Follow-up Duration, Competing Risk and Lifespan-Gain From ICDs
Follow-up Duration, Competing Risk and Lifespan-Gain From ICDs
Lifespan-gain from ICD implantations is underestimated when benefit is assessed only over short-time periods such as the duration of RCTs. Compared with lifespan-gain at 3 years (the duration of most RCTs), benefit is —two to three times greater at 5-year follow-up and 6- to 9-fold greater at 10-year follow-up.
The majority of ICDs are now implanted in lower risk patients, for primary prevention. Naive inspection of RCT results might suggest maximal benefit in high-risk patients, but this is because follow-up has to be relatively short. In contrast, taking a longer term view by the analysis processes described in this manuscript reveals that lower risk patients stand to gain the greatest lifespan. Increasing risk from non-cardiac disease causes attenuation in lifespan-gain in both the trial and post-trial period.
Survival modelling allows projection of lifespan-gain from an intervention over a much greater time-frame than is possible with a clinical trial. The model uses trial data to establish the hazard ratio associated with an intervention such as ICD implantation. The Gompertz–Makeham formulation adjusts for the effect of increasing age and co-morbidities with time. As the patient becomes older, there is an increased probability of non-cardiac pathologies and the device therefore becomes less effective at reducing total mortality due to an increase in non-cardiac death.
Modelling allows assessment of lifespan-gain over follow-up periods of 20 years or more. It cannot replace real-life long-term follow-up data but has the advantage of being able to control for the effects of an intervention and confounding factors such as non-cardiac disease. It is also dependent on the trial data used to derive hazard ratios. While this means that the model is based on a strong body of evidence, the hazard ratio for ICD implantation will reflect disease management at the time.
Lifespan-gain can be observed directly from trial data, but the short duration of most trials leads to gross underestimation of a patient's total lifespan-gain. Survival modelling allows projection of lifespan-gain from an intervention over a much greater timeframe. It cannot replace real-life long-term follow-up data but has the advantage of being able to control for the effects of an intervention and confounding factors such as non-cardiac disease.
For short durations, such as the 3- to 5-year follow-up of many of the landmark ICD trials, patients with the greatest lifespan-gain are those at highest risk of having a defibrillator-preventable event. Lower risk populations obtain little benefit over short follow-up due to their low event rate. In these patients, it is only when follow-up continues for the lifetime of a patient that the true benefit from ICD implantation is revealed. Our analysis suggests in low-risk patients that lifespan-gain may be up to nine times greater after 10-year follow-up than after 3-year follow-up. Our results are consistent with previous economic analysis using data from SCD-HeFT, and MADIT II trials. This has large clinical implications as the majority of ICDs are now being implanted in the lower risk primary prevention populations.
Economic analyses are more dependent on a 'per ICD' metric for lifespan-gain, whereas the 'per patient' data are more relevant for day-to-day clinical practice. For lifetime analysis, lifespan-gain per device was greatest in the higher risk patients (due to the lower number of replacement devices required) whereas lifespan-gain per patient was greatest in the lower risk patients. We suggest that presentations of benefit, and analyses of cost effectiveness, should quote both values.
Lifespan-gain from ICD implantation is attenuated in the trial and the post-trial period by competing risks that are not accessible to ICD therapy. This was recognized in the formulation of 'years needed to treat', a measure that accounts for potential long-term benefits and costs of pharmacological therapy.
In real life, heart failure patients often have multiple comorbidities. Within SCD-HeFT and MADIT-II, older subpopulations with renal dysfunction and other non-cardiac comorbidities were identified who did not show survival benefit from ICD implantation ( Table 2 ). Patients with heart failure requiring diuretics are more likely to die prior to the first appropriate ICD therapy and patients with significant co-morbidities are known to have a higher proportion of non-sudden death. Our analysis indicates that device implantation confers little gain in lifespan within the trial and post-trial period when the patient is at high-risk of non-cardiac death.
