|
End-Of-Life Care At Academic Medical Centers: Implications For Future Workforce Requirements
The expansion of U.S. physician workforce training has been justified on the basis of population growth, technological innovation, and economic expansion. Our analyses found threefold differences in physician full-time-equivalent (FTE) inputs for Medicare cohorts cared for at academic medical centers (AMCs); AMC inputs were highly correlated with the number of physician FTEs per Medicare beneficiary in AMC regions. Given the apparent inefficiency of current physician practices, the supply pipeline is sufficient to meet future needs through 2020, with adoption of the workforce deployment patterns now seen among AMCs and regions dominated by large group practices.
THE ASSOCIATION OF AMERICAN MEDICAL COLLEGES (AAMC) recently joined forces with a growing number of organizations that are calling for an increase in the number of physicians to address a perceived future workforce shortage.1 Advocates for expanded physician training point to population growth, particularly among the elderly; increases in age-specific rates of physician utilization; and economic expansion. In response to these trends, the AAMC advocates a 15 percent increase in total U.S. medical school enrollment (leading to an additional 3,000 graduates annually) over the next decade and a removal of Medicare funding limits for graduate medical education. But how many more physicians are required to improve the future health and well-being of the population? Previous research has shown a complex relationship between physician supply, medical care use, and health outcomes. Physician supply and utilization rates, for example, are known to vary across regions, and these variations cannot be explained by differences in population age structure or health status.2 Furthermore, both cross-sectional and cohort studies have failed to show an enduring link between greater supply and improved population health.3 Indeed, regions and states with more medical specialists and general internists appear to have lower quality of care as measured by mortality and common performance measures endorsed by the National Committee for Quality Assurance (NCQA).4 In light of these facts, we argue in this paper that regions and academic medical centers (AMCs) with low physician labor input rates should be viewed as useful benchmarks of overall efficiency in the allocation of physician labor.5 These benchmarks, in turn, can shed light on the increment in the current U.S. supply of physicians required to meet the needs of tomorrow’s elderly populations.
Although there is value in measuring variation in physician labor inputs across entire populations, such analyses could be criticized for possible differences in health status and adequacy of medical care. Instead, we first examined physician inputs to a more restricted but standardized population: Medicare patients living in Hospital Referral Regions (HRRs) and those who receive most of their care from AMCs, arguably sites providing the highest quality of care. For AMCs, we report physician full-time-equivalents (FTEs) during patients’ last six months of life, which greatly reduces the possibility that the differences in physician deployment relate to differences in illness or inadequacies in the quality of care. To establish benchmarks for efficient physician workforce allocation to this elderly population, we also examined the entire eligible Medicare populations living in HRRs. Medicare study populations. We studied patients enrolled in both Parts A and B of fee-for-service (FFS) Medicare. The regional comparisons are among the 306 HRRs featured in the Dartmouth Atlas of Health Care for the resident population during 2000.6 The population served by AMCs consists of beneficiaries who died during 1999–2001 and who (1) received the majority of their hospitalized days during their last two years in an AMC and (2) were admitted for one or more of twelve chronic disease conditions that have a high probability of hospital death.7 National Medicare claims files that account for physician services are limited to a 20 percent beneficiary sample; our study cohort consisted of this sample. We identified 108 major AMCs integrated with medical schools; of these, our sample was restricted to the 79 AMCs with at least 100 chronically ill decedents during the study period.8 The percentage of hospitalized days in the last twenty-four months of life that occurred at the designated AMCs for these seventy-nine hospitals ranged from 77 percent to 94 percent of medical admissions, with a median of 87 percent. For each of these seventy-nine patient cohorts, we measured the total Medicare physician use over the last twenty-four months of life. Measuring physician labor inputs. Physician inputs to populations who receive most of their inpatient care at a given hospital cannot be measured using traditional physician supply data such as the American Medical Association (AMA) Physician Masterfile.9 To develop such estimates, we used Medicare claims from 20 percent Part B and Outpatient Files to identify physician services provided to each region or each cohort assigned to a given hospital.10 Current Procedural Terminology (CPT) codes for each claim were linked to Work Relative Value Units (WRVUs), a standardized method of estimating physician labor for each medical service. We summed WRVUs by specialty for each region or hospital and then divided the totals by the specialty-specific WRVUs per FTE physician as measured in two large surveys of medical clinics.11 Physician categories were primary care, medical specialists, surgical specialists, and hospital-based specialists.12 The total was the sum of these categories in addition to psychiatry. Anesthesiology reimbursements are not based on WRVUs, so their labor could not be included. Standardized FTEs. It would not be surprising if the mix of patients with terminal chronic conditions differed across hospitals and regions. To provide hospital-specific rates that were standardized for case-mix, estimates for FTEs per 1,000 patients were adjusted using a fixed-effects linear regression model. A separate model was run for each specialty and aggregation of specialties (primary care and medical, surgical, and hospital-based specialties). The dependent variable was specialty-specific FTEs per decedent. Independent variables included the patient’s age (67–69, 70–74, 75–79, 80–84, and 85–99), sex, race (nonblack and black), and a dummy variable for each of the twelve chronic conditions. To calculate standardized hospital-specific rates, we included a dummy variable for each of the hospitals. Regression coefficients were centered to the national rate to represent average FTEs per 1,000 or 10,000 beneficiaries by hospital or region.13 Regional rates were calculated for the sixty HRRs with one of the above AMCs and adjusted for age, sex, and race differences using the indirect method. End-of-life cohort members were excluded from the regional rates.
