OUP user menu

★ editor's choice ★

The long-term cost-effectiveness of cardiac resynchronization therapy with or without an implantable cardioverter-defibrillator

Guiqing Yao , Nick Freemantle , Melanie J. Calvert , Stirling Bryan , Jean-Claude Daubert , John G.F. Cleland
DOI: http://dx.doi.org/10.1093/eurheartj/ehl382 42-51 First published online: 16 November 2006


Aims Cardiac resynchronization therapy (CRT-P) is an effective treatment for patients with heart failure and cardiac dyssynchrony with moderate or severe symptoms despite pharmacological therapy. The addition of an implantable cardioverter-defibrillator (ICD) function may further reduce the risk of sudden death. We assessed the cost-effectiveness of CRT-P compared with medical therapy (MT) alone, and the cost-effectiveness of CRT–ICD + MT compared with CRT-P + MT, on incremental cost per quality adjusted life year (QALY) and life year using data from two landmark clinical trials.

Methods and results A Markov model with Monte Carlo simulation to assess costs, life years, and QALYs associated with CRT (± ICD) and MT in patients with heart failure and cardiac dyssynchrony, on the basis of a UK healthcare perspective was constructed. NYHA class distribution and transitions, associated health utilities, rates and cause of hospitalization and death were estimated from individual patient data from the CArdiac REsychronization in Heart Failure (CARE-HF trial). The estimated additional benefit on survival of an ICD was based on results from COMPANION. The base case analysis used 10 000 individual life-time simulations assuming a battery life of 6 years for CRT-P and 7 years for CRT–ICD. From a life-time perspective in a 65-year-old patient, the incremental cost-effectiveness of CRT-P compared with MT is €7538 (95% CI €5325–€11 784) per QALY gained and €7011 (95% CI €5346–€10 003) per life year gained. The incremental cost-effectiveness of CRT–ICD compared with CRT-P is €47 909 (95% CI €35 703–€79 438) per QALY gained, and €35 864 (95% CI €26 709–€56 353) per life year gained.

Conclusion Long-term treatment with CRT-P appears cost-effective compared with MT alone. From a life-time perspective, assuming a reasonable life expectancy when receiving effective treatment for heart failure, CRT–ICD may also be considered cost-effective when compared with CRT-P + MT.

  • Cost effectiveness
  • Cardiac resynchronization therapy
  • Implantable cardioverter defibrillator
  • Markov model
  • Individual simulation


Heart failure is a common disease and costly in terms of morbidity, mortality, and resources consumed.13

Randomized controlled trials have demonstrated that cardiac resynchronization therapy (CRT-P) and cardiac resynchronization with an implantable cardioverter-defibrillator (CRT–ICD) improve symptoms, exercise capacity, ventricular function, quality-of-life, and reduces mortality in patients with heart failure due to cardiac dyssynchrony who have persistent moderate or severe symptoms, despite standard pharmacological therapy.48

A within-trial cost-effectiveness analysis based on individual patient data from the CARE-HF trial and UK cost structures showed that CRT-P was associated with increased costs (€4316, 95% CI: €1327–€7485), increased survival (0.10 years, 95% CI − 0.01–0.21), and increased quality adjusted life years (QALYs) (0.22 95% CI 0.13–0.32).9 The incremental cost-effectiveness ratio (ICER) was €19 319 per QALY gained (95% CI: €5482–€45 402) and €43 596 per life-year gained (95% CI: − €146 236–€223 849). The results were sensitive to the costs of device and procedure, and indicate that treatment with CRT-P is cost-effective at the notional willingness to pay threshold of €29 400 (£20 000) per QALY gained. The within-trial analysis suggests that CRT-P might be cost-effective over a patients lifetime, but this has not been established with trial evidence.

Previous evaluations have provided varying estimates of the cost-effectiveness of CRT-P and CRT–ICD relative to medical therapy (MT).1012 However, none has addressed the incremental cost-effectiveness of CRT-P vs. CRT–ICD, a more relevant and appropriate question, and none has been completed since the availability of the results from the CARE-HF trial.7 It is quite possible that CRT–ICD may appear cost-effective compared with MT, but the incremental benefit of ICD in addition to CRT-P might be beyond the threshold of willingness to pay (UK perspective). This could occur if the additional costs associated with the ICD component are high compared with any additional benefits gained.

