How to present treatment effects of implantable defibrillators? Statistics, baseline risk, and meaningful benefit

This editorial refers to ‘Absolute risk reduction in total mortality with implantable cardioverter defibrillators: analysis of primary and secondary prevention trial data to aid risk/benefit analysis’ by T.R. Betts et al., on page 813.

Patients with left ventricular dysfunction are prone to sudden cardiac death, presumably from ventricular tachyarrhythmias. Randomized controlled trials (RCTs) demonstrated that prophylactic implantable cardioverter-defibrillators (ICDs) improve survival among selected patients with left ventricular dysfunction considered at high risk of sudden cardiac death. Based on these clinical trial data, the indications for ICD therapy rapidly expanded from restricted therapy of ‘last resort’ (secondary prevention) to a broad-reaching pre-emptive therapy (primary prevention). However, the survival benefit of prophylactic ICD therapy is not uniform across the population with implants. For evidence to be of value, healthcare professionals, patients, and policy makers are faced with the challenge to interpret and apply the data of RCTs in daily practice. Simplified, they need to understand whether one treatment is better than another or better than no treatment at all. Clinicians have the task to communicate risk of a particular treatment to their patients. So, do we understand the concept of risk of a particular treatment, and how do we communicate this to our patients?

In recent years, the amount of medical literature has increased rapidly, and with the Internet era information on medical research has become more easily accessible. The problem is how to interpret the results of several studies and decide whether it justifies changing the current treatment. The poor presentation of medical statistics of risk associated with a particular treatment can lead to poor decision-making. There are several statistical formats to present risk and risk reductions. Formats for presenting risk include frequency, percentage, and probability. Formats for presenting risk reduction include relative risk reduction (RRR), absolute risk reduction (ARR), and numbers needed to treat (NTT).

Example illustrating the statistical formats

The SCD-HeFT (Sudden Cardiac Death in Heart Failure Trial) study found that the ICD reduces the risk of all-cause mortality by 23% over the 5-year follow-up.¹ Specifically, 36.1% of patients without an ICD died at 5 years, compared with 28.9% of patients who received an ICD. Thus, 7.2% (36.1 − 28.9%) fewer patients would die if they received an ICD during the 5-year follow-up. In other words, 14 patients need to receive an ICD to prevent one death over 5 years. ‘Reduces the risk of all-cause mortality by 23%’ represents a RRR. ‘Seven percent less would die’ represents the ARR. ‘Fourteen patients need to receive an ICD in order to prevent one death’ represents the NTT.

The question faced by the clinicians is then ‘Which one will help me in choosing the best treatment for my patient?’ In terms of risk reduction, clinicians might be more willing to recommend, and patients are more willing to accept, an intervention when its benefits are presented in relative compared with absolute terms. In this respect, the RRR can be regarded as a more persuasive summary statistic. But what really counts in decision-making is the quantitative impact of a treatment, as shown by the ARR and NNT, since RRR can lead to errors.^2,3 For example, an intervention is more effective if mortality is reduced from 30 to 20% (ARR = 10%, NNT = 10) than from 3 to 2% (ARR = 1%, NNT = 100). In both scenarios, the RRR is 33%, but the difference in baseline risk between both scenarios is a factor of 10. In this example, it is clear that relative statistical formats are misleading because they ignore the baseline risk.

In the current issue of the Journal, Betts et al.⁴ report on alternative statistical formats to present the effect of ICDs on all-cause mortality. At first glance, one might expect another review of published data on primary and secondary prevention ICD trials, which does not add much to our knowledge. But the novel contribution of this paper is the use of ARR and NNT to present the effect of prophylactic ICD therapy on all-cause mortality. The authors took the varying lengths of follow-up of the published RCTs into account and standardized the findings for the length of follow-up. The reported ARR in all-cause mortality at 3-year follow-up ranges from 4.0 to 24.6% in primary prevention. Of note, the earlier primary prevention trial, e.g. the first Multicenter Automatic Defibrillator Implantation Trial (MADIT) showed an impressive RRR (58.6%), an important appearing ARR (24.6%), and a very low NNT (4) at 3-year follow-up. In contrast, later trials as MADIT-II and SCD-HeFT had a lower ARR and a corresponding higher NNT. The use of ARR is more appropriate and clinically relevant to express risk and benefit from ICD implantation.⁵ For a better understanding of risk and benefit of ICD implantation, all risk measures as RRR, ARR, and NNT need to put in context against the total mortality in patients without ICD implantation, i.e. the untreated group.

Are we there yet? Definitely not, ICDs seem to offer a simple solution in a very complex situation. Implanting a device that is very effective in terminating life-threatening ventricular arrhythmias would be life-prolonging in the majority of patients with left ventricular dysfunction. But as mentioned earlier, not all patients benefit equally from prophylactic ICD implantation. The devices cannot prevent deaths from causes other than lethal ventricular arrhythmias. The ICDs exert their survival benefit through a reduction of arrhythmic death with a magnitude for RRR of ∼60%.⁶ From this viewpoint, the number of patients needed to treat to reduce all-cause mortality by a meaningful amount is directly related to the risk of arrhythmic death, as opposed to the risk of overall death.⁷ Thus, to ensure effective and efficient use of ICDs, it is important to understand the baseline risk of arrhythmic death in patients. In addition, we need to understand the patients' overall burden of cardiovascular disease and the presence of comorbidities. Some patients have low risk of arrhythmic death that ICD implantation is not warranted. When the burden of cardiovascular disease increases, the use of an ICD may become increasingly effective in terms of ARR, until comorbidities are more prevalent and the risk of death from causes other than lethal ventricular arrhythmias become high resulting in ineffective use of ICDs. Betts et al. provide an example with the SCD-HeFT study: patients with more severe heart failure [(New York Heart Association III (NYHA III)] received less benefit from ICD implantation (ARR = −1.8%) compared with patients with NYHA II (ARR = 9.2%).

In conclusion, appropriate decision-making regarding the use of ICD requires a thoughtful consideration of both the evidence of RCTs and the clinical context. The use of ARR is more appropriate and clinically relevant to express risk and benefit from ICD implantation. The balance between risk and benefit is dependent on the baseline risk of the patient, both the risk of arrhythmic death and the overall risk of death. The baseline risk is defined by aetiology of cardiovascular disease, degree of left ventricular dysfunction, pharmacological therapy, the presence of comorbidity, and other concomitant therapies.

Conflict of interest: none declared.

References

Author notes