It is the interplay of cardiac sudden and non-sudden death, as well as competing risks, determines whether a patient will ultimately benefit from ICD implantation. The proportion of sudden and non-sudden cardiac death varies depending on annual baseline mortality. Patients with high baseline mortality tend to have a much greater proportion of non-sudden cardiac death (50% at 25% annual mortality in the Seattle Heart Failure data compared with 10% at 5% annual mortality). In contrast, the efficacy of ICD therapy in reduction of sudden cardiac death is in the lower risk patient population (Figure 4). When a more realistic variable hazard ratio for ICD-efficacy was used, as seen from the SCD-HeFT data, this effect was even more dramatic (Figure 5). Use of scoring systems such as the Seattle Heart Failure score allows estimation of likely benefit from ICD implantation.
There are multiple reasons why current-day ICD implantation may be able to deliver more benefit in lifespan than calculated using data from the original landmark trials:
It is therefore important to be cautious in interpreting our results. Exactly how much more beneficial current-day ICDs might be is difficult to state with confidence. We can estimate using, for example, the hazard ratio of MADIT-RIT (0.45; 95% CI, 0.24–0.85) for the frugal therapy compared with historically conventional therapy. If this tremendous step-change is applicable across all ICD recipients, lifespan gained could be almost twice as large as we have calculated.
Avoidance of RV pacing will contribute further, as demonstrated by the more efficacious hazard ratio from our analysis and as seen in the DAVID trial. While the exact effect of all these features combined cannot be stated exactly, we suggest a potential estimate of the overall hazard ratio for modern ICD strategies might be 0.45× that seen in the landmark trials. This would mean an overall hazard ratio of 0.45 × 0.41 = 0.18 (Figure 7C).
(Enlarge Image)
Figure 7.
Potential lifespan-gain with different defibrillator strategies. (A) A population with a 10% annual mortality and 15% competing risk as per the original model. (B) The impact of not replacing the implantable cardioverter defibrillator at end of battery life if, for example, the patient has improved such that they no longer met criteria for implantable cardioverter defibrillator implantation. The period of improved survival from the single device results in sustained lifespan benefit. In (C), we have incorporated the effect of optimal implantable cardioverter defibrillatorprogramming using the data from MADIT RIT, resulting in an even greater lifespan-gain.
While absolute survival benefit in lower risk populations is modest within the short time-frame of clinical trials, our results using a longer time-horizon suggest substantially greater benefit (Figures 4 and 5). This is concordant with the findings when the SHFM was applied to SCD-HeFT with an ~4–5 years increase in estimated lifespan with an ICD in patients with an annual mortality of 5%. These results suggest that perhaps we should not delay consideration of ICD therapy in lower risk populations.
Improved heart failure therapy through widespread utilization of CRT, more finely tuned coronary interventions both in acute myocardial infarction and chronic ischaemic heart disease, and greater uptake of drugs including angiotensin-converting enzyme inhibitors, β-blockers and mineralocorticoid antagonists, may lead to progressive improvement in ventricular function so that a patient eligible for an ICD at the outset might not meet criteria for implantation at the end of battery life. Nevertheless, it should be remembered that the analysis must not stop at that time point, since that would be only be valid if all patients who had their lives saved by ICD therapy were to die on that day. In reality, the higher survival in those that had the ICD would continue after its battery life, as shown in Figure 7B. Thus again the lifespan-gain per ICD implanted may be larger than calculated here.
We utilized data from multiple landmark ICD trials which have varying inclusion criteria. Some of these trials, particularly those of secondary prevention, are now over 10 years old and the standard of optimal medical therapy and revascularization decisions in the control arm will therefore be different to contemporary management. This may mean that the hazard ratio for sudden death prevention from an ICD may differ if such trials were performed using contemporary management in the control arm, although there is no evidence to suggest an attenuation of benefit from ICD implantation in patients with milder heart failure.
