Journal of evidence-based medicine · 2025
Development of a Novel Interpretable Transformer-Based Deep Learning Model for Predicting Postoperative Hypokalemia in Pituitary Adenomas.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Li Zhuoyuan, Wang Jie, Wang Yunfeng, Peng Zheng, Li Xiaojian, Chen Chunlei, Yan Huiying, Jin Wei, Lu Yue, Zhuang Zong, Li Wei, Hang Chunhua
Аннотация
AIM: Hypokalemia is a prevalent complication following pituitary adenoma (PA) surgery, adversely impacting patient prognosis. Identifying predictors of early postoperative hypokalemia is crucial for managing patients effectively. This study aims to develop an interpretable predictive model to predict postoperative hypokalemia in patients with PA and recognizes individualized significant parameters contributing to the predictive outcomes, thereby facilitating early intervention.
METHODS: This retrospective cohort study investigated postoperative hypokalemia in 280 patients with PA. We developed a Transformer-based predictive model, CliTab-Transformer, and compared it with an XGBoost-based model and a multilayer perceptron (MLP)-based deep learning model. Model performance was evaluated using metrics such as accuracy, precision, recall, F1 score, and ROC/PR curves, based on five-fold cross-validation. Model interpretability was assessed using a novel Transformer-Explainability method to identify significant parameters contributing to individual predictions.
RESULTS: CliTab-Transformer outperformed XGBoost and MLP in predicting postoperative hypokalemia, showing higher accuracy (0.836 vs. 0.775 vs. 0.771), F1 score (0.845 vs. 0.792 vs. 0.766), sensitivity (0.838 vs. 0.833 vs. 0.771), and AUC (0.835 vs. 0.807 vs. 0.676). The model's interpretability analysis revealed that the preoperative factors, including gender, hypertension, serum potassium concentration, and disease duration, are significantly predictive of postoperative hypokalemia.
CONCLUSIONS: Our model outperforms XGBoost in predicting early postoperative hypokalemia in patients with pituitary adenomas. We systematically explore attention mechanisms in clinical tabular data, demonstrating their effectiveness in capturing complex feature interactions, leading to more individualized, interpretable, and clinically meaningful insights.
METHODS: This retrospective cohort study investigated postoperative hypokalemia in 280 patients with PA. We developed a Transformer-based predictive model, CliTab-Transformer, and compared it with an XGBoost-based model and a multilayer perceptron (MLP)-based deep learning model. Model performance was evaluated using metrics such as accuracy, precision, recall, F1 score, and ROC/PR curves, based on five-fold cross-validation. Model interpretability was assessed using a novel Transformer-Explainability method to identify significant parameters contributing to individual predictions.
RESULTS: CliTab-Transformer outperformed XGBoost and MLP in predicting postoperative hypokalemia, showing higher accuracy (0.836 vs. 0.775 vs. 0.771), F1 score (0.845 vs. 0.792 vs. 0.766), sensitivity (0.838 vs. 0.833 vs. 0.771), and AUC (0.835 vs. 0.807 vs. 0.676). The model's interpretability analysis revealed that the preoperative factors, including gender, hypertension, serum potassium concentration, and disease duration, are significantly predictive of postoperative hypokalemia.
CONCLUSIONS: Our model outperforms XGBoost in predicting early postoperative hypokalemia in patients with pituitary adenomas. We systematically explore attention mechanisms in clinical tabular data, demonstrating their effectiveness in capturing complex feature interactions, leading to more individualized, interpretable, and clinically meaningful insights.