What is the clinical evidence from this study?

Study design: Observational. Population: Adult patients (aged ≥18) with left bundle branch block (LBBB) and paired echocardiogram within 14 days (n=2813). Intervention: AiTiALVSD AI-enabled ECG model vs. No AI-ECG assessment (standard care). Primary outcome: Presence of left ventricular systolic dysfunction (LVSD) defined as echocardiographic LVEF ≤40% (AUROC 0.930, 95% CI 0.924–0.937).

What question did this study set out to answer?

To determine the effectiveness of an AI-ECG model for detecting LVSD and predicting outcomes in LBBB patients.

February 17, 2026Open Access

Detection and prognostic stratification of left ventricular systolic dysfunction in left bundle branch block using an artificial intelligence–enabled electrocardiography

Q: What does this research mean for the field?

The AiTiALVSD AI-enabled ECG model accurately detects left ventricular systolic dysfunction with an AUROC of 0.930 and stratifies long-term cardiovascular risk in patients with left bundle branch block. Novelty: ClaimNovelty.CONFIRMATORY. Consensus alignment: ConsensusAlignment.NEUTRAL.

Key Result

AiTiALVSD AI-enabled ECG model detected LV systolic dysfunction with an AUROC of 0.930 and 97.9% sensitivity in patients with LBBB.

Key Points

To determine the effectiveness of an AI-ECG model for detecting LVSD and predicting outcomes in LBBB patients.
Analyzed 5,689 validated LBBB ECGs from 2,813 patients
Defined LVSD as ejection fraction of ≤ 40%
Assessed model performance using AUROC, AUPRC, sensitivity, and specificity
Stratified patients into risk groups based on 90% sensitivity threshold
Conducted Kaplan-Meier analysis for clinical outcomes comparisons.
AiTiALVSD model achieved AUROC of 0.930 and sensitivity of 0.979 for LVSD detection
High-risk patients had significantly higher hazards for all-cause mortality (HR 1.87)
The model showed notable predictive power for ICD/cardiac resynchronization therapy (HR 15.2)
Higher rates of cardiovascular hospitalization were observed in high-risk groups (HR 1.11).

Study Design

Type

Observational (n=2,813)

Multicenter

Yes

Structured PICO

Does the AiTiALVSD AI-ECG model accurately detect LVSD and predict future clinical outcomes in patients with left bundle branch block?

Population

2,813 adult patients (aged 18 years or older) with a 12-lead ECG showing LBBB and a paired echocardiogram performed within 14 days, mean age 70.7 years, 43.7% male, from two government-designated heart specialty hospitals in the Republic of Korea.

Intervention

AiTiALVSD ver. 2.00.00 AI-ECG model analyzing raw 12-lead ECG data to detect left ventricular systolic dysfunction

Outcome

Presence of LVSD, defined as an LVEF of 40% or less on echocardiography and measured using the biplane Simpson methodsurrogate

The AiTiALVSD AI-ECG model accurately detects left ventricular systolic dysfunction and stratifies long-term cardiovascular risk in patients with left bundle branch block, offering a scalable screening tool.

Main Result

Effect estimate: AUROC 0.930 (95% CI 0.924–0.937)

Limitations

Retrospective design
Relatively low specificity (47.3%) at sensitivity threshold
Study limited to Korean population
Possible bias due to inclusion of patients from development hospital despite exclusion from training data

Abstract

Abstract Background Left bundle branch block (LBBB) significantly increases the risk of left ventricular systolic dysfunction (LVSD) due to cardiac dyssynchrony. Although artificial intelligence–enabled electrocardiography (AI-ECG) models show promise in detecting LVSD, their performance in LBBB patients remains underexplored. We hypothesized that an AI-ECG model clinically validated for detecting LVSD would accurately detect LVSD and predict future clinical outcomes in LBBB patients. Methods In this retrospective multicenter study, 5,689 expert-validated LBBB ECGs collected from 2,813 patients between 2016 and 2024 were analyzed using a previously developed and validated AI-ECG model. LVSD was defined as an ejection fraction of ≤ 40%. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), sensitivity, and specificity. Patients were stratified into high- and low-risk groups based on a threshold that achieved 90% sensitivity. A Kaplan–Meier analysis was used to compare clinical outcomes. Results Among the 2,813 LBBB patients (mean age, 70.7 years; male sex, 43.7%), hypertension and a history of heart failure were common. The AiTiALVSD model showed strong diagnostic performance for LVSD (AUROC, 0.930 95% CI, 0.924–0.937; AUPRC, 0.913 95% CI, 0.902–0.923; sensitivity, 0.979; specificity, 0.473). During the mean follow-up of 4.1 years, high-risk patients had significantly higher hazards than low-risk patients for all-cause mortality (adjusted hazard ratio HR, 1.87; 95% CI, 1.53–2.28), implantable cardioverter defibrillator/cardiac resynchronization therapy implantation (adjusted HR, 15.2; 95% CI, 7.51–30.77), and cardiovascular hospitalization (adjusted HR, 1.11; 95% CI, 0.96–1.28). Conclusions AiTiALVSD effectively detects LVSD and stratifies long-term cardiovascular risk in LBBB patients, supporting its clinical utility for early detection and patient management.

Bookmark

View Full Paper