Evidence-Based Medicine and the Changing Nature of Health Care: 2007 IOM Annual Meeting Summary (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

2 The Need for Better Medical Evidence INTRODUCTION Many Americans assume that the health care they receive is based on strong medical evidence of intervention and treatment effectiveness. How- ever, as suggested by regional analyses, recommended care is often not de- livered and insufficient evidence often leads to wide practice variations with little to no health benefit to patients (Fisher et al., 2003b; McGlynn et al., 2003). In addition to negatively impacting health outcomes, practice incon- sistencies have dramatic effects on the overall costs of health care—costs which represent the most pressing fiscal challenge to the nation. Papers in this chapter examine the drivers of practice variations and healthcare costs and suggest the potential for an improved evidence base to improve the efficiency and effectiveness of healthcare services. The majority of U.S. healthcare expenditures today are related to the care and treatment of chronic conditions such as heart disease, diabetes, and asthma, which affect almost half of the U.S. population. Evidence for effective strategies for care delivery in these areas is limited, resulting in care that is fragmented, uncoordinated, and characterized by unnecessary duplication of services. In Elliott S. Fisher’s paper, small-area analyses reveal that differences in care delivery explain almost all of the geographic varia- tions in spending across the United States, and that higher-spending regions of the country perform worse in measures of technical quality than regions that spend less money. When there is strong medical evidence, physicians tend to agree on courses of treatment across regions of different spending levels. Building an evidence base for areas in which physicians currently 49 

50 EVIDENCE-BASED MEDICINE use their own discretion, such as the comparative effectiveness of various treatment options, or decisions about how often to see a patient with well- controlled hypertension and when to order certain medical tests, could greatly improve the quality of care and reduce costs. In fact, if all regions adopted the practice patterns of the most conservatively spending regions of the country, health outcomes could be significantly improved and U.S. healthcare spending could decline by as much as 30 percent. Peter R. Orszag’s presentation illustrates the serious consequences that excessive healthcare spending poses to the nation’s economic well-being. If spending growth trends continue as they have over the past four decades, by 2050 Medicare and Medicaid spending will account for 20 percent of the total U.S. economy. Although often ascribed to effects of an aging popula- tion, lower fertility rates, and longer life expectancies, this long-term fiscal challenge is driven almost entirely by excessive healthcare costs—or costs per beneficiary. Slowing overall healthcare cost growth without limiting access will require changes that impact the overall healthcare system and providing better evidence to inform decision making will be an important first step. Comparative effectiveness research that draws upon the emerging electronic health record and clinical registry data resources may be the only cost-effective and feasible mechanism for bringing about the evidence-base expansion needed. Real gains in improving the quality of health care and reducing costs will come when the evidence of medical effectiveness is tied to incentive payments for healthcare providers. A combination of increased cost sharing on the consumer side combined with changes in the incentive system for providers informed by best evidence offers an important oppor- tunity to substantially reduce healthcare costs and improve quality. HEALTH CARE AND THE EVIDENCE BASE Elliott S. Fisher, Dartmouth Medical School The U.S. healthcare system faces serious challenges and the Institute of Medicine (IOM) has played a critical role in calling for fundamental transformation of the delivery system to achieve the vision of a patient- centered, high-quality, equitable, and effective delivery system (IOM, 2001, 2006). There is also a growing recognition that our current delivery system is failing to deliver on the promise of improved health offered by advances in biomedical knowledge, and the future pace of change may widen this gap substantially. The IOM Roundtable on Evidence-Based Medicine was established “to help transform the way evidence on clinical effectiveness is generated and used to improve health and health care” (IOM, 2007a). The key notions are to provide better evidence about the risks and benefits of interventions and 

THE NEED FOR BETTER MEDICAL EVIDENCE 51 to support better application of that knowledge to clinical practice. Several recent reports highlight the growing consensus on the need for expanded support for comparative effectiveness research that provides better informa- tion about the risks and benefits of specific treatments (IOM, 2007b). This paper draws on the traditions of small-area analysis to under- score the scope of the challenge faced in bringing evidence to bear on current practice and to point to the opportunity for improving both the costs and the quality of care by ensuring a broad definition of the need for evidence. Categories of Care: Biologically Targeted Interventions Versus Care Management Strategies As we consider the relationship between evidence and clinical practice, it is worth considering two broad categories of interventions: discrete, bio- logically targeted interventions and care delivery strategies. Biologically targeted interventions are focused on a specific anatomic problem or disease process. Examples include the decision about whether to adopt a specific screening test for cancer or whether to treat a patient with prostate cancer with surgery or radiation therapy. Such interventions can be well specified not only in terms of the underlying anatomic or physiologic problem to be addressed, but also in terms of the expected intermediate and long-term outcomes and how these vary across clinical subgroups. Many of the dramatic improvements in health outcomes achieved over the past decades are a result of the advances in biomedical knowledge and the devel- opment of such biologically targeted interventions. These are the traditional focus of technology assessment, clinical guidelines, and a narrow definition of “evidence-based practice.” A second category of “decisions,” which I refer to as care delivery strate- gies, is rarely considered explicitly in the day-to-day practice of clinical medi- cine. This category refers not to what care is provided (what drug, what device, what surgical procedure) or to whom (which patients should be offered the intervention), but to how a specific biologically targeted therapy is delivered: who should provide the care (patients themselves, advanced practice nurses, primary care physicians, or specialists); where care should be delivered (home, outpatient facility, or hospital); and how intensively patients should be moni- tored and reevaluated. Questions about care delivery also encompass system- and policy-level issues, such as how care should be organized, what kinds of resources should be deployed, how care should be paid for and financed, and how to improve the quality of care. There are three reasons to distinguish these two categories of decisions. First, the effectiveness of many (if not most) discrete biologically targeted interventions can depend critically on how care is provided: risk-adjusted 

