P900 The Heart Model: Automated LVEF calculation in the real world
Abstract Background Heart Model (HM) is a proprietary, model-based algorithm for measurement of left ventricular ejection fraction (LVEF) in a 3D dataset acquired from an apical 4-chamber view by transthoracic echocardiography (TTE). There is evidence for superior reproducibility compared to 2D echo methods for LVEF measurement. Objective To assess the correlation of LVEF by HM with conventional, 2D LVEF methods. Methods All TTEs performed between 04-08/02/2019 by 2 HM-trained sonographers were included. Demographic characteristics, indication for TTE, LVEF by Simpson"s (LVEF_S), by "eyeballing" (LVEF_EB) and by HM (LVEF_HM), were recorded; LVEF for each study was also estimated by eye-balling by an experienced observer unaware of the reported LVEF (LVEF_IND). We compared LVEF by each method, their reciprocal correlations and their correlation with LVEF_HM. Image quality was rated excellent (endocardial border visible for all segments in the 3 apical views), good (< 1 segment was not visible / view), adequate (< 3 segments were not visible) and limited (<4 segments were not visible). Indications for TTE were: assessment of LVEF in 1/3 of the studies, murmurs in 1/5, and other indications in the rest. Results We included 74 patients (42 M, mean age (SD) 69.8(13.9), range 18-92 years). Forty-nine (66%) patients were in sinus rhythm, 23 (31%) were in AF, and the rest were in various paced rhythms. Fifty patients (68%) had excellent, good or adequate images. The EFs calculated by different methods are shown in the Table (p > 0.05 for all), an the Bland Altman plot (LVEF_EB vs HM) in the figure. LVEF_HM correlated modestly with the other methods if all studies were included (r = 0.535 LVEF_HM vs. LVEF_EB); the correlation improved if only good-quality studies were included (r = 0.769, p < 0.001 for both). All combinations of LVEF_IND, LVEF_EB and LVEF_S had correlation coefficients >0.93. Conclusions The Heart Model algorithm for LVEF measurement correlates well with traditional 2D methods in patients with good endocardial border definition, where its use can potentially improve reproducibility and reduce exam duration. LVEF by method (good-quality studies) N = 50 Simpson"s Eyeballing HeartModel Independent Mean(%) 48.6 48.9 50.9 48.8 SD(%) 17.8 16.6 14.9 15.7 Median(%) 54.5 52.5 53 50 Range(%) 16-74 17.5-72.5 19-88 15-75 SD - standard deviation Independent - LVEF estimate (eyeballing) by independent observer unaware of reported EFs. LVEF_EB was chosen as it was available inall the reports. Abstract P900 Figure. Bland Altman Plot (LVEF_EB vs LVEF_HM)