Labor inputs. The amount of physician labor used to care for chronically ill Medicare beneficiaries at the end of life differed widely among the AMCs in this study (Exhibit 1  ). The total number of standardized physician FTEs per 1,000 decedents varied from 6.0 to 28.3, with a median of 11.2 and an interquartile range (25th–75th ) of 9.3 to 13.7. A similar degree of variation was observed for primary care physicians (range, 1.5–8.8; median, 3.7; and interquartile range, 2.9–5.0) and medical specialists (range, 2.4–15.0; median, 4.9; and interquartile range, 4.0–6.3). During the last six months of life, surgical and hospital-based specialists contributed a relatively smaller proportion of the total physician labor (median, 1.0 and 1.1 each), but the variation of standardized FTEs per 1,000 beneficiaries was similarly high across the hospitals.14
Among the seventy-nine AMCs (Exhibit 2 ), the New York University (NYU) Medical Center Medicare cohort ranked highest in total labor input, with 28.3 standardized FTEs per 1,000 decedents—1.4 times more than the second-ranked Robert Wood Johnson University Hospital and 4.7 times more than the labor input to patients in the lowest-ranked cohort Medical College of Georgia. Although many of the seventy-nine AMCs in the lower quintiles of physician labor input were in smaller urban areas, there were notable exceptions. University of Cincinnati Hospital; University of California, San Francisco; and University of Minnesota are all in major urban areas and still have low rates of physician labor input (Exhibit 2).
NYU Medical Center patients also received the most labor input from medical specialists: 15.0 standardized FTEs per 1,000 decedents. They used 6.3 times more than the lowest-ranked hospital, Strong Memorial Hospital (2.4 FTEs); the low rate of medical specialists there is associated with a low rate of use of primary care physicians (3.8 FTEs) compared with NYU Medical Center (8.8 FTEs). Indeed, we found no evidence that AMCs substituted primary care physicians for medical specialists. In general, centers with high FTEs for medical specialists had similarly high FTEs for primary care (Pearson’s correlation, 0.50). Input rates were also correlated between specific medical specialists. For example, labor inputs for cardiologists correlated with those of hematologist/oncologists (Pearson’s correlation, 0.56), pulmonologists (0.64), and neurologists (0.70).
These results show that there are wide differences in physician effort in caring for cohorts of similar patients. The patterns of labor input during patients’ last six months of life were closely correlated with the overall level of effort toward the entire FFS Medicare population living in the same region (end-of-life cohort members were excluded from these regional rates) (Exhibit 3
Benchmarking physician requirements for Medicare patients. The variation among HRRs in physician inputs exhibited by Manhattan (NY) and Rochester (MN) provides an opportunity to examine the number of physicians "needed" to meet the 56 percent increase in the number of elderly Americans predicted for 2020.15 The first step is to calculate the number of physicians that would be required under the Manhattan or Rochester (Mayo Clinic) standard for resource allocation to populations served by FFS Medicare; the second is to estimate the additional supply required to meet the predicted increase in this cohort. We assumed that U.S. physician supply would continue to grow at current rates, leading to the 24 percent increase projected by the Council on Graduate Medical Education (COGME).16 Exhibit 4 provides these estimates. Simply to meet the Manhattan standard in 2000, the United States would need an additional 27,274 physicians just to care for the FFS Medicare population age sixty-five and older beyond the actual physician count of 104,834, a 26 percent difference. To meet the 56 percent increase predicted for 2020, the Manhattan standard identifies the need for 111,558 more physicians nationally to avoid a projected physician deficit of 44,162. Applying the Rochester standard to the nation’s elderly, the United States has an excess of physician input; it needs 30,163 fewer FTE inputs than were allocated in 2000. Indeed, the current rate of supply growth along with excess capacity is sufficient to accommodate the 56 percent increase predicted for 2020, with 49,917 physicians to spare.
