中国组织工程研究 ›› 2021, Vol. 25 ›› Issue (36): 5792-5797.doi: 10.12307/2021.344

• 人工假体 artificial prosthesis • 上一篇    下一篇

基于机器学习算法预测全膝关节置换后是否需要输血的可能性

陈潮锋1,2,何达东1,梁锦成1,何智军1   

  1. 1广州市番禺区中医院 (广州中医药大学番禺中医院),广东省广州市   510000;2广州中医药大学,广东省广州市   510000
  • 收稿日期:2021-03-23 修回日期:2021-03-25 接受日期:2021-04-15 出版日期:2021-12-28 发布日期:2021-09-17
  • 通讯作者: 陈潮锋,广州市番禺区中医院 (广州中医药大学番禺中医院),广东省广州市 510000;广州中医药大学,广东省广州市 510000
  • 作者简介:陈潮锋,男,1985年6月生,医学学士学位,广州中医药大学在职研究生,广州市番禺区中医院骨关节科医疗组长,副主任医师,主要研究领域:中西医结合骨伤科学,人工关节置换术及人工关节翻修术。
  • 基金资助:
    广东省中医药局科研项目(20171202),项目负责人:陈潮锋

Predicting the possibility of blood transfusion after total knee arthroplasty based on machine learning algorithm

Chen Chaofeng1, 2, He Dadong1, Liang Jincheng1, He Zhijun1   

  1. 1Panyu Hospital of Chinese Medicine (Panyu Hospital of Chinese Medicine of guangzhou unversity of Chinese Medicine ), Guangzhou 510000, Guangdong 
  • Received:2021-03-23 Revised:2021-03-25 Accepted:2021-04-15 Online:2021-12-28 Published:2021-09-17
  • Contact: Chen Chaofeng, Panyu Hospital of Chinese Medicine (Panyu Hospital of Chinese Medicine of guangzhou unversity of Chinese Medicine ), Guangzhou 510000, Guangdong Province, China; guangzhou unversity of Chinese Medicine, guangzhou 510000, guangzhou province, china
  • About author:Chen Chaofeng, master, Associate chief physician, Panyu Hospital of Chinese Medicine (Panyu Hospital of Chinese Medicine of guangzhou unversity of Chinese Medicine ), Guangzhou 510000, Guangdong Province, China; guangzhou unversity of Chinese Medicine, guangzhou 510000, guangzhou province, china
  • Supported by:
    the Scientific Research Project of Guangdong Bureau of Chinese Medicine, No. 20171202 (to CCF)

摘要:


文题释义:

机器学习:机器学习方法是通过监督、无监督或半监督方法反映高维数据的性质,可以有效降低数据维数,提高数据的理解能力,能够对高维数、大量和复杂关系数据进行分析。
预测模型:预测模型是用统计模型来估计某种临床结局的概率,作为卫生诊疗策略风险和获益评估的量化工具,临床预测模型可以为医生、患者以及卫生政策制定者提供直观、理性的信息。

背景:为了维持全膝关节置换患者的血液动力学稳定,有必要进行输血,但这常伴随着不良反应的发生。通过研究全膝关节置换后输血的危险因素可以帮助确定哪些患者需要进行输血治疗,有利于术前评估及临床决策。
目的:该研究基于机器学习算法建立预测模型,探讨其对预测全膝关节置换后是否需要输血的预测价值。
方法:回顾性分析广州市番禺区中医院2012年1月至2019年12月的全膝关节置换后患者的临床资料,根据术后是否进行输血,将患者分为未输血组和输血组。比较两组患者性别、年龄、体质量指数、术前血红蛋白水平、ASA麻醉评分、麻醉方式、手术时长、手术类型、吸烟史、既往病史、有无使用胰岛素等临床数据,将上述潜在影响因素分别导入逻辑回归、支持向量机、随机森林和极限梯度提升算法建立4种预测模型,获得预测变量重要性和绘制受试者工作曲线,检验模型的预测价值。

结果与结论:①共收集634例患者资料,其中包括未输血的患者527例和需要输血的患者107例;②结合4个模型,预测重要性评分排名前5位血红蛋白、年龄、手术时长、体质量指数和手术类型为相关性最高的前5位变量;②逻辑回归、支持向量机、随机森林和极限梯度提升算法曲线下面积分别为0.816,0.864,0.773和0.888,通过比较,极限梯度提升算法的表现最佳;③上述数据证实,基于极限梯度提升算法建立的机器学习模型可准确预测全膝关节置换后患者是否有需要输血的可能性,有利于术前评估及临床决策,血红蛋白、年龄、手术时长、体质量指数和手术类型可能是影响全膝关节置换后患者是否需要输血的预测因素。

https://orcid.org/0000-0002-4158-6816 (陈潮锋) 

中国组织工程研究杂志出版内容重点:人工关节;骨植入物;脊柱;骨折;内固定;数字化骨科;组织工程

关键词: 机器学习, 全膝关节置换, 输血, 危险因素, 逻辑回归, 预测模型, 回顾性分析, 影响因素

Abstract: BACKGROUND: To maintain the hemodynamic stability of patients with total knee replacement, blood transfusion is necessary, but this is often accompanied by adverse reactions. Studying the risk factors of blood transfusion after total knee replacement can help determine which patients need blood transfusion, which is conducive to preoperative evaluation and clinical decision-making.  
OBJECTIVE: To establish a prediction model based on a machine learning algorithm and explore its predictive value in predicting the possibility of blood transfusion after total knee replacement.
METHODS:  The clinical data after total knee arthroplasty in the panyu hospital of chinese medicine from January 2012 to December 2019 were retrospectively analyzed, and divided the patients into a non-transfusion group and a blood transfusion group according to whether blood transfusion was performed after the operation. The data of sex, age, body mass index, preoperative hemoglobin, ASA anesthesia score, anesthesia mode, operation duration, operation type, smoking history, past medical history, and the use of insulin were compared between the two groups. The above-mentioned potential influencing factors were incorporated into logistic regression, support vector machine, random forest and XGBoost algorithm to establish four kinds of prediction model, obtain the importance of predictive variables and draw receiver working curve, and test the predictive value of the model.  
RESULTS AND CONCLUSION: (1) We included a total of 634 samples, including 527 untransfused total knee arthroplasty patients and 107 total knee arthroplasty patients requiring transfusion. (2) Combining the four models, the top five prediction importance scores were hemoglobin, age, operation length, body mass index and operation type were the top five variables with the highest correlation. (3) The areas under the curve of logistic regression, support vector machine, random forest and XGBoost algorithm were 0.816, 0.864, 0.773 and 0.888, respectively. By comparison, the XGBoost algorithm performed best. (4) It is concluded that the machine learning model based on XGBoost algorithm can accurately predict the risk of blood transfusion in patients after total knee arthroplasty, which is conducive to preoperative evaluation and clinical decision-making. Hemoglobin, age, length of surgery, body mass index, and type of surgery may be important predictors of the risk of transfusion after total knee arthroplasty.

Key words: machine learning, total knee arthroplasty, blood transfusion, risk factors, logistic regression, forecast model, retrospective analysis, influencing factors

中图分类号: