Eligibility for subcutaneous implantable cardiac defibrillator utilising artificial intelligence and deep learning methods for prolonged screening: where is the cut-off?

Elrefai, M., Abouelasaad, M. and Dunn, A. , et al. (2022) Eligibility for subcutaneous implantable cardiac defibrillator utilising artificial intelligence and deep learning methods for prolonged screening: where is the cut-off? EP Europace, 24 (Supplement_1), [euac053.447]. (doi:10.1093/europace/euac053.447).

Abstract

Funding acknowledgements: type of funding sources: Private grant(s) and/or Sponsorship. Main funding source(s): Main author is receiving an unrestricted grant by Boston Scientific

Background: S-ICD eligibility is determined by a single surface ECG analysis in which the suitability of an individual’s ECG vector morphology is assessed. A major predictor of eligibility is the T:R ratio. Current screening tools proposes T: R of 1:3 as a cut-off for eligibility. Inappropriate shocks due to T-wave oversensing (TWO) remains an issue despite screening. EFFORTLESS and PRAETORIAN trials reported inappropriate shock rates of 11.4% and 9.7% respectively, most frequently caused by cardiac oversensing.

Purpose: the cut-off T: R of 1:3 currently used incorporates a safety margin to accommodate for ECG signal amplitudes fluctuations without affecting S-ICD sensing. Prolonged screening using our tool accurately measures the T: R fluctuations. However, utilising a T: R of 1:3 for prolonged screening can unnecessarily exclude appropriate S-ICD candidates. The purpose of our study is to provide groundwork for future trials to find the optimal ratio that identifies patients at risk of TWO and inappropriate shocks while not excluding true S-ICD candidates after prolonged screening.

Methods: patients were fitted with 24-hour Holter monitors with leads placed to correspond to the vectors of an S-ICD. We used our tool to assess T: R over the recordings utilising Phase Space Reconstruction matrices - to convert the ECG signal into compressed pixel images. A Convolutional Neural Network (CNN) model was trained to accurately predict the T: R from these images resulting in a T: R variation plot for each vector. We then applied multiple T:R ratio cut-offs on the recordings to identify patients at risk of inappropriate shocks due to TWO at each proposed value. A vector with a T: R above the cut-off for 20 consecutive seconds was deemed to have failed screening, the time determined by the current detection, charge, and redetection time of the current S-ICD system. A patient has to have at least one suitable vector to pass the screening at the selected threshold.

Results: 37 patients (mean age 54.5 years,64.8% male) were included. 14 had Heart failure, 7 Hypertrophic cardiomyopathy, 7 normal hearts, 6 Adult congenital heart disease and 3 patients who received inappropriate S-ICD shocks due to TWO. Overall, 20 (54%) of patients passed prolonged screening using a 1:3 ratio. All of the patients passed screening with a T: R of 1:1. The only subgroup to wholly pass the screening for all the proposed ratios are the normal hearts group.

Conclusion: we propose adopting prolonged screening to select S-ICD eligible patients with low probability of TWO and inappropriate shocks. However, utilising T: R of 1:3 can unnecessarily exclude otherwise S-ICD eligible patients. The appropriate ratio likely lies between 1:3 - 1:1. Further studies are needed to identify the optimal screening thresholds, particularly in patients that have had inappropriate shocks due to TWO.

This record has no associated files available for download.

Contributors

Author: S. Coniglio

Author: A. Zemkoho

Download statistics