52 EVIDENCE-BASED MEDICINE surgical mortality rates, for example, vary severalfold across hospitals (Birkmeyer et al., 2002; O’Connor et al., 1991). Second, as discussed below, regional and provider-specific differences in costs are largely due to differ- ences in the intensity of care delivery: addressing the rapidly rising costs of care will require much better evidence about how to organize and deliver care effectively and efficiently. Finally, I argue that the infrastructure re- quired to provide better evidence on discrete, biologically targeted interven- tions is fundamentally the same as the infrastructure required to improve care delivery. As we build our capacity to improve evidence, we should be careful to address the need for better evidence along both dimensions. Current Practice and the Evidence Base: Biologically Targeted Interventions The recent IOM workshop on evidence-based medicine highlighted the many limitations of the current evidence base, focusing primarily on the challenges surrounding biologically targeted therapies (IOM, 2007a). Highlights include the lack of any evidence on the efficacy or effectiveness of many interventions, the difficulty of extrapolating from trials carried out on selected populations to those with multiple chronic conditions, and the growing recognition that the benefits of interventions vary according to the underlying risk of the population: trials that show benefit for the aver- age patient may not reveal that many lower-risk patients may be harmed by receiving the procedure while those at greater risk receive substantial benefits. The growing recognition of the importance of comparative effec- tiveness research can be attributed to the increasing attention focused on these issues. The relative magnitude of the uncertainty surrounding the use of se- lected, discrete, biologically targeted therapies can be illustrated by the regional variations in rates of these services among the Medicare population (Figure 2-1). We have found it useful to distinguish effective care (treat- ments where the evidence of benefit is strong and no trade-offs among benefits and harms are involved) from preference-sensitive care (treatments where patients’ values about the different outcomes may vary) (Wennberg et al., 2002). An example of the former would be hospitalizations for hip fracture: the diagnosis is straightforward and the therapy (inpatient surgi- cal repair of the fracture) is required. Variations in utilization rates are due entirely to underlying variations in the incidence of the disease. Examples   The importance of ensuring that care is aligned with patients’ well-informed preferences applies not only to discrete, biologically targeted interventions, but also to care delivery strategies. 

THE NEED FOR BETTER MEDICAL EVIDENCE 53 5.0 Standardized ratio (log scale) 1.0 0.2 Hip fracture PSA Cardiac PCI Radical discharges catheterization prostatectomy Coefficient of variation 12.7 25.2 28.0 34.0 38.5 Interquartile ratio 1.19 1.40 1.44 1.43 1.71 Extremal ratio 3.95 4.33 4.20 11.50 8.69 FIGURE 2-1  Variation in utilization rates of specific, biologically targeted interventions. NOTE: Each dot represents the ratio of the rate of the specified intervention in one of the 306 U.S. Hospital Referral Regions to the U.S. average for that intervention 2-1.eps (log scale). All rates are calculated on an annual basis for fee-for-service Medicare enrollees age 65 and over. PSA refers to prostate-specific antigen testing at least once during the year. PCI refers to percutaneous coronary interventions. Data are from the Dartmouth Atlas of Health Care. of the latter would include screening for prostate cancer (where patient attitudes toward the risks of treatment must be weighed against the still unproven benefits of screening) or percutaneous coronary interventions for stable angina (where the modest benefit in terms of angina relief must be weighed against the lifelong need for anti-platelet therapy among other risks). When we look at common biologically targeted interventions—both diagnostic and therapeutic—we see dramatic variability across the United States. Addressing these variations will require not only better informa- tion about risks and benefits (comparative effectiveness research), but also ensuring that treatment decisions reflect the well-informed judgments of patients rather than the opinions of providers (O’Connor et al., 2007; Wennberg et al., 2007). 

54 EVIDENCE-BASED MEDICINE Current Practice and the Evidence Base: Care Delivery Strategies There are also marked differences across regions in the way care is delivered (Figure 2-2). Although virtually all Medicare beneficiaries have access to care (defined as at least one physician visit during the year) and there is thus little regional variation in the age-, sex-, or race-adjusted rate of at least one physician visit, we see marked variability in the use of other care delivery strategies. Because variation in use of these services is associated with the local capacity of the delivery system (how many physi- cians, how many hospital beds), we have long referred to these services as “supply-sensitive.” One of the fundamental reasons for distinguishing care delivery strat- egies from the use of biologically targeted interventions is their distinct 5.0 Standardized ratio (log scale) 1.0 0.2 At least Primary Medical 10 or more Medical Inpatient ICU days one visit care visits specialist MDs (L6M) discharges days (L6M) visits Coefficient of variation 5.1 14.2 33.6 35.9 18.7 20.5 41.7 Interquartile ratio 1.05 1.21 1.56 1.64 1.30 1.36 1.75 Extremal ratio 1.48 2.80 5.61 6.12 3.37 2.88 11.65 FIGURE 2-2  Variation in utilization rates of care delivery strategies. NOTE: Each dot represents the ratio of the rate of the specified service or strategy 2-2.eps in one of the 306 U.S. Hospital Referral Regions to the U.S. average for that service (log scale). Visits, medical discharges, and inpatient days are calculated on an annual basis for fee-for-service Medicare enrollees age 65 and over. Data for the proportion of enrollees seeing 10 or more physicians and for intensive care unit (ICU) days are for Medicare enrollees with chronic illness who are in their last 6 months of life (L6M). Data are from the Dartmouth Atlas of Health Care. 

