|
|
| |||||||||||||||||||||||||||||||||||||||
Background Physician profiling is a method of cost control that focuses on patterns of care instead of on specific clinical decisions. It is one cost-control method that takes into account physicians' desire to curb the intrusion of administrative mechanisms into the clinical encounter. To provide a concrete example of profiling, we analyzed the inpatient practice patterns of physicians in Florida and Oregon.
Methods Data for 1991 from Medicare's National Claims History File were used to profile 12,720 attending physicians in Florida and 2589 in Oregon. For each attending physician, we determined the total relative value of all physicians' services delivered during each patient's hospital stay. Relative value was measured in relative-value units (RVUs), according to the resource-based relative-value scale used by Medicare in determining payments to physicians. The mean number of RVUs per admission was then adjusted for the physician's case mix according to the patients' assigned diagnosis-related groups. The influence of the physician's specialty and of selected types of services (such as imaging and endoscopy) was also examined.
Results Florida physicians used markedly more resources, on average, than their colleagues in Oregon (46 vs. 30 case-mix-adjusted RVUs per admission). The difference was apparent for all specialties and all types of service. To illustrate the profiling data potentially available to the medical staffs of individual hospitals, we examined specific data on individual attending physicians and for various types of service for three hospitals' staffs. Despite similar overall profiles that fell below the national mean, each staff had a different practice pattern and would require different efforts to improve efficiency.
Conclusions In an effort to encourage further debate, we have described one method of physician profiling. Profiling data help identify and characterize differences in practice style to which individual physicians or hospital staffs can respond. Because profiling is not based on rigid rules, it is a cost-containment strategy that can easily accommodate legitimate exceptions; it is therefore preferable to methods in which the appropriateness of each clinical decision is judged separately.
The concept of focusing on patterns of care instead of specific clinical decisions has been labeled "physician profiling"4. The practice pattern of a single physician or a group is expressed as a rate: some measure of the use of resources during a defined period for the population served5. The resulting profile can then be compared with a norm that is either based on practice (such as profiles of other physicians) or based on standards (such as practice guidelines). Both the American College of Physicians, in its proposal for universal insurance,6 and Congress, in enacting Medicare physician-payment reform in 1989,7 have advocated physician profiling.
The Physician Payment Review Commission recently convened a conference on the topic and found most of the existing studies limited to a single facility and to the use of a specific service or treatment (generally laboratory tests or pharmaceutical agents) instead of covering all services associated with an episode of care8. The problem of high costs, however, is one that involves multiple services. Although more comprehensive profiling of interns in the Midwest has been reported,9 most profiling of this type is being carried out by insurers and is thus proprietary10,11,12. Because insurers' methods are not peer-reviewed and the details are not publicly available, it is difficult for physicians to learn how more comprehensive profiling might actually be put into practice.
The Physician Payment Review Commission conference also identified some important methodologic criteria for physician profiling8. First, profiles must be analyzed for a well-defined population. Second, a sufficient number of observations should be included to ensure that differences are not due to chance. Third, adjustment should be made for differences in case mix. In addition, profiles must be analyzed for a small enough organizational unit so that members can feel responsible for the results and can work as a group to identify any necessary courses of action13.
The purpose of this study was to demonstrate a method of physician profiling that takes these issues into account. We considered a clearly demarcated population (inpatient Medicare beneficiaries) for whom one physician (the attending physician) was primarily responsible during a discrete episode of care (the hospital stay). Because Medicare beneficiaries are frequently hospitalized, it was possible to amass data on the hospital stays of multiple patients for each of the physicians we studied. The assignment of a diagnosis-related group (DRG) to each patient at discharge allowed each physician's admissions to be adjusted for case mix. Finally, the medical staffs of hospitals are well-defined groups that can both assess and act on profiling data. We profiled physicians who provided services in Florida and Oregon, two states that we knew from a previous study14 were at the extremes of the spectrum in terms of the use of services.
Methods
To investigate the feasibility of large-scale physician profiling, we analyzed all Medicare claims for physicians' services for beneficiaries who were not enrolled in a health maintenance organization and who resided in Florida or Oregon (approximately 2.4 million beneficiaries) for the period July through December 1991. In Florida, 12,720 attending physicians, who provided care during 221,940 admissions, were profiled. In Oregon, 2589 physicians were profiled (32,110 admissions). Medicare data were used both because they contain standardized information on a large number of hospital admissions and because their geographic scope allows local observations to be compared with national norms.
