Pitfalls in health-economic evaluations: the case of cost-effectiveness of prophylactic implantable cardioverter-defibrillator therapy in Belgium

Health-economic modelling studies with a short-term time horizon have produced huge incremental cost-effectiveness (CE) ratios (ICERs) for prophylactic implantable cardioverter–defibrillator (ICD) therapy, ranging from €81 000 to €348 000 per life year gained (LYG). Model-based studies simulating lifetime survival reported more favourable though inconsistent ICERs ranging from €17 100 to €60 200 per LYG. 1 Recently, an industry-sponsored CE analysis on prophylactic ICD therapy was published, ending up with an ICER of €31 717 per quality-adjusted life year (QALY) gained. 2 On the basis of cost data originating from the same Belgian population, the Belgian Health Care Knowledge Centre (KCE) calculated an ICER of €71 428 per QALY gained. 3 We discuss the reasons for such discrepancy. Cowie et al.2 calculated the increase in life expectancy following prophylactic ICD insertion from a meta-analysis of six clinical trials: MADIT I, CAT, MADIT II, AMIOVIRT, DEFINITE, and SCD-HeFT. In our opinion, pooling of these data is inappropriate because of the heterogeneity between those trials. There is a marked difference in baseline mortality between these studies. In MADIT I, the annual mortality rate in the control group was 17.1%, and in MADIT II, it was 11.6%, whereas in SCD-HeFT, it was 7.9%. Given the fact that the presence of clinical heart failure was a prerequisite for enrolment in SCD-HeFT and not in both MADIT trials, these differences in observed mortality suggest a selection towards patients with more severe illness in the MADIT trials. Variations in the medical therapy directed at the underlying disease among the trials also play a role. In the MADIT I trial, 65% of patients took an ACE-inhibitor, and only 14% of patients were on β-blocking therapy by the end of the study. In MADIT II, 70% of patients took an ACE-inhibitor, and 70% took a β-blocker, whereas in SCD-HeFT, 94% took an ACE-inhibitor or an angiotensin-II-receptor blocker, and 69% took a β-blocker. Apart from medical treatment of acute myocardial infarction (AMI), there is evidence from clinical trials that early reperfusion reduces the risk of sudden cardiac death (SCD) in selected AMI patients. This may have influenced the occurrence of sudden cardiac arrest in the trials and could have inflated the perceived efficacy of the ICD in the older studies. Pooling the results from trials that also considered asymptomatic patients with non-ischaemic disease is inconsistent. The clinical effectiveness of ICD therapy in primary prevention of death seems best reflected by the results of the SCD-HeFT trial. In a German series of 3000 ambulatory patients with heart failure and a low ejection fraction, 17.0 and 71.1% of patients fulfilled MADIT II and SCD-HeFT criteria, respectively. 4 SCD-HeFT is the most recent and the largest randomized trial that has been published, with the longest follow-up duration, and enrolling heart failure patients with ischaemic or non-ischaemic cardiomyopathy who were receiving up-to-date medical treatment for the underlying disease. In SCD-HeFT, the hazard ratio for all-cause mortality was 0.77 (95% CI 0.64–0.93), whereas Cowie et al. calculated a relative risk of 0.72 (95% CI 0.64 – 0.82). The vast difference in incremental effects (IE) calculated in the aforementioned CE studies cannot solely be explained by this difference. It also depends on the initial absolute benefit and lifetime extrapolations. In the model of Cowie et al. , the annual risk of death from the meta-analysis was 8.3 and 11.8% for the ICD and conventional therapy groups, respectively, whereas in SCD-HeFT, it was 6.59 and 8.57%, respectively. Remarkably, the undiscounted life expectancy of ICD patients is slightly higher in the study of Cowie et al.2 compared with the KCE study 1 (i.e. 9.52 vs. 9.38 years), in spite of a higher initial annual risk of death. A possible reason for this difference is that the KCE model increased the probability of all-cause death to reflect ageing, whereas Cowie et al. only adjusted death from non-cardiac causes. These differences result in a discounted (1.5%) IE of 1.22 and 1.88 LYG in the KCE's and Cowie et al. 's study, respectively. The scientific literature on the impact of ICD therapy on quality of life (QoL) in primary prevention is scarce. In both Cowie et al. 