Tracking the Preclinical Progression of Transthyretin Amyloid Cardiomyopathy Using Artificial Intelligence-Enabled Electrocardiography and Echocardiography

Abstract

Background and aims: The diagnosis of transthyretin amyloid cardiomyopathy (ATTR-CM) requires advanced imaging, precluding large-scale pre-clinical testing. Artificial intelligence (AI)-enabled transthoracic echocardiography (TTE) and electrocardiography (ECG) may provide a scalable strategy for pre-clinical monitoring.

Methods: This was a retrospective analysis of individuals referred for nuclear cardiac amyloid testing at Yale-New Haven Health System (YNHHS, internal cohort) and Houston Methodist Hospitals (HMH, external cohort). Deep learning models trained to discriminate ATTR-CM from age/sex-matched controls on TTE videos (AI-Echo) and ECG images (AI-ECG) were deployed to generate study-level ATTR-CM probabilities (0-100%). Longitudinal trends in AI-derived probabilities were examined using age/sex-adjusted linear mixed models, and their discrimination of future disease was evaluated across preclinical stages.

Results: Among 984 participants at YNHHS (median age 74 years, 44.3% female) and 806 at HMH (69 years, 34.5% female), 112 (11.4%) and 174 (21.6%) tested positive for ATTR-CM, respectively. Across cohorts and modalities, AI-derived ATTR-CM probabilities from 7,352 TTEs and 32,205 ECGs diverged as early as 3 years before diagnosis in cases versus controls (p _{time(x)group interaction}≤0.004). Among those with both AI-Echo and AI-ECG available one-to-three years before nuclear testing (n=433 [YNHHS] and 174 [HMH]), a double-negative screen at a 0.05 threshold (164 [37.9%] and 66 [37.9%], vs all else) had 90.9% and 85.7% sensitivity (specificity of 40.3% and 41.2%), whereas a double-positive screen (78 [18.0%] and 26 [14.9%], vs all else) had 85.5% and 88.9% specificity (sensitivity of 60.6% and 42.9%).

Conclusions: AI-enabled echocardiography and electrocardiography may enable scalable risk stratification of ATTR-CM during its pre-clinical course.

Keywords: artificial intelligence; cardiac amyloidosis; echocardiography; electrocardiography; screening; transthyretin.

Figure 1 |. Study Population.

Deep learning algorithms were trained to discriminate nuclear cardiac amyloid imaging-positive cases of ATTR-CM from age- and sex-matched controls using standard TTE videos or ECG images. These were subsequently deployed across independent sets of patients with longitudinal monitoring by TTE or ECG pre-dating their referral for nuclear cardiac amyloid testing. The overall objective was to examine the ability of the AI models to detect changes in TTE or ECG signatures that precede clinical disease and diagnosis. Such AI-enabled TTE or ECG signatures may be used to forecast the development of ATTR-CM, thus offering a standardized and scalable platform for longitudinal monitoring and screening in the community. AI: artificial intelligence; ATTR-CM: transthyretin amyloid cardiomyopathy; ECG: electrocardiography; TTE: transthoracic echocardiography.

Figure 2 |. Study Design.

AI-models were trained on transthoracic echocardiograms (TTE) and 12-lead electrocardiographic (ECG) images from patients with ATTR-CM (based on a positive nuclear cardiac amyloid test performed no more than 12 months after the index study) as well as age- and sex-matched controls without known disease across the Yale-New Haven Health System (YNHHS). Models were subsequently deployed across independent sets of patients in YNHHS, as well as an external set of patients from Houston Methodist Hospitals (HMH) who had sequential TTE or ECG performed in the years leading up to their referral for nuclear cardiac amyloid imaging. This design allowed us to evaluate the progression of AI-Echo or AI-ECG probabilities as non-invasive markers of pre-clinical ATTR-CM progression. AI: artificial intelligence; ATTR-CM: transthyretin amyloid cardiomyopathy; ECG: electrocardiography; HMH: Houston Methodist Hospitals; Tc^99m-PYP: pyrophosphate; TTE: transthoracic echocardiography; YNHHS: Yale-New Haven Health System.

Figure 3 |. Longitudinal patient-level changes in AI-Echo and AI-ECG ATTR-CM probabilities across cohorts based on nuclear cardiac amyloid positivity.

The panels illustrate the mean (with error bars denoting the 95% confidence interval of mean) of the AI-Echo and AI-ECG-derived probabilities across patients who went on to have a positive (blue color) vs negative (orange color) nuclear cardiac amyloid test across YNHHS (A, B) and HMH (C, D), respectively. The x axis denotes the time between the TTE/ECG and the timing of nuclear cardiac amyloid testing summarized across discrete time groups (negative time differences suggest that the TTE/ECG was performed before the nuclear cardiology exam). The brackets below each period along the x axis denote the number of unique positive and negative individuals in each period. If more than one study was available in a given period, we derived the median of all predictions in that timeframe. ATTR-CM: transthyretin amyloid cardiomyopathy; ECG: electrocardiography; HMH: Houston Methodist Hospitals; PYP: Tc^99m-pyrophosphate; TTE: transthoracic echocardiography; YNHHS: Yale-New Haven Health System.

Figure 4 |. Sensitivity and specificity of AI-Echo and AI-ECG for ATTR-CM across preclinical stages.

Each plot represents the sensitivity and specificity at a given threshold (0.015: blue, 0.05: orange, 0.1: green, 0.25: red, 0.5: purple) and a given time point (circle: 5 to 3 years; rectangle: 3 to 1 year(s) before; and triangle: year before nuclear cardiac amyloid testing across YNHHS (A-C) and HMH (D-F). The panels depict the evolution in predictions for AI-Echo (A&C), and a AI-ECG (B&D). The error bars in either direction denote the 95% confidence intervals for sensitivity (x axis), or specificity (y axis) derived from bootstrapping with 1000 replications. ATTR-CM: transthyretin amyloid cardiomyopathy; ECG: electrocardiography; HMH: Houston Methodist Hospitals; TTE: transthoracic echocardiography; YNHHS: Yale-New Haven Health System.

Figure 5 |. Discrimination performance of joint AI-Echo and AI-ECG testing for future ATTR-CM across pre-clinical stages.

We present the observed counts, and discrimination (sensitivity, specificity, positive [PPV] and negative predictive value [NPV]) for joint testing by both AI-Echo and AI-ECG across thresholds and timepoints. We present two operating points, a sensitive one where we compare double-negatives versus everyone else, and a more specific one where we compare double-positives against everyone else. We also present 95% confidence intervals derived from bootstrapping with 1000 replications. AI: artificial intelligence; ATTR-CM: transthyretin amyloid cardiomyopathy; ECG: electrocardiography; HMH: Houston Methodist Hospitals; TTE: transthoracic echocardiography; YNHHS: Yale-New Haven Health System.