| 马攀,贾运涛,刘芳,等.基于支持向量机技术预测丙戊酸钠血药浓度[J].安徽医药,2021,25(1):35-39. |
| 基于支持向量机技术预测丙戊酸钠血药浓度 |
| Prediction of valproate blood concentration based on support vector machine technology |
| |
| DOI：10.3969/j.issn.1009?6469.2021.01.009. |
| 中文关键词: 丙戊酸钠 精准医学 机器学习 线性模型 支持向量机 血药浓度 个体化用药 |
| 英文关键词: Sodium valproate Precision medicine Machine learning Linear models Support vector machine Blood concentration Individualized drug administration |
| 基金项目:陆军军医大学优秀人才库?苗圃项目(XZ?2019?505?073)；重庆医科大学智慧医学培育项目?智慧药学(ZHYX2019022) |
|
| 摘要点击次数: 2169 |
| 全文下载次数: 757 |
| 中文摘要: |
| 目的基于支持向量机(SVM)技术，建立丙戊酸钠的血药浓度预测模型。方法收集陆军军医大学第一附属医院 2015年 1月至 2018年 12月确诊为癫痫且服用丙戊酸钠缓释片的病人的血药浓度及 16个血药浓度影响因素指标数据。利用随机数字表法将收集的 206例病人共 271个样本数据分为 190个构成训练样本集以及 81个构成测试样本集。基于 SVM技术对 190个训练样本进行训练，建立预测模型。再用外部验证法将 81个测试样本的血药浓度模型预测值与实际观测值进行对比。结果训练样本集和测试样本集中病人的各临床指标除胱抑素 C外，其余指标差异无统计学意义(P≥0.05)训练样本集中病人胱抑素 C为(1.17±1.23)mg/L，明显高于测试样本集中病人的(0.93±0.84)mg/L(P＝0.012)。基于 SVM技术的血，药浓度预测模型取得了较好的预测效果，模型预测值与实际观测值相对误差:小于 5%的 12个； 5%～10%(含)的 23个； 10%～15%(含)的 21个； 15%～20%(含)的 13个； 20%～25%(含)的 4个，超过 25%的 8个；平均相对误差为 12.12%，相对误差小于 20%(含)的样本占比达到 85.18%。平均绝对误差为 9.98 mg/L，绝对误差小于 20 mg/L的样本占比达到 95.06%。模型预测值与实际观测值的相关系数为 0.788。结论 SVM技术在血药浓度预测方面具有良好的应用前景，基于该技术的丙戊酸钠血药浓度预测模型准确度较好，模型预测值与实际观测值的相关性较好，相对误差较小，可为临床制定个体化给药方案提供参考。 |
| 英文摘要: |
| Objective To establish a prediction model for the blood concentration of valproic acid based on the Support Vector Machine(SVM)technology.Methods Blood concentration of patients diagnosed with epilepsy from January 2015 to December2018 who took valproate sustained release tablets in The First Affiliated Hospital of Army Medical University，as well as data of 16 influencing factors of blood concentration were collected.A total of 271 samples from 206 cases were randomly divided into 190constituent training samples and 81 constituent test samples by random number table method.Based on the SVM technology，190 constituent training samples were trained and the prediction model was established.The predicted values of the blood concentrationmodel in 81 samples were compared with the observed values by external validation.Results Except for Cystatin C，there was no significant difference in the clinical indicators of the patients in the training samples and the test samples(P≥0.05).The Cystatin C of the patients in the training samples was 1.17±1.23 mg/L，which was significantly higher than the test samples’s 0.93±0.84 mg/L(P＝0.012).The prediction model for the blood drug concentration of SVM technology achieves a good prediction effect，with 12 rel? ative errors less than 5%，23 of 5%?10%，21 of 10%?15%，13 of 15%?20%，4 of 20%?25%，and 8 more than 25%；The mean rela? tive error was 12.12% and the samples with relative error less than 20% were accounted for 85.18%.The mean absolute error was 9.98 mg/L and the proportion of samples with an absolute error of less than 20 mg/L reached 95.06%，and the correlation coeffi? cient between the prediction value and the actual observation value was 0.788.Conclusion SVM technology has a good applica?tion prospect in the prediction of blood drug concentration.The prediction model of sodium valproate blood concentration based onthis technology has good accuracy，and the predicted value of the model has high correlation with the actual observation value，with relatively small errors.Thus，reference for clinical individualized drug delivery scheme can be provided. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