We developed an economic model populated with data from CARE-HF7 to evaluate the long-term incremental cost-effectiveness of CRT-P and MT compared with MT alone, on incremental cost per QALY and life-year gained. In addition, we evaluated the cost-effectiveness of CRT–ICD + MT vs. MT and the relative cost-effectiveness of CRT-P and CRT–ICD, incorporating estimates of the proportion of sudden deaths that might be prevented with CRT–ICD from the COMPANION study.13 The incremental cost-effectiveness of CRT-P or CRT–ICD in different patient subgroups was also evaluated.


Overview of CARE-HF

The design and results of the CARE-HF study have been reported previously.7 In brief, eligible patients were at least 18 years of age, had evidence of heart failure for at least 6 weeks, and were in New York Heart Association (NYHA) class III or IV despite receipt of standard pharmacological therapy, with a left ventricular ejection fraction of < 35%, a left ventricular end-diastolic dimension of ≥ 30 mm (indexed to height), and a QRS interval of > 120 ms on the electrocardiogram. Patients with a QRS interval of 120–149 ms were required to meet two of three additional criteria for dyssynchrony: an aortic pre-ejection delay of more than 140 ms, an interventricular mechanical delay of more than 40 ms, or delayed activation of the posterolateral left ventricular wall. A total of 813 patients were randomly assigned to receive MT alone (n = 404) or with cardiac resynchronization (n = 409).

Model structure

We constructed a Markov model with Monte Carlo simulation to describe the clinical history of patients with NYHA class III/IV treated with CRT-P plus MT compared with MT alone, CRT–ICD + MT compared with MT alone, and CRT–ICD plus MT compared with CRT-P plus MT. Health states were defined by NYHA functional class and death. Mortality was sub-classified by cause including: death due to worsening heart failure; sudden death; death due to all other causes. As simulated patients pass through the model, cost, and QALYs associated with each state, their experiences are accumulated.

The initial distribution of the NYHA classes, age, and gender, and subsequent transition probabilities and costs associated with treatment by MT or CRT-P + MT were based on the intention-to-treat analysis of the CARE-HF trial.7

The additional effect of ICD on sudden death was based on the observed and projected rate in patients assigned to CRT-P in CARE-HF and the proportional reduction in sudden death observed in COMPANION in patients assigned to CRT–ICD compared with CRT-P.13 Mortality for other causes was derived from the UK population.14

Model description

The model had two components: the short-term representing changes in health status and the costs and consequences of the process of device implantation, and the long-term effects of the device after successful implantation (Figures 1 and 2). In the model, MT patients did not receive CRT-P or CRT–ICD during follow-up. The CRT-P and CRT–ICD groups received treatment with their assigned therapy in accordance with the successful device implantation rates observed in the CARE-HF trial.7 In the long-term phase, patients faced different risks of sudden death and unplanned hospitalization depending on their health state, treatment group, and duration of treatment. During each cycle of the model, patients could move between the four NYHA health states, experience sudden death, or have an unplanned hospitalization for a major cardiac event. We assumed that transition probabilities between NYHA classes only differed by CRT and were independent of the time in that health state, based on data from CARE-HF.7 Patients faced treatment and time-dependent risks for experiencing sudden death and unplanned hospitalization. Unplanned hospitalizations were categorized by type: procedure-related; non-procedure-related; and those leading to death due to worsening heart failure.

Figure 1

Structure of short term model. Patients had a maximum of three implant attempts. Those patients who received a successful implant moved to the long-term model with an NYHA class according to the transition probabilities observed in the CARE-HF trial. Where implants were unsuccessful, the patient followed the clinical pathway according to the transition probabilities for the MT group.

Figure 2

Structure of long-term model (NYHA class I). The structure of the model for other NYHA classes was identical but with different transition probabilities and risk of unplanned hospitalization. Each clone indicates that the patient will follow the pathway indicated at point A on the figure.

Input data

Estimates of the rate of successful implantation

Table 1 shows the rates of successful device implantation derived from total implantation experience (in both the CRT-P and MT groups) in the CARE-HF trial.7

View this table:
Table 1

Input value and distributions


Effectiveness is expressed as transition probabilities between Markov states. The transition probabilities among NYHA classes differed in the short- and long-run. In the short-run, we assumed there was an immediate response to treatments at the end of the first month.

Table 1 shows the estimated transition probabilities among NYHA classes based on available data derived from 388 (94.1%) and 380 (94.9%) patients in the CARE-HF CRT-P and MT groups, respectively.

