目的 探讨多层感知机（ MLP）预测模型和他克莫司血药浓度监测在肾移植患者的临床应用价值。方法 收集亳州市人民医院2023年1月—2025年3月185例进行989例次他克莫司血药浓度监测的肾移植患者，分析其临床疗效和安全性。选取其中78例进行CYP3A5^*3基因多态性检测的198例次血药浓度监测数据，建立MLP血药浓度预测模型。结果 临床疗效方面，高治疗窗组［谷浓度（ C_min） ＞ 15 ng·mL^-1］平均血药浓度（ 18.82±4.42） ng·mL^-1；治疗窗组（ 5 ng·mL^-1≤C_min≤15 ng·mL^-1）平均血药浓度（ 7.86±3.48） ng·mL^-1；低治疗窗组（ C_min＜5 ng·mL^-1）平均血药浓度（ 2.85±2.16） ng·mL^-1，达标比例65.42%。安全性方面，不良反应发生组的血药浓度范围高于未发生组。多重线性回归分析显示，患者年龄、体质量、给药剂量、肌酐和总胆红素是影响他克莫司血药浓度的主要相关因素。MLP预测模型的损失函数为0.82，预测值与实测值平均误差绝对值为0.162 ng·mL^-1，相关系数R²为0.985，具有良好的预测效果。结论 血药浓度监测影响肾移植患者的临床疗效和安全性，MLP预测模型预测血药浓度准确率高，预测值与实测值之间误差小，效果理想，可辅助临床他克莫司个体化用药。

Objective To explore the clinical value of multilayer perceptron (MLP) prediction model and tacrolimus (FK506) blood concentration monitoring in kidney transplant patients. Methods Concentration monitoring data of FK506 from 185 kidney transplant patients at People’s Hospital of Bozhou were collected from January 2023 to March 2025, and the efficacy and adverse reactions of the monitoring results were analyzed. Using 198 concentration monitoring data from 78 kidney transplant patients with CYP3A5^*3 genetic information, an MLP concentration prediction model was established. Results In terms of clinical efficacy, the average therapeutic drug concentration of high therapeutic window group (C_min> 15 ng·mL^-1) was (18.82 ±4.42) ng·mL^-1, (7.86 ±3.48) ng·mL^-1 in the therapeutic window group (5 ng·mL^-1 ≤ C_min ≤ 15 ng·mL^-1) and (2.85 ±2.16) ng·mL^-1 in the low therapeutic window group (C_min < 5 ng·mL^-1). The proportion of meeting the standards was 65.42%. In terms of security, the concentration range of the occurring group was higher than that of the non occurring group. Multiple linear regression analysis showed that patient age, body weight, drug dose administered, Cr and TBIL were the main correlated factors affecting FK506 plasma concentration. The MLP prediction model has a loss function of 0.82, an absolute average error of 0.162 ng·mL^-1 between the predicted value and the measured values, and a scatter plot correlation coefficient R² of 0.985. The model has achieved good prediction performance. Conclusion Blood concentration monitoring has an impact on the clinical efficacy and Security of kidney transplant patients. The use of MLP model has high accuracy in predicting concentration, small error between predicted values and measured values, and ideal results which can assist clinical FK506 individualized medication.

R983

安徽省重点研究与开发计划项目（2022e07020066）；亳州市卫生健康科研项目（bzwj2023b001）