1 INTRODUCTION

Heart failure (HF) affects ~38 million people worldwide, with the incidence expected to rise by 46% by 2030 in the United States alone (Atherton et al., 2018; Mozaffarian et al., 2016). Half of all HF patients die within 5 years of diagnosis due to pump failure associated with reduced left ventricular ejection fraction (LVEF≤35%) (Ponikowski et al., 2016), or sudden cardiac death (SCD) (Ponikowski et al., 2016). Modern treatment strategies include risk factor modification, medications to enhance heart function, early revascularization, and implantable cardiac defibrillators (ICD) (Atherton et al., 2018; Ponikowski et al., 2016).

While most HF patients have no previous history of documented ventricular arrhythmias (VA), they have a fivefold increased risk of developing them (Priori et al., 2015). Primary prevention ICDs protect HF patients against ventricular arrhythmias that cause SCD. However, a number of studies show that up to 80% of ICD patients never experience sustained arrhythmias (Engstrom et al., 2020), suggesting the current guidelines about who should receive a device may need refinement (Disertori et al., 2020). New criteria are required to identify patients who are at risk for VA and require an implantable device.

One potential method is the detection of fragmented QRS (fQRS) on the electrocardiogram (ECG). This notching and slurring in the QRS, first described in 1969 (Flowers et al., 1969), represents inhomogeneous ventricular activation and conduction due to scar/fibrosis (Das & Zipes, 2009). The resultant slowing of terminal conduction promotes re-entrant circuits and a substrate for VA to occur (Das et al., 2009). fQRS has previously been shown to be an arrhythmogenic marker in congenital and familial acquired cardiomyopathies and syndromes (Supreeth & Francis, 2020). However, the use of fQRS as a VA marker in HF patients is unclear (Supreeth & Francis, 2020). The aim of this systematic review and meta-analysis was to assess the predictive capacity of fQRS for VA and its association with mortality in primary prevention HF patients.

2 METHODS 2.1 Systematic review

This systematic review was conducted and is reported using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Appendix S1). The protocol was registered and published with PROSPERO, an international register for systematic reviews (CRD42021226505).

2.2 Search strategy

All studies that examined fQRS and VA (ventricular tachycardia (VT) or ventricular fibrillation (VF)), in primary prevention HF patients with or without an ICD in situ, were included. An independent search was conducted in Scopus, CINAHL, EMCARE, and MEDLINE from commencement to October 2020 (Appendix S2). Reference lists of full-text studies were hand-searched to identify additional studies, and corresponding authors of two papers were contacted for additional data.

2.3 Inclusion criteria Studies were included if they met the following: Retrospective or prospective cohort, cross-sectional and longitudinal studies that described the occurrence or frequency of VA and the presence of fQRS at baseline, with a follow-up period ≥12 months. Ventricular arrhythmias included VT and VF or SCD classified as arrhythmic where (i) appropriate ICD therapy was delivered including shock and/or anti-tachycardia pacing (ATP) or shock alone, or (ii) unexpected death occurring within 1 h of cardiac symptoms in the absence of progressive cardiac deterioration, or (iii) unexpected death during sleep, or (iv) unexpected death within 24 h after the patient had been seen alive based on a modified Hinkle-Thaler system (Hinkle & Thaler, 1982). ECG analysis showing fQRS as defined by (Das et al., (2009), that is, QRS <120 ms with an additional R wave or notching at the lowest point of the S/R wave, or the existence of >1 R wave in two or more successive leads corresponding to a coronary artery territory (Das et al., 2009). Fragmented wide-QRS (f-wQRS) >120 ms as described above with the additional presence of two or more notches in the R or S wave were also included (Das et al., 2009). Primary prevention indication, that is, reduced LVEF ≤40% with no previous history of sustained VA, with or without an ICD or cardiac resynchronization therapy (CRT) in situ, and with either ischemic (ICM) and/or non-ischemic cardiomyopathy (NICM) (JCS Joint Working Group, 2012). Studies that included secondary prevention patients that had documented sustained VA, or a history of unexplained loss of consciousness with or without an ICD/CRT in situ, were included if primary prevention patients made up at least 70% of the total study population, which is representative of the current ICD population seen clinically (Kremers et al., 2013). 2.4 Exclusion criteria

Studies that focused on hypertrophic obstructive cardiomyopathy, Brugada, congenital heart disease, arrhythmogenic right ventricular cardiomyopathy, long QT, short QT, noncompaction cardiomyopathy, and Chagas were excluded. Other methodology of fQRS such as vectorcardiography, magnetocardiography, magnetic field imaging, signal-averaged ECG, and 120-lead body surface potential mapping was also excluded. Non-English language publications, review articles, case studies, conference abstracts, and animal studies were not included.

