# Efficacy of AI Models in Detecting Heart Failure Using ECG Data: A Systematic Review and Meta-Analysis

## TL;DR

This study reviews how well AI models can detect heart failure using ECG data, finding them generally accurate but needing more testing for reliability.

## Contribution

A systematic review and meta-analysis of AI model performance in detecting heart failure using ECG data.

## Key findings

- AI models showed high pooled sensitivity (0.93) and specificity (0.95) in detecting heart failure from ECG data.
- Significant variability in model performance was observed across studies, with sensitivity ranging from 0.12 to 1.00.
- AI-based ECG analysis is promising for HF screening but requires further validation in diverse populations.

## Abstract

Heart failure (HF) is the most common cause of death worldwide, characterized by low ejection fraction, substantial mortality, morbidity, and poor quality of life. Recent advancements in artificial intelligence (AI) present a promising avenue for enhancing diagnostic precision, particularly in the analysis of electrocardiogram (ECG) data. This systematic review and meta-analysis aim to synthesize current evidence on the diagnostic performance of AI models in detecting HF using ECG data. PubMed and Google Scholar databases were systematically searched from inception up to July 1, 2023, to identify original articles assessing the predictive ability of AI in HF diagnosis. A total of 218,202 participants were included, with individual studies ranging from 59 to 110,000 participants. The pooled sensitivity, specificity, and diagnostic odds ratio (DOR) for the 13 included studies, with a 97.5% confidence interval (CI), were 0.93 (CI: 0.81-0.98), 0.95 (CI: 0.89-0.97), and 303.65 (CI: 53.12-1734), respectively. The sensitivity and specificity ranged from 0.12 to 1.00 and 0.66 to 1.00, respectively, indicating substantial variability in AI model performance, which may impact their generalizability and clinical reliability. AI-based algorithms utilizing ECG data are a reliable, accurate, and promising tool for the screening, detection, and monitoring of HF. However, further prospective studies are needed, particularly randomized controlled trials and large-scale longitudinal studies across diverse populations, to evaluate the long-term clinical impact, generalizability, and real-world applicability of these AI-driven diagnostic tools.