A Meta-Regression Analysis of Atrial Fibrillation Ablation in Patients with Systolic Heart Failure

Mohammed Ruzieh1, Morgan K. Moroi2, Nader M. Aboujamous3, Mehrdad Ghahramani1, Gerald V. Naccarelli1, Matthew Nudy3, Khalil Kanjwal4, Andrew J Foy1

1Penn State Heart and Vascular Institute. Hershey, PA.2Penn State College of Medicine. Hershey, PA.3Penn State Department of Internal Medicine. Hershey, PA.4Michigan State University. McLaren Greater Lansing, MI.

Abstract

Background

Meta-analyses of randomized controlled trials comparing atrial fibrillation (AF) ablation to medical therapy in patients with heart failure (HF) reported improvement in left ventricular ejection fraction (LVEF), quality of life using the Minnesota Living with HF Questionnaire (MLWHFQ), and 6-minute walk test (6MWT). Nonetheless, there was significant heterogeneity not accounted for suggesting that not all HF patients derive the same effect from AF ablation.

Objectives

To evaluate if baseline LVEF or the etiology of the cardiomyopathy would moderate the efficacy of AF ablation.

Methods

We performed random effects meta-regression using the mean baseline LVEF and total percentage of patients with non-ischemic cardiomyopathy (NICMP) in the placebo arms as moderator variables.

Results

Six trials with a total of 687 patients were included. The baseline LVEF in the control arm of trials ranged from 25% - 42.9%, and the percentage of patients with NICMP within each trial varied from 35% to 100%. When baseline LVEF was used as the moderator variable, no significant change in heterogeneity was observed for any of the outcomes of interest (R2 0.00 – 0.02). However, when controlling for NICMP, heterogeneity dropped substantially for the outcomes of LVEF (I2 44.7%, R2 0.91), and MLWHFQ (I2 0.00%, R2 1.00) but not 6MWT (I2 67.4%, R2 0.00). This indicates that improvement in LVEF and MLWHFQ was greater in the AF ablation group when more patients with NICMP were included in the trials.

Conclusions

In patients with systolic HF, AF ablation may be more beneficial in patients with NICMP.

Key Words : Atrial fibrillation, Heart failure, Non-ischemic, Cardiomyopathy.

Correspondence to: Mohammed RuziehPenn State Heart and Vascular Institute500 University Drive PO Box 850, MC H047Hershey, PA 17033.

Introduction

Atrial fibrillation (AF) and heart failure (HF) are two common conditions that often coexist and can predispose each to one another.[1-3] Guidelines do not provide a clear consensus regarding the best approach for management of AF in patients with HF. Multiple randomized controlled trials (RCTs) examined the role of catheter ablation in AF patients with HF and demonstrated improvement in left ventricular function (LVEF) and quality of life.[2-8] Recently, several meta-analyses [9-11] have analyzed these trials and reported improvement in the pooled outcome of LVEF, 6-minute walk distance and quality of life. However, wide variation for the difference in each of these outcome measures was noted between trials suggesting that not all HF patients with AF derive the same effect from ablation and there is a need to better understand which patients with HF are most likely to benefit from AF ablation.

When conducting meta-analyses, some variation in treatment effect between trials is expected due to differences in study quality (e.g. potential bias in design, acquisition and adjudication of specific data elements) which may become evident when performing bias assessment with validated instruments such as the Cochrane Bias Assessment tool. In other cases, variation may be related to differences in patient sampling, the application of the intervention and management strategies in the control groups. This may be intentional on the part of investigators and meant to address gaps or areas of uncertainty. In each of the cases above, differences in trial-level effects may be expected and intuitively understood and contribute to the overall understanding of an interventions effect. However, in other cases, variation may not be easily explained by these factors and, when this is the case, confidence in the generalizability of a summary effect measure should decrease and the evidence-based community should seek to understand the source of variability in efforts to better target the intervention to those most likely to benefit or not.

