Article Text

This article has a correction. Please see:

other Versions

Original article
Application of a mortality risk score in a general population of patients with an implantable cardioverter defibrillator (ICD)
  1. Beat A Schaer,
  2. Michael S Kühne,
  3. David Blatter,
  4. Stefan Osswald,
  5. Christian Sticherling
  1. Department of Cardiology, University of Basel Hospital, Basel, Switzerland
  1. Correspondence to Dr Beat Schaer, Department of Cardiology, University Hospital, Petersgraben 4, Basel 4031, Switzerland; bschaer{at}


Objective The implantable cardioverter defibrillator (ICD) is very effective in the prevention of sudden cardiac death, but its benefit is impaired by competing risks. A simple risk model to predict mortality was designed for patients with primary prevention and ischaemic cardiomyopathy. We aimed to apply this score to a general ICD population.

Methods This retrospective registry study included all patients in whom an ICD was implanted at a tertiary referral hospital. Risk factors were age >70 years, QRS width >120 ms, atrial fibrillation, New York Heart Association Functional Classification class >2 and glomerular filtration rate <60 mL/min/1.73 m2. Kaplan–Meier curves were constructed according to the presence of 0, 1, 2 and >2 risk factors.

Results The cohort consists of 1032 patients, 881 (86%) were men, mean age was 61±14 years and mean follow-up 66±46 months. 256 patients (25%) died 58±41 months after implant. The setting was secondary prevention in 498 patients (48%). No risk factors was present in 32% of patients, 1 in 27%, 2 in 20% and >2 in 21%, respectively. There was a significant and comprehensible relation between risk score and mortality. Cumulative survival was 82% in patients with 0 risk factors, 63% in those with 1, 41% in those with 2 and 23% in those with >2 risk factors (p < 0.0001). ICD therapies were documented in 421 patients (41%) without correlation to risk factors.

Conclusions In a mixed population of primary and secondary preventive ICD carriers, application of a simple risk score predicts long-term mortality but not appropriate use of the ICD.

  • Implantable cardioverter defibrillator
  • ventricular fibrillation
  • ventricular tachycardia
  • sudden cardiac death

Statistics from



The efficacy of the implantable cardioverter defibrillator (ICD) for primary and secondary prevention of sudden cardiac death in patients with ischaemic and non-ischaemic heart disease has been shown in several randomised trials.1–4 Non-arrhythmic cardiac and non-cardiac death is recognised as a competing risk and impairs the benefit of ICD therapy.5 In Sudden Cardiac Death in Heart Failure (SCD-HeFT),6 for example, both accounted for about 10% each to the overall mortality rate of 35% after 5 years. Therefore, research has been carried out to determine which patients might not be suitable for ICD implantation. Using data from the Multicenter Automatic Defibrillator Implantation Trial II (MADIT-II) trial, Goldenberg developed a relatively simple mortality risk score with only five items. Mid-term as well as long-term results showed7 ,8 that patients with up to two risk factors took a benefit from ICD therapy, whereas those with more than two risk factors did not. However, only patients with ischaemic cardiomyopathy in a primary prevention setting were included in MADIT-II.3 We performed a risk analysis using a similar approach and data from a single-centre ICD registry that includes patients with different cardiomyopathies both in primary and secondary preventions.


All patients in whom an ICD was implanted at the University of Basel Hospital from 1994 onward are included in a registry that was started in 1999. Besides specific cardiological data, multiple comorbidities and laboratory values are collected at implant. The registry is continuously updated for all appropriate ICD therapies and death. Four of the five risk factors established in the paper by Goldenberg et al8 are included (age >70 years, QRS width >120 ms, presence of atrial fibrillation, New York Heart Association Functional Classification (NYHA) class >2). Blood urea nitrogen (BUN), however, is available in a minority of our patients only and has therefore been replaced by the estimated glomerular filtration rate (GFR) calculated with the Modification of Diet in Renal Disease formula.9 ,10 A GFR of <60 mL/min/1.73 m2 was used to replace a BUN of >26 mg/dL as a risk factor in the formula. In the original paper of Goldenberg,8 the HR of five risk factors was determined (1.87 for NYHA class and presence of AF, 1.65 for QRS width >120 ms and 1.57 for age >70 and BUN). Due to this small range of HR, they obviously decided not to weigh the five factors and we adopted this. Creatinine levels were not available in nine patients and therefore imputed with the cohort mean of the respective continuous variable in order to include all patients into the study. Patients were censored either at time of death (256 patients), month of the last visit if lost to follow-up (22 patients), time of heart transplantation (8 patients), time the ICD was not replaced or deactivated (12 patients) or at the time of last registry access on the 5th of December 2012.

