OUP user menu

Absolute risk reduction in total mortality with implantable cardioverter defibrillators: analysis of primary and secondary prevention trial data to aid risk/benefit analysis

Timothy R. Betts, Praveen P. Sadarmin, David R. Tomlinson, Kim Rajappan, Kelvin C.K. Wong, Joseph P. de Bono, Yaver Bashir
DOI: http://dx.doi.org/10.1093/europace/eus427 813-819 First published online: 30 January 2013


Aims Absolute risk reduction (ARR) and number needed to treat (NNT) are considered by many to be the most appropriate figures to use for the informed consent process, yet the results of published implantable cardioverter defibrillators (ICD) trials are frequently presented as relative risk reduction or odds ratio. The period over which risk reduction is calculated also varies between trials, making comparison difficult.

Methods and results Published ICD trials used to formulate national and international guidelines were examined for 1, 2, and 3 year total mortality in ICD and medically treated patients. The number of patients enrolled and at risk at these time points were also sought. Where the raw data were not included in the original text, estimates were taken from published Kaplan–Meier graphs. Eight primary prevention (PP) trials, three secondary prevention (SP) trials, and one SP meta-analyses were included. For PP, ARR at 3-year follow-up ranged from 0 (no benefit) to 24.6% (NNT = 4). For SP, ARR at 3-year follow up ranged from 3.7% (NNT = 27) to 11.3% (NNT = 9). Absolute risk reduction increased with follow-up in PP trials, whereas there was considerable variation in SP trials. Overall, very few trial patients received 3-year follow-up, giving wide confidence intervals (CIs).

Conclusion Absolute risk reduction from ICD trials varies significantly depending upon trial entry criteria, subgroup characteristics, and duration of follow-up. The relatively small number of patients followed for 2 or more years leads to wide CIs. Despite these limitations, the standardized ARR and NNT data presented may give a more individualized estimate of risk/benefit that could potentially aid an informed consent process.

  • Absolute risk reduction
  • Relative risk reduction
  • Number needed to treat
  • Implantable cardioverter defibrillator
  • Primary prevention trials
  • Secondary prevention trials
  • Patient consent


Randomized controlled trials (RCTs) and meta-analyses have demonstrated implantable cardioverter defibrillators (ICD) reduce total mortality in selected high-risk patients by treating life-threatening ventricular tachyarrhythmias.18 This has led to national and international guidelines for the provision of ICDs in primary prevention (PP) as well as secondary prevention (SP) settings.9,10 Published ICD trials have differing inclusion criteria, varying lengths of follow-up and not all have demonstrated a statistically significant reduction in all-cause mortality. Subgroup analyses from some of the larger trials have also been published.1117 In the literature the benefit gained from an ICD is invariably presented in terms of relative risk reduction (RRR); however, it is the absolute risk reduction (ARR) and number needed to treat (NNT) that are considered the most appropriate forms of data presentation when explaining risks and benefits to patients and determining cost effectiveness.1823 The aim of this study was to establish 1-, 2-, and 3-year ARR and NNT data from published ICD RCTs to provide an aid for risk/benefit analysis and informed consent, and comparison between different patient groups.


The RCTs included in the analysis were identified through the Medline database using the search criteria RCT, implantable cardioverter defibrillator and sudden cardiac death, and from citations in national guidelines.24,25 Landmark trials that compared non-thoracotomy ICD implantation with standard medical therapy were included if the publication contained annual survival data in the form of percentage all-cause mortality or survival, or Kaplan–Meier graphs of total mortality or survival. Subgroup analyses of RCTs with annual survival data or Kaplan–Meier total mortality graphs were also included and are presented as supplementary data. Annual mortality rates for the ICD and medically treated patient groups in each trial were taken from the text, and ARR and NNT calculated for 1, 2, and 3 years of follow-up. Where the percentage or actual patient numbers were not available in the text, the annual mortality was estimated from Kaplan–Meier curves using the method described by Salukhe et al.26 Published graphs were enlarged and the estimated survival or mortality measured at the 1, 2, and 3 year points by five blinded observers. Inter- and intra-observer error was calculated by performing estimates on Kaplan-Meier curves from five publications in which the annual survival data were also available. When also presented, the number of patients at risk at each annual measurement was used to calculate the 95% CIs.27