THE NEED FOR BETTER MEDICAL EVIDENCE 55 relationship to variations in spending. Figure 2-3 displays the relationship between spending and the utilization rates of specific types of services across U.S. regions. Each dot represents the ratio of the utilization rates of the specific service in regions that fall in the highest quintile to the utiliza- tion rate in the lowest-spending quintile of regional per capita Medicare spending. To control for potential differences in the underlying health status of populations across regions, these analyses are based on long-term follow-up of patients initially hospitalized with hip fracture, colon cancer, or acute myocardial infarction (Fisher et al., 2003a). Higher spending is not associated with greater use of biologically targeted interventions: whether these are treatments that all patients should receive (effective care) or in- Utilization ratio of highest to lowest quintile 0.5 1.00 1.5 2.0 2.5 3.0 Discrete: strong evidence Reperfusion in 12 hours (Heart attack) Aspirin at admission (Heart attack) Mammogram, Women 65-69 Pap Smear, Women 65+ Pneumococcal Immunization Discrete: major discretionary procedures Total Hip Replacement Total Knee Replacement Back Surgery CABG following heart attack Care management: who / how often / where Total Inpatient Days Inpatient Days in ICU or CCU Evaluation and Management (visits) Imaging Diagnostic Tests 0.5 1.00 1.5 2.0 2.5 3.0 Use Rate Lower in High-Spending Regions Use Rate Higher in High-Spending Regions FIGURE 2-3  Ratio of utilization rates for selected specific services among cohorts of Medicare beneficiaries in high- 2-3 new.eps versus low-spending regions. NOTE: High- and low-spending regions were defined as the U.S. Hospital Referral Regions in the highest and lowest quintiles of per capita Medicare spending. Data for mammograms, Pap smears, and pneumococcal immunizations were ascertained from a representative sample of the Medicare population. Data for all other utiliza- tion rates reflect either acute care for patients with heart attacks (reperfusion and aspirin administration) or the weighted average of utilization rate ratios during one year follow-up after initial hospitalization for acute myocardial infarction, hip fracture, or colorectal cancer. All data are from Fisher and colleagues (Baicker et al., 2007). 

56 EVIDENCE-BASED MEDICINE terventions where patients’ judgments about how they value the risks and benefits should determine the treatment choice (preference-sensitive care). Higher spending, however, is largely due to differences in care delivery: how frequently patients are seen (evaluation and management services), how much time they spend in the hospital, and the intensity with which they are monitored (diagnostic tests and imaging). Spending, the Intensity of Care Delivery, and Health Outcomes The critical question underlying the variations in practice and spending is their relationship to health outcomes. Over the past 10 years, a number of studies have explored the relationship between higher spending and the quality and outcomes of care (Table 2-1). Patients’ Experiences and Outcomes Whether the study was carried out at the state level (Baicker and Chandra, 2004), across hospital referral regions (Fisher et al., 2003a), or across the major academic medical centers within the United States (Fisher et al., 2004), a consistent pattern is found: the quality of care as reflected in process measures of care is worse when spending—and the intensity of care delivery—is greater. Among patients hospitalized with hip fractures, colon cancer, and acute myocardial infarction who were followed for up to five years, mortality rates in higher-spending regions and hospitals were no better or slightly worse than in lower-spending delivery systems (Fisher et al., 2003a). In regions where spending growth was greatest, survival following myocardial infarction improved more slowly than in regions where spending growth was slower (Skinner et al., 2006). Finally, Medicare beneficiaries’ overall satisfaction with care was no better in higher-spending regions and their perceptions of the accessibility of care were somewhat worse (Fisher et al., 2003a). Physician Attributes, Practice Settings, and Perceptions of Care On a per capita basis, the highest-spending quintile of hospital refer- ral regions have 65 percent more medical specialists per capita, 75 percent more general internists, and 25 percent fewer family practitioners than the lowest-spending quintile. A substantially higher proportion of physicians are foreign medical graduates, fewer are board certified, and they are much more likely to practice in small groups than physicians in lower-spend- ing regions (Sirovich et al., 2006). When surveyed, physicians in higher- spending regions are more likely to report that the continuity of care with their patients is inadequate to support high-quality care and that the quality 

