Safety, Efficiency, and Accuracy of Pre-procedural Predictive Modeling Algorithm for Left Atrial Appendage Closure

Kirollos Gabrah, DO – RESIDENT PHYSICIAN, INTERNAL MEDICINE, ST BERNARDS HEART AND VASCULAR; brandon Doty, OMS – Medical Student, NYITCOM; Arun Mahtani, MD – RESIDENT PHYSICIAN, INTERNAL MEDICINE, NYU; johnathan Cain, DO – RESIDENT PHYSICIAN, ST BERNARDS HEART AND VASCULAR; Bridget lee, DO – RESIDENT PHYSICIAN, INTERNAL MEDICINE, ST BERNARDS HEART AND VASCULAR; Jamarcus Brider, DO – RESIDENT PHYSICIAN, INTERNAL MEDICINE, ST BERNARDS HEART AND VASCULAR

Purpose: Despite technological advances in left atrial appendage closure (LAAC), anatomical variations challenge appropriate device sizing and position. FEops HEARTguide (FH) is a CT pre-planning software that uses an artificial intelligence (AI) algorithm to analyze CTs to predict optimal LAAC device size, position, and apposition. We aim to assess the clinical utility of FH software in LAAC.

Material and Methods: 200 patients who underwent LAAC with FH software preplanning were analyzed. This Data was compared to the historical standard CT preplanning data obtained from the PREDICT-LAA trial. Correlation of predicted pre-procedural and observed intra-procedural LAA measurements (LAAM) and percent of device size prediction were analyzed.

Results: 58.5% female (mean age of 74±9 years) underwent LAAC with pre-planning using the software. The average procedure time was 20.7±12.3 min compared to 55.2±24.7 min (95% CI: -38.71 to -30.09, p-value < 0.0001) in the control group. Average fluoroscopy time was 4.28±2.99 min compared to 17.6±11.4 min (95% CI: -15.04 to -11.58, p-value < 0.0001). Average contrast used during the procedure was 20.45±8.97 mL compared to 80±65 mL (95% CI: -69.09 to -50.23, p-value < 0.0001).
In 96% cases, a single LAAC device was successfully implanted with a single transseptal in all patients guided by the software. Overall correlation between predicted and observed LAAM was 0.89 (95% CI: 0.86-0.92, p-value = 3.75 e-14) using Pearson’s correlation testing. Bland-Altman analysis showed agreement between software predicted and procedural LAAM (p-value = 0.93). In 95%, the software predicted the correct device.

Conclusions: This first US comprehensive single center experience demonstrates the clinical utility of FEops HEARTguide software for LAAC preplanning with statistically significant shorter procedure and fluoroscopy times, and lower contrast use. Predictive modeling using AI software also accurately predicted the LAAM and device size.