【摘要】 目的 考察肾病综合征（NS）患者使用他克莫司后的血药浓度监测情况，同时建立他克莫司血药浓度预测模型。方法 收集厦门大学附属中山医院2020年1月1日至2023年8月31日166例NS患者使用他克莫司的血药浓度监测数据（共计509次），并对其疗效、药物不良反应（ADR）与血药浓度的相关性进行分析。利用其中109例含有基因信息的NS患者的302次血药浓度监测数据建立多层感知机（MLP）预测模型，并对其进行验证。结果 在疗效方面，未缓解组患者的中位血药浓度为2.20 ng/mL，显著低于部分缓解组的4.00 ng/mL(P<0.001)和完全缓解组的3.60 ng/mL(P=0.002)。在ADR方面，发生ADR组患者的中位血药浓度为5.01 ng/mL，显著高于未发生ADR组的3.37 ng/mL(P=0.001)，且经受试者工作特征曲线亚组分析后可知，他克莫司血药浓度≥6.65 ng/mL时，患者更易发生血肌酐升高[曲线下面积（AUC）为0.764,P<0.001]；他克莫司血药浓度≥6.55 ng/mL时，患者更易发生血糖升高（AUC=0.615,P=0.005）。所建立的MLP预测模型的损失函数值为0.9，预测值与实测值的平均误差绝对值为0.279 5ng/mL，验证散点图的决定系数为0.984，说明模型取得了良好的预测效果。结论 他克莫司血药浓度对NS患者的疗效和ADR均有影响。利用MLP模型进行血药浓度预测的准确率高，预测值与实测值之间误差小，该模型可作为临床个体化用药方案中的重要工具。更多还原

【Abstract】 OBJECTIVE To investigate the monitoring of tacrolimus blood concentration in patients with nephrotic syndrome(NS),and to establish a prediction model for tacrolimus blood concentration. METHODS Data from 509 concentration monitoring sessions of 166 NS patients using tacrolimus were collected from January 1, 2020 to August 31, 2023 in Zhongshan Hospital Affiliated to Xiamen University. The relationship of efficacy and adverse drug reaction(ADR) with blood concentration was analyzed. A multilayer perceptron(MLP) prediction model was established by using the blood concentration monitoring data of 302times from 109 NS patients with genetic information, and then verified. RESULTS In terms of efficacy, the median blood concentration of tacrolimus in the non-remission group was 2.20 ng/mL, which was significantly lower than that in the partial remission group(4.00 ng/mL, P<0.001) and the complete remission group(3.60 ng/mL, P=0.002). In terms of ADR, the median blood concentration of tacrolimus in the ADR group was 5.01 ng/m L, which was significantly higher than that in the non-ADR group(3.37 ng/m L)(P=0.001). According to the subgroup analysis of the receiver operating characteristic curve, when the blood concentration of tacrolimus was ≥6.65 ng/m L, patients were more likely to develop elevated blood creatinine [area under the curve(AUC) was 0.764, P<0.001); when the blood concentration of tacrolimus was ≥6.55 ng/m L, patients were more likely to develop blood glucose(AUC=0.615, P=0.005). The established MLP prediction model has a loss function of 0.9, with an average absolute error of 0.279 5 ng/m L between the predicted and measured values. The determination coefficient of the validation scatter plot was 0.984, indicating an excellent predictive performance of the model. CONCLUSION Tacrolimus blood concentration has an impact on both efficacy and ADR in NS patients. The use of the MLP model for predicting blood concentration exhibits high accuracy with minimal error between predicted and measured values. The model can be used as an important tool in clinical individualized medication regimens.更多还原