Prognostic value of artificial intelligence-derived echocardiographic measurements in ATTR cardiomyopathy

Introduction Transthyretin cardiomyopathy (ATTR-CM) is a progressive and potentially fatal disease, requiring accurate risk stratification. Echocardiography is the first-line imaging modality, and AI-based tools are increasingly applied for automated analysis. However, the prognostic value of these measurements remains unclear. Purpose To examine the prognostic value of AI-derived echocardiographic measurements and their incremental value over existing biomarker-based staging systems in patients with ATTR-CM. Material and Methods This retrospective cohort study included patients from two registries with ATTR-CM. The AI-based software US2AI quantified fully automated left ventricular ejection fraction (LVEF), global longitudinal strain (GLS) and tricuspid annular plane systolic excursion (TAPSE) from baseline echocardiogram. Prognostic performance was assessed using Kaplan–Meier analysis and Cox regression along with ROC curves for 1-year event prediction. Cutoffs stratifying abnormal values into moderate and severe were determined by log-rank maximization. An echo staging was developed by combining LV-GLS and TAPSE, stratifying patients into low (both normal), intermediate (one abnormal), and high risk (both abnormal). Results The cohort includes 347 patients (91% male, median age 78 years) with ATTR-CM. During follow-up of 2.4 years [IQR 1.1–4.1], 141 patients (41%) experienced the composite endpoint: 60 (17%) deaths and 81 (23%) heart failure hospitalizations. In univariate Cox regression, GLS (1.16 [1.10-1.22] per 1%, p<0.001), TAPSE (0.88 [0.84-0.92] per 1 mm, p<0.001) and LVEF (0.95 [0.94-0.97] per 1%, p<0.001) provided significant prognostic stratification. Kaplan-Meier analysis stratified by optimal cutoffs confirmed discrimination across three risk groups (overall p<0.001). Echo staging stratified risk with HRs of 4.14 and 7.81 (intermediate and high vs. low; p<0.001) and provided incremental value over age and NAC staging (χ² 87 vs. 65, p<0.001; C-index 0.790 vs. 0.759), respectively (Figure). One-year event prediction did not differ between AI and human measurements for TAPSE (AUC 0.64 vs. 0.58, p=0.21), GLS (AUC 0.64 vs. 0.62, p=0.65) and LVEF (AUC 0.66 vs. 0.69, p=0.40). Conclusion AI-derived echocardiographic measurements provide robust prognostic stratification in cardiac amyloidosis, with performance comparable to manual measurements, supporting their clinical utility for risk stratification.