Eight PP RCTs (MADIT,6 MUSTT,2 MADIT2,7 SCD-HeFT,1 DEFINITE,28 AMIOVIRT,29 CAT,30 and DINAMIT31), three SP RCTs (AVID,8 CIDS,3 and CASH5) and one SP meta-analysis4 were included in the data analyses. The CABG Patch trial32 was excluded due to the sole use of thoracotomy ICD implantation with concomitant coronary artery bypass graft surgery and the high in-hospital mortality. Three of the PP RCTs, two of the SP RCTs, and the SP meta-analysis also had published subgroup analysis available.1,1117,28 Trial data, including inclusion criteria and patient characteristics are shown in Tables 1 and 2. Data from the SCD-HeFT trial1 aetiology subgroups were used in preference to the overall combined statistics as the trial was powered for post hoc aetiology substudy analysis. The ICD recipients in SCD-HeFT were compared with the placebo group and the amiodarone group was excluded. Annual percentage survival data were available from the publication text in five RCTs and six substudies. The number of patients at risk was also identifiable in five RCTs and nine substudies. Estimation of survival was obtained from Kaplan–Meier graphs for four trials and five substudies. Inter- and intra-observer error for estimates on Kaplan–Meier curves was 0.82 ± 0.38% and 1.14 ± 0.54%, respectively.

View this table:
Table 1

Inclusion criteria, patient characteristics, and outcomes of primary prevention ICD trials

Entry criteriaIHD, EF ≤ 35% (nsVT and +ve EPS)IHD, EF ≤ 40% (nsVT and +ve EPS)IHD, EF ≤ 30%IHD, EF ≤ 35% (MI–6–40 days, impaired cardiac autonomic function)IHD or DCM, EF ≤ 35% (NYHA II or more)DCM, EF ≤ 35% (frequent PVCs or nsVT)DCM, EF ≤ 35% and NSVTDCM (<9 months), EF ≤ 30% (NYHA II or III)
No. of pts (ICD/Med)95/101351/353742/490332/344829/1692229/22951/5250/54
Ave age in years (ICD, Med)62 ± 9, 64 ± 966 ± 6, 65 ± 764 ± 10, 65 ± 1061.5 ± 11, 62.1 ± 1160.5 ± 9 (all pts)58.3 (all pts)58 ± 11, 60 ± 1252 ± 12, 52 ± 10
Mean LVEF % (ICD, Med)27 ± 7, 25 ± 727.5 ± 8, 28.5 ± 723 ± 5, 23 ± 628 ± 5, 28 ± 525, 2521 (all pts)22 ± 10, 23 ± 824 ± 6, 25 ± 8
Mean FU (months)27392030 ± 1345.529 ± 14.420.1 ± 12.622.7 ± 5, 22.9 ± 4
Med Group 3 year mortality (%)4235.43117.52219.42018
Outcome as reportedHR: 0.46 (0.26–0.82)
P = 0.009
RR: 0.40 (0.27–0.59)
P = <0.001
HR: 0.69 (0.51–0.93)
P = 0.016
HR: 1.08 (0.76–1.55)
P = 0.66
HR: 0.77 (0.62–0.96)
P = 0.007
HR: 0.65 (0.4–1.06)
P = 0.08
No diff P = 0.6No diff P = 0.554
  • EF, left ventricular ejection fraction; nsVT, non-sustained ventricular tachycardia; EPS, electrophysiological study; IHD, ischaemic heart disease; MI, myocardial infarction; DCM, dilated cardiomyopathy; PVCs, premature ventricular contractions; NYHA, New York Heart Association class; HR, hazard ratio; FU, follow up; pts, patients; diff, difference; Med, medical; Ave, average.