Post-trial survival was assessed using a Gompertz–Makeham function. This has been well validated as a descriptor of long-term survival patterns but in our study only provides a projection of potential benefit rather than direct observation. Other models including the DEALE method may be employed but may overestimate mortality over longer time horizons. It is unlikely that long-term survival data from ICD trials can be totally relied up to provide an unbiased estimate of lifespan-gain from ICD implantation as patients will cross over from control to active arms following completion of the trial. This artificially attenuates benefit, through equalization of therapy. Therefore, methods such as those we present may be a useful alternative approach to assess lifetime benefit.
We assessed lifespan-gain but did not assess quality-adjusted life years (QALYs) or other measures of health related quality of life. As the patients with the most to gain from device implantation were those with milder disease who usually have a numerically higher quality of life using such scoring systems, it is likely that the gain in the lower risk population would exceed that in the higher risk population by even more in QALY terms than in plain lifespan terms.
It should not be forgotten that some ICDs or leads can suffer recalls, which contribute additional cost to the healthcare system and inconvenience and worry to the patient. Although rates of requiring replacement of device or lead are low (2–3% of all ICD implants), any formal cost-effectiveness analysis should address this element. Routine end-of-life replacement of the ICD generator is also not without risk, including infection or haematoma in the pocket, or trauma or displacement of the lead. While the balance of these complications is different between first and replacement devices, the total rates of device-related complications are similar between initial and follow-up device replacement. The impact of these complications on lifespan-gain should therefore be accounted for using the data derived from the landmark trials.
Discussion
Lifespan-gain from ICD implantations is underestimated when benefit is assessed only over short-time periods such as the duration of RCTs. Compared with lifespan-gain at 3 years (the duration of most RCTs), benefit is —two to three times greater at 5-year follow-up and 6- to 9-fold greater at 10-year follow-up.
The majority of ICDs are now implanted in lower risk patients, for primary prevention. Naive inspection of RCT results might suggest maximal benefit in high-risk patients, but this is because follow-up has to be relatively short. In contrast, taking a longer term view by the analysis processes described in this manuscript reveals that lower risk patients stand to gain the greatest lifespan. Increasing risk from non-cardiac disease causes attenuation in lifespan-gain in both the trial and post-trial period.
Modelling Survival in the Post-trial Period
Survival modelling allows projection of lifespan-gain from an intervention over a much greater time-frame than is possible with a clinical trial. The model uses trial data to establish the hazard ratio associated with an intervention such as ICD implantation. The Gompertz–Makeham formulation adjusts for the effect of increasing age and co-morbidities with time. As the patient becomes older, there is an increased probability of non-cardiac pathologies and the device therefore becomes less effective at reducing total mortality due to an increase in non-cardiac death.
Modelling allows assessment of lifespan-gain over follow-up periods of 20 years or more. It cannot replace real-life long-term follow-up data but has the advantage of being able to control for the effects of an intervention and confounding factors such as non-cardiac disease. It is also dependent on the trial data used to derive hazard ratios. While this means that the model is based on a strong body of evidence, the hazard ratio for ICD implantation will reflect disease management at the time.
Impact of Follow-up Duration on Who Gains Most Lifespan: Modelling Survival in the Post-trial Period
Lifespan-gain can be observed directly from trial data, but the short duration of most trials leads to gross underestimation of a patient's total lifespan-gain. Survival modelling allows projection of lifespan-gain from an intervention over a much greater timeframe. It cannot replace real-life long-term follow-up data but has the advantage of being able to control for the effects of an intervention and confounding factors such as non-cardiac disease.
For short durations, such as the 3- to 5-year follow-up of many of the landmark ICD trials, patients with the greatest lifespan-gain are those at highest risk of having a defibrillator-preventable event. Lower risk populations obtain little benefit over short follow-up due to their low event rate. In these patients, it is only when follow-up continues for the lifetime of a patient that the true benefit from ICD implantation is revealed. Our analysis suggests in low-risk patients that lifespan-gain may be up to nine times greater after 10-year follow-up than after 3-year follow-up. Our results are consistent with previous economic analysis using data from SCD-HeFT, and MADIT II trials. This has large clinical implications as the majority of ICDs are now being implanted in the lower risk primary prevention populations.