Kaplan–Meier curves regarding survival were constructed for the whole population and separately for patients with ischaemic, dilated and miscellaneous cardiomyopathies. Another curve was constructed for the occurrence of any appropriate ICD therapy. As, in contrast to the MADIT-II study, we do not have a control group of patients without an ICD, we aimed to estimate the ‘benefit’ of ICD therapy at least in the subgroup of patients with ischaemic cardiomyopathy. We therefore included in a separate Kaplan–Meier curve only patients with ICD therapies delivered for ventricular tachycardia (VT) > 240 bpm or true ventricular fibrillation (VF) as a surrogate marker for sudden death without ICD.

Patients were stratified according to the presence of 0, 1, 2 and >2 (3–5) risk factors.

IBM SPSS V.21 was used. A log-rank test was applied to determine differences within the groups, and a p value <0.05 was considered statistically significant.


Patient population

The study cohort consists of 1032 patients, 881 (86%) of whom were men, mean age was 61±14 years, and mean follow-up 66±46 months. A single chamber ICD was implanted in 636 patients (62%), a CRT-ICD in 225 patients (22%) and a dual chamber ICD in 171 patients (16%), respectively. The setting was secondary prevention in 498 patients (48%) with an index arrhythmia being VF in 138 (28%), sustained VT in 241 (48%) and syncope associated with poor left ventricular ejection fraction in 119 (24%) patients. NYHA class was I in 246 (24%), II in 443 (43%), III in 312 (30%) and IV in 31 (3%) of patients, respectively. Detailed baseline characteristics are shown in table 1. Table 2 depicts the number of patients in the different score groups.

Table 1

Baseline characteristics

Table 2

Number of patients in the different score groups


In the 256 patients (25%) who died, time of death was 58±41 months after implant (median 49 months). Figure 1A–D shows the Kaplan– Meier curves of patient survival according to risk factors and underlying cardiomyopathy. There is a significant and comprehensible relation between risk score and mortality, both overall and in the two subgroups of ischaemic and dilated cardiomyopathy. However, the differences in dilated cardiomyopathy are less prominent. Cumulative survival overall was 82% in patients with 0 risk factors, 63% in those with 1, 41% in those with 2 and 23% in those with >2 risk factors. The corresponding numbers were 79%, 54%, 36% and 12% in ischaemic and 80%, 85%, 61% and 44% in dilated cardiomyopathy. The group of miscellaneous cardiomyopathies is too small beyond year 5 to make a meaningful analysis.

Figure 1

Kaplan–Meier curves of the cumulative survival according to four risk groups: (A) overall cohort; (B) ischaemic heart disease; (C) dilated cardiomyopathy; (D) miscellaneous cardiomyopathies.

In total, 154 patients died of cardiac reasons, mostly terminal heart failure and 102 patients of other reason (including vascular). Based on this, cardiac mortality was 54% with 0 RF, 48% with 1 RF, 63% with 2 RFs and 68% with >2 RFs (p=0.08).

ICD therapy

During long-term follow-up, appropriate ICD therapies were documented in 421 patients (41%). Figure 2 shows the corresponding Kaplan-Meier curve. Even though in the overall cohort as well as in the subgroup of ischaemic cardiomyopathy (not shown) statistically significant differences (p 0.005 and 0.0001) were seen according to the amount of risk factors, the split was not meaningful. Patients with 0 and those with >2 risk factors had a similar risk of a first ever ICD therapy. The cumulative incidences after 10 years were 45–60% in the overall cohort and 45–70% in the ischaemic subgroup. In patients with dilated cardiomyopathy, the range of the cumulative incidence was very small (40–50%, p 0.83, curve not shown). Whereas in patients with ischaemic cardiomyopathy, there was a continuously increase of the overall chance of a first ever ICD therapy over time, hardly any new first ever ICD therapy was seen beyond year 5 in patients with dilated cardiomyopathy. There was no correlation between risk factors and the occurrence of life-threatening arrhythmias (figure 3).

