Electrocardiographic predictors of ventricular arrhythmic events in ICD recipients: signal beyond the left ventricular ejection fraction threshold

Abstract

Background and objectives

Selection for implantable cardioverter-defibrillator (ICD) therapy based on left ventricular ejection fraction (LVEF) may miss at-risk patients with LVEF > 35%. We tested whether a simple electrocardiogram (ECG) burden adds discriminatory power for appropriate ICD therapy beyond clinical factors and LVEF.

Methods

Single-center retrospective cohort of 236 ICD recipients. Six routinely reported ECG abnormalities (heart rate > 75 bpm; QRS duration > 110 ms; obtuse QRS–T angle; delayed transition zone; prolonged QTc; prolonged Tp–Te) were prespecified, summed to form an ECG burden (0–6), and categorized as 0–1, 2–3, or ≥ 4 abnormalities. Logistic regression estimated odds ratios (OR) with 95% CIs for appropriate ICD therapy. Multivariable models included LVEF category (≤ 35% vs > 35%), ischemic vs non-ischemic cardiomyopathy, ICD indication (secondary vs primary prevention), and ECG burden category. Discrimination of the clinical + LVEF model with and without ECG burden was assessed using the AUC (C-statistic) and the DeLong test.

Results

Mean age was 62.1 ± 16.2 years; 71% were male; 61% had ischemic cardiomyopathy; 72% underwent primary-prevention ICD implantation; and mean follow-up was 7.6 ± 2.9 years. Appropriate ICD therapy was delivered to 26/236 (11.0%) patients. A high ECG burden (≥ 4 abnormalities) was associated with appropriate ICD therapy after adjustment (OR 3.6, 95% CI 1.8–7.2; p < 0.001). Adding ECG burden to a clinical + LVEF model modestly improved discrimination in this cohort, with the AUC (C-statistic) from 0.67 to 0.72 (DeLong p = 0.01). Kaplan–Meier curves showed graded separation across ECG-burden categories (log-rank p < 0.001). Among patients with LVEF > 35%, therapy was observed more frequently in the high-burden group than in the low-burden group.

Conclusions

In this single-center retrospective cohort of ICD recipients, a higher ECG abnormality burden was associated with a higher risk of appropriate ICD therapy and modestly improved discrimination when added to a clinical + LVEF model. These findings are hypothesis-generating and warrant prospective, multicenter external validation using standardized ECG measurements before any clinical application.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12872-026-05554-w.

Introduction

Sudden cardiac death (SCD) remains a leading cause of death. Current selection for implantable cardioverter-defibrillators (ICDs) relies heavily on left ventricular ejection fraction (LVEF), which reflects systolic dysfunction but does not fully capture the arrhythmic substrate. This LVEF-centered approach reflects contemporary guideline-based ICD selection criteria for primary prevention, which continue to use an LVEF threshold (typically ≤ 35%) alongside clinical context to guide eligibility [1]. Consequently, many ICD recipients never receive appropriate therapy, while some patients with preserved LVEF experience ventricular arrhythmias.

The 12-lead electrocardiogram (ECG) reflects conduction delay and repolarization heterogeneity, physiological indicators of arrhythmic risk. Individual ECG markers (e.g., prolonged QRS, abnormal QRS–T angle, prolonged QT, increased Tp–Te) have been linked to SCD-related outcomes. However, the performance of single parameters varies across cohorts and is often modest when applied in isolation. These measures capture complementary mechanistic domains, namely ventricular activation/conduction delay and repolarization heterogeneity, that may jointly reflect an arrhythmogenic substrate.

We evaluated whether a compact, six-parameter ECG “burden” (heart rate > 75 bpm; QRS > 110 ms; abnormal/obtuse QRS–T angle; delayed transition zone; prolonged QT; prolonged Tp–Te) is associated with appropriate ICD therapy beyond LVEF and key clinical factors, and whether adding ECG burden improves discrimination in this cohort. We also examined performance among patients with preserved LVEF (> 35%), where additional risk stratification could complement guideline thresholds.

