Prospective use of artificial intelligence–enabled electrocardiogram to assist in clinical decision making regarding defibrillator therapy

Key Teaching Points.

• •Artificial intelligence electrocardiogram (AI ECG) may help predict a cardiac resynchronization therapy (CRT) super-response.
• •Left bundle branch area pacing may help reverse cardiomyopathy related to left bundle branch block.
• •AI ECG may assist in decision making regarding the defibrillator component during CRT.

Introduction

Conduction system pacing, especially left bundle branch area pacing (LBBAP), is rapidly emerging as a promising method to deliver cardiac resynchronization therapy (CRT) to prevent or reverse dyssynchrony in patients with cardiomyopathy.^1,2 Multiple studies have demonstrated that LBBAP provides remarkable lead stability, excellent sensing, and low pacing threshold compared with conventional coronary sinus biventricular pacing.^3,4 A subset of patients who undergo CRT are super-responders. These patients typically experience substantial improvement in left ventricular systolic function in addition to clinical improvement in heart failure symptoms.^5,6

Most patients who are candidates for CRT may also qualify for implantable cardioverter-defibrillator (ICD) therapy for primary prevention because of low ejection fraction (EF) and heart failure. However, in CRT super-responders, the EF typically recovers fully or nearly fully, raising questions about the net benefit of ICD implantation at the time of CRT. The addition of a defibrillator component increases procedural complexity and long-term risks, while the actual risk of sudden cardiac death may be low, as suggested by recent studies.^6,7

Recently, using the Mayo Clinic artificial intelligence electrocardiogram (AI ECG) dashboard, we demonstrated a positive correlation between AI ECG–predicted low EF probability (AILEF) and the actual EF change after LBBAP, with a decrease in AILEF by ≥60% associated with the greatest EF increase.^8,9 In this study, we describe the prospective use of this AI tool to facilitate a successful shared clinical decision-making process.

Case report

The patient is a 69-year-old man with a history of essential hypertension, type 2 diabetes, hyperlipidemia, and frequent premature ventricular contractions (PVCs). He underwent PVC ablation 4 years before the evaluation for CRT. Ablation was successful in eliminating a PVC focus arising from the left coronary cusp area. Postablation, the patient was noted to have left bundle branch block on the ECG, which was absent at baseline. The patient returned to the cardiology clinic because of increasing heart failure symptoms, including fatigue and dyspnea on exertion, over the course of a year. He was found to have left ventricular systolic dysfunction, with echocardiography showing dilated cardiomyopathy and an EF of 33%. Coronary angiography was performed, which showed moderate, nonobstructive coronary artery disease. His heart failure and EF did not improve after >6 months of guideline-directed medical therapy, and he was referred for CRT. Because of his desire to continue using power tools such as chain saw, the patient wishes to avoid the defibrillator component, if possible. After reviewing the pros and cons of CRT-pacer (CRT-P) vs CRT-defibrillator (CRT-D) and discussing a recent Mayo Clinic study suggesting that AI ECG may predict a CRT response, with a large drop in AILEF indicating a possible CRT super-response, the patient and his family elected to proceed with CRT-P only, without the defibrillator component, if the AI ECG showed features predicting a CRT super-response. At baseline, his AI ECG showed an AILEF probability of 93% (Figure 1A).

Figure 1.

Twelve-lead ECG with an AI-predicted low EF probability before the LBBAP procedure (A), during the LBBAP procedure (B), and at 3-month follow-up (C). The actual EF was 33% at baseline (panel A) and 59% at 3-month follow-up (panel C). AI = artificial intelligence; ECG = electrocardiogram; EF = ejection fraction; LBBAP = left bundle branch area pacing.

