GA-BP神经网络、多元线性回归预测新疆癫痫患儿拉莫三嗪血药浓度  

Prediction of the Blood Concentration of Lamotrigine in Xinjiang Epileptic Children by GA-BP Neural Network and Multiple Linear Regression

作　　者：杨泽萍 马雪 李红健[2] 赵婷[2] 王婷婷[2] 冯杰[2] 于鲁海 Yang Zeping;Ma Xue;Li Hongjian;Zhao Ting;Wang Tingting;Feng Jie;Yu Luhai(College of Pharmacy,Shihezi University,Xinjang Shihezi 832000,China;Department of Pharmacy,People’s Hospital of Xinjiang Uygur Autonomous Region)

机构地区：[1]石河子大学药学院,新疆石河子832000 [2]新疆维吾尔自治区人民医院药学部

出　　处：《中国药师》2022年第9期1598-1602,共5页China Pharmacist

摘　　要：目的:建立遗传算法-反向传播(GA-BP)神经网络模型、多元线性回归模型预测新疆癫痫患儿拉莫三嗪血药浓度。方法:回顾性收集自治区人民医院2015年8月~2021年10月份就诊的296例癫痫患儿口服拉莫三嗪的稳态血药浓度以及临床资料,建立GA-BP人工神经网络模型、多元线性回归模型预测拉莫三嗪血药浓度,比较两种方法对拉莫三嗪血药浓度的预测能力。结果:多元线性回归分析结果显示影响拉莫三嗪血药浓度的因素有民族、合并用药、给药剂量、疗效以及部分生化指标(P<0.05);GA-BP神经网络模型对60个验证组浓度预测误差均小于10%,误差小于15%的比率是100.00%,平均预测误差(MPE)为0.00072%,平均绝对误差(MAE)为0.45%,血药浓度预测值与实测值之间的相关系数r=0.9998,预测结果较理想。多元线性回归模型60个浓度预测误差小于15%的比率是20.00%,MPE为2.54%,MAE为4.74%,血药浓度的预测值与实测值之间的相关系数r=0.2252,预测结果较差。结论:与多元线性预测模型相比,GA-BP神经网络预测模型根据患者临床资料可较好的预测新疆癫痫患儿拉莫三嗪血药浓度。Objective:To establish GA-BP neural network model and multiple linear regression model to predict the blood concentration of lamotrigine in children with epilepsy in Xinjiang.Methods:The steady blood drug concentration of lamotrigine and clinical data of 296 cases of children with epilepsy were retrospectively collected from Autonomous Region People’s Hospital in August 2015 to October 2021,the GA-BP artificial neural network model and the multiple linear regression model were set up to predict the blood concentration of lamotrigine,and the prediction ability of the two methods for lamotrigine blood concentration was compared.Results:Multiple linear regression analysis showed that the factors affecting the blood concentration of lamotrigine were ethnic group,drug combination,dose,curative effect and some biochemical indexes(P<0.05).The concentration prediction errors of GA-BP neural network model for 60 validation groups were all less than 10%,and the ratio of error less than 15%was 100.00%,MPE was 0.00072%,MAE was 0.45%,and the correlation coefficient between the predicted values and measured values was 0.9998,indicating that the prediction results were satisfactory.In the multivariate linear regression model,the ratio of the 60 concentration prediction errors less than 15%was 20.00%,MPE was 2.54%,MAE was 4.74%,and the correlation coefficient between the predicted values and the measured values was 0.2252,indicating poor prediction results.Conclusion:Compared with multiple linear prediction model,GA-BP neural network prediction model can better predict the blood concentration of lamotrigine in Xinjiang epileptic children based on clinical data.

关 键 词：遗传算法-反向传播神经网络 多元线性回归 拉莫三嗪 血药浓度预测 

分 类 号：R969.1[医药卫生—药理学]