Methods

Study design and population

This was an exploratory, single-center retrospective cohort study conducted at King Abdulaziz Cardiac Center, King Abdulaziz Medical City, National Guard Health Affairs, in Riyadh, Saudi Arabia. We identified 236 consecutive adults who underwent implantable cardioverter-defibrillator (ICD) implantation between 2010 and 2015 and had an analyzable pre-implant 12-lead ECG (Supplementary Table S1).

Inclusion criteria.

• Primary prevention: LVEF ≤ 35% in patients with ischemic or non-ischemic cardiomyopathy and no history of sustained VT/VF.

• Secondary prevention: prior sustained VT/VF, aborted SCD, or syncope with documented ventricular arrhythmias.

• Availability of a pre-implantation 12-lead ECG and follow-up device interrogations.

Exclusion criteria.

• Patients with AF/flutter or a paced rhythm on the index ECG were excluded because these rhythms can distort the QRS–T angle and repolarization measures (QTc, Tp–Te). Pacing burden (e.g., ≥ 50% paced beats) was not available in the source data; therefore, any paced rhythm was excluded.

• Incomplete ECG data (see Missing data) or lost to follow-up.

Screening, exclusions, and the final analytic cohort are shown in Supplementary Figure S1 (flow diagram). Baseline characteristics are presented in Table 2.

Table 2.

Baseline characteristics

Characteristics	Value
Mean age	62.1 ± 16.2 years
Male sex	71%
Primary prevention ICDs	72%
Secondary prevention ICDs	28%
Ischemic cardiomyopathy	61%
Follow-up duration	7.6 ± 2.9 years

Mean ± SD. Mean LVEF in the overall cohort was 30.1 ± 9.6%. Among patients with definitive binary EF classification, mean LVEF was 26.4 ± 4.1% in the ≤ 35% group (n = 198) and 51.9 ± 4.6% in the > 35% group (n = 30)

ECG data and risk burden construction

Six predefined ECG parameters were extracted from the pre-implantation 12-lead ECG (Table 1):

• Heart rate (abnormal if > 75 bpm)

• QRS duration (abnormal if > 110 ms)

• QRS–T angle (abnormal/obtuse)

• QRS transition zone (delayed)

• QT interval (prolonged)

• Tp–Te interval (prolonged)

Table 1.

ECG parameters and criteria

Parameter	Criteria for abnormality
Heart rate	> 75 bpm
QRS duration	> 110 ms
QRS–T angle	Abnormal/obtuse on automated ECG report (report-generated QRS–T angle)
QRS transition zone	Delayed
QT interval	Prolonged (QTc, ms; Bazett-corrected)
Tp–Te interval	Prolonged (ms)

QT prolongation was defined using sex-specific QTc thresholds (QTc ≥ 450 ms for men and ≥ 470 ms for women), QT was corrected for rate using Bazett’s formula. See Methods for details and references

Cutoffs were predefined based on existing literature linking these markers to the risk of sudden cardiac death. A cumulative ECG burden was defined as the number of abnormalities and classified as 0–1 (low), 2–3 (intermediate), or ≥ 4 (high) (Fig. 1).

Fig. 1.

Distribution of patients by number of ECG abnormalities and corresponding risk categories (0–1, 2–3, ≥ 4)

Measurement reproducibility. ECG parameters were extracted from standard automated 12-lead ECG reports by trained reviewers not involved in outcome adjudication. The QRS–T angle abnormality was taken directly from the automated ECG report (reported as abnormal/obtuse); a separate spatial QRS–T angle was not available in the source data. We did not perform formal manual re-measurement or reproducibility validation of these automated parameters in this retrospective dataset.

Outcomes and follow-up

The primary endpoint was appropriate ICD therapy for ventricular tachyarrhythmia (anti-tachycardia pacing or shock), as determined by routine device checks and clinical records. The time origin was ICD implantation; patients were followed until the first appropriate therapy, death, last contact, or study conclusion.

Endpoints were adjudicated. Appropriate therapies were confirmed by the treating electrophysiology team using device logs, in accordance with standard institutional procedures.

Statistical analysis

Continuous variables are presented as mean ± SD (or median [IQR], as appropriate), and categorical variables as counts (%). Group comparisons used the t test or Wilcoxon rank-sum test for continuous variables and the χ² test or Fisher’s exact test for categorical variables, as appropriate. Two-sided p values < 0.05 were considered statistically significant. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals. Pairwise AUC differences were compared using DeLong’s test.

