medicare hospital inpatient payments. The 1997 estimates range from $18.1 billion to $22.8 billion, including direct GME costs of $7.2 billion. The national explanatory model provides the lowest estimate of GME-related costs. Explanatory models that allow GME coefficients to vary across states produce estimates that are 12% higher than the national model. Switching from an explanatory to a payment model increases the state model results by an additional 5%. Finally, excluding the staffing variable from the state payment increases the predicted indirect teaching hospital costs by 7%. Considerations for Public Policy The high cost of teaching hospitals justifies concern about the effect of price-competitive markets on educational and research missions. The effect of managed care on AMC financeshasbeenverymarketdependent. 17 However,datafrom 9 states indicate that patients in health maintenance organizations are less likely than other types of patients to receive care in teaching hospitals. 18 Intensified competition (or its perceived threat) continues to create pressure for AMCs to control costs. Concern over the effect this may have on AMC social missions has led some policymakers to suggest new allpayer funding mechanisms. 19 Our cost estimates provide a starting point for considering all-payer funding levels. The estimates are based on current costs of residency training. But some analysts question whether current methods of financing GME have produced an appropriate workforce mix or whether they have done so efficiently. 20 Our estimates also do not account for financial benefits of residency programs that may offset some GME-related costs for sponsoring institutions. Therefore, the figures presented herein may not reflect what the public or the US Congress is willing to support. Financial Benefits of Residency Programs A publicly administered, all-payer GME trust fund would require new revenues or the transfer of resources from other programs. Proposals such as the administration s Health Security Act of 1994 proposed to finance new GME support through a dedicated tax on health insurance premiums. One way to limit the need for new revenue is to reduce funding to account for the current financial benefits that residency programs provide hospitals. In 1996, Medicare issued roughly $4.3 billion in IME payments and $2.4 billion in direct GME payments to approximately 1100 teaching hospitals. 21 According to 1 estimate, state Medicaid programs paid hospitals $1.0 billion for IME and $1.4 billion for direct GME in 1995. 22 Other potential financial benefits are difficult to measure. Among those are the ability to negotiate higher reimbursement because of perceived teaching hospital quality, the reduction of staffing costs through use of residents, and the ability to earn revenues from services provided by residents. Many private payers implicitly support teaching hospital missionswithrateshigherthanwhattheypaycommunityhospitals. This differential is reported to be shrinking in many markets. 23,24 The degree to which teaching hospitals can continue to command higher payments than nonteaching hospitals is unclear. Hospital and faculty physician billings for services provided by residents are another source of revenue for some AMCs, but future revenue may also be limited by Medicare s recent ruling that a teaching physician must be present for a key portion of the service for which payment is sought. Does the US Medical Education System Meet the Country s Needs? The public s investment in teaching hospitals should ideally result in a physician workforce consistent with community needs. One could argue that GME funding mechanisms should be based on the cost of efficient training for an appropriate mix of physicians, but there is substantial room for disagreement about what these terms mean and the costs they represent. Pressure to reduce the size of medical residency programs and shift the focus of training from specialty to primary care is growing. 25-28 Such changes will clearly alter teaching costs but will also create new analytic challenges such as measuring the indirect cost of ambulatory GME-related activities. 29 1018 JAMA, September 16, 1998 Vol 280, No. 11 Policy Perspectives

Informing Future Policy Decisions This study has a number of limitations that must be considered in future research and policy development. First, as in other studies, our model has focused on costs associated with teaching intensity. This variable is thought to capture costs related to a range of AMC missions beyond GME. The difficulty of quantifying how distinct teaching hospital activities contribute to cost variation is frustrating, since the concept of supporting specific academic missions is much more appealing than just paying for higher teaching hospital costs. Future research should concentrate on developing variables that reflect clinical research intensity, undergraduate medical education, standby capacity, and indigent care to improve the usefulness of these models in designing targeted payment mechanisms. A second limitation is the lack of a national hospital database that can be used to calculate an all-payer case-mix index. with more precise case-mix coefficients could be developed for a limited number of states using hospital discharge data collected by the federal Agency for Health Care Policy and Research. However, our state models suggest substantial variation in teaching hospital cost structures. These may reflect differences in program size, program efficiency, specialty mix, the degree to which institutions attract sick patients as national or regional centers of excellence, and related activities, such as clinical research. Thus, models based on selected states may not be generalizable for national policy. Even if complete data were available, it is unclear whether developing national policy based on state models is desirable. National payment systems, such as Medicare, create efficiency incentives for programs in high-cost states. Although one could argue that national systems are less equitable for institutions in highly concentrated teaching states, changing payment policy from the current national approach to a system based on unique state characteristics would substantially redistribute public GME funding. A third limitation is the model s exclusive focus on inpatient costs. Outpatient care now makes up about 30% of hospital revenue. 