A method to understand the source of variability in clinical trials is to perform meta-regression analyses where the treatment effect of the trial is measured against one or more moderator variables. Intuitively, this approach is not that different from performing regression analyses of single datasets to determine how the presence of baseline covariates contributes to the outcome of interest. Similarly,meta-regression of clinical trials seeks to understand how differences in covariates among the study groups of individual clinical trials contribute to the observed treatment effects.

We hypothesized that the wide variation in the treatment effects of AF ablation for HF patients observed in the published clinical trials is due to differences in the patient populations of the individual trials and that understanding this may contribute to better application of AF ablation in HF patients. Specifically, we hypothesized that baseline LVEF and the etiology of the cardiomyopathy (i.e. ischemic versus non-ischemic) would moderate the efficacy of ablation on outcomes of LVEF improvement, 6-minute walk distance and quality of life. To test this hypothesis, we conducted a meta-regression analysis of several covariates, which we felt could contribute to the heterogeneity of effects observed between trials.

Methods

Data collection and extraction

We searched Medline, Google Scholar, the Cochrane Central Register for RCTs, and ClinicalTrials.gov for studies that examined AF catheter ablation in patients with HF (latest search date: Dec 1, 2018). Three authors (M.R., M.M. and A.F.) drafted the study protocol which was then revised by all coauthors. Two authors (M.R. and M.M.) independently reviewed all articles and abstracts for inclusion, and independently extracted information on patient's characteristics, study design, intervention, follow-up, and outcomes in a standardized manner. Discrepancies were discussed and resolved by consensus.

Trials that randomized patients with AF and systolic HF to catheter ablation versus medical therapy were included.

Outcome and quality assessment

The primary outcomes of interest were LVEF, Minnesota Living with Heart Failure Questionnaire (MLWHFQ) scores, and 6-minute walk distance.We used the Cochrane Risk of Bias table and theGrading of Recommendations, Assessment, Developmentand Evaluation (GRADE) system, to report risk of bias and quality of study outcomes in each study, respectively.

Statistical analysis

The primary analyses were performed using RevMan version 5.3 (The Nordic Cochrane Center, The Cochrane Collaboration; Copenhagen, Denmark).We used the inverse variance random effects model to calculate the pooled mean difference in the outcomes of interest. Sensitivity analysis was performed as following: (i) comparing trials that randomized patients to AF catheter ablation vs. rate control, (ii) comparing AF catheter ablation to medical therapy in patients with persistent AF only, and (iii) individually eliminating studies to detect if any is the cause of heterogeneity.

To examine whether baseline LVEF or etiology of cardiomyopathycontributed to theheterogeneity in the outcomes, we performedrandom effects meta-regressionusing comprehensive Meta-Analysis Version 3, Borenstein, M., Hedges, L., Higgins, J., and Rothstein, H.Biostat, Englewood, NJ 2013. For these regression analyses, the mean baseline LVEF and total percentage of patients with non-ischemic cardiomyopathy in the placebo arms of trials were used as moderator variables.

We created meta-regression linear prediction graphs by plotting the moderator variable on the x-axis and treatment effect measure on the y-axis. The bubbles were plotted in proportion tothe contribution of each study to the regression model.

The following parameters were used to test the model of heterogeneity: (i) Tau2 which is the estimate of the true variance among studies, (ii) I2which represents the percentage of variability in the effect risk estimate among studies due to heterogeneity rather than chance (with I2<25% considered as low, I2>75% considered as high, and in between [25% to 75%] as intermediate), and (iii) R2 which representsthe proportion of between-study variance explained by the moderator.

A two-sided p-value of <0.05 was considered to be statistically significant.

Results

Qualitative Synthesis

We included six trials in our analysis, [Figure 1]. A total of 687 patients were included (342 patients randomized to catheter ablation and 345 patients randomized to medical therapy alone). The mean age in the trials ranged from 55 to 64 years,andthe mean follow-up time ranged from six months to 38 months.The average baseline LVEF was 33.2% in the ablation arm and 34.0% in the control arm. Non-ischemic cardiomyopathy was present in 199 (58.2%) patients in the ablation arm and 170 (49.3%) patients in the control arm. 297 (86.8%) patients in the ablation arm and 286 (83.0%) in the control arm, had persistent AF. Further patients'characteristics are shown in [Table 1].

