Physician profiling: trends and implications.
A number of earlier reports of hospitals' practices in profiling the practice habits of physicians were reviewed in 1993.1 A 1989 survey of 3,400 hospitals determined that only 5 percent of hospitals then had programs in operation to review physicians' utilization of resources for inpatients and their associated costs. A 1991 survey of 500 hospitals found that 40 percent would consider physicians' economic performance in credentialing; 2 percent indicated they regularly prepared and used physician-specific profiles for educational purposes. A 1993 survey of 55 member hospitals of the University Hospital Consortium found that 66 percent developed and reviewed profiles of physicians. However, only two-thirds of those that did so (44 percent of the total) shared the profiles with their medical staffs, and only 35 percent of them collected and included economic data elements. As late as 1993,40 of the 48 academic medical centers indicated their boards of trustees were very remote from issues of physician profiling and economic credentialing.
While we expect the number of facilities investing in physician profiling and continuing education for medical staff members based on the results of such profiling to continue to increase as hospital operating margins decline, few hospitals, even in 1995, seem prepared to analyze the clinical and financial outcomes for patients of physicians' practice patterns and to share those data with clinical departments. Most hospitals do not have the epidemiological expertise or budgets for health data analysis, practice profiling, and risk adjustment that would allow them to develop practice guidelines with their clinicians. Nor do they have clinical, financial, and functional status data on processes of care and outcomes in an electronic format that would permit rapid practice profiling and outcomes analysis. Most hospitals would face the daunting task of marshaling resources for retrospective, manual chart review if clinical departments were interested in systematic study of the clinical and financial outcomes of their patients. Clinical departments of most hospitals include competitors who have less incentive to invest their time to help colleagues improve practice habits than they would if they were all in the same economic unit, a group practice. But most group practices also do not have the data and have less money for practice profiling than do hospitals.
Nevertheless, there has been growing hospital interest in profiling physicians' practice habits. All the commercial information systems most hospitals acquire for such profiling use the same basic data set - the UB-92 data set. The data elements include demographic identifiers of the patient (birth date, sex, race, discharge status); up to nine diagnoses and up to six procedures; and the sum of charges by major type of service, such as pharmacy, laboratory, radiology, physical therapy, dietary, nursing, and operating room. These data are very limited for practice profiling. There are no findings from physical examinations, no laboratory results, no pharmaceuticals identified by name, and no measures of patients' functional status before or after treatment. Risk adjustment in most of these commercial systems currently depends on age, sex, diagnosis, and procedure codes alone.
None of the most popular commercial systems for profiling physicians' practice habits in hospitals explains substantially more than half the variation in costs of care, length of stay, or mortality rates, the three major outcomes measured by these systems. Much of the variation in outcomes is not explained by the limited data collected for profiling. In the future, these systems will be far less important than those now emerging that purport to adjust for risk and to profile physicians' practice habits across the continuum of care.
In the past 10 years, the proportion of U.S. health care expenditures spent in hospitals has declined from 50 percent to less than 40 percent, much less in areas where physicians have strong financial incentives to manage their patients out of hospitals. The average number of hospital bed days for a representative population of people under 65 years of age has declined in the past 10 years from more than 700 days per 1,000 people per year to less than 200 in many areas where health maintenance organizations and capitation are prevalent. In regions where competition is most fierce among health plans for members and among providers for patients, far less than half the total per capita funds for health care services are spent in hospitals. And hospitals rarely have data on what happens to patients outside their facilities, unless they have self-insured for the health care benefits of employees and can analyze claims data for a few thousand lives.
For all these reasons, hospitals have not been, and probably will never be, the center of attention for profiling physicians' performance. The center of attention in health care policy and economics has shifted from the hospital to the health plan and the integrated delivery system, and funds devoted to physician profiling are flowing to institutions that can collect and analyze data on members (those who use clinical resources and those who do not) and on patients who receive care in many settings, over time, usually for chronic illnesses.
In the mid-1980s, interest in provider profiling emerged in the employer community. It was focused on the outcomes of physicians and hospitals and was led by employer coalitions concerned about rising health care costs and impressed with small area variation studies. Employer coalitions convinced, or coerced, hospitals to fund the collection and analysis of discharge abstract data for public distribution of profiles. Several states joined the action, mandating public profiling of hospitals on the basis of risk-adjusted outcomes. In all, more than 30 states began collecting UB-92 data on inpatients, but what they did with those data, and how they risk-adjusted the outcomes (length of stay, charges, and mortality rates) defined by those data varied widely from state to state.
