| 引用本文： |
-
谢秋华,鲁作华,邓生琼,等.基于机器学习算法的NT-proBNP灰值患者心力衰竭判别模型评价[J].同济大学学报（医学版）,2021,42(3):375-380. [点击复制]
- XIE Qiu-hua,LU Zuo-hua,DENG Sheng-qiong,et al.Evaluation of heart failure discriminant model in patients with NT-proBNP gray value based on machine learning algorithm[J].同济大学学报（医学版）,2021,42(3):375-380. [点击复制]
|
|
| 摘要: |
| 目的应用机器学习算法构建氨基末端脑钠尿肽（N-terminal pro-brain natriuretic peptide, NT-proBNP）灰值患者心力衰竭判别模型并评价。方法收集2013年1月至2018年12月在上海市浦东新区公利医院进行NT-proBNP实验室检测的患者临床资料和实验室检测信息，数据清洗后纳入研究对象，并按7∶3的比例划分训练集和测试集。用L1范数正则化和递归特征消除方法对特征进行筛选。应用基于机器学习的Logistic回归、随机森林、梯度提升树和XGBoost算法构建模型，比较4种方法构建的模型对NT-proBNP灰值患者心力衰竭判别价值。结果按重要性筛选出模型因子年龄、性别、肌酸激酶同工酶、肌酐、肌红蛋白、肌钙蛋白Ⅰ、血红蛋白、白细胞计数。Logistic回归、随机森林、梯度提升树和XGBoost四种模型灵敏度分别为58.42%、56.83%、65.74%和60.04%；特异度分别为57.47%、68.18%、60.13%、65.93%。结论基于机器学习建立的NT-proBNP灰值患者心力衰竭判别模型有一定临床价值，本研究结果应用价值有待于更大样本进行验证。 |
| 关键词: NT-proBNP 心力衰竭 机器学习算法 |
| DOI：10.12289/j.issn.1008-0392.20433 |
| 投稿时间：2020-10-10 |
| 基金项目:浦东新区卫生系统重点学科群建设资助项目（PWZxq2017-15） |
|
| Evaluation of heart failure discriminant model in patients with NT-proBNP gray value based on machine learning algorithm |
| XIE Qiu-hua,LU Zuo-hua,DENG Sheng-qiong,LIU Qian-qian,XU Li-min,ZHANG Deng-hai,LIU Xing-hui |
| (Dept. of Clinical Laboratory, Gongli Hospital of Shanghai Pudong New Area, Shanghai 200135, China) |
| Abstract: |
| ObjectiveTo construct and evaluate a heart failure discrimination model in patients with N-terminal pro-brain natriuretic peptide(NT-proBNP) gray value based on machine learning algorithms. MethodsClinical data of patients who underwent NT-proBNP laboratory testing at Gongli Hospital from January 2013 to December 2018 were retrospectively analyzed. Data were divided into training set and a test set with a ratio of 7∶3. L1 paradigm regularization and recursive feature elimination methods were used to filter the features. Machine learning-based logistic regression, random forest, gradient boosting tree, and XGBoost algorithms were applied to construct models to compare the discriminatory value of the models constructed by the four methods for heart failure in patients with NT-proBNP gray values. ResultsThe model factors filtered by importance were age, sex, creatine kinase isoenzyme, creatinine, troponin, troponin I, hemoglobin, and leukocyte count. The sensitivities of logistic regression, random forest, gradient boost tree, and XGBoost models were 58.42%, 56.83%, 65.74%, and 60.04%, and the specificities of them were 57.47%, 68.18%, 60.13%, and 65.93%, respectively. ConclusionThe machine learning-based NT-proBNP gray value model of heart failure has some clinical value, and the application of the results of this study needs to be validated in a larger sample. |
| Key words: NT-proBNP cardiac failure machine learning algorithm |