The long-term treatment effect on NYHA class was assumed to follow constant transition probabilities. This is supported by the CARE-HF trial data.7 In the CARE-HF trial, outcomes including NYHA class have been measured at months 1, 6, 9, and 12, and every 6 months thereafter. The monthly transition probabilities from one NYHA class to another for the long-term were derived from NYHA classes assessed at month 1 and 6. We estimated monthly transition probability on the basis of the 5 month data by matrix algebra on the assumption of a constant Markov chain property during this period (Table 1). Each NYHA class was assigned a utility score independent of treatment which was estimated from quality-of-life assessments made during CARE-HF using the EQ-5D at baseline and 90 days (Table 2). Utility weights were combined with life-years to estimate QALYs.

View this table:
Table 2

Costs and utilities

View this table:
Table 3

Probabilities of events and associated distributions

Estimating time to sudden death

Estimates of the time to sudden death were based on the Kaplan–Meier survival curve extrapolated using a Weibull distribution from the CARE-HF trial (Table 3).7

Estimating time to unplanned hospitalization

The baseline survival function for time to cardiovascular unplanned hospitalization was derived from the time to first unplanned hospitalization using a Weibull survival function. Table 3 shows the estimated hazards ratios by different NYHA class with NYHA class I as reference case.

Estimating the risk reduction from ICD

The estimated additional benefit of ICD in reducing sudden death was based on the observed rate of sudden death in patients assigned to CRT-P in CARE-HF and the difference in sudden death rates in the COMPANION trial between the CRT-P and CRT–ICD groups, based on a median follow-up in that trial of 16 months in the device therapy groups.6,13 We assumed no additional benefit apart from preventing sudden death attributable to ICD. The monthly probability of hospitalization has been reported to be similar for CRT-P and CRT–ICD in the COMPANION trial12 (0.098 and 0.097 respectively), so no further penalty was applied to CRT–ICD patients for hospitalization rates due to the presence of the ICD component.

Cost analysis

The economic analysis was conducted from a UK NHS perspective, including device cost of CRT-P and CRT–ICD, implantation procedure cost, cost of hospitalization (hospital stay during implantation and unplanned hospitalization), medical care cost, and drug costs, and were converted to euros based on a conversion rate of €1.47 = £1. Implantation cost included device cost, procedure cost, intravenous medication, and hospital stay (including ICU and CCU). Table 2 summarizes the cost data by different categories.

Medical care cost included outpatient visits, cardiology or primary care visits, and length of time spent in nursing or residential homes or rehabilitation centres. Cost per patient per day for medical care and drug cost were estimated from CARE-HF.7 We assumed the same drug and medical care costs per day for all treatment groups.

Unplanned hospitalization for a major cardiac event was characterized by the presence or absence of a procedure cost. Procedure costs included ICU, CCU, CABG, PTCA, and heart transplantation. Procedure costs were based on the frequency and cost of events, and average costs for ICU and CCU. The unit costs employed have been previously reported.9

Battery life

On the basis of product specifications, we assumed that batteries were replaced for surviving patients in the CRT-P group every 6 years, and every 7 years in the CRT–ICD group.15,16 In order to examine the influence of battery life on the cost-effectiveness of the CRT–ICD device, which will vary with the device used and the specific programming employed, we also examined the cost-effectiveness of CRT–ICD using a device life of 4, 5, 6, and 8 years. The cost associated with battery replacement was the device cost plus one cardiac out-patient visit day.

Patient age

While the base case assumed a starting age of 65, this was varied in sensitivity analyses to examine the extent to which patient starting age (and thus life expectancy) affected the results of the model.


We adopted a discount rate of 3.5% annually for costs and benefits.

Probabilistic sensitivity analysis

We conducted probabilistic sensitivity analysis across all input values, together with scenario analysis of the assumptions within the model. Tables 13 list all input values and their respective distributions used to examine second-order uncertainty.17 Each set of random input values was drawn on the bass of their specific distributions for every 10 000 patients and the results were iterated 1000 times. We constructed cost-effectiveness acceptability curves to illustrate uncertainty.

Model validation

We validated the model using the observed results of CARE-HF and the published results from COMPANION.9,13



For the base-case analysis, where all mean input values were used based on 10 000 individual simulations, and patients started at a fixed age of 65, the predicted median survival was 7.44, 10.53, and 11.98 years for MT, CRT-P and CRT–ICD, respectively and 75% of patients were dead by 11.33, 15.92, and 17.92 years (Table 4, Figure 3). The undiscounted life gained for CRT-P vs. MT was 3.09 years and for CRT–ICD vs. CRT-P was 1.45 years.

