|
Evidence-Based Medicine And Policy: The Case Of The Implantable Cardioverter Defibrillator
The implantable cardioverter defibrillator (ICD) is a costly new treatment for patients at high risk of sudden cardiac death. Randomized trials of the ICD showed it to be effective in some groups of patients but not in others. While new trials testing the ICD were ongoing to clarify the evidence, policymakers faced immediate decisions about providing insurance coverage for the device. The high cost of ICDs, the large population of patients potentially eligible to receive them, the potential to reduce preventable deaths, and the unsettled state of the medical evidence provided a challenge to evidence-based medicine and to policymakers.
Heart disease may be complicated at any moment by a sudden, unexpected, and fatal abnormality of heart rhythm. A rapid rhythm originating from the heart’s main pumping chamber (the left ventricle) may be initially organized as "ventricular tachycardia" (VT) but often degenerates into a disorganized, chaotic rhythm termed "ventricular fibrillation" (VF). A person who develops VF has no effective circulation of blood to the body, which leads to brain damage within minutes and death shortly thereafter. The risk of this catastrophe is a dominant issue in the management of patients with heart disease, and concern about it affects almost every major decision about treatment. Until the late 1950s development of VF was inevitably fatal. Physicians then discovered that an electrical discharge applied to the chest of the victim could break the chaotic rhythm and restore a normal heartbeat. The new technology of external cardiac defibrillation led directly to the development of the cardiac intensive care unit and modern management of critically ill patients. Although external defibrillation was a huge advance, a defibrillator and qualified operator are rarely present when a person develops cardiac arrest. Even with the development of emergency medical systems, 911 calls, and public training in cardiac resuscitation, a victim of cardiac arrest has less than a 5 percent chance of survival in the United States today. Given the limitations of responding to cardiac arrest, an innovative physician, Michel Mirowski, believed that the only effective strategy to combat sudden cardiac death would be to develop a cardiac defibrillator small enough to be permanently implanted within the patient. This device could constantly monitor the heart’s rhythm, quickly detect development of VF, and deliver a life-saving shock without need for outside intervention. After more than a decade of development, the first prototype was implanted in humans and dramatically demonstrated the effectiveness of this strategy for detecting and treating VF.1 Early devices were extremely bulky and required open-heart surgery to implant. However, once Mirowski had proved the concept of an implantable cardioverter defibrillator (ICD), newer and better devices were developed by reducing their size to the point that open-heart surgery was no longer necessary and by improving their capabilities to detect and treat serious ventricular arrhythmias. With increased ease and reduced risk of implantation, ICDs can now be considered for many patients at risk of cardiac arrest. Several device manufacturers now offer ICDs, which cost upward of $25,000 to implant.
One of the major developments in medicine during the past twenty years has been the recognition that randomized controlled trials (RCTs) are the best method to evaluate the effectiveness of therapy. The RCT is an experiment, and rigorous experiments are the hallmark of all branches of science. The need for evidence provided by RCTs has been underscored by dramatic examples in which treatments widely considered to be effective actually led to worse outcomes when formally tested. One of the most famous such reversals was the use of drugs to prevent sudden cardiac death in patients with a prior heart attack. This treatment was based on the observation that abnormal premature ventricular beats were strongly associated with sudden cardiac death. Anti-arrhythmic drugs can suppress premature beats, and most physicians believed that these drugs would therefore prevent sudden cardiac death. A large clinical trial showed that this belief was wrong: Anti-arrhythmic drug therapy actually doubled mortality in patients with a prior heart attack.2 This unexpected result, and others like it, makes many physicians wary of accepting new treatments without solid data from RCTs. This skepticism spawned the movement that has been termed "evidence-based medicine." The ICD appeared to be an exception to the general rule about needing a randomized trial to prove the effectiveness of therapy. Since all patients with VF die within minutes if not treated, when an ICD converts VF to a normal rhythm, it seemed clear to many observers that it must be an effective, life-saving therapy. Many physicians argued that RCTs of ICDs were not only unnecessary but also basically unethical. The counterargument in favor of RCTs of the ICD was based on the observation that many patients who are successfully converted from VF to a normal rhythm die shortly thereafter because of their underlying heart disease. If the ICD only postponed death by a few hours or days, it might not be really worthwhile. Indeed, it might be inhumane to convert a sudden and painless death from VF into a prolonged and miserable ordeal in intensive care. The arguments in favor of performing RCTs ultimately won the day, perhaps because of the earlier humbling experience with drug therapy to prevent sudden cardiac death. RCTs of the ICD were initiated in two distinct groups of patients. The first consisted of patients who had survived an episode of VT or VF and thus were at very high risk of sudden cardiac death due to recurrent arrhythmia. The second consisted of patients who had not had an episode of VT or VF but were judged to be at risk of sudden cardiac death because of preexisting heart disease such as a prior heart attack or heart failure. The first group has been termed "secondary prevention"; the second, "primary prevention" or "prophylactic ICD implantation."
