Inconvenient truths about supplier induced demand and unwarranted variation in medical practice

Practice variation and the complexity of medical decisions

The first systematic account of practice variation was Glover’s 1938 report on the incidence of tonsillectomy among schoolchildren in England and Wales.4 He found a tenfold variation in tonsillectomy rates from one region to another. Although he did not comment on the economic consequences of this variation, he did observe that for every death caused by complications attributable to enlarged tonsils there were at least eight deaths caused by tonsillectomy. Economics aside, there was a clear ethical imperative to question and confront practice variation.

Decades after Glover’s report, regional variations in tonsillectomy and other surgical procedures were rediscovered in the United States (figure⇓).1 2 5 Wennberg and colleagues showed that, although the incidences of their studied surgical procedures often varied between countries, the degree of variation within countries for each procedure was similar. Furthermore, the degree of variation was associated with the level of uncertainty about the probabilities of surgical outcomes.5

Rates of common surgical procedures among Medicare patients for 306 referral regions1

• Download figure
• Open in new tab
• Download powerpoint

Many of the procedures with high variation were aimed at improving quality of life by reducing symptoms associated with conditions such as benign prostatic hyperplasia, benign uterine conditions, disorders of the lumbar spine, and coronary artery disease. In each case, evidence to compare effectiveness of treatments was inadequate; surgeons who preferred action to inaction for whatever reason could choose to believe in untested hypotheses that supported their recommendations for intervention.

Uncertainty about outcomes was not the only source of complexity. Patients had variable reactions to the same level of symptoms and felt differently about the probability and degree of symptom relief afforded by treatments. They also felt differently about the probability and impact of side effects and risk of death.6 For example, men with prostatic hyperplasia had different views about the inconvenience of urinary dysfunction and the prospect of sexual dysfunction. When informed and given help to reflect on how they felt about the outcomes, they made choices that were concordant with those feelings: men who were most bothered by their symptoms were more likely to choose surgery; men who were worried about the prospect of sexual dysfunction were less likely to choose surgery. When decisions were informed by both evidence and personal knowledge with the help of a decision aid, the rate of surgery fell 40% from baseline levels, suggesting that surgeons had systematically overestimated patients’ preferences for symptom relief and underestimated their preferences to avoid sexual dysfunction.7 These faulty assumptions about patients’ preferences might vary from one region to another, contributing to regional variation in rates.

Randomised trials of similar decision aids showed declines in demand for surgery for benign uterine conditions in the United Kingdom, for coronary artery disease in Canada, and for herniated lumbar disc in the United States.8 9 10 Rates of surgical intervention for spinal stenosis increased.10 An international Cochrane review of more than 50 randomised trials of decision aids included seven trials involving major elective surgery; demand was decreased by 21-44% as knowledge was increased.11 A small trial among men with prostate disease in England showed an increase.11 When expert opinion was used to define ideal candidates on clinical grounds for joint replacement in Canada, 85% in high rate regions and 92% in low rate regions did not want surgery.12

Practice variation and effect of regional supply

What about the decisions that clinicians make daily in the course of clinical practice—the interval between a patient’s visits for managing a chronic problem, when to refer a patient to a specialist, or when to request diagnostic imaging and laboratory tests? There are few or no randomised trials to guide such decisions, and if clinicians paused to ponder, professional uncertainty would bring clinical care to a standstill. Instead, clinicians cope by adopting routines or standard clinical policies.13 Over time, these clinical policies serve to mask the underlying uncertainty for clinicians and their patients.

Countries with different degrees of reliance on government and private insurance all show strong regional associations between per capita supply and per capita use of physician visits, diagnostic imaging, hospital bed days, and intensive care.1 2 14 Once heavy investment has been made in building capacity, the marginal cost perceived at the point of service can be low enough to encourage adoption of increasingly service intensive standard clinical policies, even when the potential for benefit is recognised as modest.

The full implications of these service intensive policies are often not considered. Many studies have shown that clinicians are not adept at dealing with the conditional probabilities that determine the predictive value of test results or communicating about uncertainty.13 Because there is little room for uncertainty in the “if, then” standard clinical policies, clinicians often respond reflexively to positive results with more tests or treatment. This diagnostic-therapeutic cascade is especially pronounced when radiographic images show abnormalities that are incidental to presenting symptoms and would be of no consequence to future health and wellbeing.15

The allure of early detection of disease evokes a special case of the clinical cascade. Clinicians and patients want to believe in the value of screening and often pay too little attention to the potential for harm from misdiagnosis, overdiagnosis, and overtreatment. That information can have a negative effect on health is counterintuitive to both clinicians and patients. And after the fact, cognitive dissonance often keeps both from acknowledging that the harms were avoidable. For the man left impotent and incontinent by treatment for prostate cancer, and for the clinicians who screened and treated him, it is easier to believe that early detection saved his life.16

Referral to specialists often precipitates another diagnostic-therapeutic cascade. Specialists, by definition, focus on a limited range of conditions. This narrow view affects the way they frame their research, which in turn affects the way they frame decisions for patients.17 Cardiologists and cardiovascular surgeons have for decades focused on cardiovascular outcomes of coronary artery bypass surgery, paying relatively little attention to the neurological complications of stroke and cognitive impairment.

