应用人工神经网络法预测地高辛血药浓度

目的:探讨应用人工神经网络(ANN)预测地高辛血药浓度的效果和特点。方法:回顾性收集我院2008年3月-2010年3月长期口服地高辛患者的临床资料,其中84例为建模数据集,10例为验证数据集。采用反向传播人工神经网络(BPANN)法预测地高辛血药浓度,并与非线性混合效应模型(NONMEM)法预测结果进行比较,计算影响血药浓度各因素的平均影响值(MIV)并排序。结果:NONMEM和ANN法预测地高辛浓度值与测定值的相关性分别为0.851及0.946,各因素中患者血清肌酐、体重、年龄较其他因素MIV大。结论:ANN法可较好地处理各因素间复杂的关系,为血药浓度预测提供了一条有效思路。

Prediction of Blood Concentration of Digoxin by Artificial Neural Network

OBJECTIVE： To explore the effects and characteristics of artificial neural network （ANN） on predicting blood concentration of digoxin. METHODS： Clinical information was collected from patients who administrated digoxin for a long time in our hospital during Mar. 2008-Mar. 2010, among which 84 examples for modeling and the other 10 examples for validating. BPANN was applied to predict blood concentration of digoxin, and the results were compared with NONMEM. The mean impact value （MIV） for each factor was calculated and factors were sequenced according to their absolute MIVs. RESULTS： The correction between measured value and predicted value got by NONMEM and ANN were 0.851 and 0.946, respectively. The MIVs of serum creatinine, weight and age wes higher than other factors. CONCLUSION： ANN can be used to process complicated relationships. It provides a beneficial clew to predict blood concentration of drugs.