The patient was brought to the electrophysiology laboratory and underwent standard left precordial preparation and draping. Venous access was obtained by left axillary vein puncture guided by fluoroscopy. The LBBAP lead was then implanted using the same technique as previously reported.¹⁰ Briefly, a C315HIS delivery sheath (Medtronic) was advanced into the right ventricle over a standard J wire. The 3830 lead (Medtronic) was then advanced through the delivery sheath into the right ventricular septum ∼1.5–2 cm distal to the His bundle region. Once the lead was pointed perpendicular to the septum, manual rotations were applied with forward pressure. Pacing from the 3830 lead was then performed in a unipolar fashion after every 1–2 rotations until the paced morphology demonstrated a narrow right bundle branch block pattern, indicating capture of the left-sided conduction system. At the target position, the R-wave peak time in lead V₆ was 80 ms, with a QRS duration of 105 ms (measured manually from the beginning to the end of the QRS complex). The pacing threshold was <0.75 V at 0.5 ms. An intraoperative AI ECG was performed during LBBAP immediately after implantation of the LBBAP lead. This demonstrated a large drop in AILEF to 3.7% (Figure 1B). Decisions were then made to proceed with CRT-P through LBBAP using a dual-chamber device instead of CRT-D. Postoperative chest radiography demonstrated a satisfactory lead position (Figure 2). The total procedure time from lidocaine injection to pocket closure was 59 minutes, with a fluoroscopy time of 5.7 minutes.

Figure 2.

Postoperative chest radiograph and intraoperative fluoroscopic images in AP (A), RAO (B) and LAO (C) views, demonstrating lead position of an LBBAP dual-chamber pacemaker. AP = anteroposterior; LAO = left anterior oblique; LBBAP = left bundle branch area pacing; RAO = right anterior oblique.

At 3-month postoperative follow-up, the patient’s heart failure symptoms greatly improved after LBBAP. His New York Heart Association class improved by about 2 classes, and his EF normalized to 59%. AI ECG at that time showed a very low probability of low EF at 1.8% (Figure 1C).

Discussion

The case presented here demonstrates that the AI-ECG tool may be invaluable in identifying patients who are super-responders to CRT using LBBAP and therefore potential candidates for CRT-P alone instead of CRT-D. Avoiding the defibrillator component not only has lifestyle implications, as in this patient, who desires to continue frequent use of a chain saw and other power tools, but also eliminates potential procedural risks and long-term complications associated with ICD therapy.

A CRT super-response is well-documented in the literature, with contributing factors including the underlying etiology (ischemic vs nonischemic), the morphology of left bundle branch block (typical vs atypical), and the duration of heart failure.⁵ These patients typically experience a substantial improvement in left ventricular function, including EF. In some cases, the EF may fully normalize, significantly lowering the risk of sudden cardiac death. In a registry study, super-responders (left ventricular EF ≥ 50%) had an annual cardiac death rate and an appropriate CRT-D intervention rate of zero over a median follow-up of 2.8 years, suggesting that the risk of significant ventricular tachyarrhythmia in super-responders is effectively eliminated by the normalization of left ventricular EF due to CRT.⁷ Another study suggests that normalization of left ventricular EF after CRT also normalizes survival.⁶ As such, the net benefit of CRT-D in super-responders, especially those with normalized EF, remains uncertain and may not justify the procedural and long-term risks associated with ICD therapy. Compared with initial CRT-D therapy, performing CRT-P alone using LBBAP has several advantages, including fewer vascular access, shorter procedure time (which typically leads to improved patient comfort), fewer leads in the vasculature, smaller device volume, and the ability to eliminate inappropriate shocks. Additionally, if follow-up EF assessment does not show a CRT super-response or normalization of EF, upgrading the CRT-P system to a CRT-D device can still be an option.

The AI-enabled ECG algorithm at Mayo Clinic has been shown to predict both the onset of low EF and the EF response to CRT through LBBAP.^8,9,11,12 Recent data from our group have demonstrated a >90% specificity of AI ECG in predicting a CRT super-response when using a 60% drop in AILEF as a cutoff (unpublished observations). This finding, together with other favorable conditions predicting a CRT super-response in this patient, such as nonischemic cardiomyopathy, a typical left bundle branch block pattern, a relatively short duration of cardiomyopathy, and the patient’s strong wish to avoid defibrillator therapy, all contributed to the shared decision making to proceed with CRT-P rather than CRT-D implantation. A larger prospective study is required to determine whether such an AI tool can be used to help determine whether the defibrillator component is beneficial in all patients undergoing CRT.

Disclosures

The authors have no conflicts of interest to disclose.