Primary modeling

Logistic regression was used to estimate odds ratios (ORs) with 95% CIs for appropriate ICD therapy. Prespecified covariates included LVEF category (≤ 35% vs > 35%), ischemic versus non-ischemic cardiomyopathy, and ICD indication (secondary vs primary prevention). ECG burden was modeled as an ordinal categorical variable (0–1, 2–3, ≥ 4), with ≥ 4 defined a priori as the high-burden category of interest. Analyses requiring dichotomous EF classification were restricted to patients with definitive EF categories; LVEF ranges spanning the 35% threshold (e.g., 34–45% or 35–45%) and those with missing LVEF were excluded to avoid misclassification.

Time-to-event analyses

Kaplan–Meier curves were used to compare therapy-free survival across ECG burden categories, and differences were assessed with the log-rank test (Fig. 2). As a secondary analysis, Cox proportional hazards models were fit for time to first appropriate therapy, and proportional hazards assumptions were evaluated using Schoenfeld residuals.

Fig. 2.

Kaplan–Meier curves for ICD therapy-free survival stratified by the number of ECG abnormalities (0–1, 2–3, ≥ 4), showing separation of risk strata during follow-up

Missing data

No imputation was performed. All six ECG parameters required to compute the ECG burden were available in the automated ECG reports for the included ECGs. Values were abstracted by trained reviewers, and automated misreads cannot be fully ruled out.

Discrimination and calibration

We assessed discrimination using the area under the receiver operating characteristic curve (AUC) for the clinical + LVEF model and for the model additionally including ECG burden. Analyses requiring dichotomous EF classification (≤ 35% vs > 35%) were restricted to patients with definitive EF categories; patients with LVEF ranges spanning 35% or missing EF were excluded from binary EF analyses to avoid misclassification around the 35% threshold (binary EF available in 228/236 patients). 95% CIs for AUCs and pairwise DeLong tests were used to evaluate differences between models; calibration was assessed using the calibration slope and intercept (Supplementary Figure S2).

Subgroup and interaction

A prespecified subgroup included patients with unequivocal LVEF > 35% (n = 30). We examined the interaction between ECG burden (≥ 4) and LVEF categories (≤ 35% vs. > 35%) on the logit scale among patients with definitive binary EF classification (n = 228). Observed rates by LVEF and ECG categories are shown in Fig. 4.

Fig. 4.

Rates of appropriate ICD therapy by LVEF and ECG burden category (High ECG burden: ≥ 4 abnormalities)

Sensitivity analyses

Sensitivity analyses included: (1) primary prevention only; (2) secondary prevention only; (3) excluding ECGs flagged for borderline quality; and (4) modeling ECG burden as a continuous 0–6 count.

Software

Analyses were performed in R (version ≥ 4.3) or Stata (version ≥ 17). All tests were two-sided at α = 0.05.

Results

Binary EF analyses

Binary EF classification was available in 228/236 patients (198 with LVEF ≤ 35% and 30 with LVEF > 35%); 7 patients (3.0%) had LVEF reported as a range crossing the 35% threshold (e.g., 34–45% or 35–45%), and one patient (0.4%) had missing LVEF. These were excluded from analyses requiring dichotomous EF classification.

ECG Parameter Distribution (Table 3)

Table 3.

Distribution of ECG abnormalities

Number of abnormalities	Patients (%)
0–1	27%
2–3	45%
≥ 4	28%

• ≥ 4 ECG abnormalities: 28% of patients have four or more ECG abnormalities.

• Most common abnormalities:

  • HR > 75 bpm (47.3%)
  • Prolonged QRS duration (46.8%)
  • Obtuse QRS–T angle (41.4%)
  • Delayed QRS transition (40.9%)

Primary outcome

Appropriate ICD therapy was delivered to 26/236 (11.0%) patients. Event rates increased with higher ECG abnormality burden (Fig. 1). Among patients with unequivocal LVEF > 35% (n = 30), events still occurred (4/30; 13.3%), with higher observed therapy rates in the high-burden group than in the low-burden group (Fig. 4).

Multivariable association (Table 4)

Table 4.