2.5 Study selection

Two investigators (NE and HL) screened the titles and abstracts of all retrieved citations to identify studies meeting the inclusion criteria. Full texts of eligible studies were retrieved and reviewed by the same two investigators for inclusion and relevance with mutual agreement.

2.6 Data extraction

Data were extracted for general characteristics (authors, year, title, journal, publication type); study characteristics (design, sample size, follow-up time, fQRS definition); patient characteristics (age, gender, comorbidities, medications); clinical characteristics (cardiomyopathy type, New York Heart Association (NYHA) class, LVEF, ICD status); and outcome data (VA, ICD therapy, mortality). When assessing VA, if appropriate shock only was reported, this was combined with ICD therapy of both shock and ATP.

2.7 Quality assessment

A modified Newcastle-Ottawa scale, including assessments of indication/etiology, representativeness of patient cohort, research methodology, detail of ECG analysis, VA definition, adequacy of follow-up, reporting of loss to follow-up, and detail of coronary artery territory location, was used for quality assessment (Appendix S3). Each study was assessed as low, moderate, or high risk of bias.

2.8 Meta-analysis

The meta-analysis was conducted in accordance with the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) Group guidelines (Appendix S4), using Review Manager software (V5.4.1). A random-effects meta-analysis was undertaken to account for inherent variability. Proportions or count data were converted to hazard ratios (HR) using the methodology of Parmar and colleagues (Parmar et al., 1998). Outcome measures analyzed between fQRS and non-fQRS patients were as follows: (1) VA (including appropriate ICD shock), (2) all-cause mortality, and (3) composite endpoint of VA and/or all-cause mortality. Subgroup analyses included (1) primary prevention patients vs primary and secondary prevention patients, (2) NICM vs ICM patients, and (3) fQRS 12-lead ECG coronary artery territory location. Outcomes are reported as HR with 95% confidence intervals. Statistical significance was defined as p < .05. Statistical heterogeneity was determined by I2 statistic (I2 < 25%, low; I2= 25–50%, moderate; I2 > 50%, substantial).

3 RESULTS 3.1 Study characteristics

The search strategy yielded 1,111 articles of which 211 were duplicates, leaving 900 for title/abstract assessment. No additional articles were obtained through contact with authors or reference list searching. Based on eligibility criteria, 848 articles were excluded by title/abstract, leaving 52 for full-text evaluation. A total of 26 were excluded because of specific etiology and 10 due to ineligible methodology and outcomes. Five studies were excluded due to the patient cohort having predominantly secondary prevention indications (≥30%), while one study used the same patient cohort in two separate articles, leaving 10 studies involving 3,885 patients for analysis (Figure 1). A full description of included studies is shown in Table 1. According to the modified Newcastle-Ottawa quality assessment scale, seven studies had a moderate risk of bias, and three were low risk. Heterogeneity was high for both the primary and secondary outcomes of VA (I2 = 92%) and all-cause mortality (I2 = 91%). Follow-up time ranged from 14 to 50 months (Table 1).

PRISMA flow diagram of study selection process. LVEF, left ventricular ejection fraction; VA, ventricular arrhythmias

TABLE 1. Study characteristics and risk of bias Author, Year Study Type Sample Size Follow-up (months) Risk of bias Kucharz and Kułakowski, 2020 Retrospective, single-center cohort study 365 34.5 ± 18 Low Claridge et al., 2017 Prospective, single-center cohort study 130 33.5 ± 24 Moderate Vandenberk et al., 2017 Retrospective, single-center cohort study 407 50.5 ± 38 Moderate Igarashi et al. 2017 Retrospective, multi-center cohort study 137 18 Moderate Ozcan et al., 2014 Retrospective, single-center cross-sectional 215 23.5 ± 12.1 Moderate Ozcan et al., 2013 Retrospective, single-center cohort study 227 44.8 ± 16.9 Moderate Brenyo et al., 2012 Retrospective, RCT study 1040 20 Low Forleo et al., 2011 Retrospective, single-center cohort study 394 23.6 ± 17.5 Moderate Sha et al., 2011 Retrospective, single-center cohort study 128 14 ± 5 Moderate Cheema et al., 2010 Retrospective, multi-center cohort study 842 40 ± 17 Low Abbreviation: RCT, Randomized controlled trial. 3.2 Patient cohort