View this table:
Table 2

Inclusion criteria, patient characteristics, and outcomes of secondary prevention ICD trials

Inclusion criteriaVF; sustained VT with syncope; VT with LVEF ≤40% and haemodynamic compromiseVF; sustained VT with syncope; VT with LVEF ≤35% and haemodynamic compromise; unmonitored syncope with inducible VT at EPSCardiac arrest survivorsCriteria from all 3 trials
No. of patientsICDMedICDMedICDMedICDMed
Age (yrs)65 ± 1165 ± 1063.3 ± 9.263.8 ± 9.959 ± 1056 ± 1163 ± 1164 ± 10
LVEF (%)32 ± 1331 ± 1334.3 ± 14.533.3 ± 14.144 ± 1747 ± 1734 ± 1533 ± 14
VF or cardiac arrest as index arrhythmia (%)44.64545.150.11001005152
IHD (%)818182.982.2777081
Mean FU (months)18 ± 12.23557 ± 34 
3-year mortality in Med group (%)35.927.33932
Outcome as reportedARR: 8.2%
RRR: 39 ± 20%, 27 ± 21% and 31 ± 21% at 1, 2, and 3 years
P = 0.02
RRR: 19.7% (−7.7 to 40%)
P = 0.142
HR: 0.766
P = 0.081
HR: 0.72 (0.6 to 0.87)
P = 0.0006
  • Med, medical; ARR, absolute risk reduction; RRR, relative risk reduction; HR, hazards ratio; FU, follow-up.

Primary prevention trials

Results from PP trials are summarized in Table 3. The 3-year ARR for PP trials varied significantly between heart failure aetiologies, high-risk patients with ischaemic heart disease (IHD) enrolled in MADIT6 and MUSTT2 having the greatest reductions and smallest NNTs (ARR 24.6% and 19%, NNT 4 and 5, respectively). In the MADIT2 trial,7 where patient selection was solely based on left ventricular ejection fraction (LVEF) and limited to IHD, the 3-year RRR was 29% and the ARR was 9% (NNT 11). Patients enrolled in the SCD-HeFT trial with IHD had a 3-year RRR of 20% and an ARR of 5.6% (NNT 18), whereas patients in the same trial with non-ischaemic dilated cardiomyopathy (DCM) had a 3-year RRR of 25.3% and ARR of 4% (NNT 25). Of those trials that exclusively enlisted DCM patients, the DEFINITE trial28 showed a 3-year RRR of 32% and ARR of 6.2% (NNT 16), whereas the CAT and AMIOVERT trials showed a 2% absolute risk increase in mortality in the ICD group.

View this table:
Table 3

Mortality, ARR with 95% CI, RRR, and NNT at 1-, 2-, and 3-year follow-up from primary prevention trials

TrialFUMed (%)ICD (%)ARR and CI (%)NNTRRR (%)
MADITYear 123419 ± 10.9682.6
Year 233.412.620.8 ± 16562.3
Year 34217.424.6 ± 22.4458.6
MUSTTYear 115.6411.6 ± 4.6974.4
Year 228.21117.2 ± 6.7661
Year 335.416.419 ± 8.6553.7
MADIT2Year 11091 ± 47110
Year 222166 ± 7.51727.3
Year 331229 ± 13.71129
SCD-HeFT (IHD)Year 19.28.21 ± 3.910010.9
Year 218.215.23 ± 5.43316.5
Year 32822.45.6 ± 7.71820
SCD-HeFT (DCM)Year 13.63.60
Year ± 4.13130.2
Year 315.811.84 ± 5.92525.3
DEFINITEYear ± 3.82958.1
Year ± 7.41644
Year 319.413.26.2 ± 12.11632
CATYear 13.68−4.4
Year 27.88−0.2
Year 31820−2
AMIOVERTYear 1411−7
Year 2711−4
Year 32022−2
  • Med, medical mortality; ICD, ICD mortality; ARR, absolute risk reduction; NNT, number needed to treat; RRR, relative risk reduction; FU, follow-up; CI, confidence interval.