I would predict that none of these programs has produced improvements in the costs or the quality of care greater than improvements that occurred in communities without such programs. In most of the programs, providers did not take the initiative to measure themselves. The providers did not keep the data for analysis; employer coalitions and state agencies did. Most important, the data were not complete enough for the keepers of the data to say why wide variation existed in the costs of health care for apparently similar patients. The processes for data collection were set in motion in board rooms far removed from where the care was delivered to patients, and physicians were put in the position of defending the differences, usually by criticizing the methodologies for data collection and analysis. Occasional reminders from some distant evaluative organization that length of stay or costs of care are too high does little to motivate a clinical department of competitors to invest their time and resources in systematic evaluation and modification of the processes of care. Receiving general and dated messages that there is a wide variation in practice styles and costs of care will not induce a clinical department of competitors to invest hundreds of hours in thorough data analysis and consideration of alternative processes to improve clinical care and hospital financial performance. They need to have good data, and the fear of God that if they don't invest the time to improve their clinical practices, very bad things will happen to them.
In 1994, the senior physician executive of Providence Medical Center (PMC), Seattle, Wash., described in detail the successful implementation of a physician profiling program that led to quantifiable benefits that substantially exceeded the costs of the program.2 Seattle hospital managers felt financial pressure from managed care plans and the prospective payment system in the late 1980s. PMC administration first presented profiling data to the medical staff at that time, but physicians resented the intrusion into their domain and rejected the process, complaining that the data were not valid because they were based on financial claims data, were not risk-adjusted, included outlier cases that ought to have been excluded from the analysis, were not accompanied by local benchmarks for comparisons, produced no summary scoring or relative ranking of physicians that they could understand easily, and were presented without graphics. The medical staff did not receive any training in risk adjustment, in the methods used for profiling, or in interpretation of the data presented to them. After meeting unexpected resistance and contempt from the medical staff, administration abandoned its attempt to influence medical staff practice habits.
Two years later, the competitive climate in Seattle had grown much less comfortable for clinicians. They felt under siege and were aware that they might lose a substantial proportion of their patients to other providers contracting with managed care plans if their patients' outcomes were not as favorable, in the eyes of the plans, as those of the other providers. Medical staff leaders in 1991 asked administration to organize an outcomes assessment and practice profiling program. With the support of medical staff leaders, administration, established a profiling methodology and process that avoided the mistakes made two years earlier. The medical staff approved the selection of All-Payer-Defined DRGs (APR-DRGs) as the risk-adjustment methodology for the program. After careful study, 8 percent of patients' records were removed from study as outliers. The hospital acquired comparable data on patients from all other state hospitals from the state discharge abstract database.
Every member of each clinical specialty department received his or her data compared to all other physicians in the same department and also compared to peer groups in other hospitals. Every physician had several opportunities to attend a 90-minute presentation on profiling in general and on the data and profiling methodology selected by PMC in particular, followed by a more thorough presentation to each clinical department of its own data and of valid ways of interpreting the data. Each physician received a unique number by which all data were reported, so the data were anonymous. The range of variation discovered for most common DRGs was considerable, more than members of the medical staff expected. Individual counseling and follow-up data analysis were offered to any physician who wanted them.
After release of the first physician profiles, interest in development of practice guidelines appeared for the first time among members of the medical staff. Since then, the hospital has released profiling data every six months, more often for physicians who request it. After one year, a retrospective assessment of the program disclosed a surprisingly large decline in average length of stay (from 5.3 to 4.8 days), accounting for an estimated savings of more than 7,700 hospital bed days and an observation that the hospital appeared to health plans to be a more attractive facility with which to contract. No physicians lost admitting privileges as a result of this profiling project, and most physicians expressed interest in having the program continue.
The profiling programs that will have a much larger (and potentially devastating) influence on physicians are those created by health insurance carriers to winnow their networks of participating physicians by factors of 33 to 66 percent. Insurers are profiling physicians' practice habits and selecting physicians whose habits appear most consistent with "cost-effective" practice. Their means of measuring cost-effective practice (they scrupulously avoid any mention of quality of care) may be challenged, but the effects they can have on physicians' practices are indisputable. Some physicians find themselves losing 10 percent or more of their practices immediately after an insurer drops them from its managed care panel of providers. For instance, in 1993, Blue Cross and Blue Shield of the National Capital Area (BCBSNCA, Washington, D.C.) eliminated one-third of participating physicians from its new preferred provider plan. BCBSNCA used Pro/File, a physician profiling system it developed under the direction of Dr. Ron Klar and now sells to other Blue Cross and Blue Shield plans. King County Medical Blue Shield, Seattle, Wash., removed 40 percent of participating physicians from its approved networks of providers using Pro/File.
Pro/File aggregates claims data into a patient-centered database and allows comparisons of physicians' practice habits over time across multiple outpatient and inpatient visits for all the physicians' patients. Pro/File is used to calculate an average consumption of resources for each physician, including diagnostic studies and consultations and hospitalizations, for the most common types of patients he or she treats. Then outcomes (defined in terms of resource consumption) are compared to the norm for the physician's specialty based on the types and numbers of patients treated. The system finds an expected resource consumption for each type of patient, based on practice habits in the entire metropolitan area, and assigns those expected values to each physician's patient population to determine whether expected resource consumption is higher or lower than actual resource consumption for each physician.
