【摘要】 目的 分析儿科重症患者丙戊酸钠血药浓度超标的影响因素及预测血药浓度。方法 回顾性收集浙江大学医学院附属儿童医院儿科重症医学数据库（PIC）中重症患儿丙戊酸钠血药浓度的监测数据，检索时限为从2010年1月1日至2018年12月31日，采用logistic回归分析，建立丙戊酸钠血药浓度超标的预测模型。采用极限梯度提升树法预测血药浓度，经Passing-Bablok回归和Bland-Altman散点图法比较预测与实测的数据。结果共收集259例儿科重症患者丙戊酸钠血药浓度数据，非超标组209例，超标组50例。获得4个潜在影响因素（年龄、体质量、白蛋白和白细胞计数），建立预测丙戊酸钠血药浓度超标的列线图模型。曲线下面积、一致性指数、校准曲线、决策曲线分析法和临床影响曲线显示该模型良好。纳入4个影响因素建立预测血药浓度模型，模拟预测的血药浓度具有和实际血药浓度一致的检测效能。结论 所建立针对儿童丙戊酸血药浓度超标的列线图模型能够有效预测患者类别，极限梯度提升树法能够有效预测血药浓度，对指导临床个体化给药具有一定意义。更多还原

【Abstract】 Objective To analyze and predict the excessive blood sodium valproate concentrations in pediatric critically ill patients. Methods We retrospectively collected sodium valproate blood concentration monitoring data at the Children’s Hospital of Zhejiang University School of Medicine in Pediatric Intensive Care(PIC) database from January 1,2010 to December 31,2018. Logistic regression analysis was used to establish the prediction model of excessive serum concentration of sodium valproate. The extreme gradient boosting(XGBoost) method was used to obtain the formula of blood drug concentration,and the predicted and measured data were analyzed by Passing-Bablok regression and Bland-Altman scatter plot method.Results There were 259 serum concentration data of sodium valproate in pediatric critically ill patients,209 children in normal range group and 50 children in excessive concentration group. We obtained four potential factors(age,weight,albumin and white blood cell count) to establish a nomogram model for predicting the excessive blood concentration of sodium valproate. Area under curve,consistenty index,calibration curve,decision curve analysis and clinical influence curve showed that the model was good. The four factors were taken into account to establish a model for predicting serum drug concentration.The simulated predicted blood drug concentration had the same detection efficiency as the actual blood drug concentration.Conclusion The nomogram model established can predict critically ill pediatric patients with valproic acid concentration exceeding the normal range,and XGBoost method can help predict serum drug concentration,which has certain significance for guiding clinical individualized drug administration.更多还原