Primary prevention patients comprised 100% of the study cohort in six studies (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Özcan et al., 2014; Ozcan et al., 2013; Vandenberk et al., 2017), with the remaining four having 71%–91% primary prevention indications (Table 2) (Claridge et al., 2017; Kucharz & Kułakowski, 2020; Igarashi et al., (2017); Sha et al., 2011). ICM was the sole etiology in Brenyo et al. (2012), while Igarashi et al. (2017) and Sha et al. (2011) included only NICM patients. The remaining studies had mixed indications with ICM accounting for ~47%–77% of the study population (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Kucharz & Kułakowski, 2020; Özcan et al., 2014; Ozcan et al., 2013; Vandenberk et al., 2017). In six studies, ICDs were implanted in the whole cohort (Cheema et al., 2010; Claridge et al., 2017; Forleo et al., 2011; Kucharz & Kułakowski, 2020; Özcan et al., 2014; Ozcan et al., 2013; Vandenberk et al., 2017), while three studies had a mix of ICD and non-ICD patients (Brenyo et al., 2012; Cheema et al., 2010; Sha et al., 2011).

TABLE 2. Patient Characteristics Author, Year Male Gender Primary Prevention ICM NICM NYHA Class LVEF (%)* ICD implanted Kucharz and Kułakowski, 2020 306 (83.4%) 259 (70.6%) 273 (74%) 94 (26%) 10 (6%) 68 (41%) 84(50.6%) 4 (2.4%) 27.7±9.5

Yes

100%

Claridge et al., 2017 58(80.6%) 93 (71.5%) 72 (55.4%) 58 (44.6%) N/A N/A

Yes

100%

Vandenberk et al., 2017 343 (84.3%) 407 (100%) 215 (52.8%) 192 (47.2%) 95 (23.4%) 156 (38.3%) 156 (38.3%) 28.3±10.3

Yes

100%

Igarashi et al., 2017 92 (67.2%) 137 (79.6%) 0 (0%) 137 (100%) 0 25 (18.2%) 84 (61.3%) 23 (16.8%) 29.2±9.7

Yes

CRT-P or CRT-D

(no breakdown)

Ozcan et al., 2014 156 (72.5%) 215 (100%) 102 (47.4%) 113 (52.5%)

76 (35.3%)

112 (52.1%)

27 (12.5%)

27.7±3.5

Yes

100%

Ozcan et al., 2013 156 (68.7%) 227 (100%) 142 (62.5%) 85 (37.4%) 0 83 (36.5%) 104 (45.8%) 40 (17.6%) 26.5±0.06

Yes

100%

Brenyo et al., 2012 1009 (97%) 100% 1040 (100%) 0 (0%) N/A N/A

Yes

ICD−693 (66.6%)

Non-ICD 347 (33.3%)

Forleo et al., 2011 334 (84.8%) 100% 242 (61.4%) 115 (29.2%) 3 (2–3)† 27±9

Yes

100%

Sha et al., 2011 87 (68%) 90.6% LVEF<40% 0 (0%) 128 (100%) IDCM N/A 30±6

Yes

ICD−10 (7.8%)

Non-ICD 118 (72.2%)

Cheema et al., 2010 655 (77.8%) 842 (100%) 644 (76.5%) 198 (23.5%) N/A 27±6.3

Yes

ICD−435 (51.7%)

Non-ICD 407 (48.3%)

Abbreviations: CRT-D, cardiac resynchronization therapy defibrillator; CRT-P, cardiac resynchronization therapy pacemaker; ICD, implantable cardiac defibrillator; ICM, ischemic cardiomyopathy; IDCM, idiopathic dilated cardiomyopathy; LVEF, left ventricular ejection fraction; N/A, not applicable; NICM, non-ischemic cardiomyopathy; NYHA, New York Heart Association. *Mean ± standard deviation. †Median (interquartile range).