The use of the physician as the unit of analysis in this study was made possible by the unique physician-identification numbers (UPINs) in the Health Care Financing Administration's new data system, the National Claims History File. This file includes records on both hospital admissions (including the DRG, admission date, discharge date, and other information) and physicians' services (UPIN, procedure code, dates of service, and so on).
For each attending physician, we determined the total relative value of all services delivered by a physician during a given hospital stay (including services delivered by other physicians, such as radiologists). Relative value was expressed in relative-value units (RVUs), according to the resource-based relative-value scale used by Medicare for calculating reimbursement15. Each service has been assigned a relative value that reflects the physician's work as well as practice overhead and malpractice-insurance costs. The number of RVUs for each service is determined in relation to an intermediate (routine) office visit (1 RVU). The relative value for a hernia repair, for example, is 10.9; the relative value for reading a chest film is 0.28. After determining the average number of RVUs per admission, we adjusted for case mix according to the national average number of RVUs per admission for a patient's DRG.
Assigning Resource Use
The number of RVUs for each service had to be assigned first to a specific inpatient stay and then to an individual attending physician. Using the beneficiary's identification number and the dates of service, we linked physician's services to admission claims. The relative value of services provided between admission and discharge was summed to obtain the total number of RVUs for each hospital stay.
Because the UPINs of attending physicians were not in this Medicare data set, the total number of RVUs for an admission was attributed to an attending physician according to the following algorithm. For admissions assigned a surgical DRG, the attending physician was defined as the provider who performed the surgical procedure with the highest relative value. For admissions assigned a medical DRG, the attending physician was defined as the provider with the most RVUs for services designated as hospital visits (including visits in the intensive care unit). Because the initial hospital visit is assigned a relative value approximately three times that of follow-up visits, this method generally identified the admitting physician as the attending physician. For patients with long stays whose care was transferred to a second physician early on, however, this method correctly attributed attending status to the second physician.
Adjustment for Case Mix
Our method of adjusting for case mix was similar to that used in our earlier work14,16. We assigned relative values to the hospital stays of a national random sample consisting of 5 percent of hospitalized Medicare beneficiaries in 1989 (approximately 500,000 admissions). The mean number of RVUs per admission was calculated for each DRG. Relative weights were created by dividing the mean number of RVUs for each DRG by the overall average for all DRGs (38.9 RVUs, or about $1,200 per admission based on the 1992 dollar conversion factor). The ratio of the observed number of RVUs for an admission to the relative weight for the particular DRG is the case-mix-adjusted relative value per admission.
Analysis of Practice Patterns
We examined the profiles of attending physicians at two levels of aggregation: the population of physicians in each of the two states and the medical staffs of hospitals. Statewide averages are all-inclusive. In order to mitigate the effect of outlier admissions and to characterize the pattern of practice more accurately, we included only physicians who provided care during at least 10 admissions in our analysis of individual attending physicians' profiles.
To analyze the relevant components of the total relative value of the physicians' services per admission for medical staffs and for each state as a whole, we examined practice patterns for different specialties and for selected types of service. Physicians' specialties were identified through the UPIN master file. We grouped Current Procedural Terminology (CPT-4) codes into categories for type of service by the method developed by Berenson and Holahan17. Considerable effort was made to group about 7000 CPT-4 codes into 6 major categories and 23 subcategories that reflected the clinical content of the service, as opposed to the specialty of the physician who primarily provided it. Thus, cardiac ultrasonography and coronary angiography were categorized as imaging, despite the fact that they are not typically performed by radiologists.
We focused on the seven subcategories of service that were most often responsible for differences in the case-mix-adjusted relative value per admission: hospital visits, consultations, endoscopy, standard imaging, ultrasonography, computed tomography (CT) and magnetic resonance imaging (MRI), and imaging procedures. The first two are self-evident. Endoscopy comprised all procedures involving fiber-optic instruments and included bronchoscopy, cystoscopy, and all gastrointestinal endoscopy. The last four subcategories made up the imaging category. Standard imaging referred to plain films, barium studies, and nuclear medicine. Ultrasound is self-explanatory, as are CT and MRI. Imaging procedures included cardiac catheterization, other angiography, and venography. Because a case-mix adjustment incorporating the average relative value of all services per admission does not apply to a specific type of service (for example, two DRGs with similar overall numbers of RVUs may have very different averages for endoscopy), we developed a case-mix adjustment specific to each type of service we studied18.
