基于XGBoost-SHAP模型的重度抑郁症复发风险预测-现代医学

基于XGBoost-SHAP模型的重度抑郁症复发风险预测

单位：南京脑科医院院长办公室, 江苏南京 210000

分类号：R749

出版年·卷·期（页码）：2026·54·第二期（302-307）

摘要：

目的： 构建适用于临床实践的重度抑郁症复发的预测模型，以便识别高危患者并采取有针对性的临床管理干预策略。方法： 选取2022—2025年在南京脑科医院诊治的612例重度抑郁症患者为研究对象，按7∶3比例随机分为训练组和验证组。通过R软件“XGBoost”包构建极端梯度提升(XGBoost)构建分类器以实现复发风险预测模型构建。通过受试者工作特征曲线(ROC)曲线对模型预测性能进行评价，并利用沙普利加性解释(SHAP)对模型的特征进行解释。结果： 612例重度抑郁症患者中有297例重度抑郁症患者复发，复发率为48.5%。XGBoost模型预测重度抑郁症复发风险的ROC曲线下面积(AUC)为0.947，模型预测价值较高。SHAP值解释变量在模型中的重要性排序由高至低依次为收缩压、舒张压和疼痛评分，其预测贡献度分别为0.183、0.135和0.129。结论： 本研究采用XGBoost算法构建的临床重度抑郁症患者复发风险的预测模型，便于临床医护人员依据该预测模型准确识别重度抑郁症复发高风险患者，为针对性的临床管理策略和干预措施实施提供依据。

Objective: To develop a clinically applicable prediction model for relapse in patients with major depressive disorder(MDD), enabling the identification of high-risk individuals and the implementation of targeted clinical management and intervention strategies. Methods: A total of 612 patients diagnosed with MDD and treated at Nanjing Brain Hospital between 2022 and 2025 were enrolled and randomly divided into a training set and a validation set at a ratio of 7:3. An eXtreme gradient boosting(XGBoost) classifier was constructed using the “XGBoost” package in R software to develop a relapse risk prediction model. Model performance was evaluated using the receiver operating characteristic(ROC) curve, and SHapley Additive exPlanations(SHAP) were applied to interpret the contribution of individual features to the model. Results: Among 612 patients included in the analysis, the relapse rate was 48.5%. The XGBoost model demonstrated excellent predictive performance for relapse risk in patients with MDD, with an area under the ROC curve(AUC) of 0.947. SHAP-based feature importance analysis indicated that systolic blood pressure, diastolic blood pressure, and pain score were the top three contributors to the model, with contribution values of 0.183, 0.135, and 0.129, respectively. Conclusion: This study employs the XGBoost model for predicting a relapse in patients with MDD, which can help clinical staffs to accurately identify patients at high risk and implement targeted clinical management strategies.

参考文献：

[1] 李丹.复发性抑郁障碍患者抑郁严重程度、睡眠质量和认知功能的相关研究[J].心理月刊,2024,19(10):53-54,63.
[2] SU Y A,YE C,XIN Q,et al.Major depressive disorder with suicidal ideation or behavior in Chinese population:a scoping review of current evidence on disease assessment,burden,treatment and risk factors[J].J Affect Disord,2023,340:732-742.
[3] 张惟.正念认知疗法在双相抑郁复发患者中的应用[J].中国城乡企业卫生,2025,40(1):102-104.
[4] 于丽燕,严保平,张丽丽,等.首发与复发抑郁障碍临床特点及治疗现况对比分析[J].四川医学,2023,44(9):938-943.
[5] TAGLIAFERRI S D,HAN L K M,KHETAN M,et al.Systematic review and meta-analysis:predictors of relapsing,recurrent,and chronicdepression in young people[J].J Am Acad Child Adolesc Psychiatry,2025,S0890-S8567(25)00166-2.
[6] TEUTENBERG L,STEIN F,THOMAS-ODENTHAL F,et al.Machine learning-based prediction of illness course in major depression:The relevance of risk factors[J].J Affect Disord,2025,374:513-522.
[7] UHER R,PAYNE J L,PAVLOVA B,et al.Major depressive disorder in DSM-5:implications for clinical practice and research of changes from DSM-IV[J].Depress Anxiety,2014,31(6):459-471.
[8] HJERMSTAD M J,FAYERS P M,HAUGEN D F,et al.Studies comparing Numerical Rating Scales,Verbal Rating Scales,and Visual Analogue Scales for assessment of pain intensity in adults:a systematic literature review[J].J Pain Symptom Manage,2011,41(6):1073-1093.
[9] 沈路凡,王晓科,王鸿,等.肾素-血管紧张素系统在高血压抑郁中的研究进展[J].中国药理学通报,2025,41(9):1629-1635.
[10] 秦丽娟,汪丽佳,刘丹,等.大鼠抑郁行为的性差异与下丘脑-垂体-肾上腺轴功能及脑源性神经营养因子表达的相关性[J].中国生物制品学杂志,2014,27(4):524-528.
[11] 凌佳,吴梦瑶,杨琴,等.下丘脑-垂体-肾上腺轴紊乱对焦虑性抑郁模型大鼠海马结构的影响[J].基础医学与临床,2018,38(9):1231-1238.
[12] 李文涛,刘长红,周晓华,等.下丘脑-垂体-肾上腺轴紊乱对脊髓损伤大鼠抑郁样行为的影响及其机制[J].吉林大学学报(医学版),2019,45(6):1395-1400.
[13] 张月良.抑郁症和高血压病共病患者临床对照研究[J].中国健康心理学杂志,2011,19(4):397-399.
[14] BREEDEN M,GILLIS A,SALAS J,et al.Antidepressant treatment and blood pressure control in patients with comorbid depression and treatment resistant hypertension[J].J Psychosom Res,2022,153:110692.
[15] SOUERY D,OSWALD P,MASSAT I,et al.Clinical factors associated with treatment resistance in major depressive disorder:results from a European multicenter study[J].J Clin Psychiatry,2007,68(7):1062-1070.
[16] 罗涛,刘翠珍.抑郁症个体的疼痛感知异常:来自元分析的证据[J].科学通报,2025,70(8):1048-1062.
[17] 成坤.术前抑郁状态对术后疼痛评分影响的研究[D].济南:山东大学,2015.
[18] 江小建,林承凤,余英,等.C反应蛋白与白蛋白的比值联合T淋巴细胞线粒体损伤指数预测老年重度抑郁障碍临床疗效的价值[J].中国老年学杂志,2025,45(21):5197-5201.
[19] 石艳芬,陈源清.甲状旁腺激素+尿素氮+肌酐联合检测在肾功能衰竭早期诊断中的应用效果观察[J].现代诊断与治疗,2014,25(12):2761-2762.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录