基于人工神经网络的新疆维吾尔族癫痫患儿奥卡西平血清浓度预测研究  

Prediction of Serum Concentration of Oxcarbazepine in Uygur Children with Epilepsy in Xinjiang Based on Artificial Neural Network

作　　者：赵婷[1] 李红健[1] 翁振群[1] 章立华 王婷婷[1] 冯杰[1] 孙力[1] 郭喜红[1] 于鲁海[1] ZHAO Ting;LI Hong-jian;WENG Zhen-qun;ZHANG Li-hua;WANG Ting-ting;FENG Jie;SUN Li;GUO Xi-hong;YU Lu-hai(Department of Medicine,People's Hospital of Xinjiang Uygur Autonomous Region,Urumqi 830001,China;College of Pharmacy,Jiangnan University,Wuxi 214000,China)

机构地区：[1]新疆维吾尔自治区人民医院药学部,乌鲁木齐830001 [2]江南大学药学院,江苏无锡214000

出　　处：《中国药学杂志》2020年第16期1376-1380,共5页Chinese Pharmaceutical Journal

基　　金：新疆维吾尔自治区自然科学基金项目资助(2016D01C097)。

摘　　要：目的利用人工神经网络对新疆维吾尔族癫痫患儿奥卡西平稳态血清药物浓度进行预测,为临床奥卡西平个体化给药提供理论依据。方法测定270例新疆维吾尔自治区人民医院维吾尔族癫痫患儿奥卡西平稳态血清药物浓度,提取相关数据,采用Matlab(R2018a)编程软件,结合深度学习网络构建奥卡西平血药浓度预测模型。结果模型建立的网络参数为:初始学习率为0.001,最终学习率为0.0001,动量系数为0.90,最大训练次数为1000,遗传代数为6000,其他参数为默认值。模型验证结果显示,45例维吾尔族癫痫患儿奥卡西平血清谷浓度中,45个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0.01%,平均预测绝对误差(MAE)为1.21%。预测的血药浓度和实际测定浓度之间的相关系数为0.997,预测结果较理想。结论应用人工神经网络预测新疆地区维吾尔族癫痫患儿奥卡西平血清药物浓度是可行的,可将其用于奥卡西平个体化给药的研究,促进临床合理用药。OBJECTIVE To predict the steady-state serum concentration of oxcarbazepine in Uygur children with epilepsy in Xinjiang by artificial neural network,thus to provide a theoretical basis for individualized administration of oxcarbazepine.METHODS The steady-state serum concentration of oxcarbazepine was measured in 270 Uygur children with epilepsy in the People's Hospital of Xinjiang Uygur Autonomous Region,and the relevant data was extracted.The prediction model of plasma concentration of oxcarbazepine was constructed by using Matlab(R2018a)programming software and deep learning network.RESULTS The network parameters of the model were as follows:the initial learning rate was 0.001,the final learning rate was 0.0001,the momentum coefficient was 0.90,the maximum training times was 1000,the genetic algebra was 6000,and the other parameters were default values.The results of model verification showed that among the 45 Uygur children with epilepsy,the prediction errors of 45 oxcarbazepine serum trough concentrations were all less than 10%,and the rate of error of less than 15%was 100.00%.The mean prediction error(MPE)was 0.01%and the mean absolute prediction error(MAE)was 1.21%.The correlation coefficient between the predicted blood concentration and the actual determined concentration was 0.997,and the predicted result was ideal.CONCLUSION It is feasible to use artificial neural network to predict the serum concentration of oxcarbazepine in Uygur children with epilepsy in Xinjiang.It can be used in the study of individual administration of oxcarbazepine to promote the rational use of oxcarbazepine in clinic.

关 键 词：深度学习网络 奥卡西平 血药浓度 癫痫 

分 类 号：R969.3[医药卫生—药理学]