Results
Statewide Data
Florida physicians, on average, practiced a style of medicine characterized by the use of more resources than their colleagues in Oregon (case-mix-adjusted relative value per admission, 46 vs. 30 RVUs). Although we expected this mean difference on the basis of our previous work, the differences were pervasive. Figure 1 shows the distribution of the number of RVUs per admission, after adjustment for case mix, for individual attending physicians in each state. Instead of reflecting the practice style of a few physicians, the Florida average is the result of a shift to the right of the entire distribution. For example, although three quarters of the Florida physicians averaged between 35 and 65 RVUs per admission, only one quarter of Oregon physicians were in this range.
|
|
|
There are two major limitations to the use of statewide data. First, the aggregated data are likely to obscure important variations at the local level. Second, the state is such a large administrative unit that it is difficult to conceive how a state's physicians as a group could react to the information. We therefore examined a smaller administrative unit, the medical staff of a hospital, in which physicians are better able to assess and act on profiling data.
To demonstrate the type of data that could be provided to a specific medical staff, we profiled the medical staffs of three hospitals in our data base according to type of service and for individual attending physicians. To demonstrate the capacity of profiling data to distinguish underlying practice patterns, we selected the staffs because of their similar overall profile (case-mix-adjusted relative value per admission, approximately 33 RVUs). All three staffs are primarily devoted to clinical service (i.e., the physicians have no major teaching responsibilities) and serve short-stay community hospitals (100 to 400 beds) that do not meet the Medicare criteria for serving a disproportionate share of the poor.
Figure 4 shows the relative values per admission for Staff A and Staff B according to the type of service. Although the two groups of physicians had similar overall profiles (and were below the national mean in RVUs per admission), they might direct efforts to improve their efficiency differently. After adjusting for case mix according to the type of service, we found that Staff A used endoscopy more frequently than the average staff, whereas Staff B used more imaging procedures such as cardiac catheterization, other types of angiography, and venography than average.
|
|
The endocrinologist on Staff A used endoscopy three times more frequently than average and requested consultations twice as often. Although the orthopedic surgeon was an infrequent user of imaging and evaluation and management services, he or she was performing twice as many major surgical procedures per admission as would be expected given the patients' discharge diagnoses. The family practitioner on Staff C used 20 percent more standard imaging procedures (plain films, barium studies, and nuclear-medicine studies), used endoscopy 70 percent more often, and had 40 percent more RVUs for hospital visits than an average physician with the same case mix. The cardiologist on Staff C was similarly high on the measure for endoscopy, using 180 percent of the national mean for the same case mix.
Discussion
Although physicians have always known something about the practice styles of their colleagues, three factors have made it difficult to make judgments about these patterns. First, the observations are haphazard and are subject to sampling error. Second, little is known about what constitutes average behavior nationally because only the practice styles of local physicians are accessible for comparison. Third, the variation in the type of patients being treated has made it nearly impossible to establish a typical practice style.
Our study was intended to introduce physicians to profiling and to demonstrate one way to approach the problem of comparing individual physicians' practice styles to national or statewide averages or to ideal patterns. By analyzing patterns of behavior over a period of time, it is possible to obtain a sufficient number of observations to minimize sampling error. Nationwide data allow local observations to be compared with norms for the country (or state) as a whole. Finally, the adjustment for case mix begins to take into account the fact that different doctors treat different types of patients.
Profiling data could be used by a number of participants in health care reform. The most obvious are payers. Medicare, for example, might profile medical staffs and set volume-performance standards based on the case-mix-adjusted relative value of services provided per admission. Health alliances (as conceived in President Bill Clinton's plan for reform) and integrated systems of care (such as health maintenance organizations and the Department of Veterans Affairs) might choose to use profiling to identify overused services and to investigate the contribution of specific investments (such as the decision to purchase an MRI scanner or to hire another gastroenterologist) to rates of use.
