Prediction of Stroke Outcome After Endovascular Thrombectomy From Clinical Characteristics: A Machine Learning Approach

BACKGROUND: Predicting outcomes after endovascular thrombectomy (EVT) for acute ischemic stroke is challenging. This study aimed to develop a machine learning model to predict outcomes after EVT using only clinical characteristics available shortly after hospital arrival, excluding imaging, laboratory, and angiographic data. METHODS: We analyzed data from the JSDB (Japan Stroke Data Bank), a multicenter, prospective registry of patients with acute stroke. A total of 4353 patients who underwent EVT between 2016 and 2020 were included, with those from 2016 to 2019 assigned to the derivation cohort and those from 2020 to the validation cohort. The model was developed using 132 routinely collected clinical variables that can be rapidly obtained on patient arrival. Feature selection was performed using the Boruta algorithm, and the model was built using CatBoost. The primary outcome was a modified Rankin Scale (mRS) score of 0 to 2 at discharge. RESULTS: Among 4353 patients treated with EVT, the derivation cohort included 2764 patients and the validation cohort, 1589 patients. In these cohorts, 973 (35.2%) and 547 (34.4%) achieved mRS 0 to 2, respectively. Boruta selected 47 predictors. CatBoost achieved an area under the receiver operating characteristic curve of 0.744 (95% CI, 0.710-0.776) in the derivation test set and 0.781 (95% CI, 0.755-0.803) in the validation cohort. Validation areas under the receiver operating characteristic curves were 0.792 (95% CI, 0.765-0.815) for mRS 0 to 1, 0.766 (95% CI, 0.744-0.787) for mRS 0 to 3, and 0.680 (95% CI, 0.631-0.729) for mRS 6. CONCLUSIONS: Our machine learning model predicted outcomes following EVT using only admission clinical data. This imaging-independent machine learning approach may support early prognostication and facilitate discussions of expected clinical courses.

Authors

Institutions

Publication Details

Journal
Journal of the American Heart Association
Published
2026-11-12
DOI
https://doi.org/10.1161/jaha.125.047479
Primary Topic
Acute Ischemic Stroke Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Prediction of Stroke Outcome After Endovascular Thrombectomy From Clinical Characteristics: A Machine Learning Approach

Sohei Yoshimura, Ḱazunori Toyoda, Junji Miyazaki, Ken Johkura et al.
Journal of the American Heart Association
Acute Ischemic Stroke Management
article

Prediction of Stroke Outcome After Endovascular Thrombectomy From Clinical Characteristics: A Machine Learning Approach

Sohei Yoshimura, Ḱazunori Toyoda, Junji Miyazaki, Ken Johkura, Masatoshi Koga, Nobuyuki Sakai, Kazuo Minematsu, Junpei Koge, Tomohide Yoshie, Yoshihiro Miyamoto, Koji Iihara, Shinichi Wada, Toshiharu Yanagisawa, Kaori Miwa, Toshiichi Watanabe, Makino Sakuraba, Kazuo Kitazawa, The Japan Stroke Data Bank Investigators
article en

Abstract

BACKGROUND: Predicting outcomes after endovascular thrombectomy (EVT) for acute ischemic stroke is challenging. This study aimed to develop a machine learning model to predict outcomes after EVT using only clinical characteristics available shortly after hospital arrival, excluding imaging, laboratory, and angiographic data. METHODS: We analyzed data from the JSDB (Japan Stroke Data Bank), a multicenter, prospective registry of patients with acute stroke. A total of 4353 patients who underwent EVT between 2016 and 2020 were included, with those from 2016 to 2019 assigned to the derivation cohort and those from 2020 to the validation cohort. The model was developed using 132 routinely collected clinical variables that can be rapidly obtained on patient arrival. Feature selection was performed using the Boruta algorithm, and the model was built using CatBoost. The primary outcome was a modified Rankin Scale (mRS) score of 0 to 2 at discharge. RESULTS: Among 4353 patients treated with EVT, the derivation cohort included 2764 patients and the validation cohort, 1589 patients. In these cohorts, 973 (35.2%) and 547 (34.4%) achieved mRS 0 to 2, respectively. Boruta selected 47 predictors. CatBoost achieved an area under the receiver operating characteristic curve of 0.744 (95% CI, 0.710-0.776) in the derivation test set and 0.781 (95% CI, 0.755-0.803) in the validation cohort. Validation areas under the receiver operating characteristic curves were 0.792 (95% CI, 0.765-0.815) for mRS 0 to 1, 0.766 (95% CI, 0.744-0.787) for mRS 0 to 3, and 0.680 (95% CI, 0.631-0.729) for mRS 6. CONCLUSIONS: Our machine learning model predicted outcomes following EVT using only admission clinical data. This imaging-independent machine learning approach may support early prognostication and facilitate discussions of expected clinical courses.

Journal of the American Heart Association
National Cerebral and Cardiovascular Center (JP), SoftBank Group (Japan) (JP), Nakamura Memorial Hospital (JP), Aizawa Hospital (JP), Omagari Kousei Medical Center (JP), Yokohama Brain and Spine Center (JP), SoftBank Corp. (Japan) (JP), The University of Osaka (JP)
Good health and well-being
Acute Ischemic Stroke Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon