Physician-Level Practice Variation
by Brian Powers, Sachin Jain, David Cutler & Ziad Obermeyer
Published on Nov 26, 2016
Variation in the cost, quality, and intensity of clinical services remains a challenge for health system performance in the United States. Successfully reducing unwarranted variation will be necessary to achieve the shared goals of better health, higher quality care, and lower overall costs. This article was originally published in Health Affairs Blog Sept 23, 2015
Variation in the cost, quality, and intensity of clinical services remains a challenge for health system performance in the United States. Despite increased awareness over its scope and implication, little progress has been made in explaining—and still less, reducing—unwarranted variation. Successfully doing so will be necessary to achieve the shared goals of better health, higher quality care, and lower overall costs.
Variation Is Local
Researchers and policymakers have traditionally studied variation at the geographic level, demonstrating striking regional differences in the intensity and cost of services. More recent analyses have gone further, finding just as much, if not more, variation within these regions as among them.
For example, within a given community, costs associated with treating some conditions (e.g., heart disease) are significantly above average, but those for many other conditions (e.g., diabetes) are below average. There is even tremendous variation in the delivery of clinical services among patients treated within the same health system. Regional demographic and market factors are insufficient to explain observed variation in practice patterns.
An alternative explanation is emerging from new research: that variation has its roots not in static demographic or geographic factors, but rather from the very local contexts within which physicians—and patients—negotiate, weigh risks, and adopt individual treatment plans. This view, while intuitive, has been difficult to substantiate empirically. Although pioneering researchers unearthed enduring small-area variations and the impact of physician preference on local practice patterns decades ago, there remains little data on contribution of physician practice pattern to variation at the national level.
Emerging Evidence Of Physicians’ Impact
Bridging this gap requires the creative application of new quantitative methods to the analysis of administrative claims data. Two recent analyses of the Medicare population demonstrate the promise of this approach.
The first examined variation in preoperative testing before cataract surgery, a practice widely considered to be of no benefit to patients. The authors found that the likelihood of undergoing testing was more strongly associated with the ophthalmologist who managed the preoperative evaluation than any specific characteristics of the patient. In fact, 36 percent of ophthalmologists were responsible for 84 percent of preoperative screening tests — despite treating only 26 percent of patients in the study.
The second study analyzed variation in hospice use among cancer patients with a poor prognosis, a service consistent with the wishes of most patients in this population. Physicians in the study had wide discrepancies in the proportion of patients under their care who enrolled in hospice. The proportion of a physician’s patients enrolled in hospice predicted whether or not other patients under that physician’s care would also enroll. Here too, the impact of the treating physician was considerably stronger than well-established predictors of hospice enrollment including patients’ medical comorbidity, age, race, and sex.
That physician factors can explain both the overuse of unnecessary services and the underuse of clinically indicated services across clinical entities suggests physician-level variation is not only pervasive, but substantially impacts the cost, quality, and value of care delivered across a wide spectrum of clinical services.
Implications For Researchers And Policymakers
Going forward, it will be important for researchers to extend variation analyses to other metrics of quality, cost, and outcomes. Information from claims data should be merged with clinical data from electronic health records in order to expand the scope, granularity, and nuance of analyses.
Further research is also needed to clarify the source of this variation. Physician-level variation could result from physician preference, as suggested by some small, survey-based studies, but could also stem from factors such as local practice environments. If patient-provider decision making is the root of variation, then it will be necessary to move beyond geographic analyses, and characterize in detail the physician and patient-dependent variables, preferences, and values that shape care decisions.
An enduring challenge for health services research has been translating findings into policy. Over the past few decades, the literature on geographic variation has opened our eyes to striking differences in patient care. But the simple truth is that knowledge of regional patterns offers little in the way of understanding the realities of clinical decision making or developing strategies to change practice. It’s time to move on.
Policies that address variation must target decision-making, not geography. Analyses of physician-level variation offer a novel, nimble, and adaptable method for extending variations research into the local practice environments in which patients and physicians make care decisions. Furthermore, these data can be calculated at the provider and the facility level even without the use of large datasets or complex statistical models, delivering actionable information to local administrators and practice leaders. Doing so is a necessary step toward helping doctors deliver the high-quality care their patients deserve.