MADIT II (Second Multicenter Automated Defibrillator Implantation Trial) Debate
Risk Stratification, Costs, and Public Policy
After extensive deliberation, on June 6, 2003, the Centers for Medicare and Medicaid Services (CMS) reached a long-awaited decision on coverage of implantable cardioverter defibrillators (ICDs) for patients meeting the clinical profile of the second Multicenter Automated Defibrillator Implantation Trial (MADIT II).1 CMS offered a split decision, extending coverage only to Medicare patients meeting the MADIT II inclusion and exclusion criteria and having a QRS duration >120 ms. Information on and a timeline of the CMS coverage determination process are publicly available at their World Wide Web site,2 which has a link to a detailed decision memorandum.3 Although versions of the “MADIT II debate” have already taken place in many forums, we believe several issues regarding both the trial and the CMS decision require further discussion. We will focus on the intertwined issues of risk stratification and cost that have arisen repeatedly in this debate.
The MADIT II debate boils down to the question posed by Dr Alfred Buxton in a recent editorial4: Should all patients with an ejection fraction <30% because of coronary artery disease receive ICDs, regardless of other risk-stratifying tests? Dr Buxton suggested the answer to this question should be “no”: “I believe it is premature to recommend ICDs on the basis of only one trial that may not be generalizable to all patients.”4 Dr Buxton’s remarks and the CMS decision stand in contrast to the otherwise unimpeded march the MADIT II results have made toward clinical acceptance. After publication of MADIT II, the Food and Drug Administration quickly approved the Guidant ICD used in the trial for this new indication, and updated American Heart Association/American College of Cardiology/North American Society of Pacing and Electrophysiology (AHA/ACC/NASPE) consensus guidelines5 recommended the trial’s inclusion/exclusion criteria as a class IIa indication for ICD implantation. Furthermore, the Medicare Coverage Advisory Committee (MCAC) that was convened in February 2003 to help CMS with their decision voted unanimously (7 to 0) in favor of Medicare funding for the MADIT II indication.6
Why have clinicians like Dr Buxton taken a position on MADIT II that on the surface appears to contradict most other advice regarding the interpretation of this trial? Why did CMS agree to only limited coverage despite lobbying from national cardiology organizations and the vote of its own advisory committee? There are several possible reasons. First, as observed in a recent meta-analysis,7 the magnitude of benefit in MADIT II was clearly smaller than that seen in primary prevention trials that randomized only patients who had inducible ventricular arrhythmias in response to invasive electrophysiology (EP) studies.8,9 As a result, the number needed to treat to prevent 1 premature death in 2 years, as calculated by Ezekowitz et al,7 was 18 for MADIT II but closer to 4 for primary prevention trials that used EP testing for risk stratification. From the outset, there has been concern that compared with current risk-stratification strategies, less selective criteria for ICD implantation could result in many patients receiving ICDs who do not stand to benefit from them, exposing some patients to unnecessary risks and using societal resources less efficiently.
Supporters of MADIT II counter this criticism by pointing out that the absolute mortality benefit seen in the trial—5.6% at an average of 20 months of follow-up—was not trivial, particularly compared with other primary prevention measures, and that ejection fraction alone is a sufficient risk stratifier in this population. This leads to a common question that accompanies clinical trials: Which patients received the benefit? This question is particularly important in MADIT II with respect to the issue of risk stratification.
A subtle feature of MADIT II is that by not performing EP studies on patients before randomization, the trial in effect enrolled a combination of subjects who would have had inducible or noninducible ventricular arrhythmias had they been tested beforehand. However, because ICDs had already been shown to provide substantial benefit for patients with inducible ventricular arrhythmias,8,9 the subjects of real interest were those who would have been noninducible. Unfortunately, because inducible and noninducible subjects were lumped together, and the trial was stopped early by its data safety and monitoring board, we are left with uncertainty as to whether the noninducible patients in MADIT II received the same mortality benefit from ICDs as the inducible patients. This uncertainty was clearly a stumbling block for CMS as it considered the MADIT II coverage request, because Medicare already covered ICDs for EP-inducible patients.