Table 1. Characteristics of patients included in the studies
MacDonald Jones Hunter Di Biase Prabhu Marrouche
Ablation arm Rate control Ablation arm Rate control Ablation arm Rate control Ablation arm Amiod-arone Ablation arm rate control Ablation arm Medical therapy
Mean age (yrs) 62.3 ± 6.7 64.4 ± 8.3 64 ± 10 62 ± 9 55±12 60±10 62±10 60±11 59 ± 11 62 ± 9.4 64 64
Female gender 23% 21% 19% 8% 4% 4% 25% 27% 6% 12% 13% 16%
No. of patients 22 19 26 26 26 24 102 101 33 33 179 184
Follow up (months) 9.7 6.9 12 12 12 6 24 24 6 6 37.6 ± 20.4 37.4 ± 17.7
Persistent AF 100% 100% 100% 100% 96% 88% 100% 100% 100% 100% 70% 65%
NYHA class II and III II and III II and III II and III II and III II and III II and III II and III ≥II ≥II I-IV I-IV
ICMP 50% 47% 38% 27% 23% 29% 62% 65% 0% 0% 40% 52%
NICMP 50% 53% 62% 73% 77% 71% 38% 35% 100% 100% 60% 48%
LVEF % 36.1 ± 11.9 42.9 ± 9.6 22 ± 8 25 ± 7 31.8±7.7 33.7±12.1 29±5 30±8 32 ± 9.4 34 ± 7.8 32.5 31.5
LA diameter (mm) N/A N/A 50 ± 6 47 ± 7 52±11 50±10 47±4 48±5 48 ± 6 47 ± 8 48 49.5
6 min walk distance (meters) 317.5 ± 125.8 351.8 ± 117.1 416 ± 78 411 ± 109 N/A N/A 348±111 350±130 491 ± 147 489 ± 132 N/A N/A
Quality of life 55.8 ± 19.8 59.2 ± 22.4 42 ± 23 49 ± 21 N/A N/A 52±24 50±27 N/A N/A N/A N/A
Diabetes Mellitus 32% 21% N/A N/A N/A N/A 22% 24% 4% 5% 24% 36%
HTN 64% 58% N/A N/A 30% 33% 45% 48% 13% 12% 72% 74%
CAD 50% 53% 42% 50% N/A N/A 62% 65% N/A N/A 27%* 36%*

AF: Atrial fibrillation, CAD: Coronary artery disease, HTN: Hypertension, ICMP: Ischemic cardiomyopathy, NICMP: Non-ischemic cardiomyopathy, N/A: Not available. *History of myocardial infarction.

Figure 1. PRISMA diagram showing search strategy results.



Risks of bias and quality assessment

Study limitations and biases (per Cochrane and GRADE criteria) are summarized in [Table 2]. Randomization was performed using random number generation in all trials.None of the trials tested AF ablation against a sham procedure and thus patients and their treating physicians were not blinded.This creates performance and outcomes assessment bias. Therefore, our confidence in the outcome assessment is moderate. Assessment ofLVEF was blinded in four trials.[2,3,5,6]

All studies appropriately described crossovers and dropouts. Crossover occurred in two patients in the study by Jones et. al.[3] and in 46 patients in the CASTLE AF trial. [8] Loss to follow-up was most prevalent in the CASTLE AF trial (33 [9.1%] patients).Further details on interventions and follow up are provided in [Table 3].

Evaluation of the funnel plots revealed no evidence of publication bias.