Figure 2

Kaplan–Meier curve of the cumulative incidence of first implantable cardioverter defibrillator therapy according to four risk groups: overall cohort.

Figure 3

Kaplan–Meier curve of the cumulative incidence of serious arrhythmias in ischaemic patients.


There are four major findings in this retrospective database study: (a) the ‘Goldenberg score’, originally derived for patients with ischaemic cardiomyopathy in a primary prevention setting, is applicable in a meaningful way to predict mortality of all patients with an ICD, irrespective of their cardiomyopathy; (b) patients with a higher ‘Goldenberg score’ had a higher mortality, indicating that the competing risks precluded them from fully benefiting from an ICD. Conversely, the group with the lowest score had the best survival. (c) No meaningful correlation between risk score and incidence of appropriate ICD therapies was seen and (d) mortality was much higher in patients with ischaemic than in those with non-ischaemic cardiomyopathy, even though the latter group encompassed a lower percentage of patients with a risk score of 0 and 1 (59% vs 36%). Findings can be summarised as ‘the sicker the patients, the higher their mortality’.

The ‘Goldenberg Score’7 was created with data from patients of the MADIT-II study and a relatively short follow-up of median 1.5 years. Patients with one or two risk factors showed a benefit from ICD therapy compared with the control group (‘optimal medical therapy’), whereas those with 0 or >2 risk factors were too healthy or too sick to benefit. These results were slightly different in the extension study8 with a median follow-up of 7.6 years. During very long-term follow-up also patients without risk factors at inclusion did benefit now, whereas those with >2 risk factors were still too sick to benefit. The authors concluded that “a simple risk score can identify patients who derive significant long-term benefit from primary ICD therapy”.

In the present study, we could not assess the effect of an ICD in different risk groups as there is no control group, but used the score as a predictor of overall mortality, the most important competing risk for a beneficial effect of device therapy. For this purpose, the ‘Goldenberg score’ is much more user-friendly than the Charlson Comorbidity Index (CCI), the score by Buxton derived from the Multicenter Unsustained Tachycardia Trial (MUSTT) database11 and the Seattle Heart Failure Model (SHFM). The ‘Buxton score’ has never been widely accepted due to two specific reasons: first, patients enrolled in MUSTT nowadays represent a small subgroup of our patients and, second, parameters included in the score as, for example, inducibility of ventricular arrhythmias or non-sustained arrhythmias are not routinely tested for anymore. The SHFM incorporates almost 20 clinical and laboratory parameters that render the score not very feasible to use in daily practice. It has been tested in two large populations of ICD patients12 ,13 and showed a good correlation between the severity of comorbidities and mortality. However, only patients with a CRT device were studied, which limits its applicability in a general ICD population. In the CCI, 17 parameters are incorporated, which also limits its ease of use. It has also been tested in a CRT population, where it showed a very good correlation adjusted for age and dichotomised with an index </> 5.14 The CCI has also been applied to a population of both primary and secondary prevention patients with ischaemic or non-ischaemic cardiomyopathy.15 It showed a very good correlation to mortality at 1 year, with 5% in patients with a CCI of 0 up to 78% in patients with a CCI ≥5. The five parameters needed for the ‘Goldenberg score’ are determined within 1 or 2 min, which makes this score so compelling and, as shown in our study, reliable for prediction of long-term mortality.

The accuracy of the score to predict mortality was best in the subgroup of patients with ischaemic cardiomyopathy. Ten-year mortality was 20% in patients with no risk factor, and each risk factor increased mortality by about 20%. Of note, in the low-risk group, mortality was very low up to 5 years of follow-up and started to matter only during the next 5 years. Even though we included 48% patients with secondary prevention who might be sicker than patients with primary prevention, the mortality rate at 10 years was similar to the MADIT-II population at 8 years (eg, 21% vs 24% with 0 risk factors and 82% vs 80% with >2 risk factors).