Patients in the non-fQRS cohort were significantly younger than fQRS patients (62.6 ± 13.7 vs 60.7 ± 12.9 years), with a mean difference of ~1.5 years (−1.5[−2.66, −0.42], p = .007). All studies had a higher proportion of males (67.2%–97%), and males were more likely to have fQRS (Table 2). There was no difference in LVEF, or incidence of coronary artery disease, hypertension, or renal failure between fQRS and non-fQRS cohorts. The non-fQRS patients were 28% more likely to have a history of atrial fibrillation; however, this was not statistically significant (0.82[0.67, 1.00], p = .05). Diabetes was significantly more likely if fQRS was present (1.12[1.01, 1.25], p = .03). Medications including beta-blockers, angiotensin-converting enzyme inhibitors, aspirin, and Class III antiarrhythmic drugs showed similar use in both groups; however, fQRS patients were 27% more likely to be on a statin (1.27[1.05, 1.55], p = .02).

3.3 Ventricular arrhythmias

The association between fQRS and incidence of VA was reported in eight studies (Figure 2a) (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Igarashi et al., 2017; Kucharz & Kułakowski, 2020; Özcan et al., 2014; Ozcan et al., 2013; Vandenberk et al., 2017). Of the 3,627 patients where VT/VF occurred, arrhythmias were significantly ~1.5 times more likely in fQRS patients (1.51[1.02, 2.25], p = .04). A sensitivity analysis omitting studies that also included secondary prevention patients resulted in the same hazard ratio, however, did not reach statistical significance (1.51[0.98, 2.31], p = .06) (Figure 2b).

Forest plot demonstrating the association between fQRS and ventricular arrhythmias in heart failure patients (a), including subgroup analysis of primary prevention only compared to primary and secondary prevention patients (b). CI, confidence interval; fQRS, fragmented QRS

3.4 All-cause mortality

All-cause mortality was reported in seven studies (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Özcan et al., 2014; Ozcan et al., 2013; Sha et al., 2011; Vandenberk et al., 2017) and was significantly 1.7 times more likely in fQRS patients (1.68[1.13, 2.52], p = .01) (Figure 3a). When fQRS was isolated to ECG lead territories, fQRS found in the lateral leads was associated with 39% increased mortality risk, while inferior and anterior lead fQRS showed 21% and 33% increases, respectively, with no territory demonstrating a statistical association with all-cause mortality (Figure 3b). A comparison of NICM and ICM groups showed NICM patients with fQRS had a significant 2.6-fold increased risk of death (2.55[1.63, 3.98], p < .0001), whereas in ICM patients the presence of fQRS did not increase mortality (1.10[0.79, 1.53], p = .58) (Figure 4).

Forest plot demonstrating the association between fQRS and all-cause mortality in heart failure patients (a), including subgroup analysis of 12-lead ECG coronary artery territory location (b). CI, confidence interval; ECG, electrocardiogram; fQRS, fragmented QRS

Forest plot demonstrating the association between fQRS and all-cause mortality in NICM versus ICM heart failure patients. CI, confidence interval; fQRS, fragmented QRS; ICM, ischemic cardiomyopathy; NICM, non-ischemic cardiomyopathy

3.5 All-cause mortality and ventricular arrhythmias

The composite endpoint of all-cause mortality and VA was assessed in 817 patients across four studies (Claridge et al., 2017; Forleo et al., 2011; Özcan et al., 2014; Sha et al., 2011). Patients with fQRS were ~2.2-times more likely to have VA or die of any cause; however, this association was not significant (2.17 [0.95, 4.98], p = .07) (Figure 5).

Forest plot demonstrating the association between fQRS and the composite endpoint of ventricular arrhythmias or all-cause mortality in heart failure patients. CI, confidence interval; fQRS, fragmented QRS

4 DISCUSSION

Despite significant advances in cardiac research, imaging and testing, the identification of patients at risk of sudden death from ventricular arrhythmias (VAs) remains challenging (Priori et al., 2015). One possible independent risk parameter that has attracted much interest is fragmentation of the QRS (fQRS). The complex originates from a conduction delay and disrupted ventricular depolarization due to regional myocardial scarring that can form an arrhythmogenic substrate for lethal VA (Das & Zipes, 2009). Our meta-analysis indicates that fQRS is significantly associated with VA in HF patients with ischemic and non-ischemic cardiomyopathy (Figure 2a). Patients exhibiting fQRS were also significantly 1.7 times more likely to die of any cause (Figure 3a), with the incidence of death significantly higher in NICM patients (Figure 4). Patients with and without fQRS were comparable with regards to EF, comorbidities, and medications, except for diabetes which was significantly more likely in fQRS patients, and a 28% increased likelihood of atrial fibrillation in non-fQRS patients. These results will now be discussed in terms of the structure of the meta-analysis with ventricular arrhythmias and all-cause mortality as primary endpoints.

