利培酮联合用药安全性的神经网络评价模型研究
A Neural Network-Based Safety Evaluation Model for Risperidone Combination Therapy
摘要
目的 基于真实世界电子医疗记录数据,利用神经网络构建利培酮临床联合用药安全性评价模型,为保障临床安全联合用药提供证据.方法 提取湖北省某三级甲等精神专科医院 2017-2022 年主要诊断为精神分裂症并使用利培酮的住院患者病历.采用 Louvain 算法进行联合用药网络分析以确定核心用药组合,并利用全面触发工具主动识别药品不良事件.通过单因素及多因素 Logistic 回归分析筛选关键影响因素,并以此构建利培酮多层感知器神经网络(MLP-ANN)安全性评价模型.结果 共纳入患者4 494例,确认药物不良事件(ADE)710 例.MLP-ANN 模型训练集与验证集的总体正确预测率分别为84.4%和84.0%,受试者工作特征曲线下面积(AUC)为0.721.神经网络模型自变量重要性从高至低依次为住院天数、药物联用天数、出院 31 d 内再住院计划、院内手术、就业状况、年龄、性别、合并症数量、药物联用情况.结论 该研究构建的 MLP-ANN 模型对利培酮联合用药安全性具有良好的评价效能.住院及用药时长是首要风险因素,而利培酮与氯氮平联用显示出相对较低的风险.该模型可为临床识别利培酮联合用药高风险患者、实现用药安全早期预警.
Objective To develop a safety evaluation model for risperidone-containing drug combinations based on a neural network using real-world electronic medical record data,so as to provide evidence and research support for ensuring clini-cal safety in combination therapy.Methods Medical records of inpatients primarily diagnosed with schizophrenia and treated with risperidone from 2017 to 2022 were extracted from a Grade A Class III psychiatric hospital in Hubei Province.The Louvain algorithm was employed for drug combination network analysis to identify core medication patterns,and the Global Trigger Tool was used to proactively identify adverse drug events(ADEs).Key influencing factors were screened through univariate and multi-variate logistic regression analyses,and these were used to construct a risperidone multilayer perceptron neural network(MLP-ANN)safety evaluation model.Results A total of 4 494 patients were included,and 710 ADEs were confirmed.The overall correct prediction rates of the MLP-ANN model were 84.4%for the training set and 84.0%for the validation set,with an area un-der the receiver operating characteristic curve(AUC)of 0.721.The normalized importance of independent variables in the neural network model,in descending order,were:duration of hospitalization,duration of drug combination,presence of a readmission plan within 31 days after discharge,presence of in-hospital surgery,employment status,age,gender,number of comorbidities,and drug combination pattern.Conclusions The MLP-ANN model developed in this study demonstrates good performance in e-valuating the safety of risperidone-containing drug combinations.Duration of hospitalization and drug combination were identified as the primary risk factors,while the combination of risperidone and clozapine showed a relatively lower risk.This model can pro-vide data-driven support for clinically identifying high-risk patients and achieving early warning of medication safety.
年,卷(期):
2026, 45(10)
分类号:
R971.4
R951
机标分类号:
U416.14
X936
TV66
在线出版日期:
2026年11月2日
基金项目:
72274065:国家自然科学基金CDR2024R01008:国家药品监督管理局药物警戒研究与评价重点实验室开放课题2025AFD713:湖北省自然科学基金创新发展联合基金项目,