Accuracy was less good, even though still significant, in the group of patients with dilated cardiomyopathy. However, mortality was homogenous with about 20% in patients with 0–2 risk factors but increased to 55% in patients with >2 risk factors. In part, this might be explained by the fact that 52% of patients in this subgroup were implanted with a CRT device indicated for QRS prolongation, probably alleviating one risk factor. CRTs were used in only 16% of patient with ischaemic cardiomyopathy.

In contrast, the ‘Goldenberg score’ was not useful to predict use of ICDs, that is, the occurrence of VT or VF treated by the device. Overall and in the different subgroups the rate was always about 50% after the follow-up of 10 years, and no meaningful ranking according to the risk factors was perceptible. At first glance, it might seem contradictory that the score does not predict ICD use. However, data regarding this are scarce16 and did not identify factors similar to the ones used in the ‘Goldenberg’ score and the MADIT-II group never published on this. ICD therapies are probably not triggered by factors used in the ‘Goldenberg’ score but by the underlying cardiomyopathy (scar, potential of ischemia) and the prevention mode. One must also bear in mind that the score was constructed to predict mortality and not ICD therapies.

The study has some limitations that need to be addressed. Twenty-nine per cent of patients had a left bundle branch block as one of the factors adding to the score. Of those, 64% received a CRT-D. This could have interfered with the risk prediction model and make this component debatable.

Another limitation is that we did not choose the same parameter for assessing kidney function as in the original score. Instead of BUN measurements, we calculated the GFR and considered <60 mL/min as pathological. However, there is no reason to believe that this changes the predictive value of kidney dysfunction in the score. Finally, the score was constructed in a population of patients with a probably lower general risk of death as they all were implanted in primary prevention and are now tested in a larger population.


In this mixed population of primary and secondary preventive ICD carriers, the ‘Goldenberg’ score is able to predict mortality, but not appropriate use of the ICD. The decision to implant an ICD in patients with multiple comorbidities should be balanced against the considerable risk of death derived from these comorbidities. It is not because the ICD did not discharge that the initial indication was not appropriate, but the comorbidity burden was responsible for that the patient did not reach the moment to benefit from prevention against SCD as his life was shortened by collateral, sometimes non-cardiac conditions.

Key messages

What is already known on this subject

  • Comorbidities and resulting death are recognised as competing risks of implantable cardioverter defibrillator (ICD) therapy. Several risk score relying on baseline characteristics have been constructed to successfully predict mortality in different settings (primary prevention in ischaemic cardiomyopathy, cardiac resynchronisation therapy, etc.). A simple score was developed with data from the Multicenter Automatic Defibrillator Implantation Trial II study using only five items.

What this study adds

  • This study demonstrated that this simple score can not only be applied to predict mortality in subgroups of ICD patients but also to a general ICD population. In contrast, the score cannot be used to predict need of ICD therapy.

How might this impact on clinical practice

  • If results of this and probably subsequent similar studies were trusted and applied, ICD implantation in patients with a high risk of mortality should be questioned.


View Abstract


  • Contributors BAS designed this study. BAS and DB carried out the data analysis. BAS wrote the first version of the manuscript. All authors contributed to the interpretation of the data, revised the manuscript critically and approved the final manuscript. BAS is the guarantor.

  • Competing interests BAS has served on the speakers’ bureau for Medtronic. MSK has served on the speakers’ bureau for Boston Scientific, St. Jude Medical and Biotronik. DB: none declared. SO has served on the speakers’ bureau for Medtronic, Boston Scientific, Biotronik, St. Jude Medical, Sanofi-Aventis and Astra Zeneca and has received unrestricted grants from Medtronic, Boston Scientific, Biotronik and St. Jude Medical. CS has served on the speakers’ bureau for Medtronic, Biotronik and Sorin and had scientific support from Medtronic, Biotronik, Boston Scientific, Bisoense, St. Jude Medical and Sorin.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles

  • Correction
    BMJ Publishing Group Ltd and British Cardiovascular Society