Profiling serves another purpose of managed care firms besides reducing the ranks of contracting physicians. The firms use profiling systems to answer a question posed frequently by employers: "How do you know your doctors are any good?" The only source of information insurers and employers have on the practice habits of physicians is the database of historical claims data insurers maintain for retrospective analysis and profiling of physicians. Insurers with indemnity health insurance business use the claims databases built by those indemnity claims to profile community physicians before they select a subset of them for new managed care plans.
Using claims data presents many challenges to insurers, the largest of which involve reliably identifying patients and providers and tracking all claims filed by patients over time to identify resources consumed and ailments treated during an episode of illness. Physicians' complete claims more accurately than ever before and insurers process them with more edits for reliability than ever before, but claims still too often cannot be linked reliably to the appropriate physicians or patients. Physicians are suspicious of any evaluation of their practice habits based on claims data, because they know claims are often completed haphazardly and may include only one diagnosis code when a variety of services were rendered.
Recognizing these limitation, insurers have still moved ahead to profile physicians using claims data, knowing that the data are more reliable in tightly managed health plans that assign patients to primary physicians and penalize them financially for obtaining services out of network. Insurers profile primary care physicians by calculating expected patterns of outpatient resource consumption by common types of patients (sometimes stratified into sex and age categories) and by deriving a separate model for each type of primary care physician. Once expected patterns of resource consumption are developed for each specialty for common types of patients, usually defined by common ailments, each physician's results are compared to expectations. Because most insurers cannot be confident that they can track all the services rendered to patients over time, or which physician is responsible for which services, profiling is done by visit, comparing physicians' use of resources by type of visit for type of patient. Physicians are compared on the basis of their relative total outpatient charges per type of office visit, per diagnosis code, per period (when all claims are submitted to the insurer for any given patient). Other comparisons include relative need for follow-up appointments and additional treatments within 90 days of the first office visit, relative use of consultations during a defined treatment process, relative use of laboratory and radiology services, relative number of visits per person per year, and relative level of intensity of services provided in office visits.
The use of the word relative calls attention to the need for some kind of risk adjustment. There are two general types of risk adjustment for outpatient services - visit-based and person-based. When the insurer is confident that all claims for services to individual patients are filed and it has a complete database of claims for specific populations of patients, person-based risk adjustment, also called population-based risk adjustment, leads to more accurate adjustment for risk. The two best known systems for population-based risk adjustment are Ambulatory Care Groups (ACGs) from Johns Hopkins University and Peer-A-Med from HealthChex. When payers do not have all the claims for individuals, they use visit-based risk adjustment, such as Ambulatory Visit Groups, Products of Ambulatory Care, Ambulatory Severity Index, Products of Ambulatory Surgery, and Ambulatory Patient Groups. The latter system, a methodology developed by 3M Healthcare, has been selected by the Health Care Financing Administration for reimbursement of Medicare outpatient services in the near future.
The focus of cost containment and outcomes management has expanded to the entire continuum of care. More and more Americans each month receive their care from health care systems of providers that accept a global capitation for their services for a defined population over a specified period. Those delivery systems ant their physicians to understand how they compare to their peers, and they want them to work together to improve the effectiveness and efficiency of their work. The average total costs of care for a physician's panel of patients is much more important to a capitated delivery system than the hospital, outpatient, or consultants' costs.
Insurers and health plans that have access to their own claims data are licensing profiling systems so they can measure and predict the resources needed to care for specific populations of patients. With these models of expected resource consumption, health plans are profiling physicians, comparing their results to those the model would predict, budgeting for staff and resources for ambulatory treatment facilities, and credentialing physicians.
In summary, physicians should expect to be profiled and to receive profiling results from hospitals, group practices, PHOs, and health plans. I suggest that physicians try to get profiled and insist that the organizations in which they work or the systems of care in which they see patients invest in staff and information systems to profile them. Outcome measurement for patients and subsequent profiling of physicians by the outcomes of their patients are the first two steps on the road to continuous quality improvement. Without comparative data, physicians and their organizations will compete for managed care contracts without a clear sense of how competitive they are.
Once your clinical community has selected a methodology for collection and analysis of patient data, a team consisting of clinicians trained in epidemiology and a statistician will prove invaluable in helping clinicians whose care is profiled to understand the methodology used, its weaknesses and strengths, and how to inquire into the data set beyond the initial reports a commercial system may produce. Sharing the data with clinicians in an educational way is key to successful outcomes studies and physician profiling. Physicians are learners. They want to interrogate data and see how their performance compare to that of their peers. The process needs grounding in valid clinical research methods and in continuing education for professionals.
[1.] Riley, D. "Economic Credentialing Survey of University Teaching Hospitals." Healthcare Financial Management 47(12):42-54, Dec. 1993. [2.] "Case Study of Physician Profiling.: Managed Care Quarterly 2(4):60-70, 1994. [3.] Ron Klar, MD, Blue Cross and Blue Shield of the National Capital Area, personal communication.
|Printer friendly Cite/link Email Feedback|
|Date:||Nov 1, 1995|
|Previous Article:||The role of health care ADR in reducing legal fees.|
|Next Article:||The pitfalls of managing contingency workers.|