Three major RCTs tested ICDs in the secondary prevention population: one in Germany, one in Canada, and one in the United States.3 These secondary prevention trials were consistent in showing a 28 percent reduction in all-cause mortality attributable to ICD implantation (Exhibit 1
In contrast, the trials of prophylactic ICD implantation have been more variable in their design and their results. One key study sponsored by the National Institutes of Health (NIH)—the Coronary Artery Bypass Graft Patch (CABG-Patch) trial—found no benefit of ICDs placed prophylactically at the time of CABG surgery in patients with a prior heart attack and extensive heart damage (as judged by a measure of the pumping capacity of the left ventricle—the ejection fraction—which was reduced below 36 percent compared with a normal level of 50 percent or greater).5 A second large, industry-sponsored study (the Multicenter Automatic Defibrillator Implantation Trial, or MADIT-I) found that prophylactic ICD implantation reduced mortality by 46 percent in patients with prior heart attack, a reduced ejection fraction, and an abnormal electrophysiologic study.6 These inconsistent results dampened enthusiasm for prophylactic ICD implantation until 2002, when a second large, industry-sponsored study (MADIT-II) was completed.7 This study was conducted in a much less selected population: patients who had a prior heart attack and a low ejection fraction (30 percent or less). The MADIT-II report of a 31 percent reduction in mortality spawned great interest and controversy regarding use of prophylactic ICDs. After MADIT-II, the evidence about the efficacy of prophylactic ICDs was interpreted broadly by ICD proponents and narrowly by ICD skeptics. A formal overview of the data found significant heterogeneity among trial results but an overall 28 percent lower risk of death in patients given a prophylactic ICD.8 Numerous editorials and commentaries weighed the evidence and drew variable conclusions. In 2002 the major cardiac professional societies rated prophylactic ICD use in patients with prior heart attack and reduced ejection fraction as a "Class IIA indication" (meaning that there is conflicting evidence or divergence of opinion, or both, but leaning toward use of the treatment).9 In 2004 additional studies were released testing the value of prophylactic ICD implantation. The largest study, the Sudden Cardiac Death in Heart Failure Trial (SCD-HeFT), which was sponsored by the NIH, found 23 percent lower total mortality in patients with heart failure and an ejection fraction of 35 percent or less.10 The weight of evidence has shifted gradually toward showing a significant reduction in mortality from prophylactic ICDs in patients with impaired functioning of the left ventricle.
ICDs are a very costly technology, with implantation costs usually exceeding $25,000 per patient. Even if the ICD is effective in reducing the risk of sudden death, it may do so at a cost that is simply unaffordable. The Stanford University Patient Outcome Research Team (PORT) on sudden cardiac death was funded by the Agency for Healthcare Research and Quality (AHRQ) to examine the effectiveness and cost-effectiveness of strategies to manage life-threatening arrhythmias in patients with heart disease. One of our main goals was to assess the economic outcomes of ICD use and determine the circumstances in which these devices would be both effective and cost-effective. We first documented the long-term cost of using an ICD because the device might "pay for itself" by reducing costly cardiac complications. We examined Medicare claims data of patients who received an ICD and of comparable patients who did not, and we found that the total medical costs of the ICD-treated patients remained roughly $30,000 higher throughout seven years of follow-up.11 We also reviewed the RCTs of ICDs that reported economic data, which were consistent in showing that little of the initial cost of an ICD ($25,000–$30,000) was recouped by reducing later costs of care. These studies confirmed that implanting an ICD would add greatly to the nation’s health care bill. The added expense of the ICD might be readily justified if it either prolongs life expectancy or improves quality of life. A formal cost-effectiveness analysis requires a projection of lifetime outcomes, so we developed a model to simulate the clinical course of patients who might benefit from an ICD.12 The model incorporated estimates of the rate of sudden cardiac death and of the effectiveness, risks, and costs of ICD treatment, as well as other parameters needed to simulate the "natural history" of patients with heart disease. Using this model, we projected that a patient who had survived a previous episode of VT or VF would live an average of 5.64 years if given an ICD, compared with an average of 4.95 years if given the best available medical therapy. We also projected that the lifetime medical costs would be higher after an ICD implantation ($88,400) than under medical management ($51,000). We estimated the cost-effectiveness of the ICD in this model of secondary prevention at $54,000 per year of life added, which is within the commonly cited benchmark range of $50,000–$100,000 per year of life added. We subsequently modified this model to estimate the cost-effectiveness of prophylactic ICD use in patients with a history of heart attack but no history of VT or VF.13 This group of patients is far larger and more heterogeneous with respect to their risk of subsequent sudden cardiac death. Depending on markers of risk (the level of left ventricular ejection fraction) and the effectiveness of the ICD in reducing sudden cardiac death, we estimated that the life expectancy of patients given an ICD would range between 6.70 and 11.82 years, whereas using optimal medical treatment without specific antiarrhythmic therapy, life expectancy would range between 6.07 and 11.44 years. Lifetime medical costs showed a similar broad range: $119,600–$152,600 if an ICD were used versus $70,100–$91,700 if best medical therapy were used. Cost-effectiveness ratios ranged very widely depending on the patients chosen: from $24,100 per life year added in high-risk patients to higher cost and worse life expectancy in low-risk patients. Therefore, a prophylactic ICD can be a good value for the money spent in properly selected groups of patients, yet a waste of resources in less-well-chosen groups of patients.