The DINAMIT study31 enrolled patients with low LVEF early after myocardial infarction (MI). The ICD group failed to show a benefit in reduction of total mortality at 1, 2, or 3 years of follow-up. As current guidelines exclude this group of patients from PP ICD implantation the DINAMIT study results were not included in further analysis. The total mortality in the medically treated groups was higher in trials of patients with IHD. These same patients also had the most to gain over 3 years in terms of ARR and NNT from an ICD, particularly those in MADIT where Holter monitoring and provocative electrophysiological testing was used to select patients. Total mortality in trial patients with DCM was less than those with IHD and as a consequence resulted in smaller ARR and failure for three trials with smaller patient numbers to show statistically significant reductions in mortality with the use of an ICD.

Primary prevention subgroup analysis

Results from published PP substudies are summarized in Supplementary material online, Tables S1–S3. In MADIT2 patients the ARR was greatest in those with New York Heart Association (NYHA) class III symptoms (13% at 3 years, NNT 8). Patients with NYHA class III symptoms in the DEFINITE trial also had the greatest ARR (18.2% at 3 years, NNT 5). Subgroup analysis of the SCD-HeFT trial, which combined IHD and non-IHD patients, showed the greatest benefit in NYHA class II patients (ARR 9.2% at 3 years, NNT 11) with no benefit in the NYHA class III group.

In the MADIT2 trial the greatest benefit at 3 years was seen in patients more than 18 months from their last MI (ARR 9.4%, NNT 11), with an LVEF < 21% (ARR 12%, NNT 8), age >75 years (ARR 13%, NNT 8), a risk score of 2 (ARR 29%, NNT 3) and of Caucasian origin (ARR 7%, NNT 14).

Secondary prevention trials and subgroup analysis

Results from published SP trials and the SP meta-analysis are summarized in Table 4. The three SP trials contained a mix of patients with IHD and non-IHD as well as different presenting arrhythmias. The ARR at 3 years ranged from 3.7 to 11.3% with the meta-analysis revealing an overall ARR of 8% (NNT 13). Meta-analysis subgroup analysis by LVEF showed greatest ARR at 3 years in those with LVEF < 35% (11.4%, NNT 9) compared to those with LVEF > 35% (ARR 2.2%, NNT 45).

View this table:
Table 4

Mortality, ARR with 95% CI, RRR and NNT at 1, 2, and 3-year follow-up from secondary prevention trials

TrialFUMed (%)ICD (%)ARR and CI (%)NNTRRR (%)
AVIDYear 117.710.77 ± 5.31439.4
Year 225.318.46.9 ± 8.81427.3
Year 335.924.611.3 ± 17.5931.5
CASHYear 11587 ± 7.81446.7
Year 225178 ± 10.81332
Year 33931.37.7 ± 14.21319.7
CIDSYear ± 4.95815.4
Year 220.914.76.2 ± 7.31629.7
Year 32723.33.7 ± 9.32713.7
Meta-analysisYear 115105 ± 3.42033.3
Year 224168 ± 5.21333.3
Year 332248 ± 7.71325
  • Med, medical mortality; ICD, ICD mortality; ARR, absolute risk reduction; NNT, number needed to treat; RRR, relative risk reduction; FU, follow-up; CI, confidence interval.

Benefit over time

In the PP RCTs, the ARR increased with length of follow-up (see Supplementary material online, Figure S1). In CAT and AMIOVERT there was less harm with increasing follow-up. In the SP trials only the AVID trial showed a notable increase in benefit between 2 and 3 years follow-up, yet as this is by far the largest trial it heavily influences the SP meta-analysis, offsetting the reduction in benefit in CIDS seen with longer follow-up (see Supplementary material online, Figure S2).

