Predictive models for secondary epilepsy within 1 year in patients with acute ischemic stroke: a multicenter retrospective study DOI Creative Commons
Jinxin Liu,

Haoyue He,

Yanglingxi Wang

и другие.

eLife, Год журнала: 2024, Номер 13

Опубликована: Июль 8, 2024

Background: Post-stroke epilepsy (PSE) is a critical complication that worsens both prognosis and quality of life in patients with ischemic stroke. An interpretable machine learning model was developed to predict PSE using medical records from four hospitals Chongqing. Methods: Medical records, imaging reports, laboratory test results 21,459 stroke were collected analyzed. Univariable multivariable statistical analyses identified key predictive factors. The dataset split into 70% training set 30% testing set. To address the class imbalance, Synthetic Minority Oversampling Technique combined Edited Nearest Neighbors employed. Nine widely used algorithms evaluated relevant prediction metrics, SHAP (SHapley Additive exPlanations) interpret assess contributions different features. Results: Regression revealed complications such as hydrocephalus, cerebral hernia, deep vein thrombosis, well specific brain regions (frontal, parietal, temporal lobes), significantly contributed PSE. Factors age, gender, NIH Stroke Scale (NIHSS) scores, like WBC count D-dimer levels associated increased risk. Tree-based methods Random Forest, XGBoost, LightGBM showed strong performance, achieving an AUC 0.99. Conclusions: accurately predicts risk, tree-based models demonstrating superior performance. NIHSS score, count, most crucial predictors. Funding: research funded by Central University basic young teachers students ability promotion sub-projec t(2023CDJYGRH-ZD06), Emergency Medicine Chongqing Key Laboratory Talent Innovation development joint fund project (2024RCCX10).

Язык: Английский

Predictive models for secondary epilepsy within 1 year in patients with acute ischemic stroke: a multicenter retrospective study DOI Creative Commons
Jinxin Liu,

Haoyue He,

Yanglingxi Wang

и другие.

eLife, Год журнала: 2024, Номер 13

Опубликована: Июль 8, 2024

Background: Post-stroke epilepsy (PSE) is a critical complication that worsens both prognosis and quality of life in patients with ischemic stroke. An interpretable machine learning model was developed to predict PSE using medical records from four hospitals Chongqing. Methods: Medical records, imaging reports, laboratory test results 21,459 stroke were collected analyzed. Univariable multivariable statistical analyses identified key predictive factors. The dataset split into 70% training set 30% testing set. To address the class imbalance, Synthetic Minority Oversampling Technique combined Edited Nearest Neighbors employed. Nine widely used algorithms evaluated relevant prediction metrics, SHAP (SHapley Additive exPlanations) interpret assess contributions different features. Results: Regression revealed complications such as hydrocephalus, cerebral hernia, deep vein thrombosis, well specific brain regions (frontal, parietal, temporal lobes), significantly contributed PSE. Factors age, gender, NIH Stroke Scale (NIHSS) scores, like WBC count D-dimer levels associated increased risk. Tree-based methods Random Forest, XGBoost, LightGBM showed strong performance, achieving an AUC 0.99. Conclusions: accurately predicts risk, tree-based models demonstrating superior performance. NIHSS score, count, most crucial predictors. Funding: research funded by Central University basic young teachers students ability promotion sub-projec t(2023CDJYGRH-ZD06), Emergency Medicine Chongqing Key Laboratory Talent Innovation development joint fund project (2024RCCX10).

Язык: Английский

Процитировано

1