However, one thing has changed. Although clinicians may have had difficulty in finding implications in these studies for their personal practices, variation can no longer be seen as a mere intellectual curiosity. Almost all of the other stakeholders involved in the delivery of health care recognize variation as a fundamental challenge. Who are these stakeholders? They are insurers, employers, the government, and the populations for which physicians provide care. In this editorial, I examine what is at stake for each stakeholder and look at how the concept of variation has been "operationalized" in ways that affect physicians' day-to-day activities. I also argue that physicians must participate in the debate about variation.

Insurers' interest in variation is the most apparent. In almost every clinical context, from therapeutic interventions to outpatient diagnostic testing, physician practice styles diverge widely. Variation in practice styles has a profound effect on the cost of insuring enrollees. This observation has fueled the new industry of managing physicians. The tools that are used, from profiling to guidelines, are aimed at reducing variation.

Almost every physician involved with managed care companies now receives routine practice profiles intended to reduce variation in use of services. Most profiles assess differences in per member, per month costs across providers compared with a benchmark (often the lowest-cost provider). Currently, profiles have several deficiencies. First, cost may not be the most important measurement with which to compare clinical practices. Second, risk adjustment methods are crude (if they are used at all), and identified outliers may simply be physicians with adverse selection [7, 8]. Finally, in the predominant managed care model-the independent practice association-the numbers of patients per panel are usually so small that comparisons between providers are simply comparisons of chance events. Despite these drawbacks, insurers regularly use these tools in attempts to change physician behavior.

Physicians are also deluged by practice guidelines. Insurers use guidelines to reduce variability in practice patterns by defining "acceptable" behavior. Unfortunately, the paucity of outcome data has allowed managed care organizations to create widely varied guidelines for the same clinical situations. Thus, physicians are left in the unenviable position of having to treat patients with the same clinical presentation differently, solely on the basis of who insures the patient. Physicians view guidelines as "cookbook" medicine, but insurers correctly assert that the "art of medicine" has resulted in a disorganized delivery system.

Employers are also interested in cost issues. However, unlike insurers, whose predominant goal is to pay the lowest price, employers have a more complicated role in the delivery system. In addition to needing a motivated, healthy workforce, employers see health insurance as a benefit that helps them recruit and retain valuable employees. Businesses are therefore interested in having an effective, well-managed health system for their employees. Furthermore, businesses are often well schooled in continuous quality improvement, a major component of which is reducing variation.

What do employers see when they look at the delivery of health care? Rather than a well-ordered, knowledge-based system, they see chaos. For example, a large employer in the midwestern United States finds that despite having similar insurance, their employees in Flint, Michigan, are 70% more likely to undergo cardiac revascularization than their employees in Grand Rapids [9]. In addition to concerns about cost, this variability raises several fundamental questions for employers. Do those delivering care know about the variability? If so, can they justify these findings by showing that patients are receiving the care that they want? Finally, how can a system that produces such variation deliver high-quality care?

From purchasing coalitions to direct contracting, businesses are beginning to address these issues as they negotiate health benefit packages. The Midwest Business Group on Health, a coalition of large employers, is working with local teams of physicians and hospital executives. They examine variations in health care use, cost, and access, and they work toward implementing a system that will make it easy and convenient for patients to participate in their health care through shared decision-making (James Mortimer, Personal communication). In the northeastern United States, the Maine Health Management Coalition, a group of 30 large and small employers, is working with the Maine Medical Assessment Foundation (MMAF), a nonprofit, physician-run education and research organization. The projects that these organizations are developing aim to improve quality of care for patients with conditions ranging from asthma to chest pain.

