A review of economic evaluation models for cardiac resynchronization therapy with implantable cardioverter defibrillators in patients with heart failure

The search retrieved 1839 citations, which were reduced to 1420 after excluding for duplicates and for non-English language citations. After screening titles and abstracts, 99 articles (Fig. 1), were further scrutinized to exclude papers that fell outside the scope of the review. The remaining references (15 studies) were scrutinized to include only studies on: (1) HF patients with NYHA II, III or IV, LVEF ≤ 35 % [10], (2) treated with the CRT-D as a comparator and (3) that included decision-tree models or Markov chain models. The eight studies remaining [11‐18] included model-based economic evaluations (i.e., Markov models and decision-tree models) of CRT-D implantations and were included for further review.

Table 1 gives the general study characteristics of the selected publications for this review. The selected studies included economic models adapting perspectives of four European health care systems (Belgium, Germany, Spain and UK) [11‐15, 17, 18] as well that of a middle income country (Brazil) [16].

All models distinguish between a short-term (represented by costs and consequences of the process of device implantation) and a long-term phase (represented by the costs and consequences of the post implementation follow-up period). The initial short-term implantation phase was generally 4 weeks while the long-term maintenance phase was the lifetime of the patient (see Table 1). Four out of the eight selected studies [13, 15‐17] used a decision tree to model the short-term phase while four of the other studies [11, 12, 14, 18] included this as an initial phase into the Markov cohort models. All the selected studies [11‐18] used a Markov cohort model for the long-term maintenance period.

Paths in the short-term decision trees consisted of combinations of four treatment strategies: (1) optimal medical therapy (OPT), (2) CRT-P, (3) CRT-D and (4) implantable cardioverter defibrillator (ICD). Bertoldi et al. [16] linked successful implantation of the device (CRT-P, CRT-D or ICD) to complications or no-complications after implantation, which then led to the long-term model. Failure of implantation led to the long-term OPT strategy. This was similar to Yao et al. [12] except that these investigators allowed for up to three re-implantation attempts.

Markov states used in the eight selected studies [11‐18] can be broadly classified into four categories: (1) the short-term implantation period, (2) upgrading or switching between different implantable devices (e.g. switching from CRT-P to CRT-D), (3) the maintenance states after device implantation or when on OPT therapy; and (4) the death state. Table 2 gives a more detailed summary of these Markov states.

As mentioned, four out of the eight delected studies [11, 12, 18] included the initial phase of the device implementation in the Markov model by adding the following states: (1) surgical intervention for CRT implantation, (2) hospitalisation because of complications after CRT implantation and (3) death because of CRT implantation (or any subsequent operation). The consequent maintenance states usually referred to events like stable state (while on CRT or OPT), hospitalisation due to lead displacements or infections, hospitalisation due to HF worsening and hospitalisation due to other causes (e.g. heart transplant, CABG, ablation, etc).

All models allowed patients to move from one Markov state to another. Death states were detailed in SCD, HF death, or death from non-cardiac related (nCR) causes. In Yao et al. [12] patients that remained alive in the next time period could continue being in the same NYHA class or move to higher or lower NYHA classes. Similarly, in Bertoldi et al. [16] patients could move between NYHA classes (but with only a NHYA class at a time).

Upgrading devices from CRT-P to CRT-D was allowed in three of the selected studies [11, 15, 17]. In such cases CRT-P patients could at any time experience arrhythmia and thereafter upgrade to a CRT-D device. In addition to the above states the model by Fox et al. [11] (which consists of a synthesis of the existing models) allowed also for device explanting or replacement of the CRT-D device with a new one. Patients could also be switched to the OPT arm or get an ICD alone but only when experiencing arrhythmias or receiving a heart transplant (heart transplant state was also allowed in Colquitt [18]).