Economic analyses are more dependent on a 'per ICD' metric for lifespan-gain, whereas the 'per patient' data are more relevant for day-to-day clinical practice. For lifetime analysis, lifespan-gain per device was greatest in the higher risk patients (due to the lower number of replacement devices required) whereas lifespan-gain per patient was greatest in the lower risk patients. We suggest that presentations of benefit, and analyses of cost effectiveness, should quote both values.
Effect of Competing Risks on Lifespan-Gain From Implantable Cardioverter Defibrillator Implantation
Lifespan-gain from ICD implantation is attenuated in the trial and the post-trial period by competing risks that are not accessible to ICD therapy. This was recognized in the formulation of 'years needed to treat', a measure that accounts for potential long-term benefits and costs of pharmacological therapy.
In real life, heart failure patients often have multiple comorbidities. Within SCD-HeFT and MADIT-II, older subpopulations with renal dysfunction and other non-cardiac comorbidities were identified who did not show survival benefit from ICD implantation ( Table 2 ). Patients with heart failure requiring diuretics are more likely to die prior to the first appropriate ICD therapy and patients with significant co-morbidities are known to have a higher proportion of non-sudden death. Our analysis indicates that device implantation confers little gain in lifespan within the trial and post-trial period when the patient is at high-risk of non-cardiac death.
Mode of Death and Implantable Cardioverter Defibrillator Effectiveness: Can a High-Risk Patient Ever Benefit From Implantable Cardioverter Defibrillator Implantation?
It is the interplay of cardiac sudden and non-sudden death, as well as competing risks, determines whether a patient will ultimately benefit from ICD implantation. The proportion of sudden and non-sudden cardiac death varies depending on annual baseline mortality. Patients with high baseline mortality tend to have a much greater proportion of non-sudden cardiac death (50% at 25% annual mortality in the Seattle Heart Failure data compared with 10% at 5% annual mortality). In contrast, the efficacy of ICD therapy in reduction of sudden cardiac death is in the lower risk patient population (Figure 4). When a more realistic variable hazard ratio for ICD-efficacy was used, as seen from the SCD-HeFT data, this effect was even more dramatic (Figure 5). Use of scoring systems such as the Seattle Heart Failure score allows estimation of likely benefit from ICD implantation.
Lifespan-Gain May Be Larger in Modern Practice
There are multiple reasons why current-day ICD implantation may be able to deliver more benefit in lifespan than calculated using data from the original landmark trials:
The hazard ratios used in our calculations are based on the best available evidence from RCTs. However, many of these trials are now over 10 years old, and therefore do not capture the effects of improved clinical practice that have since occurred, e.g. device insertion aimed at minimizing isolated RV pacing with implantation of fewer dual chamber ICDs and more CRT-D in eligible patients. By reducing the risk of death from pump failure, this may potentially increase the benefit from ICD implantation.
Modern refinements to ICD programming use anti-tachycardia pacing and defibrillation more frugally. Reducing episodes of therapy in this way has been shown to improve mortality.
No RCT to date has focused on patients with channelopathies, purely genetic arrhythmias, arrhythmogenic RV cardiomyopathy, or other congenital heart disease. These conditions, which can be diagnosed at a young age and are associated with little or no co-morbidities, will likely generate a long period of large hazard ratio reduction, permitting larger lifespan-gains than calculated here.
Modern ICDs have growing capacity for home monitoring, feeding back information on episodes of tachyarrhythmia, anti-tachycardia pacing episodes, impedance changes, amounts of pacing required, lead impedance changes, and device faults. This wealth of information may assist physicians to improve outcomes further.
It is therefore important to be cautious in interpreting our results. Exactly how much more beneficial current-day ICDs might be is difficult to state with confidence. We can estimate using, for example, the hazard ratio of MADIT-RIT (0.45; 95% CI, 0.24–0.85) for the frugal therapy compared with historically conventional therapy. If this tremendous step-change is applicable across all ICD recipients, lifespan gained could be almost twice as large as we have calculated.