Specialists’ experience may be so limited to a particular disease that they neglect competing forces of morbidity and mortality that make benefit of treatment less likely or irrelevant. Urologists see many men die from prostate cancer, but they are not likely to see the men they have treated die of other causes before treatment of prostate cancer has yielded any benefit.

Yet another form of clinical cascade occurs when a patient is referred to a hospital with immediate availability of imaging and laboratory testing and myriad consultants, especially in specialised settings like intensive care units. Here too, standard clinical policies are shaped by supply. In the United States, striking differences exist among hospitals in the numbers of consultations and days in the hospital and intensive care unit in the last six months of patients’ lives that cannot be explained by differences in patients.1 Clinicians become accustomed to standard clinical policies that have been shaped over time by local capacity and are often surprised to learn that they are radically different from those in other regions. Though invisible, these local policies profoundly affect behaviour; when presented with standardised patient vignettes, clinicians from different regions make decisions consistent with their local service intensity.18 In the United States, it is the difference in rates of use of these services that explains most of the threefold regional variation in per capita expenditures for care.1

Higher intensity of visits, tests, specialist consultant referrals, and hospital admissions would not necessarily constitute inefficiency if it produced better experiences of care or health outcomes. In the United States, it does not. Doctors in the high cost regions actually perceive greater difficulty getting access to specialists and to hospital beds for their patients.19 Patients, who are also generally unaware of inter-regional differences, see no better access to or quality of care. Patients in high intensity regions do not receive more care that is known to be effective, and their outcomes, including mortality, are no better and may be worse.2 3

Responses to practice variation

There was little discernible response to Glover’s paper in 1938. The wide variations in rates of tonsillectomy, within countries and between countries, persist today.20 Efforts to reduce variation began decades later with the promotion of outcomes research and the development of guidelines by professional societies and eventually government agencies such as the National Institute for Health and Clinical Excellence. Clinicians have generally resisted measures that they perceive as limiting their professional prerogatives. Even the evidence based medicine movement has stirred controversy.

Wennberg and colleagues have focused on the distinction between unwarranted and warranted variation. They have developed decision aids to support clinicians and patients in reducing unwarranted variation while honouring valid variation, thereby making care both more evidence based and more personalised.6 An international collaboration has since evolved to develop standards for evidence based decision aids designed to inform patient choice without bias towards a particular treatment.21 But with a few exceptions, the uptake of decision aids has been limited and response to the implications of practice variation has been inadequate.

Finding a way forward

What more can be done about the unwarranted variation that exists in all healthcare economies? In the near term, more should be invested in the systems that support clinicians and patients in their decision making, including measures of decision quality and outcomes.22 Measures designed to assure that decisions are made with the personal knowledge of what the illness, treatment, and possible outcomes mean to the patient could create greater awareness among clinicians and patients of the complexity of decision making and the need for support. By focusing on the degree to which patients are informed and treatment conforms to their informed preferences rather than usage, these measures safeguard against biasing decision support to control costs.

Primary care providers are well positioned to work with patients to improve the quality of decisions across a range of illnesses because of their accessibility, the ongoing relationship, and orientation to the whole patient rather than a disease. Health services that emphasise primary care have better outcomes and lower costs.23 The decline of primary care in the United States is probably a major contributor to its poor standing in international quality and cost comparisons. But the need for systems support in primary care is universal. Well constructed systems to support shared decision making and collaborative care could improve the quality of primary care while decreasing the need for referral to specialists.

Over the longer term, primary care clinicians are also well positioned to create learning healthcare systems that capture the collective experience of patients and use the evidence about what works for whom and what is valued by whom to improve care processes. Studies can be designed that follow patients who choose or are randomly assigned to different treatments or care patterns.24 This approach has been used to improve evidence for surgical interventions.25

When randomised trials are unlikely, learning healthcare systems that take full advantage of electronic medical and health records could amass sufficient patient experience to determine the effectiveness of clinical policies. Such systems could also determine the effects of seemingly inconsequential decisions that trigger clinical cascades of great consequence. Emphasis on informed patient choice would reveal patients’ preferences and thereby inform policy makers’ decisions about expanding or contracting capacity to deliver different services.

These are not new ideas. The importance of avoiding overuse by informing patients about options and outcomes and the need to capture collective experience by measuring “end results” was articulated a century ago.26 27 Together these principles of professionalism form a constructive professional response to variation in practice. The barriers to implementation are neither conceptual nor technical.

For too long, stakeholders in health care have avoided the inconvenient truth about the real limits of medical science as the basis for decisions about diagnosis and treatment. The belief that science mediated by clinicians determines what is the best medical care for each patient confers authority for decision making on the clinician and absolution from responsibility on patients and policy makers. As comfortable as this may be for all, it puts patients at risk of care they would not choose and puts policy makers at risk of presiding over the resulting inefficient healthcare economies. Clinicians, patients, and policy makers must find the competence, curiosity, and courage to confront the implications of practice variation and improve the quality of medical decision making. At stake is nothing less than the health and wealth of nations.