Table 2. Risk of bias assessment
Bias Study Judgement Support for judgement
Random sequence generation (selection bias)
MacDonald 2011 Low risk Computer generated
Jones 2013 Low risk Computer generated
Hunter 2014 Low risk Random number generator
Di Biase 2016 Low risk Computer generated
Prabhu 2017 Low risk Computer generated
Marrouche 2018 Low risk Computer generated
Allocation concealment (selection bias)
MacDonald 2011 Low risk Computer generated randomization
Jones 2013 Low risk Computer generated randomization
Hunter 2014 Low risk Random number generator
Di Biase 2016 Low risk Computer generated randomization
Prabhu 2017 Low risk Computer generated randomization
Marrouche 2018 Low risk Computer generated randomization
Blinding of participants and personnel (performance bias)
MacDonald 2011 High risk No blinding
Jones 2013 High risk No blinding
Hunter 2014 High risk No blinding
Di Biase 2016 High risk No blinding
Prabhu 2017 High risk No blinding
Marrouche 2018 High risk No blinding
Blinding of outcome assessment (detection bias)
MacDonald 2011 Moderate risk Only scans analysis was blinded
Jones 2013 Low risk People conducting cardiopulmonary exercise test and imaging analysis were blinded
Hunter 2014 Moderate risk Only echocardiogram analysis was blinded
Di Biase 2016 Moderate risk Only echocardiogram analysis was blinded
Prabhu 2017 High risk No blinding
Marrouche 2018 High risk No blinding
Incomplete outcome data addressed (attrition bias)
MacDonald 2011 Low risk No significant attrition
Jones 2013 Low risk No significant attrition
Hunter 2014 Low risk No significant attrition
Di Biase 2016 Low risk No significant attrition
Prabhu 2017 Low risk No significant attrition
Marrouche 2018 Low risk No significant attrition
Selective reporting (reporting bias)
MacDonald 2011 Low risk
Jones 2013 Low risk
Hunter 2014 Low risk
Di Biase 2016 Low risk
Prabhu 2017 Low risk
Marrouche 2018 Low risk



Table 3. Intervention and follow-up
MacDonald Jones Hunter Di Biase Prabhu Marrouche
Ablation strategy PVI ± Linear lesions and sources of complex fractionated electrograms ± Cardioversion ± cavotricuspid isthmus ablation PVI ± Linear lesions ± left atrial complex fractionated electrograms ± Cardioversion ± cavotricuspid isthmus ablation. PVI with ablation of complex or fractionated electrograms ± Linear lesions ± Cavotricuspid isthmus ablation PVI, and left atrial posterior wall isolation ± SVC isolation ± Linear lesions ± left atrial complex fractionated electrograms ± cardioversion PVI, left posterior wall isolation ± cardioversion PVI, Additional ablation lesions were made at the discretion of the operators
Frequency of monitoring (months) 3 and 6 3,6 and 12 1, 3 and 6 3, 6, 12 and 24 3 and 6 3, 6, 12, 24, 36, 48 and 60
Method of assessing rhythm on follow up 24h Holter monitor 48h Holter monitor ± existing implantable devices 48h Holter monitor ECG, and existing implantable devices 24h Holter monitor and ILR Existing implantable devices
Repeat ablation 6 (28.6%) 5 (19.2%) 14 (53.8%) 1.4 ± 0.6 per person Repeat procedure was allowed (frequency not given) 37 (24.5%)
Crossover None 2 None None None 46
Loss to follow up 3 None 1 None 1 33
AAD on follow up Oral amiodarone for 3 months in all patients post ablation. AAD stopped post ablation unless indicated by other reasons AAD stopped post ablation unless indicated by other reasons AAD allowed for 3 months after the first ablation 12 patients post ablation 48 patients in the ablation arm and 64 in the control arm.



Outcomes

LVEF: Data for difference in change in LVEF was available from all six trials. Compared to medical therapy alone, AF catheter ablation was associated with a significant increase in LVEF (mean difference 6.4%; 95% CI: 2.8 – 10.0; P<0.001), [Figure 2].In a sensitivity analysis when including only trials that had a blinded assessment of LVEF, AF catheter ablation was not associated with a statistically significant increase in LVEF (mean difference 5.3%; 95% CI: -0.6 – 11.2; P=0.08). The heterogeneity test was significant (Tau²= 16.2; df= 5; P< 0.001, I²= 91%), and it did not improve on sensitivity analysis.

When baseline LVEF of the placebo group was used as the moderator variable, we observed no significant change in heterogeneity (Tau²= 15.9; df= 4; P< 0.001, I²= 91.1%, R2= 0.02). However, when percentage of patients with non-ischemic cardiomyopathy in the placebo group was used as the moderator variable,heterogeneity dropped significantly and a strong linear relationship was observed such that as the percentage of patients with non-ischemic cardiomyopathy increased in the trials, the difference in change in LVEF was greater with ablation (Tau²= 1.5; df= 4, I²= 44.7%; P= 0.12, R2: 0.91), [Figure 2]. This means that most of the variation observed in the treatment effect of the difference in change in LVEF between the ablation and control groups from the original meta-analysis could be explained by the percentage of patients in the trials who had non-ischemic cardiomyopathy.

Figure 2. Change in LVEF, meta-analysis (left) and meta-regression (right) results.



Quality of life based on MLWHFQ scores:Four trials reported data on MLWHFQ.There was a significant improvement in the MLWHFQ scores in the AF catheter ablation group when compared to the medical therapy group (mean difference -8.0; 95% CI: -14.3 – -1.7; P= 0.01), [Figure 3]. There was moderate heterogeneity (Tau²= 14.1, df= 3; P= 0.22, I² = 33%).

Baseline LVEF was not related to the observed treatment effects (Tau²= 35.6; df= 2; P= 0.11, I²= 54.4%, R2= 0.00). On the other hand, when the percentage of patients with non-ischemic cardiomyopathy was used as the moderator variable, heterogeneity dropped to zero and a strong linear relationship was observed such that as the percentage of patients with non-ischemic cardiomyopathy increased in the trials, improvement in MLWHFQ scores was greater with ablation (Tau²= 0.0; df= 2; P= 0.41, I²= 0.0%, R2= 1.00), [Figure 3]. This means that nearly all of the variation observed in the treatment effect of mean difference in change in MLWHFQ scores between the ablation and control groups from the original meta-analysis could be explained by the percentage of patients in the trials who had non-ischemic cardiomyopathy.

Figure 3. Change in MLWHFQ, meta-analysis (left) and meta-regression (right) results.



6-minute walk distance in meters:Data on6-minute walk distancewere reported in five trials. The mean increase in 6-minute walk distance was higher in the AF catheter ablation group compared to the medical therapy group (mean difference 24.2; 95% CI: 5.7 – 42.7; P= 0.01), [Figure 4].Heterogeneity was significant (Tau²= 235.9; df= 4, P= 0.01, I²= 70%). On sensitivity analysis, most of the heterogeneity was driven by the AATAC and CASTLE AF trials [6,8], and when excluded from the analysis, the heterogeneity became low (I2: 0.0%) without significant change in the point estimate, P< 0.05).

Controlling for baseline LVEF didn't result in significant change in heterogeneity (Tau²= 289.0; df= 3; P= 0.005, I²= 77.0%, R2= 0.00). Similarly, we didn't observe significant change in heterogeneity when the percentage of patients with non-ischemic cardiomyopathy was used as the moderator variable(Tau²= 228.2; df= 3; P = 0.03, I²= 67.4%, R2= 0.00), [Figure 4]. This means that the variation observed in the treatment effect of mean difference in change in 6-minute walk distances between the ablation and control groups from the original meta-analysis could not be explained by the percentage of patients in the trials who had non-ischemic cardiomyopathy.

Figure 4. Change in 6MWT, meta-analysis (left) and meta-regression (right) results.



Cardiovascular mortality, heart failure hospitalizations, and stroke:With the exception of the AATAC [6] and CASTLE AF trials,[8] the remaining trials were not designed nor powered to detect a difference in cardiovascular mortality orHF hospitalizations.

Unplanned hospitalizations and death were significantly higher in the amiodarone arm in the AATAC trial6 (58 [57%] vs. 32 [31%]; P < 0.001) and (18 [18%] vs. 8 [8%]; P= 0.037), respectively. In the CASTLE AF trial,[8] cardiovascular mortality and HF hospitalizations weresignificantly higher in the medical treatment arm (41 [22.3%] vs. 20 [11.2%]; P= 0.008) and (66 [37.9%] vs. 37 [23.7%] P= 0.004), respectively. Stroke occurred at higher rates in the medical treatment arm, however, this didn't reach statistical significance (11 [6.0%] vs. 5 [2.8%]; P= 0.14).

