Page 122 - 《中国药房》2025年15期

·智慧药学·

基于机器学习构建中重度抑郁症住院患者使用SNRI类抗抑郁药
的疗效预测模型
Δ

2
2
2
2
1
刘学涛 1， 2＊ ，刘 阳 ，李红建 ，吴建华 ，刘思明 ，焦 敏 ，于鲁海 （1. 石河子大学药学院，新疆 石河子
2 #
832000；2.新疆维吾尔自治区人民医院药学部，乌鲁木齐 830001）
中图分类号 R971+.43；TP181 文献标志码 A 文章编号 1001-0408（2025）15-1936-06
DOI 10.6039/j.issn.1001-0408.2025.15.21
摘 要 目的 运用机器学习方法构建中重度抑郁症住院患者使用 5-羟色胺去甲肾上腺素再摄取抑制剂（SNRI）的疗效预测模
型。方法 回顾性收集2022年1月至2024年10月在新疆某三甲医院使用SNRI类药物治疗的中重度抑郁症住院患者病历资料，根
据24项汉密尔顿抑郁量表评分标准的减分率，将患者分为有效组与无效组；经过LASSO回归筛选与SNRI类药物疗效相关的特
征变量，应用训练集构建支持向量机、k近邻、随机森林、轻量级梯度提升机和极端梯度提升5种预测模型，使用贝叶斯优化算法调
整模型的超参数，再以验证集评估模型性能，以筛选出最优模型。应用夏普利加性解释方法对最优模型进行解释。结果 共收集
到355例中重度抑郁症住院患者的病历资料，其中有效组285例、无效组70例，治疗有效率为80.28%。经过特征变量筛选，得到与
疗效相关的5个特征变量，分别为汉密尔顿焦虑量表评分、血尿素氮、合用抗焦虑药物、饮酒史、首次发病。与其他模型相比，随机
森林模型的性能表现最优，其受试者工作特征曲线下面积值为0.85，精确率-召回率曲线下面积值为0.87，准确度为0.74，召回率为
0.75。结论 基于5种特征变量建立的随机森林模型可用于中重度抑郁症住院患者使用SNRI类药物的疗效预测。
关键词 5-羟色胺去甲肾上腺素再摄取抑制剂；中重度抑郁症；疗效；机器学习；预测模型

Construction of a predictive model for the efficacy of SNRI antidepressants in inpatients with moderate
and severe depression based on machine learning
LIU Xuetao ，LIU Yang ，LI Hongjian ，WU Jianhua ，LIU Siming ，JIAO Min ，YU Luhai （1. School of
2
2
1， 2
1
2
2
2
Pharmacy， Shihezi University， Xinjiang Shihezi 832000， China；2. Dept. of Pharmacy， Xinjiang Uygur
Autonomous Region People’s Hospital， Urumqi 830001， China）
ABSTRACT OBJECTIVE To construct a prediction model for the efficacy of serotonin-norepinephrine reuptake inhibitor
（SNRI） in inpatients with moderate and severe depression by using a machine learning method. METHODS The case records of
inpatients with moderate and severe depression treated with SNRI antidepressants were collected from a third-grade class-A hospital
in Xinjiang from January 2022 to October 2024； those patients were divided into effective group and ineffective group based on the
Hamilton depression scale-24 score reduction rate. After screening the characteristic variables related to the therapeutic efficacy of
SNRI drugs through LASSO regression， five prediction models including support vector machine， k-nearest neighbor， random
forest， lightweight gradient boosting machine and extreme gradient boosting were constructed using the training set. Bayesian
optimization was used to adjust the hyperparameters of these models. The performance of the models was evaluated in the validation
set to select the optimal model. The Shapley additive explanations method was used to perform explainable analysis on the best
model. RESULTS The medical records from 355 hospitalized patients with moderate and severe depression were collected，
comprising 285 cases in the effective group and 70 cases in the ineffective group， resulting in an overall therapeutic response rate of
80.28%. After feature variable screening， five characteristic variables for therapeutic efficacy were obtained， including Hamilton
anxiety scale， blood urea nitrogen， combination of anti-anxiety drugs， drinking history， and first onset of the disease. Compared
with other models， the random forest model performed the best. The area under the receiver operating characteristic curve was
0.85， the area under the precision-recall curve was 0.87， the accuracy was 0.74， and the recall rate value was 0.75.
CONCLUSIONS The random forest model constructed based
on five characteristic variables demonstrates potential for
Δ 基金项目“天山英才”医药卫生高层次人才培养计划项目（No.
predicting the therapeutic efficacy of SNRI antidepressants in
TSYC202301A028）
＊ 第一作者 硕 士 研 究 生 。 研 究 方 向 ：临 床 药 学 。 E-mail： hospitalized patients with moderate and severe depression.
1440354272@qq.com KEYWORDS serotonin-norepinephrine reuptake inhibitor；
# 通信作者 主任药师，硕士生导师，硕士。研究方向：医院药学、 moderate and severe depression； efficacy； machine learning；
药物分析。E-mail：1523264450@qq.com predictive model

· 1936 · China Pharmacy 2025 Vol. 36 No. 15 中国药房 2025年第36卷第15期