30 Although there is a strong call for increased medical training in outpatient settings, GME-related cost models will remain focused on inpatient care until good metrics of outpatient teaching intensity can be developed. A final limitation is the lack of data for Veterans Affairs and children s hospitals, both major sites for GME and research. To the extent that these institutions are excluded, our models may understate national hospital GME costs. COMMENT Debate over funding the missions of teaching hospitals will place them under greater public scrutiny. While internal crosssubsidies are allocated at the discretion of individual institutions, publicly administered payments would be accompanied by more explicit demands for accountability. Taxpayers, employers, health plans, or health care providers tapped for financial support will demand a seat at the table in setting training and research priorities. As the recent experience of New York State demonstrates, explicit GME funding pools face intense public scrutiny, particularly when the costs are clear to those footing the bill. Although new funding mechanisms may provide financial relief, they will also create pressure for teaching hospitals to demonstrate the value of their unique contributions to education, research, and clinical service. This work was supported by grant 970 440 from the Commonwealth Fund, New York, NY. Earlier versions of this work were supported by research grant Contract No. 282-92-0041 from the US Public Health Service, Rockville, Md, and by the Association of American Medical Colleges, Washington, DC. References 1. Iglehart JK. Rapid changes for academic medical centers. N Engl J Med. 1995;332:6. 2. Salsberg E. State Strategies for Financing Graduate Medical Education. New York, NY: United Hospital Fund; 1997. 3. Healthcare Association of New York. The New York Health Care Reform Act of 1996: A Summary. Albany: Healthcare Association of New York; 1996. 4. Pettengill J, Vertrees J. Reliability and validity in hospital case-mix measurement. Health Care Financing Rev. 1982;4:101-128. 5. Lave J. The Medicare Adjustment for Indirect Costs of Medical Education: Historical Development and Current Status. Washington, DC: Association of American Medical Colleges; 1985. 6. Thorpe KE. The use of regression analysis to determine hospital payment: the case of Medicare s indirect teaching adjustment. Inquiry. 1988;25:219-231. 7. Dobson A, Coleman K, Mechanic RE. Analysis of Teaching Hospital Costs. Washington, DC: Association of American Medical Colleges; 1994. 8. Anderson G, Lave J. Financing graduate medical education using multiple regression to set payment rates. Inquiry. 1986;23:191 199. 9. House Ways and Means Committee Report No. 98-25, 98th Cong, March 4, 1983. 10. Prospective Payment Assessment Commission. Report and Recommendations March 1995. 11. Grannemann TW, Brown RS, Pauly MV. Estimating hospital costs: a multiple-output analysis. J Health Economics. 1986;5:107-128. 12. Breyer F. The specification of a hospital cost function: a comment on the recent literature. J Health Economics. 1987;6:147-158. 13. Vita MG. Exploring hospital production relationships with flexible functional forms. J Health Economics. 1990;9:1-22. 14. Dor A, Farley DE. Payment source and the cost of hospital care. J Health Economics. 1996;15:1-22. 15. Association of American Medical Colleges. AAMC Data Book: Statistical Information Related to Medical Education. Washington, DC: Association of American Medical Colleges; 1998. 16. Prospective Payment Assessment Commission. Report and Recommendations March 1996. 17. AHCs on edge as congress meets, market pressures build. Medicine and Health Perspectives. January 13, 1997:1-4. 18. ReuterJ, GaskinD. Academichealthcentersincompetitivemarkets.Health Aff (Millwood). 1997;16:242-252. 19. The Commonwealth Fund Task Force on Academic Health Centers. Leveling the Playing Field: Financing the Missions of Academic Health Centers. New York, NY: Commonwealth Fund; 1997. 20. US Congressional Budget Office. Medicare and Graduate Medical Education. Washington, DC: US Congressional Budget Office; 1995. 21. US Congressional Budget Office. January 1997 Baseline Estimates.Washington, DC: US Congressional Budget Office; 1997. 22. National Association of Children s Hospitals. Calculations based on: Plumb DN, Henderson T. Medicaid Funding of Graduate Medical Education: A Survey of the States. Washington, DC: George Washington University Press; 1995. 23. Mechanic RE, Dobson A. The impact of managed care on clinical research: a preliminary investigation. Health Aff (Millwood). 1996;15:72-89. 24. Blumenthal D, Meyer GS. Academic health centers in a changing environment. Health Aff (Millwood). 1996;15:201-215. 25. Pew Health Professions Commission. Critical Challenges: Revitalizing the Health Professions for the Twenty-first Century. San Francisco: University of California, San Francisco, Center for the Health Professions; 1995. 26. Institute of Medicine. The Nation s Workforce: Options for Balancing Supply and Requirements. Washington, DC: National Academy Press; 1996. 27. Council on Graduate Medical Education. Patient Care, Physician Supply and Requirements: Testing COGME Recommendations, Eighth Report. Washington, DC: US Dept of Health and Human Services, Public Health Service, Council on Graduate Medical Education; July 1996. 28. Prospective Payment Assessment Commission. Report and Recommendations March 1997. 29. Association of American Medical Colleges. Medical Education in Ambulatory Care Settings: Annotated Bibliography of Selected Works. Washington, DC: Association of American Medical Colleges; 1994. 30. American Hospital Association. Hospital Statistics: 1998 Edition. Chicago, Ill: American Hospital Association; 1998. JAMA, September 16, 1998 Vol 280, No. 11 Policy Perspectives 1019