THE NEED FOR BETTER MEDICAL EVIDENCE 57 TABLE 2-1  Relationship Between Regional Differences in Spending and the Content, Quality, and Outcomes of Care Higher-Spending Regions Compared to Lower-Spending Onesa Healthcare •  er capita supply of hospital beds 32% higher (Fisher et al., 2003a) P resources •  er capita supply of physicians 31% higher overall: 65% more P medical specialists, 75% more general internists, 29% more surgeons, and 26% fewer family practitioners (Fisher et al., 2003a) Content and •  dherence to process-based measures of quality lower (quality worse) A quality of care •  ittle difference in rates of major elective surgery (Fisher et al., 2003a; L Wennberg et al., 2002) •  ore hospital stays, physician visits, specialist referrals, imaging, and M minor tests and procedures (Fisher et al., 2003a) Health •  ortality up to 5 years slightly higher following acute myocardial M outcomes infarction, hip fracture, and colorectal cancer diagnosis (Fisher et al., 2003a) • No difference in functional status (Fisher et al., 2003a) Physician •  ore likely to report poor communication among physicians (Sirovich M perceptions of et al., 2006) quality •  ore likely to report inadequate continuity of patient care (Sirovich M et al., 2006) •  reater difficulty obtaining inpatient admissions or high-quality G specialist referrals (Sirovich et al., 2006) Patient-reported •  orse access to care and greater waiting times (Fisher et al., 2003a) W quality of care •  o difference in patient-reported satisfaction with care (Fisher et al., N 2003a) Trends over •  lthough all U.S. regions experienced improvements in acute A time myocardial infarction survival between 1986 and 2002, regions with greater growth in spending had smaller gains in survival than those with lower growth in spending (Skinner et al., 2006) aHigh- and low-spending regions were defined as the U.S. Hospital Referral Regions in the highest and lowest quintiles of per capita Medicare spending as in Fisher et al. (2003a). of communication is insufficient to support high-quality care. In spite of the substantially greater per capita supply of both beds and specialists, physi- cians in higher-spending regions are more likely to perceive scarcity: they are more likely to report that it is difficult to get a patient into the hospital and that it is hard to obtain adequate medical specialist referrals. These findings are consistent with the hypothesis that the lower- spending regions represent a reasonable benchmark of efficiency. In fact, if all U.S. regions could safely adopt the organizational structures and prac- tice patterns of the lowest-spending regions of the United States, Medicare spending would decline by about 30 percent (Fisher et al., 2003a; Wennberg et al., 2002). While it may not be realistic to reduce spending by that 

58 EVIDENCE-BASED MEDICINE amount, the magnitude of the differences in practice and the fact that the differences in spending are largely due to differences in care delivery point to an important opportunity: improving efficiency will require attention not only to the comparative effectiveness of biologically targeted interven- tions, but also to addressing the underlying causes of the differences in care delivery across regions and systems. Underlying Causes of the Differences in Care Delivery: Evidence and Theory The Evidence A number of studies have explored the underlying causes of the re- gional differences in spending and the intensity of care delivery. Patients’ preferences for care vary slightly across regions, but not enough to explain the magnitude of spending differences seen. For example, Medicare ben- eficiaries in high-spending regions are no more likely to prefer aggressive end-of-life care than those in low-spending regions (Barnato et al., 2007; Pritchard et al., 1998). Differences in the malpractice environment are asso- ciated with differences in both practice and spending, but explain less than 10 percent of state-level differences in spending and have a comparably small impact on differences in the growth in spending across states (Baicker et al., 2007; Kessler and McClellan, 1996). The role of capacity is clearly important, but the hospital bed supply and physician supply combined explain less than 50 percent of the difference in spending across regions (Fisher et al., 2004). The most recent studies have focused on the use of clinical vignettes to explore how physicians’ judgments vary across regions of differing spend- ing levels. These studies have found that physicians in higher-spending re- gions were no more likely to intervene in cases where evidence was strong (such as chest pain with an abnormal stress test), but were much more likely to recommend discretionary treatments (such as more frequent visits, refer- ral to a specialist, or use of imaging services) than those in low-spending regions (Sirovich et al., 2005). A Likely Diagnosis: Capacity, Payment, and Clinical Culture These findings suggest a likely explanation for the dramatic differ- ences in spending across regions and the paradoxical finding that higher spending seems to lead to worse quality and worse outcomes. Current clinical evidence is an important, but limited, influence on clinical decision making. Most physicians practice within a local organizational context and policy environment that profoundly influences their decision making, 

THE NEED FOR BETTER MEDICAL EVIDENCE 59 especially in discretionary clinical settings. Hospitals and physicians each face incentives that will in general reward expansion of capacity (especially for highly reimbursed services) and recruitment of additional procedure- oriented specialists. When there are more physicians relative to the size of the population they serve, physicians will see their patients more frequently. When there are more specialists or hospital beds available, primary care physicians and others will learn to rely upon those specialists and use those beds. (It is more efficient from the primary care physician’s perspective to refer a difficult problem to a specialist or to admit a patient to the hospital than to try to manage the patient in the context of an office visit for which payments have become relatively constrained.) The consequence is that what—given the state of current evidence—are “reasonable” individual clinical and policy decisions lead in aggregate to higher utilization rates, greater costs, and inadvertently, worse quality and worse outcomes. The key element of this theory is that because so many clinical decisions are in the “gray areas” (how often to see a patient, when to refer to a specialist, when to admit to the hospital), any expansion of capacity will result in a subtle shift in clinical judgment toward greater intensity. Implications for Evidence Development These findings and their likely explanation point to the need for much better evidence. We need evidence about the risks and benefits of discrete, biologically targeted interventions and how these risks and benefits vary across different subgroups of the population, especially those often excluded from current randomized trials (IOM, 2007b), but we also need much better evidence about care delivery. No matter how good our clinical evidence about specific interventions becomes, many—if not most—clinical decisions will still require judgment. Also, because there will always be gray areas, we will need evidence that can guide clinicians, administrators, and policy makers when they are making decisions about care delivery. Although the need for evidence may appear overwhelming, an im- portant opportunity lies in recognizing that the information systems and analytic approaches required to improve the evidence base for biologically targeted interventions and for improving care delivery are fundamentally the same (Table 2-2). In the ideal world of improved information systems and electronic records that might allow relatively routine assessment of both short- and long-term health outcomes and effective follow-up of pa- tients, the capacity to evaluate both care delivery and biologically targeted interventions would be critical, at least in part because lack of information about the local context (delivery system attributes) would sharply limit our ability to properly interpret studies of biologically targeted interventions. 