Discussion

In this meta-regression analysis of randomized controlled trials,it appears that patients with non-ischemic cardiomyopathy benefit more from AF catheter ablation compared to those with ischemic cardiomyopathy. This can be inferred from the regression analyses showing that the difference in change in LVEF and MLWHFQ scores was greater and in favor of the AF ablation group when more patients with non-ischemic cardiomyopathy were included in the trials. Each of these outcomes was found to have a strong linear relationship with the regression line,R2 values ranging from 0.91 – 1.0 when the percentage of patients with non-ischemic cardiomyopathy was plotted on the x-axis and treatment effect was plotted on the y-axis, [Figure 2-Figure 3]. On the contrary, the regression line for 6MWT was flat, indicating that the improvement in 6MWT was not affected by the percentage of patients with non-ischemic cardiomyopathy included in the trials. This is not surprising as the heterogeneity for 6MWT was driven by the AATAC and CASTLE AF trials,[6,8] and when excluded from the analysis, the heterogeneity dropped significantly (I2: 0.0%). On the otherhand, it does not appear that baseline LVEF was related to the efficacy of ablation in these trials for any of the outcomes assessed.

Multiple meta-analysis of RCTs [9,10,12] and observational studies [13] reported improvement in LVEF, 6-minute walk distance and quality of life when catheter ablation is used as a treatment strategy in patients with AF and systolic HF. Nonetheless, there was significant heterogeneity in the outcome measures that was unaccounted for. While the results of the pooled outcomes of this analysis are similar to previously published reports, this is the first meta-regression of RCTs to examine the source of heterogeneity among studies that we are aware of.

The greater benefit of AF catheter ablation seen in patients with non-ischemic cardiomyopathy may not be surprising if many of these patients have tachycardia induced cardiomyopathy that would be expected to improve with restoration of sinus rhythm. Conversely, restoration of sinus rhythm may be less likely to improve cardiac function and related outcomes whencardiomyopathy is due to ischemia.The results of this analysis support previous report by Ling et al [14] who performed AF ablation in 16 patients with cardiomyopathy and no late-gadolinium enhancement on cardiac magnetic resonance imaging. At six months follow-up, LVEF improved from 40% ± 10% to 60% ± 6% in the 15 patients who maintained sinus rhythm post ablation.

These results are novel and interesting and should be viewed as hypothesis generating.There are important limitations to this meta-regression analysis.First, each of the associations derived from the separate regressions are limited by the small number of trials. Second, the overall quality of the individual trials was assessed as moderateonly due to the potential for performance and ascertainment bias. Third, the meta-regression was performed based on the mean percentage of non-ischemic cardiomyopathy in each study. An individual level meta-analysis would more accurately address our questions. Nonetheless, these data are not available for the authors. Despite this limitation, we still find bio-plausibility in the regression results. For the regression analysis of difference in change inLVEF, where significant heterogeneity in this outcome was not reduced by sensitivity testing, the percentage of patients with non-ischemic cardiomyopathyin each trial varied from a low of 35% to a high of 100%. This is a significant spread of cardiomyopathy percentage over the trials, which could plausibly affect outcomes of the intervention. Also, the difference in change inmeanLVEF between trials ranged from nearly 0% to 15%. This spread is outside the range of typical inter-reader variability using echocardiography and enough to be considered clinically meaningful.Furthermore, when visually examining the regression plot there is no single trial that significantly deviates from the regression line intercept and its 95% confidence interval bounds. If on the other hand, the spread in non-ischemic cardiomyopathy percentage across trials ranged from 40-50% and mean LVEF difference from 0-5% it would be less credible to assert that a true relationship existed and even more so if one or more trials deviated significantly from the regression line.

Conclusions

In patients with systolic HF, AF catheter ablation appears to be more beneficial in patients with non-ischemic cardiomyopathy. More studies are needed to specifically examine this group of patients and test this hypothesis.

References

  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]