Figure 3

Model predicted survival curves for MT, CRT-P, and CRT–ICD.

View this table:
Table 4

Model predicted survival

Cost-effectiveness results

Tables 57 show the difference in costs, life years, and QALYs by group for the base case. The total cost per patient for CRT–ICD + MT was €87 350 compared with €53 996 and €39 060 for CRT-P + MT and MT, respectively. The mean life-time QALYs were 6.75, 6.06, and 4.08 and life years was 9.16, 8.23, and 6.10 for CRT–ICD + MT, CRT-P + MT, and MT, respectively.

View this table:
Table 5

Cost-effectiveness result for the base case assumptions

View this table:
Table 6

ICERs per QALY for the base case assumptions

View this table:
Table 7

ICERs per LYG for the base case assumptions

For the comparison of CRT-P + MT and MT, the probabilistic sensitivity analysis gave an incremental cost of €14 935, a QALY score of 1.98, and a life-year estimate of 2.13. This gives ICERs of €7,538 (95% CI €5325–€11 784) per QALY gained and €7011 (95% CI €5346–€10 003) per life-year gained. The incremental cost-effectiveness of CRT–ICD + MT vs. MT is €18 017 (95% CI €14 500–€25 070) per QALY gained, and €15 780 (95% CI €12 955–€20 728) per life-year gained. For CRT–ICD + MT vs. CRT-P + MT, the incremental cost is €33 354, the QALY score is 0.70, and the life years gained is 0.93. The ICER here is €47 909 (95% CI €35 703–€79 438) per QALY gained, and €35 864 (95% CI €26 709–€56 353) per life year gained.

Figures 4A and B present the cost-effectiveness acceptability curves for CRT-P + MT and CRT–ICD + MT vs. MT, and CRT–ICD + MT vs. CRT-P + MT respectively. On the basis of a willingness to pay threshold of €44 100 (£30 000)/QALY, CRT–ICD + MT has a probability of 0.40 of being cost-effective compared with CRT-P + MT treatment alone.18

Figure 4

(A) Cost-effectiveness acceptability curves of CRT (+/ − ICD) vs. MT. (B) Cost-effectiveness acceptability curves of CRT–ICD vs. CRT-P.

Analyses by cohort age

We modelled patient groups who started at age 55, 60, 70, and 75 (Table 8, Figure 5). If patients received CRT–ICD at age 60, the ICER for the comparison with CRT-P alone decreased from €47 909–€42 701 per QALY gained, and for patients starting at age 55, the ICER fell to €36 777. Similarly, for patients starting at age 75, the ICER rose to €73 299.

Figure 5

Incremental cost per QALY gained by different starting age at treatment (in EUROs).

View this table:
Table 8

Estimated mean incremental cost per QALY for different starting ages

Length of follow-up

The effect of varying the period of follow-up in the model is shown in Table 9.

View this table:
Table 9

Estimated mean incremental cost per QALY at different durations of follow-up for the base case population

Battery life

The effect of different battery life for CRT–ICD on the incremental cost per QALY is described in Table 10. Reducing battery life to 4 years, the cost per QALY for CRT–ICD + MT vs. CRT-P + MT was increased to €75 091. Conversely, increasing battery life to 8 years reduced the cost per QALY for this comparison to €43 506.

View this table:
Table 10

Estimated mean incremental cost per QALY at different battery life for CRT–ICD device for the base case assumptions

Model validity

In Table 9, we assessed the internal validity of the model output by estimating a variety of shorter-term effects to contrast with other models12 and with the within-trial analysis from CARE-HF trial.9 Figure 6 contrasts the model predicted and observed survival in the CARE-HF trial.

Figure 6

Model predicting survival with the CARE-HF trial age matched cohort and the trial based Kaplan–Meier estimates of survival curves.


For the base case, CRT-P appears a highly cost-effective addition to MT among eligible patients. CRT–ICD + MT also appears cost-effective compared with MT, although this may not be considered an appropriate comparison as the treatment with high cost relative to benefits may appear cost-effective when combined with a highly cost-effective regimen. However, from a life-time perspective, assuming a reasonable life expectancy when receiving effective treatment for heart failure, CRT–ICD + MT may still appear cost-effective, although to a lesser extent, compared with CRT-P + MT.18 CRT-P + MT appeared cost-effective in all age groups. The cost-effectiveness of CRT-P + MT for patients in the ninth decade of life may seem surprising. This gain reflects a substantial benefit on quality-of-life among survivors, and some increase in longevity. The cost-effectiveness of CRT–ICD + MT is substantially greater in younger subjects, due to the longer potential period when the subject is at risk of sudden death. The cost-effectiveness of CRT–ICD + MT compared with CRT-P + MT was lower in older people partly because these treatments exert similar effects on quality-of-life and because older patients were likely to die of other problems even if sudden death was treated. Varying the period of follow-up in the model (Table 9) indicates the sensitivity of the results for CRT–ICD + MT to the duration of follow-up being considered, effectively the duration of the patients' exposure to the risk of sudden death. It also indicates the similarity of the model results to our previously reported within-trial analysis.9