60 EVIDENCE-BASED MEDICINE TABLE 2-2  Relationship Between the Information and Approaches Required to Improve the Evidence Base Around Biologically Targeted Interventions and Care Delivery Discrete, Biologically Targeted Interventions Approaches to Care Delivery Example of research How effective are How should primary care question endovascular carotid artery offices be organized to provide stents in the prevention of care to patients with heart stroke? failure? Outcomes of interest Survival, functional status, Survival, functional status, quality of life, total costs quality of life, total costs Comparison of interest Carotid stent vs. medical Offices meeting criteria for therapy “medical home” vs. other primary care offices Important patient-level Age, sex, race, severity Age, sex, race, severity factors required for either of illness, comorbidities, of illness, comorbidities, adjustment or stratification socioeconomic status socioeconomic status Contextual factors Attributes of care delivery Attributes of care delivery required for adjustment or system system stratification Applicable methods Randomized trials Randomized trials and/or population-based and/or population-based observational studies observational studies Key notion Compare biologically Compare care delivery targeted interventions, strategies, while accounting while accounting for for patient and contextual patient and contextual factors factors Moving Forward: A Challenge to Academic Medicine The critical importance of healthcare spending to the future financial health of the U.S. government and the economy in general has received growing attention (Orszag and Ellis, 2007). Our capacity to provide af- fordable healthcare coverage to the U.S. population and our ability to pay for the new biologically targeted interventions that are under development will clearly depend not only on the costs of the interventions but also on the costs of delivering those interventions. Academic medicine—and the federal agencies that provide research support—have largely focused on improving our understanding of disease biology, while ignoring the need to understand and address the dramatic variations in care delivery among academic medical centers (Wennberg et al., 1987). 

THE NEED FOR BETTER MEDICAL EVIDENCE 61 Table 2-3 points to the magnitude of the opportunity—and the challenge—for academic medicine. The upper portion of the table focuses on the degree to which each of these five members of the U.S. News and World Report’s “Honor Roll” of academic medical centers is able to deliver proven clinical interventions to eligible patients during an acute inpatient stay. The lower portion of the table highlights the differences in spending and overall intensity of care. The specific data focus on care provided in the last six months of life, but these patterns of practice are highly predictive of how these institutions treat other seriously ill patients. All five provide TABLE 2-3  Performance of Selected Major Academic Medical Centers on Measures of Adherence to Biologically Targeted Treatments and the Intensity of Care Delivery UCLA Johns Massachusetts Cleveland Mayo Clinic Medical Hopkins General Clinic (St. Mary’s Center Hospital Hospital Foundation Hospital) Provision of discrete, biologically targeted evidence- based interventions Composite quality 81.5 84.3 85.9 89.2 90.4 score on measures of inpatient technical quality Spending and care delivery for patients with serious chronic illness during last 6 months of life Medicare spending 50,522 43,363 40,181 28,077 26,330 Physician visits 52.1 29.8 42.2 32.2 23.9 Hospital days 19.2 17.1 17.7 14.6 12.9 Intensive care days 11.4 4.3 2.8 3.5 3.9 % admitted to 26.1 31.5 19.6 34.2 25.5 hospice % seeing 10 or 57.7 44.3 54.6 46.8 43.0 more physicians NOTES: Hospitals were selected for inclusion because they were ranked as the top five aca- demic medical centers on the U.S. News and World Report’s 2007 “Honor Roll.” Utilization data are for 1999-2003. Composite quality score was calculated from CMS data for 2005, which are from the Dartmouth Atlas of Health Care. 

