Abstract 4145142: Enhancing Inpatient Mortality Prediction in TAVR Patients: A Modified CHA2DS2-VASc Approach Using Machine Learning

Introduction: Inpatient mortality remains a significant concern for patients who have undergone transcatheter aortic valve replacement (TAVR) procedures. Accurate risk prediction models are vital for optimizing patient care and outcomes in this population. The CHA2DS2-VASc score has been utilized to predict various inpatient and long-term outcomes, irrespective of patient history of atrial fibrillation. This abstract introduces a novel approach to enhance inpatient mortality prediction in TAVR patients by adapting the CHA2DS2-VASc score using Lasso regression. Methods: We used the National Inpatient Sample Database to include all patients who underwent TAVR from 2018 to 2020. All patients with and without any history of atrial fibrillation or atrial flutter were included in the analysis. We employed Lasso regression, to modify the CHA2DS2-VASc score for predicting stroke risk in patients who underwent TAVR. The data was split into a 70:30 training-testing set. The model’s performance was evaluated using the Area Under the Receiver Operating Characteristic (AUROC) curve and compared to CHADS2-VASc. Results: The study included a total population of 28,684 patients in the training set and 12,290 patients in the testing set. The median age of the entire population was 80 years (IQR: 73-85), and females accounted for 44% of the total population. The mean CHA2DS2-VASc for the entire population was 3.82 (CI: 3.81-3.84) The AUROC values were 0.8239 for the training dataset and 0.8203 for the testing dataset, indicating strong discriminative ability in both cases. This performance was significantly improved compared to the CHA2DS2-VASc score (AUROC of 0.654). Conclusion: The modified CHA2DS2-VASc score improves inpatient mortality prediction for TAVR patients. With strong discrimination (AUROC 0.8239/0.8203) compared to the CHA2DS2-VASc score (AUROC 0.654), it holds promise for tailored risk assessment and improved patient care. Prospective validation is needed and comparison with the more widely used STS score is needed to establish real-world applicability.