It turns out that 78% of the MADIT II patients randomized to the ICD arm had EP studies done in conjunction with their ICD implants, most apparently by standard catheter techniques.10 Of those patients studied, 36% were found to be inducible. It is impossible to tell how many of these inducible patients might have met the original exclusion criteria of the trial, because the requirement for preenrollment Holter monitoring was eliminated early because “almost all eligible patients” had the required “frequent or repetitive ventricular ectopic beats”1 for study entry and because the Holter requirement was “inhibiting enrollment.”11
At the MCAC hearings in February 2003, both CMS and the MADIT II investigators presented post hoc analyses that attempted to evaluate the therapeutic effect of ICDs in the noninducible patients.6 These efforts were hampered by the fact that inducibility was generally not assessed in the control group patients. The 2 analyses, perhaps not surprisingly, reached different conclusions. Such post hoc analyses are usually viewed with caution owing to well-recognized statistical limitations.12 That these analyses even became necessary in our opinion simply illustrates that the most crucial clinical question to arise from MADIT II (is EP testing necessary?) was inadequately addressed by the design of the study. In retrospect, a study enrolling only noninducible patients might have generated less controversy.
The difficulty underlying the MADIT II debate can be seen by framing the central question oppositely: Should ICDs be denied for all noninducible patients with ejection fractions <30%? Recent data published by the MUSTT (Multicenter Unsustained Tachycardia Trial) investigators raise concern about such a policy, which in fact had represented the standard of care before MADIT II. The analysis from MUSTT showed that noninducible registry patients with ejection fractions <30% had nearly identical total mortality and sudden cardiac death rates as patients who were inducible but had an ejection fraction of 30% to 40%.13 Therefore, although EP studies have long been known to predict sudden death and total mortality independently of ejection fraction in postinfarct patients,14,15 a negative EP study in a patient with an ejection fraction <30% is not fully reassuring. The troubling potential for false-negative EP studies in these patients was indeed a motivation for conducting MADIT II in the first place.
If some but not all of the MADIT II population will benefit from ICDs, how are clinicians to decide? Before the CMS decision, advice on this issue had been vague. The updated ACC/AHA/NASPE guidelines state, “the committee consensus was that further risk stratification of the referenced population might better define the benefit of an ICD in such patients.”5 Although we agree with this comment from a research perspective, it is not particularly helpful for clinicians and patients making decisions today. Which risk-stratifying tests should be performed? What decision rules should be applied? The answers are unclear.
We believe that comprehensive studies to assess both the clinical and economic impact of different risk-stratifying methods for the MADIT II population are sorely needed. Although we have focused so far on EP testing, multiple other risk-stratifying methods have been studied to varying degrees, and a number of them have shown some utility in predicting total mortality and sudden death in postinfarct patients.16,17 For example, the signal-averaged ECG—most reliably, a filtered QRS duration >114 ms—was a strong predictor of events in the MUSTT population after adjustment for a number of clinical and treatment variables, including EP inducibility and ICD therapy.18 Microvolt T-wave alternans has also shown recent promise as a risk stratifier19,20 and is being evaluated prospectively along with EP inducibility in the ongoing Alternans Before Cardioverter Defibrillator (ABCD) trial. Additional clinical factors, including single-vessel versus multivessel coronary disease, extent of revascularization, age, and comorbidity, may independently influence sudden death and total death and therefore the clinical effectiveness of ICDs. Ultimately, the best risk-stratification method may involve a combination of clinical factors and tests, a concept that has been explored in the secondary prevention setting.21 The assessment of risk stratification needs to move beyond predicting survival and on to predicting the therapeutic benefit of ICDs.22
CMS, undoubtedly influenced by the large potential financial impact of the MADIT II decision, chose what appears to have been the only risk stratifier available to them in the MADIT II study data: QRS duration. Although the decision was immediately criticized by NASPE as “not scientifically sound,”23 the importance of QRS duration is both biologically plausible and has some precedent in the literature.24,25 In control group and registry patients from MUSTT, left bundle-branch block and intraventricular conduction delay on baseline ECG predicted both sudden death and total mortality, with adjustment for other factors.25a The detrimental effect of a wide QRS in heart failure patients has also led to the development of resynchronization therapy. Nonetheless, this approach to risk stratification clearly requires further validation.
