Background: Diagnosis of cardiac amyloidosis (CA) is often delayed due to variability in clinical presentation. The electrocardiogram (ECG) is one of the most common and widely available tools for assessing cardiovascular diseases. Artificial intelligence (AI) models analyzing ECG have recently been developed to detect CA, but their pooled accuracy is yet to be evaluated. Aim: To meta-analyze the accuracy of AI-enhanced ECG in diagnosing CA. Methods: We searched the Scopus, MEDLINE, and Cochrane CENTRAL databases up until April 2024 for studies assessing AI-enhanced ECG diagnosis of CA. Studies reporting findings from derivation and validation cohorts were included. Studies combining other diagnostic modalities such as echocardiography were excluded. The outcome of interest was the area under the receiver operating characteristic curve (AUC) for overall CA and subtypes transthyretin amyloidosis (ATTR) and light chain amyloidosis (AL). Analysis was done using RevMan 5.4.1 general inverse variance random effects model, pooling data for AUC and 95% confidence intervals (CI). Results: 5 studies comprising 7 cohorts met the eligibility criteria. The total derivation and validation cohorts were 8,639 and 3,843 respectively. The AUC were 0.89 (95% CI, 0.86-0.91) for cardiac amyloidosis, 0.90 (95% CI, 0.86-0.95) for ATTR amyloidosis and 0.80 (95% CI, 0.80-0.93) for AL amyloidosis. The forest plots can be found in the Figure 1. Conclusion: AI-enhanced ECG models effectively detect CA and may provide a useful tool for the early detection and intervention of this disease.
Read full abstract