Cerebrovascular Events after Transcatheter Aortic Valve Replacement: The Difficulty in Predicting the Unpredictable

Abstract

Background: Cerebrovascular events (CVE) are feared complications following transcatheter aortic valve replacement (TAVR). We aimed to develop a new risk model for CVE prediction with the application of multimodal imaging. Methods: From May 2011 to August 2019, a total of 2015 patients underwent TAVR at our institution. The study cohort was subdivided into a derivation cohort (n = 1365) and a validation cohort (n = 650) for risk model development. Results: Of 2015 patients, 72 (3.6%) developed TAVR-related CVE. Pre-procedural factors of our risk model were history of prior CVE, a larger aortic valve area (≥0.55 cm2), a large aortic angulation (≥48.5°), and enhanced calcification of the right coronary cusp (≥447.2 AU), left ventricular outflow tract (≥262.4 AU), and ascending thoracic aorta (≥116.4 AU). Our risk model was superior for in-hospital CVE prediction following TAVR in the establishment cohort (AUC 0.73, 95% CI 0.66–0.80; p < 0.001) compared to other risk scores, such as the EuroSCORE II or the CHA2DS2-VASc score. Conclusions: Although CVE prediction in patients undergoing TAVR is challenging due to the complex nature of the TAVR procedure, our study highlights that multimodal imaging is a promising approach to generate a more accurate risk model for CVE prediction.

How AI and Robotics Will Advance Interventional Radiology: A Narrative Review and Future Perspectives

“…The DNN approach used in this model can address the limitations of traditional single-variable and multi-variable regression techniques for predicting serious but rare complications like CVEs to some extent. Similar studies, such as the one by Oliver Maier, [78] also indicate that DL methods can generate more accurate risk models for predicting CVE. Furthermore, high-degree atrioventricular block is also one of the common complications after TAVR, and AI also a high predictive value in the PPI caused by it.…”

“…Traditional statistical methods like logistic regression that incorporate a few independent risk factors, such as atherosclerosis, atrial fibrillation, balloon dilation, and device dislocation, cannot fully explain the occurrence of CVEs. [78] Taishi Okuno et al proposed using DL to develop a predictive model for CVEs after TAVR, [79] selecting a rare event autoencoder as the predictive model, with an AUC of 0.79 (0.65-0.93). The DNN approach used in this model can address the limitations of traditional single-variable and multi-variable regression techniques for predicting serious but rare complications like CVEs to some extent.…”