Aside from the issue of risk stratification, there are other potential reasons for proceeding cautiously with the MADIT II results. One cited by CMS is that MADIT II is only a single study, and additional data on prophylactic ICD implantation in overlapping patient populations will soon be available, most importantly from the federally funded Sudden Cardiac Death in Heart Failure Trial (SCD-HeFT). Such a view is consistent with the ACC/AHA/NASPE class IIa recommendation for the MADIT II criteria (“conflicting evidence and/or a divergence of opinion … weight of evidence/opinion is in favor of usefulness/efficacy”5), because class I status is typically reserved for interventions supported by more than 1 randomized trial. Aware of the MADIT II results, the SCD-HeFT data safety and monitoring board has not elected to terminate that trial prematurely.3
The results of related studies may also influence the confidence with which the MADIT II results can be generalized beyond the defined trial population. Generalizability has been raised as a concern4 because the possibility for selection bias is very difficult to eliminate from any clinical trial. The MUSTT investigators have found that even subtle factors, such as the site of identification of study patients, can have an influence on outcome, with outpatients having a much better prognosis than inpatients.26 The location and manner by which patients are identified and enrolled in a study like MADIT II can make the study sample differ from patients in general clinical practice in ways that are difficult to measure, and thus, generalizability is best assessed through the evaluation of outcomes in general practice.
Finally, it must be remembered that ICDs are not the only therapy that prevents sudden death and total mortality in MADIT II-type patients. Revascularization27 and medical therapy including ACE inhibitors and β-blockers have extremely important effects in this regard. Compliance with these therapies in the MADIT II trial was outstanding and needs to be so in general clinical practice. In addition, 2 meta-analyses of multiple randomized trials have concluded that amiodarone reduces both sudden death and total mortality in patients with ischemic left ventricular dysfunction,28,29 particularly when combined with β-blockers.30 The effects of amiodarone on mortality have not matched ICDs in prior randomized trials,8,31 but β-blockers were not administered concomitantly in those trials. Thus, the combination of amiodarone with β-blockers remains a reasonable option for patients in whom ICDs are not approved or desired.
Can Cost-Effectiveness Data Help?
CMS follows the statutory standard that a therapy is “reasonable and necessary” when making new coverage decisions. It has never been entirely clear how the notion of cost-effectiveness fits into this definition. We believe it is both reasonable and necessary for the largest single healthcare purchaser in the world32 to want to quantify the expected benefits for the money it spends, particularly when facing many competing programs in an atmosphere of cost containment. Furthermore, we believe that the academic discipline of formal cost-effectiveness analysis (CEA) is the best available approach to these questions. Apparently, some policy makers within the government agree with this, but explicit consideration of cost-effectiveness data has thus far not gained political acceptance in the United States. For example, the internal technology assessment conducted by CMS on MADIT II specifically excluded cost-effectiveness studies.3 The result is awkward, with CMS publicly pretending that their decisions are not driven in part by financial motives, and nobody really believing them.
There are additional barriers to the use of CEA to inform coverage decisions, both in general and specifically with regard to ICDs. In the past, substantial methodological variations across studies made it difficult to compare disparate health interventions with CEA. This problem was addressed with the publication of recommended methodological standards by a US Public Health Service task force.33 Furthermore, a true consensus has never been reached on a threshold ratio for defining cost-effectiveness; any such standard has a danger of being, or at least appearing to be, arbitrary. To date, ICD trials have not collected much precise cost and quality-of-life data, in part because of an appropriate focus on mortality outcomes and perhaps in part because the data were not viewed as necessary. Trial-based CEAs34,35 have therefore not been able to estimate quality-adjusted life years (QALYs), which are the recommended unit of effectiveness in CEAs.33
Apparently, a formal CEA of the MADIT II trial is in progress36 but has not yet been reported. What might we expect it to find? The results will depend in part on assumptions and extrapolations made by the investigators, because the true cost-effectiveness of ICDs in MADIT II cannot be known without looking beyond the currently reported average follow-up of only 20 months. Using the published survival data as input into a simple Markov model, we have estimated the average undiscounted gain in life expectancy with ICDs in MADIT II to be <0.15 life-years over the first 3 years of follow-up. If the incremental cost of an ICD strategy over standard medical therapy is assumed to be $30 000 per patient (a value lower than reported in prior studies37), this translates to an incremental cost-effectiveness ratio of >$200 000 per life-year saved. This ratio would appear more favorable if the cost of ICDs were lowered (eg, by market forces1) or if substantial gains in survival accrue from ICDs beyond the first 3 years of follow-up, relative to downstream costs.