Shortage or surplus? In remarkably few years, the opinion of the health policy community has changed from near-consensus that the United States has a physician surplus to one of impending physician shortage. Although a consensus may be emerging on the workforce requirements for 2020, our study demonstrates a lack of consensus, even among AMCs, on how today’s physician workforce should be deployed in the service of the Medicare population. Among Medicare patients with chronic illnesses, we found a 4.7-fold variation in the per capita number of standardized FTE physicians AMCs used in managing these patients during their last six months of life. When we examined the patterns of allocation among sixty HRRs, we found an even greater range of variation. Moreover, the variation among AMCs in their per decedent deployment of FTE physicians was highly correlated with the per capita FTE physician inputs in the region where the AMCs were located. These findings suggest that there is much "slack" in the physician labor force that could be applied to the growing and aging population. Additional needs on the horizon. Further arguments have been made that technological advances and economic expansion will require more physicians. It is hard to understand how differences in technology could explain the variations observed among AMCs. Indeed, the relatively low labor input associated with the Mayo Clinic suggests that higher levels of physician effort in other centers may be associated with a misapplication of medical technology. With respect to economic development, although it was not the primary focus of this paper, we found little correlation between the median household income of the AMC regions and their labor input rates.17 Accounting for geographic variation. Although this study is limited to differences in physician FTEs for the elderly, the magnitude of variation, two- to threefold, is similar to that observed in the regional per capita physician supply, when controlling for population differences. These differences, unexplained by any reason salient to the health and well-being of populations, dwarf the requirements in physicians attributed to temporal trends in population growth, physician utilization rates, or economic expansion. Indeed, the percentage differences in future supply and "requirements" or "needs" projected by COGME’s Sixteenth Report are less than the percentage differences in physician labor input noted between Massachusetts General Hospital in Boston and Strong Memorial Hospital in Rochester. A common explanation offered for geographic variation in resource allocation and use is differences in health status or in patients’ care preferences. For this reason, we have studied variation across cohorts with similar illness severity. Nor does it seem plausible that patients choose their hospitals or the regions they live in based on the amount of care they wish to receive in their last six months of life. Study limitations. Our study had several limitations. First, adjustments for illness using claims data miss many details important for prognosis that ordinarily require chart review. However, by studying labor inputs during the last six months of life, we narrowed possible differences in health status. Second, reliance on FFS claims omits the approximately 17 percent of beneficiaries enrolled in Medicare+ Choice (M+C, now Medicare Advantage) plans, and the percentage enrolled will vary across regions and AMCs. We expect that patients enrolled in such plans would generally experience fewer physician inputs than similarly ill patients in FFS. Third, we used surveys for estimating physician WRVUs per FTE that included a mixture of academic and nonacademic group practices; this would not have affected the relative differences we observed across centers and regions. We also note that we did not measure resource inputs of other domains, such as nonphysician labor or hospice care, and this limits the explanation of the differences observed. Although we observed that AMCs with high ratios of specialist FTEs generally also had high ratios of primary care FTEs, we do not know if this relationship holds true for both general internists and family physicians. Furthermore, the professional roles of primary care physicians at AMCs might be different from those serving broader populations.18 Policy implications. Calls for increasing the supply of physicians should not ignore the evidence that health outcomes among Medicare populations are not likely to be improved by increasing physician labor input beyond that now seen among low-rate AMCs or HRRs. Similar findings for other patient populations have been amply documented.19 Although acceptable medical care may be consistent with very different workforce levels, the primary difference between high- and low-supply areas appears to be the volume and cost of care delivered, with an apparent, albeit small, negative effect of high physician supply on the quality and outcomes of care and patients’ perceptions of barriers to access. From the point of view of gains in population health, the low-resource-input regions appear to be more efficient in improving health. We note that several low-input regions are served by well-known AMCs with strong reputations for high-quality care. We also note that in several of the low-input regions, much of the care is provided by large, multispecialty group practices (for example, the Mayo Clinic) or integrated delivery systems (for example, Intermountain Health Care). The association between large group practices and efficient use of physician workforce has been documented for prepaid group practices such as Kaiser Permanente.20 Our study suggests that even in FFS environments, group practices use fewer physicians per capita than is true in small-group or solo practices (the dominant modes in most U.S. regions). The "need" for physicians in 2020 may thus depend more on how we change the organization and financing of health care than on Americans’ health care needs. Instead of financing further growth in our medical education system, resources might be better directed to reorganizing delivery systems to models of FFS and prepaid group practice that have already demonstrated that they can deliver good care at relatively low costs.
David Goodman (david.goodman{at}dartmouth.edu) is a professor of pediatrics and of community and family medicine at the Center for the Evaluative Clinical Sciences (CECS), Dartmouth Medical School, in Hanover, New Hampshire. Thérèse Stukel is a senior scientist and vice president for research at the Institute for Clinical Evaluative Sciences in Toronto, Ontario. Chiang-hua Chang is a research associate at the CECS. John Wennberg is director of the CECS and the Peggy Y. Thomson Professor for the Evaluative Clinical Sciences at Dartmouth Medical School. This research was funded in part by National Institute on Aging (NIA) Grant no. P01 AG019783 and the Robert Wood Johnson Foundation.
This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Study Data And Methods
Results