Sensitivity analyses varying the length of CRT–ICD battery life significantly affects the cost-effectiveness of CRT–ICD + MT compared with CRT-P + MT. Our base-case analysis considered a battery life of 6 years for CRT-P and 7 years for CRT–ICD based on information from the manufacturer. Estimates of device longevity for older generation CRT–ICD products were 5–6 years, however, current developments in lead technology mean that lower battery consumption is needed to stimulate effectively the chamber with an appropriate safety margin. However, the device longevity for the CRT–ICD device also assumed an energy conserving programming specification, which may not be used routinely in practice, and the device longevity we assumed in the base case may not be achieved in practice.19,20 Longer battery life clearly results in an increased cost-effectiveness of CRT–ICD + MT compared with CRT-P + MT, but conversely shorter device life in practice will result in reduced cost-effectiveness. Further improvements in device technology, including the duration of battery life, present an important future challenge to the device manufacturers.

This model-derived analysis extends our previously published within-trial analysis based upon 29.4 months of mean follow-up.9 It also further advances the work described in the COMPANION cost-effectiveness analysis which provided estimates of benefit at 7 years which are similar to those observed in our model at 6 years. In addition, our work examines the ICER associated with adding an ICD component to CRT therapy.

Our analysis has a number of strengths. The existing clinical trials provide considerable evidence for the long-term effectiveness of both CRT-P and CRT–ICD, but most patients were alive and many felt well at the end of the trials. Patients' treatment does not cease at the end of the trial and it is inappropriate to assume that benefits cease at that point. In taking a life-time approach, we have been able to consider important issues such as device replacement, which none of the existing trials has had long enough follow-up to address. Economic modelling also enables the inclusion of data and other evidence from a range of sources in order to examine health policy questions.

The question of whether to implant CRT-P or CRT–ICD remains controversial, and as Jarcho21 has noted, a definitive trial comparing these treatments may never be conducted. In the absence of evidence from a definitive trial comparing CRT-P and CRT–ICD, economic modelling provides one useful perspective to health policy decisions.22 Our analysis indicates that CRT–ICD is cost-effective compared with CRT-P for patients with a reasonable life expectancy (when treated). However, the cost-effectiveness of CRT–ICD is sensitive to the frequency and cost of device replacements. Reduced device cost and increased device longevity both increase the cost-effectiveness of CRT–ICD.

There are a number of limitations to our analysis. The analysis is based on simulation rather than the direct observation of event rates achieved in a randomized trial, although simulation that has been constructed from a large scale long-term trial and in which the additional benefits of CRT–ICD are addressed using individual patient data from the CARE-HF trial to identify potentially preventable sudden deaths, and a further randomized trial of the effects of ICD on sudden death (COMPANION13). Thus our work may be considered a best-evidence synthesis of the likely cost-effectiveness of CRT-P and CRT–ICD, although the strength of that evidence is not as high as direct observation from sufficiently powered and appropriately designed randomized trials.


Long-term treatment with CRT-P + MT appears cost-effective compared with MT alone. CRT–ICD + MT was also cost-effective, although to a lesser extent, compared with CRT-P + MT at a willingness to pay of €44 100 (£30 000) per QALY, in the treatment of patients with moderate-to-severe heart failure characterized by dyssynchrony, except in those who have a poor life expectancy.


This study was funded by an unrestricted research grant from Medtronic Inc. N.F., J.G.F.C., and J.C.D. have received funding for research, consultancy, and travel from companies that manufacture devices for the treatment of heart failure and other related conditions. M.J.C. has received salary support from grants funded by Medtronic Inc.

Conflict of interest: Medtronic Inc., who funded this study, had the option to comment upon the final draft of the manuscript. The funders had no involvement in the design or analysis of the study, or in the drafting of the paper other than providing comments as above. The authors were not obliged to incorporate comments received from the sponsor. G.L.Y. and S.B. have no conflicts of interest.


View Abstract