Abstract WP203: Prediction of Outcome After Endovascular Therapy for Acute Ischemic Stroke: A Machine Learning Approach From Japan Stroke Data Bank

Abstract

Introduction: Previous studies have demonstrated the predictive capabilities of machine learning (ML) models using clinical, imaging, and angiographic characteristics in determining the clinical outcome of patients undergoing endovascular treatment for acute ischemic stroke. We aimed to analyze a large-scale stroke registry and develop an ML model that solely utilizes clinical characteristics, without the need for imaging or angiographic data, to predict the outcome of patients treated with endovascular therapy for acute ischemic stroke. Methods: We conducted our analysis using data from the Japan Stroke Data Bank, a nationwide acute stroke registry containing individual patient data. A total of 4353 patients who underwent endovascular treatment for acute ischemic stroke between January 2016 and December 2020 were included in the study. Patient data from 2016 to 2019 were used as the derivation cohort, while data from 2020 were used as the validation cohort. An ML models were trained on 70% of the patients randomly selected from the derivation cohort and tested on the remaining 30%. The input variables for the ML models consisted of 138 clinical characteristics including neurological examination findings. Imaging, angiographic, and laboratory data were not utilized in the ML process. The primary outcome measure was good outcome at discharge, defined as a modified Rankin Scale score of 0-2. Results: Of the 4353 patients included in the study, 1520 patients (34.9%) achieved good outcome. The area under the curve for predicting good outcome was 0.76 for the test set in the derivation cohort and 0.78 for the validation cohort. The SHAP summary plot revealed that the top five major predictive contributors for good outcomes were pre-stroke modified Rankin Scale score, level of consciousness, age, stroke classification, and NIHSS score. Conclusions: Our findings demonstrate that an ML model utilizing solely clinical characteristics, without the inclusion of imaging or angiographic data, can accurately predict good outcomes at discharge for acute stroke patients treated with endovascular therapy. This approach has the potential to provide an aid in clinical decision-making before imaging examination.