Correlation with total mortality in medically treated patients

In PP trials there was a strong correlation between medical group total mortality and ARR in the ICD group (P = 0.9, Figure 1). Patients with the highest medical mortality have the greatest benefit in ARR with an ICD. Subgroup analysis however revealed patient groups with very high total mortality and little or no apparent benefit from ICDs (see Supplementary material online, Tables S1–S4).

Figure 1

Comparison of mortality between medically treated group to ICD treatment group for the major primary prevention trials at the end of 3 years.

Number of patients at risk and confidence intervals

The number of ‘patients at risk’ used for ARR calculations at 3 years are displayed in Figure 2. Only 14% of MADIT2 and 10% of AVID patients were eligible for 3-year follow-up. The relatively short follow-up times in AVID heavily influenced the SP meta-analysis. Only CASH, MUSTT, SCD-HeFT, and CAT trials had more than half of patients enrolled eligible for 3-year mortality assessment. As a consequence, the relatively small patient numbers and magnitudes of ARR resulted in CIs of 3-year ARR being greater than the ARR itself, other than for the SP meta-analysis, MADIT, and MUSTT (Figure 3).

Figure 2

The poor follow-up of the total number of patients at risk at 1, 2, and 3 years in the major primary prevention and secondary prevention trials.

Figure 3

Three-year ARR and CIs for primary and secondary prevention trials. Circles represent primary prevention trials and squares secondary prevention trials and meta-analysis. The 95% CI bars are shown.


Communication of risk between physician and patient is an essential component of healthcare and is integral to the informed consent process.21 The results of RCTs are invariably published as RRR or odds ratio and this is the format healthcare providers are likely to use. Relative risk reduction looks at the proportional difference in risk between one alternative and another and is more effective in persuading patients to agree to a proposed treatment when compared with other risk categories.33 It may magnify the benefits gained as the numerical figures are often large and do not portray the baseline risk.19,20,27 A more appropriate and less confusing means of expressing risk and benefit from treatment is ARR, which is felt by many to be more clinically relevant.21,22,34 Absolute risk reduction also determines NNT which is a factor required for cost-effectiveness calculations.23 The RRR, ARR, and NNT need to be put into context against the total mortality in the untreated group to better understand the risks and benefit when subjecting patients to any procedure or therapy.35

Although it is possible to find NNT data in some published ICD RCTs, there is no consistency between trials regarding the time period over which the NNT is derived, making comparison between them challenging. The present study presents 3-year data including ARR, NNT, and RRR for all the published ICD RCTs that contribute to national and international guidelines (whether in their original format or as part of a meta-analysis), allowing comparison between primary and secondary prevention, aetiology of heart disease, and other patient characteristics. One of the principal findings is that patient groups with the highest mortality (e.g. those who meet MADIT or MUSTT trial entry criteria) have a small NNT and the most to gain from a PP ICD. Patients with DCM have a larger NNT. Within the SCD-HeFT trial, although the 3-year RRR was greater for DCM, the ARR was greater and the NNT smaller for those with IHD, offering ischaemic patients more ‘bang for their buck’. A comparison between primary and secondary prevention study data also shows that the NNT is smaller in the majority of PP trials (MADIT, MUSTT, and MADIT2) than the SP meta-analysis, a fact that may surprise many physicians and patients. This is likely to be the result of strict entry criteria into the early PP trials, selecting only the sickest of patients who represent a minority of patients with IHD.

The small numbers of patients in RCTs followed for a period of 2 or more years results in wide CIs and reduces the validity of risk reduction statistics over a longer time period, despite the fact that the ARR figures are larger. As the typical lifespan of an ICD is 5–7 years, follow-up data over longer time periods are required for cost–benefit analysis to account for the expensive ‘up-front’ costs.3638 Some authors have proposed that ‘life years saved’ is the most appropriate way to assess cost-effectiveness and that the benefits from an ICD increase exponentially over time.26 In this study the use of 3-year risk reduction figures are therefore a compromise between the advantages of portraying meaningful data for patient consent and cost-effectiveness analysis, and the disadvantages of a relatively small number of patients included in long-term follow-up.

