摘要
目的应用反向传播神经网络建立癫痫儿童的血清丙戊酸浓度预测模型。方法收集122例癫痫儿童的临床相关资料,根据资料获取的难易程度,应用反向传播神经网络建立建模因子分别为6及12的模型Ⅰ和模型Ⅱ,比较两种模型的预测效能。并利用模型Ⅱ前瞻性预测12例癫痫儿童的血清丙戊酸浓度。结果模型Ⅰ和模型Ⅱ测试组的预测值与实测值的相关系数分别为0.945、0.986;均方预测误差平方根(RMSE)分别为5.864、2.998;平均预测误差百分率分别为0.42%、-0.59%(P>0.05);平均绝对误差百分率分别为7.67%、4.40%(P<0.05)。前瞻性预测(12例)其误差百分率在±5%以内的7例,±5%～±10%的3例,±10%～±20%的1例,大于20%的1例。结论模型Ⅱ对癫痫儿童血清丙戊酸浓度的预测精准度优于模型Ⅰ。所建立的针对儿童的神经网络模型能够有效预测癫痫儿童的血清丙戊酸浓度,可指导临床个体化给药。
OBJECTIVE To develop a prediction model based on BP neural network for serum valproic acid concentration of epileptic children. METHODS The clinic data of 122 epileptic children were collected. Model I with 6 modeling factors and Model 11 with 12 modeling factors were established with BP neural network respectively, according to the difficulty of acquisition of the data. The prediction performances were compared between the two models. Serum valproie acid concentrations of 12 epileptic children were predicted prospectively by Model 11. RESULTS The correlation coefficients between the predicted values by Model I and Model ]1 and the observed values of the test sets were 0. 945 and 0. 986 respectively. The root mean squared errors （RMSE） were 5. 864 and 2. 998, mean percent prediction errors were 0. 42% and - 0. 59% （ P 〉 0. 05 ） , and mean percent absolute errors were 7. 67% and 4.40% （ P 〈0. 05） respectively. The percent prediction errors were within 5% for 7 cases, between 5% - 10% for 3 cases, between 10% - 20% for 1 case, and greater than 20% for the other case. CONCLUSION Model 11 has better prediction performance for serum valproic acid concentration than model I. The neural network models established in the research can effectively predict serum valproic acid concentration of epileptic children and give guidance for individual administration in clinics.
出处
《中国药学杂志》
CAS
CSCD
北大核心
2012年第10期766-770,共5页
Chinese Pharmaceutical Journal
基金
浙江省药学会医院药学科研专项基金项目(2010ZYY05)
关键词
丙戊酸钠
儿童
人工神经网络
血药浓度
癫痫
sodium valproate
children
artificial neural network
serum concentration
epileptic