4.1 fQRS is associated with ventricular arrhythmias in HF patients

Our meta-analysis provides the first synthesized evidence that fQRS may be significantly associated with VA in a cohort of 3,627 patients with reported VA. The idea of a fragmented QRS complex as a potential VA or ICD indicator was first introduced by Das and colleagues in 2009 (Das et al., 2009). However, individual studies have failed to reach a consensus on its usefulness as a VA risk factor (Brenyo et al., 2012; Cheema et al., 2010; Claridge et al., 2017; Forleo et al., 2011; Igarashi et al., 2017; Kucharz & Kułakowski, 2020; Özcan et al., 2014; Ozcan et al., 2013; Vandenberk et al., 2017). Part of the reason appears to be that many studies were underpowered and different groups used different criteria for the assessment of VA. For example, the studies of Vandenberk (Vandenberk et al., 2017) and Brenyo (Brenyo et al., 2012), comprising 1,440 patients reported ICD shock only as an endpoint. In these studies, ventricular arrhythmias that may have been treated by anti-tachycardia pacing were not included for analysis, possibly underestimating VA incidence and their association with fQRS. In contrast, Özcan et al., (2014) had considerably higher shock rates than reported by other studies (52%) and is most likely the result of short duration, low-rate detection programming of the ICDs in that study. Future studies should consider detailed specification of programming when evaluating arrhythmias in ICD patients.

Our review encompassed all VAs analyzed, including ICD shocks and ATP, shocks alone, and sudden death criteria for non-ICD patients. While the selection of eligible papers for this review appears to be representative of the current clinical population, unfortunately, not all studies included all variables required for complete analysis. Furthermore, there was no separate analysis of ischemic and non-ischemic cardiomyopathy patients. The inclusion of secondary prevention patients is also a potential confounder in the current meta-analysis; however, secondary prevention patients only represented 3.75% (136) of the total population analyzed. Future studies should include subset analysis if both ICM and NICM, and primary and secondary prevention, patients are included.

Another well-established factor promoting arrhythmias is a wide-QRS complex (>120 ms), which is a marker of slow conduction that may promote re-entrant VT (Kashani & Barold, 2005). In this meta-analysis, six of the eight studies examined the association of fQRS in both narrow and wide-QRS (>120 ms) with VA in heart failure patients (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Igarashi et al., 2017; Kucharz & Kułakowski, 2020; Özcan et al., 2014; Vandenberk et al., 2017). Two studies that reported fQRS was significantly associated with VA also reported increased wide-QRS in the fQRS group (Kucharz & Kułakowski, 2020; Özcan et al., 2014), whereas four studies showed no difference (Brenyo et al., 2012; Cheema et al., 2010; Forleo et al., 2011; Vandenberk et al., 2017), while two did not report. We conclude that future studies should include wide-QRS along with fQRS data to further evaluate the relationship in HF patients.

Our analysis also found that the fQRS patients were significantly 1.5 years older than non-fQRS patients and had a 12% increased incidence of diabetes. Multiple studies have shown that age is not a contributing factor to VA in ICD patients (Santangelo et al., 2021), (Bergau et al., 2017), and we and others have found that comorbidities are not often associated with VA or appropriate ICD therapy (Engstrom et al., 2020). While some studies have demonstrated a link between diabetes and VA (Grisanti, 2018), others have reported the opposite (Juhani Junttila et al., 2020), which may be related to HF severity and the presence of different comorbidities. Further studies involving larger populations are required to clarify the relationship between age, diabetes, comorbidities, and VA in heart failure patients with and without fQRS.

4.2 fQRS is associated with all-cause mortality in heart failure patients

Another potentially important clinical finding of our meta-analysis is that during the ~1-to-4-year follow-up, patients with fQRS were significantly 68% more likely to die of any cause (Figure 3a). Mortality is most likely a ‘systems failu