应用GA-BP人工神经网络预测丙戊酸钠血药浓度

Prediction of Serum Sodium Valproate Concentration with GA-BP Artificial Neural Network Model

目的利用GA—BP人工神经网络技术预测丙戊酸钠血药浓度。方法收集193例应用丙戊酸钠患者的血药浓度监测数据及其性别、身高、体重、肝。肾功能、用药情况等12个相关指标，构建GA—BP人工神经网络模型，预测丙戊酸钠血药浓度。结果30个病例样本的预测结果表明，与实际测定浓度相比，误差值范围为0．61—23．33μg／mL，误差百分率在±5％的为19例，±5％～±10％的为3例，±10％～±15％的为4例，±15％～±20％的为l例，超过±20％的为3例。结论应用CA—BP人工神经网络预测丙戊酸钠血药浓度是可行的，可将其用于丙戊酸钠个体化给药的研究。

Objective To predict serum concentration of sodium valproate in patients by GA-BP artificial neural network. Methods The clinical data of 193 patients were collected, which was used to develop the model. Predictive model on serum concentration of sodium valproate was based on GA-BP artificial neural network. Then, samples were forecasted using the predictive model. Results Serum concentration of sodium valproate from 30 patients demonstrated that the percent prediction errors were within ~5% in 19 cases, between ±5% and ± 10% in 3 cases, between ± 10% and ± 15% in 4 cases, between ± 15% and ±20% in 1 case and greater than 20% in 3 cases. The error range of values was 0.61 - 23.33 μg/mL. Conclusion Prediction of serum concentration of sodium valproate in patients by GA-BP artificial neural network is feasible, and can be used in individualized dosage design.