The ICD poses difficult issues for both public and private health insurers. It is an extremely costly yet potentially lifesaving device. It is also clear that the same device may be used in very different groups of patients. This is not a black-and-white situation, but rather one with many shades of gray. The decision to pay for ICD implantation in patients who have survived a previous cardiac arrest was not terribly difficult or controversial. There was consistent evidence from three RCTs that the ICD is effective in this situation, and the target population was limited in size, since so few people survive an initial cardiac arrest. Coverage was readily provided for ICD use for this indication. In contrast, prophylactic ICD implantation poses a very difficult coverage decision. There are millions of U.S. patients with heart disease, and tens of millions at risk of developing heart disease suddenly. The stakes are huge, and the evidence to guide decisions has been less clear-cut. Until 2002 the evidence supported only very narrow indications for prophylactic ICD use. This changed with the release of the large MADIT-II study showing a benefit of prophylactic ICDs in patients with a prior heart attack who had a left ventricular ejection fraction of 30 percent or less. Device manufacturers requested that the Centers for Medicare and Medicaid Services (CMS) cover ICD implantation in this population. With more than 500,000 heart attacks occurring each year in the United States, a favorable coverage decision could easily add several billion dollars to annual health care costs. In early 2003 the Medicare Coverage Advisory Committee (MCAC) held a public hearing and reviewed an independent, unpublished analysis of the MADIT-II data by an outside consultant. Two major issues were identified: (1) Evidence about the efficacy of prophylactic ICDs was heterogeneous, and large clinical trials (particularly SCD-HeFT) were still ongoing that might clarify the issue; and (2) the evidence showed uneven benefit from prophylactic ICDs across different subgroups of patients. The MCAC recommended that the CMS cover ICD implantation in patients who would have been eligible to participate in the MADIT-II trial. The CMS considered this advice for several months and in June 2003 decided to cover ICD implantation for a narrower indication than recommended by the MCAC: patients with an ejection fraction of 30 percent or less and delayed electrical conduction through the left ventricle. The CMS’s national coverage decision regarding ICDs was widely criticized. The decision was particularly difficult in view of the large financial stakes on the one hand and the incomplete state of the pertinent evidence base on the other. The inconsistent results of the major clinical trials meant that professional consensus had not yet been reached on this indication for use of prophylactic ICDs. Nevertheless, the CMS could not ignore the evidence from MADIT-II and deny all coverage. The coverage criteria chosen, however, were based on unpublished analyses, and they presumed that the entire benefit of prophylactic ICDs in MADIT-II derived from the effects in a subgroup of patients. The clinical research community is very skeptical of subgroup analyses, however, which they consider to be post hoc fishing expeditions. (A major clinical trials group ridiculed subgroup analyses by publishing an analysis showing that heart attack treatment was ineffective for patients with the astrologic birth signs of Gemini or Libra but highly effective for patients with other birth signs.)14 The CMS could have avoided the appearance of cherry-picking subgroups and followed the MCAC’s advice by covering ICDs only for patients meeting the exact inclusion and exclusion criteria for the MADIT-II trial. For instance, patients within one month of an acute myocardial infarction or three months of a coronary artery bypass or angioplasty procedure were ineligible for MADIT-II, and the CMS could have declined to cover ICD implantation in such patients, which would have greatly limited utilization while strictly adhering to the available evidence.