Even the federal government, the largest purchaser of health care services through the Medicare program, is confronting variation. Through its Health Care Quality Improvement Program, the Health Care Financing Administration (HCFA) has given up its old method of hunting for the bad apple through retrospective chart audits. Instead, HCFA has embraced a system of quality improvement [10]. This program measures population-based variability in process and quality of care, uses feedback as a mechanism to change provider behavior, and repeats measurements to assess changes in outcomes. The most visible example of this new effort is the Cooperative Cardiovascular Project, which has documented dramatic regional differences in the use of thrombolytic agents, ß-blockers, and aspirin in Medicare patients with acute myocardial infarction. Peer review organizations are now involved in the feedback phase by providing data on the variation in processes and outcomes to clinicians and hospitals. Whether this effort will improve care for Medicare enrollees remains to be seen.

What about the populations for which physicians provide care? Of all of the stakeholders, they have the most at stake. In every diagnostic and therapeutic situation, it is patients who experience the outcomes, risks, and benefits of the intervention undertaken or not undertaken. Although many still claim that patient preferences are a substantial component of measured variation, most evidence suggests that the interpretation of patient preferences by physicians drives these variations [11, 12]. Is there evidence that patients have discovered and acted on this information?

As of now, I know no instance in which a patient has moved from one geographic area to another in search of the "right rate." However, strong circumstantial evidence shows that variation in treatments and outcomes influences medical decision making by patients. Historically, patients have not shown strong preferences about which hospital they are admitted to; however, patients are beginning to choose where they are hospitalized on the basis of their perception of quality. Patients increasingly understand that the type of information they receive in the clinical encounter is predicated on whom they see, and they solicit care from alternate providers and pursue health care information on the Internet [13-15]. Unfortunately, the information that patients obtain from these sources probably has the same problem as the information that they receive from physicians: Providers of all stripes (even Web masters) provide that which they believe, not necessarily what patients want or need.

As physicians, what are we to do? First, we must take the risk of measuring the care that we deliver. Most of the tools that have been developed to study variation are applicable to the measurement of care delivered "at home." In Maine, the MMAF uses local data from statewide all-payer discharge databases and Medicaid and Medicare claims to assess delivery of care. Our findings concur with those of published studies: From hospitalizations for pneumonia to performance of colonoscopy, population-based rates of care vary greatly. We "feedback" these data to inform clinicians about how they care for their populations. These frank discussions among physicians start from the basis that we usually do not know what the "right rate" is. However, these clinicians are willing to take the risk that some rates may be too high and others too low. These efforts have led to changes in physician behavior, even though (or perhaps because) the MMAF lacks regulatory or financial power [16].

Second, where evidence-based data exists to inform clinicians about best practices, we should try to incorporate these data into daily practice. To avoid having a different guideline from each insurer, we must lead efforts to standardize the process. To set standards of care, the state medical association in Florida is developing practice guidelines that are reviewed and endorsed by the Florida Agency for Health Care. In Maine, the MMAF works with the business community to reduce variation in cardiac stress testing by creating physician-developed guidelines, providing data feedback to those who are ordering the tests, and organizing opinion leader activities.

Finally, and most important, we must be willing to reevaluate the patient–physician relationship. Although I do not believe that patient factors account for much, if any, of the variation in intervention rates across geographically defined populations, patient preferences within populations vary widely. This means, for example, that given the same information about the risks and benefits of treatment for class II to III angina and single vessel disease, one person in one region will choose medical treatment whereas another will choose percutaneous transluminal coronary angioplasty. Why? Only the person who faces the choice can assess how much their symptoms bother them and weigh the value of decreased symptoms with treatment by percutaneous transluminal coronary angioplasty (which also carries an associated 1% risk for death tomorrow) against the burden of continued angina (but no significant risk for immediate death) [17]. Although we must continue to insist on the primacy of the patient in all of our dealings, we must learn how we can best inform patients and objectively elicit their preferences [18, 19].

For physicians, these are challenging times. Insurers, employers, the government, and patients exert great pressure on us to change our practices. This pressure is often seen as coming from outside, but we must realize that we have helped fuel these forces. If we are to reassert our role in medical decision making and protect our patients from intrusion by third parties, we must take an active role in the debate about variation in practice patterns. If we do not, we will cede our role as professionals and become mere technicians.