In addition, such data could be used by physicians themselves. As has been demonstrated among surgeons in Maine, simple feedback about one's own practice style may be sufficient to bring extreme values into line19. Once extreme practice patterns are identified, an exploration of their root causes may help direct the practitioner to problem areas. For example, the endocrinologist we profiled may see patients with medical diagnoses with which he or she is unfamiliar, leading to more frequent consultation. The orthopedic surgeon may have to perform repeat operations more frequently than expected because of problems with case selection or operative technique. And the family practitioner may simply keep his or her patients in the hospital too long.
Nevertheless, the profiling of individual physicians can best be viewed as a screening technique. Confirming excessive use of services requires that the possibility of extenuating circumstances be excluded. This analysis can be done best by those close to the case under consideration, such as physician colleagues on the medical staff of a hospital. Medical staffs could consider the reasons behind the profiles of specific physicians and consider alternative explanations, such as greater severity of illness among a physician's patients.
In order for a difference in practice styles to be explained by greater severity of illness, two criteria would have to be met. First, some physicians must consistently see sicker patients within a given DRG. Although most physicians see sicker patients some of the time, only a few are likely to have sicker patients all the time. Second, the severity of illness must be a major determinant of the use of resources. Unfortunately, many measures of severity of illness incorporate measures of resource use and thus, not surprisingly, identify associations between the two. When the severity of illness has been assessed separately from resource use during the hospital stay, it has been shown to be a poor predictor of cost20,21,22. Thus, although a difference in severity of illness could be a factor in the profiles of a few physicians, it is unlikely to be a valid explanation for variation in general.
The Future of Profiling
There are three major reasons why physicians might prefer profiling to a strategy of making judgments about the appropriateness of each clinical decision. First, it is less cumbersome to have one's pattern of care reviewed at regular intervals (perhaps every six months or every year) than to have clinical decisions reviewed daily. Second, because there are fewer data to review with profiling, it is more likely that physicians would be actively engaged in the process. Third, profiling is better able to accommodate legitimate exceptions. Measurements that include a number of observations are less affected by particular extenuating circumstances than are evaluations of specific treatment decisions. Thus, profiling tends to preserve the clinical autonomy of physicians by avoiding rigid rules about the care of specific patients.
As an analytic tool, physician profiling is early in its development. Future efforts will need to move beyond the inpatient population. Answering questions about who is being admitted to the hospital and what is happening to those who are not admitted will require knowledge about which physicians are responsible for defined populations. New mechanisms to adjust for case mix will need to be developed; this effort will be particularly important if profiles are to be studied in relation to patients' outcomes. The combination of the pressure to do something about escalating costs and the desire to avoid a strategy of micromanagement is likely to encourage the further development of this analytic method.
Dr. H.G. Welch is the recipient of a Veterans Affairs Career Development Award in health services research and development. The work of Dr. Miller and Dr. W.P. Welch was supported under a cooperative agreement (18-C-90038) with the Health Care Financing Administration. The opinions expressed herein are those of the authors and do not represent the opinions or policies of the Department of Veterans Affairs, the Health Care Financing Administration, Dartmouth Medical School, the Urban Institute, or their sponsors.
We are indebted to Ellen Englert for research assistance and to Paula Beasley and her staff at Social and Scientific Systems for programming; we are also indebted to Andrew Dodds, Gregory Froehlich, and Gerry O'Connor for their thoughtful critique of the manuscript.
Source Information
From the Veterans Affairs Medical Center, White River Junction, Vt. (H.G.W.); the Center for the Evaluative Clinical Sciences, Dartmouth Medical School, Hanover, N.H. (H.G.W.); and the Urban Institute, Washington, D.C. (M.E.M., W.P.W.).
Address reprint requests to Dr. H.G. Welch at the VA Outcomes Group (111B), Veterans Affairs Medical Center, White River Junction, VT 05009.
References
| |||||||||||||||||||||||||||||||||||||||
Related Letters:
Practice Profiles
Masdeu J. C., Lossing J. H., Wears R. L., Welch H. G., Miller M. E., Welch W. P., Kassirer J. P.
Extract |
Full Text
N Engl J Med 1994;
331:201-203, Jul 21, 1994.
Correspondence
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