With the release of several major clinical trials of ICDs in 2004 (Exhibit 1
The case of the ICD illustrates several difficult challenges to the application of evidence-based medicine and cost-effectiveness analysis to public policy. Some of these challenges result from technical limitations of the methods, but others represent more fundamental issues. Technology development begets technology evaluation, but some fields move so quickly that there is a persistent gap between the evidence available and the evidence needed to set policy. The ICD field has evolved very quickly, and devices on the market today have much greater capabilities than those of only two or three years ago. It takes several years, however, to conduct RCTs, so a clear picture of clinical outcomes is available only for earlier generations of the technology. Consequently, the evidence is often criticized as being "not up-to-date" or "not relevant" to current therapies. The field of evidence-based medicine needs to develop sound methods to deal with this "moving target" problem. One potential approach is to develop decision models that can be readily updated and thereby provide a flexible framework to integrate RCT findings with other forms of evidence and keep the conclusions contemporary.16
Basing decisions on substantial, consistent evidence is desirable, yet the evidence from different clinical trials is not always consistent. For instance, one early study of ICDs (CABG-Patch) found no reduction in mortality, while another (MADIT-I) found a 54 percent lower risk of death (Exhibit 1
Therapies may work well when matched to the appropriate patients but work poorly when applied to less well-selected patients. Indeed, the outcomes of therapy generally vary depending on patient characteristics (such as age, sex, severity of disease, prior treatments), provider characteristics (such as training, skill, volume of experience), and health system characteristics (such as teaching hospital status, quality-of-care initiatives). There are rarely black-and-white clinical indications for treatment; a gradual shading across the spectrum is the norm. The experience with patient selection for prophylactic ICDs illustrates how costs and benefits of treatment can vary depending on patients’ risk levels (Exhibit 2
Cost-effectiveness analysis is predicated on the concept of diminishing returns in medicine: With continued application of resources, clinical improvements eventually will grow too small to justify the added costs. But an insured patient will want anything that may help his or her care, especially for a life-threatening condition. It is one thing for payers to deny a patient an ICD because it provides no benefit or carries too great a risk, but it is something very different to deny a patient an ICD that might extend life, on the grounds that it just costs too much. This tension between the "societal perspective" of cost-effectiveness analysis and the individual perspective of what is best for a single patient is a fundamental challenge to using cost-effectiveness criteria to set policy. This conflict is particularly acute for doctors who have a duty to do their best for individuals under their care but are constrained by reimbursement rules. One prominent cardiologist recommends that instead of practicing evidence-based medicine, doctors should practice "family-based medicine"—what treatment would you recommend for your father or your grandmother?17 Although cost-effectiveness is often discussed, it is not explicitly used as criterion for coverage by the CMS or by private insurers. By statute, the CMS does not cover services unless they are medically "necessary and reasonable," but these terms are vague.18 The policy dilemma arises when there is clear evidence from a clinical trial that an expensive therapy improves survival in a specific group of patients. In this situation, it is very hard for an insurer not to offer coverage, even if the therapy is very costly and even if the treatment may be overused as a result of "indication creep" after the coverage decision. In the case of prophylactic ICD implantation, the financial stakes are very high because of the large population of eligible patients. Insurers will clearly examine the data more closely when the stakes are higher, and they may reasonably require more solid evidence. The CMS coverage decision on ICDs was based on the evidence of variation in clinical effectiveness in different populations, however, and ICD cost-effectiveness was not explicitly considered. We believe that data on cost-effectiveness should be considered in these decisions, which would make the trade-offs between cost and outcomes open for discussion. Finally, most effective therapies do, in fact, cost more money. Extending coverage to cost-effective treatments that are either expensive or apply to large populations, or both, will add to total medical care costs. When resources are limited, cost-effective care may nevertheless be difficult to afford. Implantable cardioverter defibrillators are a dramatic new medical technology that has evolved in a very few years from an experimental therapy to a standard of care for some groups of patients. Evidence-based medicine and cost-effectiveness analysis have been applied to the ICD and provide considerable insight as to when this expensive treatment should be used. Nevertheless, there are limitations to these evaluation methods, especially when they are applied to very expensive yet very effective therapies.
Mark Hlatky (hlatky{at}stanford.edu) is a professor of health research and policy and of medicine (cardiovascular) at the Stanford University School of Medicine in Stanford, California. Gillian Sanders is an associate professor of medicine at the Duke Clinical Research Institute in Durham, North Carolina. Douglas Owens is senior investigator at the VA Palo Alto Health Care System and an associate professor of medicine at Stanford Medical School. This research was supported in part by Grant no. HS-08362 from the Agency for Healthcare Research and Quality.
This article has been cited by other articles:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Evidence And ICDs
Cost-Effectiveness Of ICDs