Avoidance of RV pacing will contribute further, as demonstrated by the more efficacious hazard ratio from our analysis and as seen in the DAVID trial. While the exact effect of all these features combined cannot be stated exactly, we suggest a potential estimate of the overall hazard ratio for modern ICD strategies might be 0.45× that seen in the landmark trials. This would mean an overall hazard ratio of 0.45 × 0.41 = 0.18 (Figure 7C).
(Enlarge Image)
Figure 7.
Potential lifespan-gain with different defibrillator strategies. (A) A population with a 10% annual mortality and 15% competing risk as per the original model. (B) The impact of not replacing the implantable cardioverter defibrillator at end of battery life if, for example, the patient has improved such that they no longer met criteria for implantable cardioverter defibrillator implantation. The period of improved survival from the single device results in sustained lifespan benefit. In (C), we have incorporated the effect of optimal implantable cardioverter defibrillatorprogramming using the data from MADIT RIT, resulting in an even greater lifespan-gain.
Implications for Heart Failure Management
While absolute survival benefit in lower risk populations is modest within the short time-frame of clinical trials, our results using a longer time-horizon suggest substantially greater benefit (Figures 4 and 5). This is concordant with the findings when the SHFM was applied to SCD-HeFT with an ~4–5 years increase in estimated lifespan with an ICD in patients with an annual mortality of 5%. These results suggest that perhaps we should not delay consideration of ICD therapy in lower risk populations.
Improved heart failure therapy through widespread utilization of CRT, more finely tuned coronary interventions both in acute myocardial infarction and chronic ischaemic heart disease, and greater uptake of drugs including angiotensin-converting enzyme inhibitors, β-blockers and mineralocorticoid antagonists, may lead to progressive improvement in ventricular function so that a patient eligible for an ICD at the outset might not meet criteria for implantation at the end of battery life. Nevertheless, it should be remembered that the analysis must not stop at that time point, since that would be only be valid if all patients who had their lives saved by ICD therapy were to die on that day. In reality, the higher survival in those that had the ICD would continue after its battery life, as shown in Figure 7B. Thus again the lifespan-gain per ICD implanted may be larger than calculated here.
Study Limitations
We utilized data from multiple landmark ICD trials which have varying inclusion criteria. Some of these trials, particularly those of secondary prevention, are now over 10 years old and the standard of optimal medical therapy and revascularization decisions in the control arm will therefore be different to contemporary management. This may mean that the hazard ratio for sudden death prevention from an ICD may differ if such trials were performed using contemporary management in the control arm, although there is no evidence to suggest an attenuation of benefit from ICD implantation in patients with milder heart failure.
Post-trial survival was assessed using a Gompertz–Makeham function. This has been well validated as a descriptor of long-term survival patterns but in our study only provides a projection of potential benefit rather than direct observation. Other models including the DEALE method may be employed but may overestimate mortality over longer time horizons. It is unlikely that long-term survival data from ICD trials can be totally relied up to provide an unbiased estimate of lifespan-gain from ICD implantation as patients will cross over from control to active arms following completion of the trial. This artificially attenuates benefit, through equalization of therapy. Therefore, methods such as those we present may be a useful alternative approach to assess lifetime benefit.
We assessed lifespan-gain but did not assess quality-adjusted life years (QALYs) or other measures of health related quality of life. As the patients with the most to gain from device implantation were those with milder disease who usually have a numerically higher quality of life using such scoring systems, it is likely that the gain in the lower risk population would exceed that in the higher risk population by even more in QALY terms than in plain lifespan terms.
It should not be forgotten that some ICDs or leads can suffer recalls, which contribute additional cost to the healthcare system and inconvenience and worry to the patient. Although rates of requiring replacement of device or lead are low (2–3% of all ICD implants), any formal cost-effectiveness analysis should address this element. Routine end-of-life replacement of the ICD generator is also not without risk, including infection or haematoma in the pocket, or trauma or displacement of the lead. While the balance of these complications is different between first and replacement devices, the total rates of device-related complications are similar between initial and follow-up device replacement. The impact of these complications on lifespan-gain should therefore be accounted for using the data derived from the landmark trials.