62 EVIDENCE-BASED MEDICINE high-quality inpatient care. The differences in care delivery, however, are substantial: patients at the University of California, Los Angeles, have twice as many visits, spend about 50 percent more time in the hospital, and cost about twice as much as those treated at the Mayo Clinic in Rochester, Min- nesota, or the Cleveland Clinic. If all U.S. delivery systems could achieve the apparent efficiency of a Mayo or a Cleveland Clinic, the resources available to expand coverage to the uninsured or to provide interventions of proven benefit to those who are not be able to afford them would be substantial. Failure to address this challenge would call into question not only the scientific integrity of the enterprise (Are we really committed to asking important questions?), but also our moral authority as healthcare providers (How can we continue to ignore obvious opportunities to improve quality and the future affordability of care?). Academic medicine has the opportunity to lead the development of a learning healthcare system. Such an effort should include a focus not only on the science of disease biology and improving the evidence to support the use of biologically targeted interventions, but also on the sciences of clinical practice and the evidence to support improvements in care delivery (Wennberg et al., 2007). THE HIGH PRICE OF THE LACK OF EVIDENCE Peter R. Orszag, Congressional Budget Office The nation’s long-term fiscal challenge has largely been misdiagnosed in popular descriptions. It typically is described as being driven mostly by the aging of baby boomers, with lower fertility rates and longer life expectancy causing most of the long-term budget problem. In fact, most of that long- term problem is driven by excess healthcare cost growth—that is, the rate at which healthcare costs grow compared to income per capita. In other words, it is the rising cost per beneficiary, rather than the number of benefi- ciaries, that explains the bulk of the nation’s long-term fiscal problem. You can see this phenomenon arising even over the next decade: Fig- ure 2-4 shows the Congressional Budget Office’s (CBO’s) projections for spending on Social Security, Medicare, and Medicaid through 2017. As the figure shows, Social Security rises by about 0.5 percentage points of gross domestic product (GDP), from 4.2 percent of GDP to 4.8 percent over that period. Medicare and the federal share of Medicaid rises from 4.6 percent of GDP to 5.9 percent of GDP—an increase of 1.3 percentage points of GDP, or roughly twice as much as Social Security even over the next decade. If you look over longer periods of time, the basic point is accentuated. 

THE NEED FOR BETTER MEDICAL EVIDENCE 63 7 Medicaid 6 Medicare 5.9 Social Security 5 1.9 4.8 4.6 4.2 4 1.4 3 4.8 2 4.2 4.0 3.2 1 0 2007 2017 FIGURE 2-4  Spending on Medicare and Medicaid and on Social Security as a per- centage of GDP, 2007 and 2017. 2-4.eps Figure 2-5 shows a simple extrapolation in which Medicare and Medicaid costs continue to grow at the same rate over the next four decades as they did over the past four. (Even with no change in federal policy, there are reasons to believe that this simple extrapolation may overstate future cost growth in Medicare and Medicaid. CBO has recently released a long-term health outlook that presents a more sophisticated approach to projecting Medicare and Medicaid costs under current law, but a straight historical extrapolation is shown here for simplicity.) Under that scenario, Medicare and Medicaid would rise from 4.6 percent of the economy today to 20 percent of the economy by 2050; 20 percent of GDP is the entire size of the federal government today. The most interesting part of Figure 2-5 is the bottom line, which iso- lates the pure effect of demographics on those two programs. The only reason that the bottom line is rising is that the population is getting older and there are more beneficiaries on the two public programs. The increase between today and 2050 in that bottom dotted line shows that aging does indeed affect the federal government’s fiscal position. Yet that increase is much smaller than the difference in 2050 between the bottom line and the top line. In other words, the rate at which healthcare costs grow—whether they continue to grow at 2.5 percentage points per year faster than per capita income, or 1 percentage point, or 0.5 percentage point, is to a 

64 EVIDENCE-BASED MEDICINE 25 Actual Projection Differential of: 2.5 Percentage Points Percent of Gross Domestic Product 20 1 Percentage Point Zero 15 10 5 0 1966 1972 1978 1984 1990 1996 2002 2008 2014 2020 2026 2032 2038 2044 2050 FIGURE 2-5  Total federal spending for Medicare and Medicaid under assumptions 2-5 new.eps about the health cost growth differential. SOURCE: Congressional Budget Office, 2007. first approximation the central long-time fiscal challenge facing the United States. It is common to say that the sooner we act the better off we are, and just to calibrate that, Figure 2-6 shows that if we slowed healthcare costs growth from 2.5 percentage points to 1 percentage point starting in 2015— which would be extremely difficult if not impossible to do, but is helpful as an illustration—the result in 2050 would be a reduction of 10 percent GDP in Medicare and Medicaid expenditures for the federal government relative to no slowing in the cost growth rate. That 10 percent of GDP difference is half of what the federal government spends today. All of this may seem pretty challenging and is further complicated by the fact that it is implausible that we will slow Medicare and Medicaid growth in a sustainable way unless overall healthcare spending also slows. The reason is that if all you did was, say, to reduce payment rates under Medicare and Medicaid, and then try to perpetuate that over time without a slowing of overall healthcare cost growth, the result would likely be substantial access problems that would be inconsistent with the underlying premise and public understanding of these programs. One therefore needs to think about changes to Medicare and Medicaid in terms of the impact they can have on the overall healthcare system. From that perspective, there appears to be a very substantial opportu- nity embedded in this long-term fiscal challenge facing the United States: 

