Assessment of HeartModel® Automated Left Ventricular Ejection Fraction for Patients with Hypertrophic Cardiomyopathy
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Cardiac myosin inhibitors (CMIs) have revolutionized care for patients with obstructive hypertrophic cardiomyopathy (HCM), however they are associated with a risk of systolic dysfunction requiring longitudinal echocardiographic monitoring. Machine learning (ML) algorithms applied to echocardiography might expand access to frequent accurate assessment of left ventricular ejection fraction (LVEF). Existing algorithms have not been tested on patients with HCM. We assess the performance of a commercial ML-based LVEF model for patients with HCM.
Methods
Single center prospective study of transthoracic echocardiography (TTE) measurements of left ventricular function by Philips HeartModel® (automated) assessment and echocardiographer assessment (standard) for patients with HCM. Assessments of LVEF, end diastolic volume (EDV) and end systolic volume (ESV) were studied across methods and modalities. Median percent differences between measurement methods and correlation coefficients were calculated. Clinical decisions about CMI dosing strategies were modeled around LVEF < 50%.
Results
50 patients with HCM were included. Median age 64 years; 64% were male. Median automated LVEF was lower than standard assessment (55.5% (IQR 9) vs 62.5% (IQR 10), p 0.002, median difference – 8% (IQR 14)). Automated assessment traced larger EDVs and ESVs compared to standard 2D tracings (141 ml (IQR 66) vs 114 ml (IQR 55), p 0.001, median difference + 20% (IQR 35), and 64 ml (IQR 35) vs. 41 ml (IQR 25), p < 0.001, median difference + 43% (IQR 58)). Correlation between automated and standard assessment of LVEF was modest (R^2 = 0.24). Automated assessment identified 11 (22%) patients as having LVEF < 50% vs. 6 (12%) patients identified using standard imaging assessment.
Conclusions
For patients with HCM, automated assessments of LV size and function differ significantly from standard assessments, raising concerns about the use of this ML-enabled LVEF software for this patient population and potential application to guiding CMI treatment decisions.