It stands to reason that the application of further risk-stratification methods to MADIT II patients could make ICDs more cost-effective than simply basing the decision on ejection fraction alone.38 Compared with any reasonably performing risk-stratification strategy, the selection of patients on the basis of ejection fraction alone is likely to result in the largest gain in average life expectancy for this population at the highest cost. Risk stratification can never be perfect, but understanding how different methods affect the tradeoff between costs and benefits might at least make the discussion of those tradeoffs more transparent. Identification of the MADIT II patients least likely to benefit from ICD implantation might allow for substantial cost savings with small or even negligible sacrifice in population life expectancy. In essence, the CMS decision to restrict approval based on QRS duration allows them to avoid coverage for a subgroup in whom cost-effectiveness might be particularly unattractive. The problem is, CMS never claimed to be evaluating cost-effectiveness.
From the above discussion, we hope to convey that we all have much to learn from the MADIT II debate about how to make appropriate decisions for patients and how as a society to make appropriate decisions about healthcare spending. Many suggestions have been made regarding how to improve these processes; we wish to highlight a few:
As mentioned, rigorous and comprehensive studies comparing the clinical and economic impact of different specific approaches to risk stratifying ICD candidates are needed. This has been recommended in published practice guidelines,5 by the ACC at the MCAC hearings, and by CMS in its decision memorandum. More detailed exploration of risk stratification is needed in future consensus guidelines.
The ACC has also recommended the establishment of a registry of patients treated with ICDs under the MADIT II indication. This would enable monitoring of compliance with coverage guidelines (and possibly “guarantee” it, if made mandatory) and also provide an opportunity to assess outcomes in the community over time. An even more ambitious registry might also collect risk-stratification data and monitor outcomes of patients denied ICDs under current policy. The design, implementation, and analysis of such a registry would pose significant administrative challenges and ideally should be conducted by an independent group, perhaps in concert with government and industry.
For now, the eligibility criteria for the MADIT II trial need to be adhered to strictly in patient selection. Relevant exclusion criteria in the trial included class IV congestive heart failure, revascularization within 3 months, myocardial infarction within 1 month, and any comorbid condition associated with an expected survival of <1 year.1 High standards of informed consent are necessary.
CMS should develop and apply clear and consistent standards for considering cost-effectiveness data in all major coverage decisions. We do not have unlimited resources to devote to health care, and difficult decisions must be made. As much as possible, these decisions should be made on rational and transparent rather than political grounds. Well-established methods for doing so exist.
Interventions that have important impacts on healthcare costs and quality of life need to measure these outcomes, in addition to mortality, and report the results in a timely fashion. This requires careful planning early in the process of trial design and execution. ICD trials have lagged behind investigations in other fields on this issue.39
As promised, CMS must revisit their decision on MADIT II in light of new data as it arises.
Healthcare practitioners, government, academia, and industry all have important roles to play in advancing the health of the population. Despite many disagreements over MADIT II, we do not believe anyone has lost sight of the importance of preventing sudden cardiac death. We must all continue to work together to ensure that our interventions are as safe, effective, affordable, and appropriately utilized as possible.
Centers for Medicare and Medicaid Services. National coverage analysis: implantable cardioverter defibrillators (#CAG-00157N). Tracking sheet. Available at: http://www.cms.hhs.gov/ncdr/trackingsheet.asp?id=39. Accessed September 5, 2003.
Centers for Medicare and Medicaid Services. Administrative file CAG: decision memorandum: implantable cardioverter defibrillators (# 00157N). June 6, 2003. Available at: http://www.cms.hhs.gov/ncdr/memo.asp?id=39. Accessed June 9, 2003.
Gregoratos G, Abrams J, Epstein AE, et al. ACC/AHA/NASPE 2002 guideline update for implantation of cardiac pacemakers and antiarrhythmia devices: summary article: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (ACC/AHA/NASPE Committee to Update the 1998 Pacemaker Guidelines). Circulation. 2002; 106: 2145–2161.
Centers for Medicare and Medicaid Services. Transcript of the Medicare Coverage Advisory Committee open session, February 12, 2003. Available at: http://www.cms.hhs.gov/mcac/id39-5.pdf. Accessed June 9, 2003.
Moss AJ. Requestor’s presentation: Guidant Corporation. Centers for Medicare and Medicaid Services: Medicare Coverage Advisory Committee open session, February 12, 2003. Available at: http://www.cms.hhs.gov/mcac/id39-5.pdf. Accessed September 25, 2003.
Buxton AE, Lee KL, Hafley GE, et al. Relation of ejection fraction and inducible ventricular tachycardia to mode of death in patients with coronary artery disease: an analysis of patients enrolled in the Multicenter Unsustained Tachycardia Trial. Circulation. 2002; 106: 2466–2472.
Richards DAB, Byth K, Ross DL, et al. What is the best predictor of spontaneous ventricular tachycardia and sudden death after myocardial infarction? Circulation. 1991; 83: 756–763.
Gomes JA, Cain ME, Buxton AE, et al. Prediction of long-term outcomes by signal-averaged electrocardiography in patients with unsustained ventricular tachycardia, coronary artery disease, and left ventricular dysfunction. Circulation. 2001; 104: 436–441.
Sheldon R, Connolly S, Krahn A, et al. Identification of patients most likely to benefit from implantable cardioverter-defibrillator therapy: the Canadian Implantable Defibrillator Study. Circulation. 2000; 101: 1660–1664.
Cain ME. NASPE/Heart Rhythm Society concerned with CMS ruling limiting ICDs for select patients. Available at: http://www.naspe.org/naspe_in_action/washington_advocacy/view/?id=8746. Accessed June 9, 2003.
Baldasseroni S, Opasich C, Gorini M, et al. Left bundle-branch block is associated with increased 1-year sudden and total mortality rate in 5517 outpatients with congestive heart failure: a report from the Italian network on congestive heart failure. Am Heart J. 2002; 143: 398–405.
Josephson ME, Buxton AE, Lee KL, et al. Prognostic value of the ECG in MUSST patients. Circulation. 2000; 102 (suppl II): II-494.
Pires LA, Lehmann MH, Buxton AE, et al. Differences in inducibility and prognosis of in-hospital versus out-of-hospital identified nonsustained ventricular tachycardia in patients with coronary artery disease: clinical and trial design implications. J Am Coll Cardiol. 2001; 38: 1156–1162.
Sim I, McDonald KM, Lavori PW, et al. Quantitative overview of randomized trials of amiodarone to prevent sudden cardiac death. Circulation. 1997; 96: 2823–2829.
Connolly SJ. Evidence-based analysis of amiodarone efficacy and safety. Circulation. 1999; 100: 2025–2034.
Boutitie F, Boissel JP, Connolly SJ, et al. Amiodarone interaction with β-blockers: analysis of the merged EMIAT and CAMIAT databases. Circulation. 1999; 99: 2268–2275.
Mushlin AI, Hall WJ, Zwanziger J, et al. The cost-effectiveness of automatic implantable cardiac defibrillators: results from MADIT. Circulation. 1998; 97: 2129–2135.
Larsen G, Hallstrom A, McAnulty J, et al. Cost-effectiveness of the implantable cardioverter-defibrillator versus antiarrhythmic drugs in survivors of serious ventricular tachyarrhythmias: results of the Antiarrhythmics Versus Implantable Defibrillators (AVID) economic analysis sub-study. Circulation. 2002; 105: 2049–2057.
Moss AJ, Hall WJ, Zareba W. Implantable cardiac defibrillators. N Engl J Med. 2002; 347: 366–367.Letter.