The wide ranges of ARR in the MADIT2 trial LVEF subgroup analysis and the ‘risk score’ data suggest that certain patient groups have much to gain whereas in others there may be no benefit. Caution must be advised however when using subgroup data to influence decision making, particularly when not proposed in the original trial design.39 The MADIT2 substudy of black vs. white ethnicity reveals an ARR at 3 years of 7% in white patients but an increased mortality of 5% in black patients; however, there were only 154 out of 1073 white patients and 10 out of 102 black patients followed up for this length of time, resulting in very wide CI.15

Clinical guidelines are based on patient groups in whom a statistically significant reduction in total mortality with the use of an ICD has been demonstrated in an RCT or meta-analysis. It is yet to be determined what magnitude of ARR is deemed to be clinically or ethically significant by patients, physicians, and healthcare providers, even though this will inevitably guide choice and decision making. The 5.6% ARR seen in the SCD-HeFT study may be regarded as small by some, yet meaningful enough to act on by most and has become incorporated in national guidelines to varying degrees.9,10,40 It has also been suggested that an NNT of 50 per year (which equates to 17 over a 3-year period) is ‘clinically significant’ and only patients whose baseline risk of arrhythmic death is greater than 3% per year (9% over 3 years) will achieve this magnitude of benefit and justify ICD implant.35 Only patients with IHD who satisfy MADIT, MADIT2, or MUSTT trial criteria meet this requirement in the setting of PP, and only those with criteria based on the AVID trial satisfy it for SP.

The data presented in this study are derived from published clinical trials. It is recognized that individuals enrolled in such trials may not represent ‘real-world’ patients, who are often older and have more co-morbidities that may potentially detract from the beneficial effect of the ICD.41 Indeed, it has been proposed that clinical trials may overestimate the benefit from an ICD due to underuse or imbalances in the prescription of B-blockers and angiotensin-converting enzyme inhibitors and the use of anti-arrhythmic medication in control arms.42 Despite this, registry data indicate that SP patients with IHD appear to derive the same amount of benefit in the real world as those enrolled in clinical trials.43

Study limitations

The ARR at 3 years that is deemed clinically and ethically, rather than just statistically significant, is yet to be determined. A questionnaire survey of cardiologists would suggest that this figure is dependent on patient age and varies between 9 and 21%.44

Many of the limitations that affect published ICD RCTs and their substudies apply to the present analysis. Small patient numbers and wide CI and the weaknesses of subgroup analyses limit the validity of the ARR and NNT figures extracted from the published data. Importantly, the findings presented in this study cannot be used to aid decision making for patients who fall outside of RCT entry criteria.


The present study reports ARR and NNT from published trials comparing ICDs to medical therapy that are used to formulate national and international guidelines. The findings are standardized for length of follow-up. There is considerable variation in the magnitude of benefit between different heart failure aetiologies and other patient characteristics. The relatively small number of patients who have follow-up periods of 2 or more years results in wide CI and highlights the difficulty in formulating robust cost-effectiveness analysis. As ARR is a preferred statistic for use in informed consent, the results of this study may help facilitate communication between physicians, patients, and healthcare providers, to better understand the risk/benefits as applicable to individual patients.

Supplementary material

Supplementary material is available at Europace online.

Conflict of interest: T.R.B. has received research funding, speaker fees, travel bursaries, and consultancy honoraria from Medtronic, Boston Scientific, and St Jude Medical. P.P.S. has received an unrestricted educational grant from Boston Scientific and travel bursaries from other device manufacturers. K.R. has received speaker fees and travel bursaries from Medtronic, Boston Scientific, and St Jude Medical. K.C.K.W. has received an unrestricted educational grant from Medtronic and travel bursaries from Medtronic, Boston Scientific, and St Jude Medical. J.P.dB. has received travel bursaries from Medtronic, Boston Scientific, and St Jude Medical. Y.B. has received travel bursaries from Medtronic, Boston Scientific, and St Jude Medical.


View Abstract