Multivariable analysis for ICD therapy

Predictor	OR (95% CI)	P-value
≥ 4 ECG abnormalities	3.6 (1.8–7.2)	< 0.001
LVEF ≤ 35%	2.2 (1.4–3.5)	0.001
Ischemic Cardiomyopathy	1.7 (1.0–2.8)	0.024
Secondary prevention	2.3 (1.3–3.9)	0.004

In sensitivity analyses, treating ECG burden as a continuous 0–6 count showed a similar direction and magnitude of association compared with the categorical specification (0–1, 2–3, ≥ 4), and the main conclusions were unchanged

In multivariable analysis, high ECG burden (≥ 4 abnormalities) remained associated with appropriate ICD therapy after adjustment for key clinical covariates, including LVEF ≤ 35%, ischemic cardiomyopathy, and secondary prevention indication (Table 4).

Model discrimination, calibration, and time-to-event analyses. ROC analyses showed higher discrimination when ECG burden was added to the clinical + LVEF model (AUC (C-statistic) 0.67 vs 0.72; DeLong p = 0.01) (Fig. 3). Calibration of the combined model is shown in Supplementary Figure S2. Kaplan–Meier curves showed graded separation across ECG-burden categories (log-rank p < 0.001) (Fig. 2).

Fig. 3.

Receiver operating characteristic (ROC) curves comparing the clinical + LVEF model with and without the ECG burden. AUCs (95% CIs) are shown; DeLong p = 0.01 for the comparison. Events: 26/236 appropriate ICD therapy events

Discussion

Principal findings

In this single-center cohort of 236 ICD recipients, a higher burden of routinely reported ECG abnormalities was associated with appropriate ICD therapy after adjustment beyond LVEF and key clinical factors. Adding ECG burden to a clinical + LVEF model modestly improved discrimination in this cohort (AUC 0.67 to 0.72; DeLong p = 0.01), and time-to-event analyses showed graded separation in therapy-free survival across ECG-burden categories (Figs. 2, 3 and 4). These findings should be considered exploratory and hypothesis-generating and require external validation before any clinical application.

Comparison with existing literature

Population-based studies have shown that composite ECG abnormalities can identify increased risk of sudden cardiac death beyond LVEF, and device-based cohorts have linked specific ECG markers (e.g., QRS prolongation, repolarization indices, and abnormal depolarization–repolarization relationships) to ventricular arrhythmias and ICD therapy. Our six-parameter ECG burden integrates multiple routinely captured signals into a pragmatic composite, suggesting potential incremental discrimination beyond LVEF in an ICD cohort and extending observations from SCD and trial populations to real-world ICD recipients (Figs. 2, 3 and 4) [2–8].

Interpretation and potential mechanisms

The six ECG parameters reflect complementary domains of electrical risk: ventricular activation/conduction (QRS duration, transition zone), global repolarization timing (QT/QTc), and repolarization heterogeneity and activation–repolarization coupling (QRS–T angle and Tp–Te), along with baseline autonomic/physiologic tone (heart rate). In particular, an obtuse QRS–T angle and a prolonged Tp–Te interval both indicate increased electrical heterogeneity and a mismatch between depolarization and repolarization, which may facilitate re-entrant and triggered ventricular arrhythmias. Accordingly, a high burden (≥ 4 abnormalities) may represent an aggregate signal of arrhythmogenic substrate and vulnerability, consistent with the graded separation observed in time-to-event analyses and the association after adjustment (Fig. 2; Table 4).

Clinical interpretation. ECG burden can be derived from routinely reported ECG features without additional testing, supporting the feasibility of standardized reporting in future studies. In this cohort, higher ECG burden was associated with higher observed rates of appropriate ICD therapy, including among patients with LVEF > 35%. These observations are hypothesis-generating and should not guide ICD decision-making without prospective multicenter external validation using standardized ECG measurements and assessment of clinical utility.

Model performance, reclassification, and subgroups

The observed improvement in discrimination (AUC) with the combined model was modest; however, discrimination alone does not establish clinical utility. Future prospective multicenter studies should evaluate calibration across diverse settings and incorporate clinical utility metrics, including decision-curve analysis and reclassification measures (e.g., NRI/IDI), to determine whether adding ECG burden meaningfully improves risk stratification at clinically relevant thresholds. Although patterns across LVEF strata suggest potential heterogeneity in effect (Fig. 4), this does not establish effect modification; adequately powered studies with prespecified interaction testing will be required. Additional robustness analyses (e.g., standardizing device programming and adjudication and using protocol-defined ECG measurements) would further strengthen inference.

