CERC2

CERC2 predicts 30-day cerebrovascular event (CVE) risk after transcatheter aortic valve replacement (TAVR) by integrating clinical and imaging data with a deep learning autoencoder for prognostic modeling.

Key Features:

• Data source and size: Prospective TAVR registry (2007–2019) including 2,279 patients with 1,492 complete datasets.
• CVE definitions: Distinguishes acute (<24 hours) and subacute (day 2–30) CVEs defined to include strokes and transient ischemic attacks (TIAs).
• Feature dimensionality: Considers over 107 clinical and imaging characteristics as candidate variables.
• Feature selection: Applies recursive feature reduction to select relevant variables from the candidate set.
• Model architecture: Trains an autoencoder-based predictive model using the selected features.
• Performance metric: Reports discrimination with area under the curve (AUC) = 0.79 (95% CI: 0.65–0.93) for 30-day CVE prediction.
• Prognostic association: Identifies CVEs as associated with increased mortality (hazard ratio 2.62, 95% CI: 1.82–3.78).

Scientific Applications:

• 30-day CVE risk estimation: Predicts individual risk of acute and subacute cerebrovascular events following TAVR.
• Risk stratification: Stratifies TAVR patients based on integrated clinical and imaging predictors for stroke and TIA events.
• Prognostic evaluation: Assesses the association between predicted CVE risk and increased post-TAVR mortality.

Methodology:

Analyzed integrated clinical and imaging data from a prospective TAVR registry (2007–2019) with 2,279 patients (1,492 complete); applied recursive feature reduction on >107 variables; trained an autoencoder-based predictive model and evaluated discrimination using AUC (0.79, 95% CI: 0.65–0.93).