The population considered in the selected studies were adults (aged 18 and over) eligible for CRT implantation. Eligibility criteria in all the studies were largely associated with the guidelines for CRT implantation in patients suffering from heart failure. These guidelines recommend implantation of a CRT device (with or without ICD) for patients who have left ventricular ejection fraction (LVEF) ≤35 %, a QRS duration ≥120 ms, sinus rhythm, and fall within NYHA functional class III or ambulatory class IV heart failure [4, 10, 19, 20]. Most of the studies complied with these criteria. Five out of the eight selected studies [11, 12, 14, 15, 18] used clinical data from the Care-HF trial, (a multicentre, international randomized trial comparing CRT-P to OPT) [2], while two other studies [13, 17] used data from the COMPANION trial (also a multicentre, international randomized trial comparing CRT-P to OPT and CRT-D to OPT) [1]. Both trials used similar eligibility criteria for the selection of patients [1, 2]. Bertoldi et al. [16] used data from an outpatient clinic in a Brazilian hospital but reported a similar target population. Colquitt et al. [18] used data from the MADIT-CRT [21] and RAFT [22] trials for the comparison of CRT-D with ICD arms. Colquitt et al. [18] distinguished between three different population groups. However, for consistency with the populations of other studies, here we focus on the analysis of group II (i.e., patients with heart failure as a result of left ventricular systolic dysfunction (LVSD) and cardiac dyssynchrony despite receiving OPT) [18].

Six [11, 12, 14‐17] out of eight selected studies included three comparators; CRT-P with OPT, CRT-D with OPT, and OPT alone. One study [13] compared only cost-effectiveness of CRT-D against CRT-P. Colquitt et al. [18] included also comparisons of ICD with OPT and CRT-D (though this was done in a different patient population).

Pharmacoeconomic guidelines agree that the time horizon of a cost-effectiveness model should extend far enough in the future to capture the major health and economic outcomes, including both the intended and unintended effects [23]. With treatment for heart failure, certain treatment outcomes can be realized over a shorter period (like the outcome of the surgical intervention) while others, such as a possible effect on survival, can only be realized over a long time horizon (extending to lifetime of the patient). The duration of clinical trials testing the cardiac resynchronization therapy in heart failure disease varies from 6 months [21, 24] to 12 months [25] or to a longer period of 29 months [2]. In the selected modelling studies, patient outcomes and costs were simulated over 20 years [16] or over the complete lifetime of the patient [11, 12, 14, 15, 17, 18], except for one study [13] which applied a time horizon of only 2 years after implantation (Table 1). This is considered rather a conservative approach, as the high costs of implantation cannot be fully recovered within such a short term. The authors justified their choice by explaining the difficulties in the extrapolation of utilities, costs and transition probabilities beyond the 2-year follow-up of the COMPANION trial [1].

Resource use and unit prices in the selected studies were predominantly obtained from the health care systems of the respective countries. All the eight selected studies [11‐18] employed a payer perspective. As such, they have included costs of CRT-P, CRT-D and ICD devices, costs of device implantation, lead replacement, heart failure hospitalisation and follow-up costs while in a stable health state. The level of detail varied substantially between the studies, and was predominantly dependent on the structure of the short-term or long-term models (see also above).

It should be noted that prices of devices differed between studies reflecting also the market value of the devices over time. Hence, earlier studies by Fox et al. and Yao et al. [11, 12] estimated prices of the CRT-D devices in the UK healthcare setting at respectively €19,196 and €19,914 (converted into 2014 prices in euros using OECD estimates of purchasing power parities (PPPs) for GDP [26] while the most recent study from Colquitt et al. [18] estimated this at only €13,170 (PPP adjusted and in 2014 prices).

The primary outcomes were quality-adjusted life years (QALYs) or life years (LYs). All the studies use the NYHA Functional Classification to assign the quality of life (QoL) scores for patients in each health state. Again, the only exception here is the study by Aidelsburger et al. [13] that also included hospitalisations as an outcome.

The calculation of QALYs in the selected studies was based on utility values from NYHA classes (Table 3). Seven out of eight studies [11‐16, 18] distinguish between utility values for each NYHA class. Neyt et al. [17] considered mean health utilities by treatment rather than by NYHA class. They argued that this was preferred, given the substantial variation of NYHA class utility estimates between publications. In fact, health utility values used per NYHA class do vary greatly between studies, as can be seen from values in Table 3.

