南方医科大学学报

南方医科大学学报 ›› 2026, Vol. 46 ›› Issue (7): 1660-1670.doi: 10.12122/j.issn.1673-4254.2026.07.19

• • 上一篇    

多维特征下的消化道恶性肿瘤患者焦虑情绪识别模型构建：基于多种可解释性机器学习算法

陈萌萌1(), 郭润芳4, 赵文洁4, 王咏琪1, 左芦根2,3, 刘牧林2(), 李彬1,4,5()   

  1. 1.蚌埠医科大学，医学信息与工程学院，安徽 蚌埠 233030
    2.蚌埠医科大学，公共卫生学院，安徽 蚌埠 233030
    3.蚌埠医科大学，医院管理研究院数智医疗研究所，安徽 蚌埠 233030
    2.蚌埠医科大学第一附属医院胃肠外科，安徽 蚌埠 233004
    3.炎症相关性疾病基础与转化研究安徽省重点实验室，安徽 蚌埠 233004
  • 收稿日期:2025-12-17 出版日期:2026-07-20 发布日期:2026-07-20
  • 通讯作者: 刘牧林,李彬 E-mail:20231002151@stu.bbmu.edu.cn;liumulin66@aliyun.com;libin@bbmu.edu.cn
  • 作者简介:陈萌萌，在读硕士研究生，E-mail: 20231002151@stu.bbmu.edu.cn
  • 基金资助:
    安徽省高等学校自然科学研究重点项目(2022AH051458);安徽省高校科研创新团队(2023AH010068);安徽省大学协同创新计划(GXXT-2022-058);蚌埠医科大学科研创新项目(Byycx24056)

Construction of an anxiety recognition model for patients with malignant gastrointestinal tumors based on multidimensional features and multiple interpretable machine learning

Mengmeng CHEN1(), Runfang GUO4, Wenjie ZHAO4, Yongqi WANG1, Lugen ZUO2,3, Mulin LIU2(), Bin LI1,4,5()   

  1. 1.School of Medical Information and Engineering, Hospital Management Research Institute, Bengbu Medical University, Bengbu 233030, China
    2.School of Public Health, Hospital Management Research Institute, Bengbu Medical University, Bengbu 233030, China
    3.Institute of Digital Healthcare, Hospital Management Research Institute, Bengbu Medical University, Bengbu 233030, China
    2.Department of Gastrointestinal Surgery, First Affiliated Hospital of Bengbu Medical University, Bengbu 233004, China
    3.Anhui Provincial Key Laboratory of Basic and Translational Research on Inflammation-Related Diseases, Bengbu 233004, China
  • Received:2025-12-17 Online:2026-07-20 Published:2026-07-20
  • Contact: Mulin LIU, Bin LI E-mail:20231002151@stu.bbmu.edu.cn;liumulin66@aliyun.com;libin@bbmu.edu.cn

RichHTML

16

PDF

79

摘要：

目的 通过机器学习算法研究消化道恶性肿瘤患者在临床诊疗过程中伴发焦虑情绪的相关危险因素。 方法 收集2022年11月~2023年4月蚌埠医科大学第一附属医院280例消化道恶性肿瘤患者数据，并于2024年11月至2025年4月对数据进行再分析，依据医院焦虑抑郁量表将焦虑分量表评分≥8分的患者纳入焦虑组，<8分纳入非焦虑组；采用逻辑回归、决策树、随机森林、K-近邻、支持向量机、轻量梯度提升机6种机器学习算法预测焦虑情绪，比较各模型性能指标，并根据最优模型的特征重要性排序及SHAP分析结果，分析焦虑情绪影响因素。 结果 在预测消化道恶性肿瘤患者焦虑情绪的6种机器学习算法中，随机森林模型表现最优，其准确率为0.73、精确率为0.48、召回率为0.86、特异度为0.69、F1-Score为0.62、ROC曲线下面积为0.85，优于其他算法，对消化道恶性肿瘤患者焦虑情绪的识别分类更精准。根据最优模型的特征重要性排序，影响患者焦虑情绪的前10位关键因素依次为总胆固醇（0.0406）、性别（0.0224）、白蛋白（0.0157）、血小板计数（0.0152）、白细胞计数（0.0121）、视黄醇结合蛋白（0.0116）、总蛋白（0.0115）、BMI（0.0098）、C反应蛋白（0.0096）、血小板分布宽度（0.0095）。 结论 随机森林模型为消化道恶性肿瘤患者焦虑情绪提供了较好的预测性能及关键因素筛选，为临床医生早期识别与干预提供辅助决策。

关键词: 消化道恶性肿瘤, 焦虑情绪, 机器学习, 影响因素

Abstract:

Objective To explore the risk factors for concomitant anxiety in patients with malignant gastrointestinal tumors based on machine learning algorithms. Methods A total of 280 patients with malignant gastrointestinal tumors admitted to the First Affiliated Hospital of Bengbu Medical University from November, 2022 to April, 2023 were enrolled, and the collected data were re-analyzed between November,2024 and April, 2025. According to Hospital Anxiety and Depression Scale (HADS-A) anxiety scores, the patients were divided into anxiety group (HADS-A score≥8) and non-anxiety group (<8). Six machine learning algorithms including Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Support Vector Machine and Light Gradient Boosting Machine were used to predict anxiety. The performance of each model was assessed, and the factors affecting anxiety in these patients were identified by feature importance ranking and SHAP analysis by the optimal model. Results Among the 6 machine learning algorithms for predicting anxiety in patients with malignant gastrointestinal tumors, the Random Forest model showed the best performance with an accuracy of 0.73, precision of 0.48, recall of 0.86, specificity of 0.69, F1-Score of 0.62, and the area under the ROC curve (AUC) of 0.85, and was more accurate for identifying and classifying anxiety status in the target patients. According to the feature importance ranking by the optimal model, the top 10 key factors affecting anxiety in these patients were total cholesterol (0.0406), gender (0.0224), albumin (0.0157), platelet count (0.0152), white blood cell count (0.0121), retinol-binding protein (0.0116), total protein (0.0115), body mass index (0.0098), C-reactive protein (0.0096), and platelet distribution width (0.0095). Conclusion The Random Forest model has good performance for predicting anxiety and screening key factors affecting anxiety in patients with malignant gastrointestinal tumors to facilitate its early identification and targeted intervention.

Key words: gastrointestinal malignant tumors, anxiety, machine learning, influencing factors