Atrial Fibrillation Associated With Ivabradine Treatment
Atrial Fibrillation Associated With Ivabradine Treatment
Our initial search strategy identified 84 published articles. We reviewed titles and abstracts and excluded papers that were not clinical trials, were not of sufficient duration of follow-up or which reported supplementary data from trials published elsewhere. We assessed the full text of the remaining 34 articles for eligibility. Eight papers were translated into English. After studies were excluded on the basis of study design (one), follow-up duration (two), randomisation (nine), study blinding (10) and duplication (two), our search strategy yielded 10 double-blind randomised controlled trials that enrolled 20 022 patients for a mean of 1.5 (range 0.076–1.88) years follow-up, for a total of 30 090 patient-years of follow-up (Table 1, figure 1). Risk of bias of the included studies is summarised in online supplementary table S1 http://heart.bmj.com/content/100/19/1506/suppl/DC1. Of the 10 studies identified, AF incidence was only reported in one, the SHIFT trial. We wrote to the authors of the other studies to request the unpublished AF data. Of the five studies not sponsored by the manufacturer, four authors replied with AF data. None of the authors of studies sponsored by the manufacturer provided AF data. We contacted the manufacturer directly who declined to provide the unpublished data, citing ongoing inhouse analyses. Of the missing studies, full AF safety data were provided for the BEAUTIFUL study on the EMeA website. The safety data from the ASSOCIATE study which is reported in the EMeA documentation do not report the incidence of AF. The data from the INITIATIVE study and from Ruzyllo et al are included in the analyses performed by the EMeA for licensing of ivabradine. The data are not available for each trial individually, however, but are combined with the safety data from a 3-month phase III clinical trial (for which we were unable to identify a published report) and two very short (four doses and 2 weeks, respectively) dose-ranging phase II studies. All of the studies included in this overall oral safety set (OOSS) were randomised, controlled and double-blinded. None of the trials for which we obtained AF incidence data separately were included in the OOSS. In order to provide the most accurate estimate of the effect of AF, we decided to include the OOSS as a single study in our meta-analysis and also to perform two sensitivity analyses. In the first, the OOSS was excluded and in the second only those trials where the drug treatment was compared with placebo, rather than an alternative such as amlodipine or a β-blocker, were included.
(Enlarge Image)
Figure 1.
Flow chart of study selection.
In our full data set for analysis including the OOSS, therefore, there were 21 571 patients, followed up for a mean of 1.43 years to provide a total of 30 755 patient-years of follow-up. When the OOSS was excluded, the data set included 17 635 patients, followed up for a mean of 1.67 years to give a total of 29 385 patient-years of follow-up. In the placebo-controlled trials only data set, there were 17 571 patients followed up for a mean of 1.67 years to give a total of 29 380 patient-years of follow-up.
Compared with controls, treatment with ivabradine was associated with a relative risk (RR) of AF of 1.15 times that of alternative treatments (95% CI 1.07 to 1.24, p=0.0027) (figure 2). The results were largely similar if the OOSS was excluded, RR=1.15 (95% CI 1.06 to 1.25, p=0.0065) and if only placebo-controlled trials were examined, RR=1.15 (95% CI 1.05 to 1.26, p=0.015). When treatment indication was included in the model, the association between treatment indication and AF risk was not significant (p=0.99), although this analysis was limited as the majority of patient-years of follow-up (17 475/17 635) were in the two trials (SHIFT and BEAUTIFUL) for which the main treatment indication was heart failure.
(Enlarge Image)
Figure 2.
Forest plot of relative risk of atrial fibrillation in all trials with available data. RE Model, random effects model.
When the absolute risk of AF was compared between ivabradine and control groups, ivabradine treatment was associated with an NNH of 208.3 per year of treatment (95% CI 122.0 to 666.7, p=0.013) (figure 3). This corresponds to an absolute risk difference of 0.48%. This effect was very similar when the OOSS was excluded, NNH=208.3 (95% CI 113.6 to 1250, p=0.028) and when only placebo-controlled trials were included, NNH=208.3 (95% CI 105.3 to 10 000, p=0.048). As ivabradine is indicated for chronic treatment of angina or cardiac failure and treatment is likely to continue for more than 1 year, this NNH represents a substantially greater risk than the very rare (less than 1 : 10 000) risk of AF reported in the product literature.
(Enlarge Image)
Figure 3.
Number needed to harm per year of follow-up. Individual studies are plotted as circles with size proportional to the weighting in the meta-analysis model of absolute risk difference. The best fit line with 95% CI (dashed lines) is shown.
The heterogeneity of RR in the trials included was low, I=0.00%, with the observed heterogeneity explained by within-study variance (p=0.96). The same was true in the analysis of absolute risk difference, I=0.00%, where the residual heterogeneity after accounting for length of follow-up was explained by within-study variance (p=0.86). This explains the similarity of the meta-analysis results between subgroups of trials.
There was no evidence of publication bias, rank test p=1.00 and regression test p=0.64, and examination of the funnel plots did not suggest any missing data (figure 4).
(Enlarge Image)
Figure 4.
Funnel plot of relative risk versus SE. The plot is largely symmetrical, suggesting that there is no publication bias.
Results
Study Selection and Characteristics of Included Trials
Our initial search strategy identified 84 published articles. We reviewed titles and abstracts and excluded papers that were not clinical trials, were not of sufficient duration of follow-up or which reported supplementary data from trials published elsewhere. We assessed the full text of the remaining 34 articles for eligibility. Eight papers were translated into English. After studies were excluded on the basis of study design (one), follow-up duration (two), randomisation (nine), study blinding (10) and duplication (two), our search strategy yielded 10 double-blind randomised controlled trials that enrolled 20 022 patients for a mean of 1.5 (range 0.076–1.88) years follow-up, for a total of 30 090 patient-years of follow-up (Table 1, figure 1). Risk of bias of the included studies is summarised in online supplementary table S1 http://heart.bmj.com/content/100/19/1506/suppl/DC1. Of the 10 studies identified, AF incidence was only reported in one, the SHIFT trial. We wrote to the authors of the other studies to request the unpublished AF data. Of the five studies not sponsored by the manufacturer, four authors replied with AF data. None of the authors of studies sponsored by the manufacturer provided AF data. We contacted the manufacturer directly who declined to provide the unpublished data, citing ongoing inhouse analyses. Of the missing studies, full AF safety data were provided for the BEAUTIFUL study on the EMeA website. The safety data from the ASSOCIATE study which is reported in the EMeA documentation do not report the incidence of AF. The data from the INITIATIVE study and from Ruzyllo et al are included in the analyses performed by the EMeA for licensing of ivabradine. The data are not available for each trial individually, however, but are combined with the safety data from a 3-month phase III clinical trial (for which we were unable to identify a published report) and two very short (four doses and 2 weeks, respectively) dose-ranging phase II studies. All of the studies included in this overall oral safety set (OOSS) were randomised, controlled and double-blinded. None of the trials for which we obtained AF incidence data separately were included in the OOSS. In order to provide the most accurate estimate of the effect of AF, we decided to include the OOSS as a single study in our meta-analysis and also to perform two sensitivity analyses. In the first, the OOSS was excluded and in the second only those trials where the drug treatment was compared with placebo, rather than an alternative such as amlodipine or a β-blocker, were included.
(Enlarge Image)
Figure 1.
Flow chart of study selection.
In our full data set for analysis including the OOSS, therefore, there were 21 571 patients, followed up for a mean of 1.43 years to provide a total of 30 755 patient-years of follow-up. When the OOSS was excluded, the data set included 17 635 patients, followed up for a mean of 1.67 years to give a total of 29 385 patient-years of follow-up. In the placebo-controlled trials only data set, there were 17 571 patients followed up for a mean of 1.67 years to give a total of 29 380 patient-years of follow-up.
AF Incidence
Compared with controls, treatment with ivabradine was associated with a relative risk (RR) of AF of 1.15 times that of alternative treatments (95% CI 1.07 to 1.24, p=0.0027) (figure 2). The results were largely similar if the OOSS was excluded, RR=1.15 (95% CI 1.06 to 1.25, p=0.0065) and if only placebo-controlled trials were examined, RR=1.15 (95% CI 1.05 to 1.26, p=0.015). When treatment indication was included in the model, the association between treatment indication and AF risk was not significant (p=0.99), although this analysis was limited as the majority of patient-years of follow-up (17 475/17 635) were in the two trials (SHIFT and BEAUTIFUL) for which the main treatment indication was heart failure.
(Enlarge Image)
Figure 2.
Forest plot of relative risk of atrial fibrillation in all trials with available data. RE Model, random effects model.
When the absolute risk of AF was compared between ivabradine and control groups, ivabradine treatment was associated with an NNH of 208.3 per year of treatment (95% CI 122.0 to 666.7, p=0.013) (figure 3). This corresponds to an absolute risk difference of 0.48%. This effect was very similar when the OOSS was excluded, NNH=208.3 (95% CI 113.6 to 1250, p=0.028) and when only placebo-controlled trials were included, NNH=208.3 (95% CI 105.3 to 10 000, p=0.048). As ivabradine is indicated for chronic treatment of angina or cardiac failure and treatment is likely to continue for more than 1 year, this NNH represents a substantially greater risk than the very rare (less than 1 : 10 000) risk of AF reported in the product literature.
(Enlarge Image)
Figure 3.
Number needed to harm per year of follow-up. Individual studies are plotted as circles with size proportional to the weighting in the meta-analysis model of absolute risk difference. The best fit line with 95% CI (dashed lines) is shown.
The heterogeneity of RR in the trials included was low, I=0.00%, with the observed heterogeneity explained by within-study variance (p=0.96). The same was true in the analysis of absolute risk difference, I=0.00%, where the residual heterogeneity after accounting for length of follow-up was explained by within-study variance (p=0.86). This explains the similarity of the meta-analysis results between subgroups of trials.
There was no evidence of publication bias, rank test p=1.00 and regression test p=0.64, and examination of the funnel plots did not suggest any missing data (figure 4).
(Enlarge Image)
Figure 4.
Funnel plot of relative risk versus SE. The plot is largely symmetrical, suggesting that there is no publication bias.