ICERs for cost per QALY gained and cost per LY gained are presented in Tables 4 and 5. After checking for transferability criteria [32, 33], all ICERs were converted in 2014 prices in euros using PPPs [26]. Seven out of the eight selected studies [11, 12, 14‐18] reported cost-effectiveness results per QALY for CRT-P compared to OPT and for CRT-D compared to CRT-P, four studies [11‐13, 18] for CRT-D compared to OPT, and two studies [16, 18] for CRT-D compared to ICD (Table 4). Despite the differences in ICERs, results were consistent in showing that CRT-P was mostly cost-effective in comparison to OPT alone. However, results were less clear for ICERs of CRT-D compared to OPT. In the study by Aidelsburger et al. [13] this ICER was much higher than in other studies (€76,350) which could relate to the time horizon of only 2 years in this study. In general, ICERs for CRT-D compared to CRT-P were considerably higher than for CRT-D compared to OPT. Hence, ICERs for CRT-D versus CRT-P ranged from €42,986 to €63,343 while ICERs per QALY for CRT-D versus OPT ranged from €16,166 to €29,889. The only exception was the study from Colquitt et al. [18] where both ICERs were comparable (€29,889 vs €30,447), but these did not apply to the same population [18]. It should be noted that most ICERs for CRT-D versus CRT-P are well above what is considered cost-effective in most countries (for instance the £20,000-£30,000 threshold applied in the UK [34]—converted to €21,427-€32,139 PPP adjusted). Bertoldi et al. [16] suggested that CRT-D therapy should not be systematically recommended for CRT-P eligible patients, while it can be an option for ICD eligible patients. Yao et al. [12] suggested that CRT-D was not cost-effective, particularly for CRT-P patients with poor life expectancy. Fox et al. [11] called for more research to explore the added value of CRT-D over CRT-P and for improving identification of non-responders among patients in the CRT-D group.

Two studies [16, 18] reported ICERs for QALYs gained for CRT-D compared to ICD. Both ICERs were comparable, ranging from €29,135 to €32,664 (even though they concern different populations). Colquitt et al. [18] found such ICERs to be robust and influenced only from all-cause mortality in the ICD-only arm and lifetime of CRT-D and ICD devices [18].

Two of the selected studies [11, 14] did not report on ICERs per LY (Table 5). As for the other studies, similar trends held as in ICERs per QALY. The ICER for the comparison between CRT-D and OPT in the Aidelsburger et al. study [13] appears to be much higher than others, while ICERs for CRT-D against CRT-P remain constantly higher than ICERs for CRT-D against OPT. Again the study from Colquitt et al. [18] is an exception here, although these results should be cautiously interpreted as the population in Colquitt et al. [18] is not the same as in other studies.

All selected studies [11‐18] reported univariate sensitivity analysis on key variables. Key determinant variables were battery longevity [12, 13, 16, 18], cost of the device [16, 18] as well as relative risk for mortality from HF (for CRT-P vs OPT or CRT-D vs CRT-P) [14, 16, 18]. Decreasing the cost of the CRT device by 50 % decreased the ICERs by 23 % for CRT-D vs CRT-P or by 40 % for CRT-P vs OPT [16]. The selected studies assumed a base case battery life that varied from 5 [16] to 6.5 years [14, 15] for CRT-P devices and from 5 [16, 17] to 5.5 years [14, 15] for CRT-D. Yao et al. [12] assumed a base case battery life of 7 years for CRT-D. Increasing battery life by 40 % decreased the ICERs by more than 20 % in Bertoldi et al. [16] and by 29 % in Colquitt et al. [18]. The reduction of mortality with CRT-D by 13.3 % decreased the ICER for CRT-D vs CRT-P by 36 % [16] while the decrease in relative risk for HF death in CRT-D patients by 20 % decreased the ICER by 58 % [15]. Five studies reported probabilistic sensitivity analyses [11, 12, 16‐18], while one study [13] stated to have performed a two-way sensitivity analysis, but did not report the results. Four out of the eight selected studies [11, 12, 17, 18] incorporated cost-effectiveness acceptability curves (CEACs), which are used to summarize the uncertainty in the cost-effectiveness estimates. The CEACs in these four studies showed that, for a willingness-to-pay (WTP) threshold between €27.000 and €44.000 (PPP adjusted) per QALY, the probability of CRT-D being cost-effective compared to OPT or CRT-P was only 26–40 %.