Strengths and limitations

Notable strengths include prespecified, interpretable ECG thresholds, clinically meaningful endpoints, and extended follow-up. The primary limitation of this study is its single-center, retrospective, exploratory design, which limits generalizability and introduces potential residual confounding despite multivariable adjustment. In addition, the number of outcome events was modest (n = 26), which may undermine the stability of multivariable estimates and discrimination metrics and increase the risk of overfitting. ECG abnormalities were derived from routine clinical records and may be subject to measurement heterogeneity. Several ECG parameters relied on automated report-derived readings; although extracted by trained reviewers, automated misclassification and measurement variability cannot be fully excluded. Device programming and detection thresholds were not standardized, which may affect the ascertainment of appropriate ICD therapy. Although we evaluated discrimination and calibration, we did not assess clinical utility metrics (e.g., decision-curve analysis) or reclassification measures (e.g., NRI/IDI), which should be addressed in future prospective studies. Finally, because LVEF was occasionally documented as a range spanning the 35% threshold (e.g., 34–45% or 35–45%), these patients, along with those with missing LVEF, were excluded from analyses requiring dichotomous EF classification to avoid misclassification. Although the number was small, this may limit inference specifically at the borderline EF threshold and further supports the need for standardized EF assessment and external multicenter validation.

Conclusion

In this exploratory, single-center, retrospective cohort, a higher six-parameter ECG abnormality burden was associated with appropriate ICD therapy after adjusting for LVEF and key clinical factors. Discrimination improved when ECG burden was added to the clinical + LVEF model, but performance estimates are preliminary given the limited number of events and the retrospective design. These findings are hypothesis-generating and require prospective, multicenter external validation with standardized ECG measurement and assessment of clinical utility before any clinical application.

Abbreviations

AF

Atrial fibrillation

AUC

Area under the receiver operating characteristic curve

AVID

Antiarrhythmics versus Implantable Defibrillators

CIDS

Canadian Implantable Defibrillator Study

CI

Confidence interval

ECG

Electrocardiogram

ICD

Implantable cardioverter-defibrillator

KAIMRC

King Abdullah International Medical Research Center

KM

Kaplan–Meier

LVEF

Left ventricular ejection fraction

MNGHA

Ministry of National Guard Health Affairs

OR

Odds ratio

ROC

Receiver operating characteristic

SCD

Sudden cardiac death

SD

Standard deviation

Tp–Te

Tpeak-to-Tend interval

TRIPOD

Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis

VF

Ventricular fibrillation

VT

Ventricular tachycardia

Authors’ contributions

Maysan Almegbel conceived the study, collected clinical data, and contributed to data interpretation. Abdulmohsen Almusaad contributed to study design, clinical adjudication, and manuscript revision. Ali Mahmoud Alzammam assisted with data collection and verification of clinical outcomes. Razan Alzayer helped with data extraction and the preparation of tables and figures. Muneera AlTaweel supervised the project, oversaw statistical analysis and data interpretation, and drafted and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This research did not receive any specific funding from public, commercial, or not-for-profit sector agencies.

Data availability

The datasets generated or analyzed in this study are not publicly available due to institutional policies and patient privacy concerns, but they can be obtained from the corresponding author upon reasonable request and approval from the Institutional Review Board.

Declarations

Ethics approval and consent to participate

The study protocol was reviewed and approved by the Institutional Review Board at King Abdullah International Medical Research Center (KAIMRC), Ministry of National Guard Health Affairs, Riyadh, Saudi Arabia (approval no. RC19/072/R). All procedures involving human participants and their data were conducted in accordance with institutional and national research standards, as well as the 1964 Declaration of Helsinki and its subsequent amendments. Given the study's retrospective design and the use of de-identified clinical data, the Institutional Review Board waived the requirement for informed consent.

Consent for publication

Not applicable. This manuscript contains no personally identifiable information in any form, including images, videos, or detailed case descriptions.

Competing interests

The authors declare no competing interests.