Abstract 18602: A Multivariable Algorithm Using Transthoracic Echocardiographic Findings to Predict Reduced Ejection Fraction in the Single Systemic Right Ventricle

    loading  Checking for direct PDF access through Ovid


Introduction: Preservation of systolic function is of paramount importance in patients with single systemic right ventricles (RV). Ejection fraction (EF) measured by cardiac magnetic resonance imaging (cMRI) is considered the “gold standard” assessment of systolic function. Alternative transthoracic echocardiography (TTE) parameters have been proposed for assessing RV systolic function; however, no single TTE variable correlates optimally with cMRI EF. Our objectives were to determine which TTE parameters best reflect systemic RV EF and if a multivariable algorithm provided improved correlation.

Methods: Subjects with hypoplastic left heart syndrome (post stage 2 or 3 palliation) underwent cMRI and TTE (median interval, 1 day). Using machine learning techniques, multiple TTE variables were evaluated for their ability to discriminate cMRI EF of more or less than 40%. To determine the ability of combined TTE values to predict reduced EF by cMRI, a lasso logistic regression model with several potential TTE parameters was used. “Leave-one-out” cross-validation was applied to select the best multivariable model.

Results: A total of 54 cMRI/TTE pairs from 42 subjects were analyzed. 15 of 54 had EF < 40% by cMRI. Median age was 10.4 years (range, 8 months - 27.9 years). The optimal cross-validated lasso model included: biplane FAC, VVI circumferential strain, indexed apical RV systolic area, and RV MPI. Individual patient TTE measurements can be put into the model for prediction of cMRI EF ≤40%. The model uses these 4 values to predict whether the patient has cMRI EF ≤40%, with a correct classification rate (CR) of 93% (95% CI=87%-100%). A second cross-validated model was created to eliminate the requirement for VVI analysis; this model included biplane FAC, indexed apical RV systolic area, indexed short-axis RV systolic area, and RV MPI, with a CR of 91% (95% CI = 84%-98%).

Conclusions: Our multivariable models are able to improve upon the single variable prediction for the single, systemic RV and accurately differentiate between those with RV cMRI EF greater than or less than 40% with a CR of 91-93%. Once validated in an independent data set, these algorithms could be made available as an online tool for clinical use or potentially expanded to apply to other cohorts.

Related Topics

    loading  Loading Related Articles