增强机器学习模型预测A型主动脉夹层患者1年死亡率
本研究纳入289例A型主动脉夹层患者,构建并验证可解释机器学习模型预测1年死亡率。Treebag模型表现最优,Brier评分0.128,AUC 0.91。高龄和白细胞升高为主要风险因素,收缩压较高、淋巴细胞、二氧化碳结合力、嗜酸性粒细胞、β受体阻滞剂使用和手术干预为保护因素。
要点
- Treebag模型预测1年死亡率表现最优,Brier评分0.128,AUC 0.91。
- 高龄和白细胞升高是1年死亡的主要风险因素。
- 较高收缩压、淋巴细胞、二氧化碳结合力、嗜酸性粒细胞、β受体阻滞剂使用和手术干预为保护因素。
核心结论
- Treebag机器学习模型可有效预测A型主动脉夹层患者1年死亡率并进行风险分层。
研究设计与样本
- 纳入289例A型主动脉夹层患者,分为训练队列202例和验证队列87例。
- 采用LASSO方法结合10折交叉验证识别与1年死亡率相关的8个关键因素。
主要发现
- Treebag模型表现优于其他模型,Brier评分为0.128,AUC为0.91。
- 主要风险因素包括高龄和白细胞计数升高。
- 较高收缩压、淋巴细胞、二氧化碳结合力、嗜酸性粒细胞、β受体阻滞剂使用和手术干预为保护因素。
关键数值与推荐
- LASSO方法结合10折交叉验证识别出8个与1年死亡率相关的关键因素。
临床怎么用
- 开发了基于网络的TAAD One-Year Prognostic Risk Assessment Web应用(https://taad-1year-mortality-predictor.streamlit.app/),可根据识别出的预测因素预测患者1年死亡率,辅助临床决策和患者管理。
局限性
- 原文未述局限性
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: The study objective was to develop and validate an interpretable machine learning model to predict 1-year mortality in patients with type A aortic dissection, improving risk classification and aiding clinical decision-making. METHODS: We enrolled 289 patients with type A aortic dissection, dividing them into a training cohort (202 patients) and a validation cohort (87 patients). The Least Absolute Shrinkage and Selection Operator method with 10-fold cross-validation identified 8 key factors related to 1-year mortality. The Treebag model's performance was assessed using accuracy, F1-Score, Brier score, area under the curve, and area under the precision-recall curve with calibration and clinical utility evaluated through decision curves. Shapley Additive Explanations analysis determined the most influential predictors. RESULTS: The Treebag model outperformed others, achieving a Brier score of 0.128 and an area under the curve of 0.91. Key risk factors included older age and elevated white blood cell count, whereas higher systolic blood pressure, lymphocyte, carbon dioxide combining power, eosinophil, β-receptor blocker use, and surgical intervention were protective. A web-based application, TAAD One-Year Prognostic Risk Assessment Web, was developed for clinical use (available at https://taad-1year-mortality-predictor.streamlit.app/). This platform allows for the prediction of 1-year mortality in patients with type A aortic dissection based on the identified predictive factors, facilitating clinical decision-making and patient management. CONCLUSIONS: The Treebag machine learning model effectively predicts 1-year mortality in patients with type A aortic dissection, stratifying risk profiles. Key factors for enhancing survival include surgical intervention, β-blocker administration, and management of systolic blood pressure, lymphocyte, carbon dioxide combining power, eosinophil, and white blood cell levels, offering a valuable tool for improving patient outcomes.
引用
- Zhang J, Xiong W, Yang J, et al. Enhanced machine learning models for predicting one-year mortality in individuals suffering from type A aortic dissection. J Thorac Cardiovasc Surg. 2024. PMID: 39303944. doi:10.1016/j.jtcvs.2024.09.019
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。