's study and ours, 1 year of life of an ICD patient treated in primary prevention and a control subject is assumed to equal 0.85. It may, however, be too optimistic to suppose that the ICD does not negatively affect QoL. Implantable cardioverter–defibrillators do not prevent illness since they are only intended to abort SCD, and an improvement in QoL is not to be expected. On the other hand, QoL in ICD recipients could be significantly impaired for several reasons, such as an increased awareness of a medical condition, the need for a surgical procedure and repeat interventions, lifelong follow-up, stress associated with (in)appropriate shocks, driving and leisure restrictions, risk of complications, and so on. Furthermore, several authors have argued that an ICD may play an unfavourable role in the progression of heart failure. The KCE study based its costs on real-world data, whereas Cowie et al. often introduced expert opinion. Both Cowie et al. 's and KCE's analyses used Belgian cost data and applied a 3% discount rate. Nevertheless, the discounted incremental costs (IC) were largely different: €46 413 in the former and €66 464 (95% CI 56 989–77 215) in the latter. The cost of an ICD was set at €16 650 in both studies. In the KCE study, other costs related to the intervention were derived from the Belgian ICD Registry. A detailed protocol for the analysis of the database is published elsewhere. 1 The average cost for a prophylactic ICD implantation was calculated to be €27 116 (95% CI 26 290–27 921). In the study of Cowie et al. , on the other hand, the combined device and procedure cost are set at only €23 072. For the procedural cost of €4650, reference is made to a government-operated website ( www.tct.fgov.be ), but there, an average cost per stay of €11 601 for ‘implantation of a heart defibrillator’ (APR-DRG 161) is mentioned. The same incongruity applies to ICD replacement procedure cost. On the basis of real-world data, the costs of the replacement were, on average, €32 664 (95% CI 25 907–39 830) in the Belgian primary prevention population in 2005. In the study of Cowie et al. , the replacement cost was much lower (about €18 000). On the basis of Belgian data, the longevity of ICDs implanted in 2001 was estimated at 4.56 years. This number reflects real-world conditions and includes not only replacements because of battery depletion, but also infections, recalls, electronic failures, and upgrading towards CRT-D devices. Because insufficient long-term follow-up data were available, a conservative estimate of 5 years was introduced in the base-case scenario. Cowie et al. , on the other hand, introduced a higher device longevity (100% replaced over 6.5 years) on the basis of a Medtronic performance report related to one specific ICD model. The device longevity from the KCE's database better corresponds to that reported in the literature. Hauser 5 found an average service life of 814 single-chamber ICDs of 4.7 ± 1 years compared with 4.0 ± 1 years for 293 dual-chamber ICDs, indicating that the shift to (newer) dual-chamber models has significantly shortened the battery life. A source of uncertainty in ICER calculations arises from parameter precision. According to NICE guidelines, ‘probabilistic sensitivity analysis (PSA) is preferred for translating the imprecision in all input variables into a measure of decision uncertainty in the CE of the options being compared’. Cowie et al. applied deterministic modelling in their base case, resulting in an ICER of €29 530 per QALY gained. PSA was performed separately. The authors state that they defined probability distributions for the key model variables, including transition probabilities between different states in the model, as well as the health utilities associated with different states. It is not clear to what extent the uncertainty around transition probabilities reflects the uncertainty surrounding the treatment effect. Furthermore, uncertainty-surrounding costs were not taken into account, and disaggregated results showing the impact on both IC and IE were not published. The difference between the outcomes of two economic evaluations, both performed on a Belgian population, is striking. An industry-sponsored study calculated an ICER of about €30 000 per QALY gained, whereas an independent KCE study resulted in an ICER of more than €70 000 per QALY gained. A combination of several factors renders the results of the former an optimistic scenario rather than a realistic base case: the treatment effect, initial mortality risk, extrapolation scenario, costs of the (re-)intervention, and imperfect probabilistic modelling are all favouring ICD therapy. When applying input data reflecting real-world costs, a realistic treatment effect, conservative extrapolation, and real-world device longevity, ICD implantation can be considered as being not cost effective (although there is no clear unique ICER threshold) for the primary prevention of death in patients with a severely reduced left ventricular function. Conflict of interest: none declared.