THE NEED FOR BETTER MEDICAL EVIDENCE 65 25 Actual Projection Differential of: Percent of Gross Domestic Product 20 2.5 Percentage Points 1 Percentage Point Beginning in 2025 15 1 Percentage Point Beginning in 2015 10 5 0 1966 1972 1978 1984 1990 1996 2002 2008 2014 2020 2026 2032 2038 2044 2050 FIGURE 2-6  Effects of slowing the growth of spending for Medicare and Medicaid. SOURCE: Congressional Budget new 2-6.eps Office, 2007. the possibility of taking costs out of the system without harming health. Perhaps the most compelling evidence underscoring this opportunity is the significant variations across different parts of the United States that do not translate into differences in health quality or health outcomes (Fig- ure 2-7). The question then becomes, why is this happening? To me, it appears to be a combination of two things. One is the lack of information specifi- cally about what works and what does not. The second thing is a payment system, on both the provider and the consumer sides, that accommodates the delivery of low-value or negative-value care. On the consumer side, despite media portrayals to the contrary, the share of healthcare expenditures paid out of pocket—which is basically the relevant factor for evaluating the degree to which consumers are faced with cost sharing—has plummeted over the past few decades, from about 33 percent in 1975 to 15 percent today (Figure 2-8). All available evidence suggests that lower cost sharing increases healthcare spending overall, and collectively we all pay a higher burden, although the evidence is somewhat mixed on the precise magnitude of the effect by which lower cost sharing raises overall spending. This observation leads some analysts to argue that the way forward is more cost sharing and a health savings account approach, and this can indeed help to reduce costs. However, two things need to be kept in mind in evaluating this approach. The first is that a significant amount of cost 

66 EVIDENCE-BASED MEDICINE $7,200 to 11,600 (74) 6,300 to < 6,800 (55) 4,500 to < 5,800 (72) 6,800 to < 7,200 (45) 5,800 to < 6,300 (60) Not Populated FIGURE 2-7  Medicare spending per capita in the United States, by hospital referral region, 2003. new 2-7.eps NOTE: The numbers in parentheses indicate the number of regions in each group. SOURCE: The Dartmouth Atlas Project, 2003. sharing is involved in existing plans. Moving to universal health savings ac- counts would thus not entail as great an increase in cost sharing, and there- fore as much a reduction in spending, as you might think. Second, there is an inherent limit to what we should expect from increased consumer cost sharing because healthcare costs are so concentrated among the very sick. For example, the top 25 percent most expensive Medicare beneficiaries ac- count for 85 percent of total costs, and the basic fact that healthcare costs are very concentrated among a small share of the population is replicated in Medicaid and in the private healthcare system. To the extent that we in the United States want to provide insurance, and insurance is supposed to provide coverage against catastrophic costs, the fact that those catastrophic costs are accounting for such a large share of overall costs imposes an in- herent limit to the traction that one can obtain from increased consumer cost sharing. In sum, increased cost sharing on the consumer side can help to reduce costs, but it seems very unlikely to capture the full potential to reduce costs without impairing health quality. This leads us to the provider side. On the provider side, the accumu- lation of additional information and changes in incentives could improve efficiency in the delivery of health care. There is growing interest in com- parative effectiveness research, and the original House version of the State 

THE NEED FOR BETTER MEDICAL EVIDENCE 67 35 30 33% 25 26% 20 Percent 15 17% 15% 10 13% 5 0 1975 1985 1995 2005 2015 FIGURE 2-8  Share of personal healthcare expenditures paid out of pocket. new 2-8.eps Children’s Health Insurance Program legislation had some additional fund- ing for comparative effectiveness research. Other policy makers seem inter- ested in expanding comparative effectiveness research. We need, though, to ask some hard questions about what we mean by comparative effectiveness research and how it would be implemented. The key issues are what kind of research is undertaken and the stan- dard of evidence used. As Mark McClellan has noted, comparative effec- tiveness research will very likely have to rely on nonrandomized evidence. The reason is that it seems implausible that we could build out the evidence base across a whole variety of different clinical interventions and practice norms using only randomized control trials, especially if we want to study subpopulations. On the other hand, economists have long been aware of the limitations of panel data econometrics, where one attempts to control for every possible factor that could influence the results—typically, that attempt is far from perfectly successful. There is thus a tension between using statistical techniques on panel data sets (of electronic health records, insurance claims, and other medical data), which seems to be the only cost- effective and feasible mechanism for significantly expanding the evidence base, and the inherent difficulty of separating correlation and causation in such an approach. In terms of the budgetary effects of comparative effectiveness research, a lot depends on both what is done and how it is implemented. If the effort only involves releasing the results of literature surveys, the effects would likely be relatively modest. If new research using registries or analysis of 

68 EVIDENCE-BASED MEDICINE electronic health records is involved, there may be somewhat larger ef- fects. The real traction, though, will come from building the results of that research into financial incentives for providers. In other words, if we move from a “fee-for-service” system to a “fee-for-value” one, where higher-value care is awarded stronger financial incentives and low-value or negative-value health care is penalized by smaller incentives, or perhaps even penalties, the effects would be maximized. The design of such a system is very complicated and difficult to implement, but this is where the greatest long-term budgetary savings could come. In conclusion, it is plausible to me that the combination of some in- creased cost sharing on the consumer side and a substantially expanded comparative effectiveness effort, combined with changes in the incentive system for providers, offers the nation the most auspicious approach to capturing the apparent opportunity to reduce healthcare costs at minimal or no adverse consequences for health outcomes. The focus of this publication is thus central to addressing the nation’s long-term fiscal challenge. REFERENCES Baicker, K., and A. Chandra. 2004. Medicare spending, the physician workforce, and beneficia- ries’ quality of care. Health Affairs (Millwood) Suppl Web Exclusives:W4-184-W4-197. Baicker, K., E. S. Fisher, and A. Chandra. 2007. Malpractice liability costs and the practice of medicine in the Medicare program. Health Affairs (Millwood) 26(3):841-852. Barnato, A. E., M. B. Herndon, D. L. Anthony, P. M. Gallagher, J. S. Skinner, J. P. Bynum, and E.S. Fisher. 2007. Are regional variations in end-of-life care intensity explained by patient preferences?: A study of the U.S. Medicare population. Medical Care 45(5):374-376. Birkmeyer, J. D., A. E. Siewers, E. V. Finlayson, T. A. Stukel, F. L. Lucas, I. Batista, H. G. Welch, and D. E. Wennberg. 2002. Hospital volume and surgical mortality in the United States. New England Journal of Medicine 346(15):1128-1137. Congressional Budget Office. 2007. Research on the comparative effectiveness of medical treatments: Options for an expanded federal role. Testimony by Director Peter R. Orszag before House Ways and Means Subcommittee on Health. http://www.cbo.gov/ftpdocs/ 82xx/doc8209/Comparative_Testimony.pdf (accessed June 26, 2008). The Dartmouth Atlas Project. 2003 (unpublished). Medicare spending per capita in the United States, by hospital referral region. Dartmouth College. Fisher, E. S., D. E. Wennberg, T. A. Stukel, D. J. Gottlieb, F. L. Lucas, and E. L. Pinder. ������� 2003a. The implications of regional variations in Medicare spending. Part 1: The content, qual- ity, and accessibility of care. Annals of Internal Medicine 138(4):273-287. ———. 2003b. The implications of regional variations in Medicare spending. Part 2: Health outcomes and satisfaction with care. Annals of Internal Medicine 138(4):288-298. Fisher, E. S., D. E. Wennberg, T. A. Stukel, and D. J. Gottlieb. 2004. Variations in the lon- ���������������������������� gitudinal efficiency of academic medical centers. Health Affairs (Millwood) Suppl Web Exclusive:VAR19-32. IOM (Institute of Medicine). 2001. Crossing the quality chasm: A new health system for the 21st century. Washington, DC: National Academy Press. ———. 2006. Performance measurement: Accelerating improvement. Washington, DC: The National Academies Press. 

THE NEED FOR BETTER MEDICAL EVIDENCE 69 ———. 2007a. The learning healthcare system: Workshop summary. Washington, DC: The National Academies Press. ———. 2007b. Learning what works best: The nation’s need for evidence on comparative effectiveness in health care. http://www.iom.edu/ebm-effectiveness (accessed January 22, 2008). Kessler, D. P., and M. McClellan. 1996. Do doctors practice defensive medicine? Quarterly Journal of Economics 111:353-390. McGlynn, E. A., S. M. Asch, J. Adams, J. Keesey, J. Hicks, A. DeCristofaro, and E. A. Kerr. 2003. The quality of health care delivered to adults in the United States. New England Journal of Medicine 348(26):2635-2645. O’Connor, A. M., J. E. Wennberg, F. Legare, H. A. Llewellyn-Thomas, B. W. Moulton, K. R. Sepucha, A. G. Sodano, and J. S. King. 2007. Toward the “tipping point”: Decision aids and informed patient choice. Health Affairs (Millwood) 26(3):716-725. O’Connor, G. T., S. K. Plume, E. M. Olmstead, L. H. Coffin, J. R. Morton, C. T. Maloney, E. R. Nowicki, J. F. Tryzelaar, F. Hernandez, and L. Adrian. 1991. A regional prospec- tive study of in-hospital mortality associated with coronary artery bypass grafting. The Northern New England cardiovascular disease study group. JAMA 266(6):803-809. Orszag, P. R., and P. Ellis. 2007. The challenge of rising health care costs—a view from the Congressional Budget Office. New England Journal of Medicine 357(18):1793-1795. Pritchard, R. S., E. S. Fisher, J. M. Teno, S. M. Sharp, D. J. Reding, W. A. Knaus, J. E. ­Wennberg, and J. Lynn. 1998. Influence of patient preferences and local health system characteristics on the place of death. Support investigators. Study to understand prognoses and prefer- ences for risks and outcomes of treatment. Journal of the American Geriatrics Society 46(10):1242-1250. Sirovich, B. E., D. J. Gottlieb, H. G. Welch, and E. S. Fisher. 2005. Variation in the ten- ��������������������� dency of primary care physicians to intervene. Archives of Internal Medicine 165(19): 2252-2256. ———. 2006. Regional variations in health care intensity and physician perceptions of quality of care. Annals of Internal Medicine 144(9):641-649. Skinner, J. S., D. O. Staiger, and E. S. Fisher. 2006. Is technological change in medicine always worth it? The case of acute myocardial infarction. Health Affairs (Millwood) 25(2): w34-w47. Wennberg, J. E., J. L. Freeman, and W. J. Culp. 1987. Are hospital services rationed in new haven or over-utilised in Boston? Lancet 1(8543):1185-1189. Wennberg, J. E., E. S. Fisher, and J. S. Skinner. 2002. Geography and the debate over Medicare ��������������������������������������������� reform. Health Affairs (Millwood) Supp Web Exclusives:W96-W114. Wennberg, J. E., E. S. Fisher, J. S. Skinner, and K. K. Bronner. 2007. Extending the P4P agenda, part 2: How Medicare can reduce waste and improve the care of the chronically ill. Health Affairs 26(6):1575-1585.