Wang Ruo-Ying, Yan Shui-Di, Zeng Jian-Qi, Mu Tong, Yan Ya, Zhao Yuan-Yi, Xie Lin, Liu Li-Li
Center of Clinical Laboratory, School of Medicine, Zhongshan Hospital of Xiamen University, Xiamen University, Xiamen, China.
Department of Neurology, School of Medicine, Zhongshan Hospital of Xiamen University, Xiamen University, Xiamen, China.
Cardiovasc Toxicol. 2025 Jul 14. doi: 10.1007/s12012-025-10026-2.
Clopidogrel is extensively utilized for the prevention and treatment of cardiovascular, cerebrovascular, and other arterial circulation disorders attributed to platelet hyperaggregation. Nevertheless, its antiplatelet efficacy displays substantial individual variability and unpredictability. Our aim was to develop a machine learning model based on clinical data, incorporating various laboratory indicators, to predict the risk of clopidogrel resistance in clinical patients. This study included 1592 cardiovascular disease patients treated with clopidogrel. Potential predictive variables included age, sex, hematological, coagulation, biochemical parameters, and CYP2C19 genetic polymorphisms. Lasso regression and multivariable logistic regression were used for variable selection. Modeling was performed using Logistic Regression, LGBM Classifier, Random Forest Classifier, and SVC machine learning models, followed by model comparison, to ultimately construct the clopidogrel resistance risk prediction model. The clopidogrel resistance rate increased year by year from 2020 to 2022, but decreased slightly in 2023. There was a significant difference in clopidogrel resistance rate among different years (χ = 49.969, P = 0.000). Predictive variables included white blood cell count, hemoglobin level, platelet count, fibrinogen, triglycerides, D-Dimer, mean platelet volume, prothrombin time ratio, uric acid, glycated hemoglobin, and apolipoprotein B. The Random Forest Classifier machine learning method yielded a CR risk prediction model with AUC = 0.8730 and accuracy = 0.8033, demonstrating good predictive capability for identifying the risk of clopidogrel resistance after clinical use of clopidogrel. This study developed a highly predictive clopidogrel resistance risk prediction model, which can assist in clinical decision-making for better treatment strategies.
氯吡格雷被广泛用于预防和治疗因血小板过度聚集引起的心血管、脑血管及其他动脉循环障碍。然而，其抗血小板疗效存在显著的个体差异和不可预测性。我们的目的是基于临床数据，纳入各种实验室指标，开发一种机器学习模型，以预测临床患者氯吡格雷抵抗的风险。本研究纳入了1592例接受氯吡格雷治疗的心血管疾病患者。潜在的预测变量包括年龄、性别、血液学、凝血、生化参数以及CYP2C19基因多态性。采用Lasso回归和多变量逻辑回归进行变量选择。使用逻辑回归、LightGBM分类器、随机森林分类器和支持向量机（SVC）机器学习模型进行建模，随后进行模型比较，最终构建氯吡格雷抵抗风险预测模型。2020年至2022年氯吡格雷抵抗率逐年上升，但在2023年略有下降。不同年份的氯吡格雷抵抗率存在显著差异（χ = 49.969，P = 0.000）。预测变量包括白细胞计数、血红蛋白水平、血小板计数、纤维蛋白原、甘油三酯、D-二聚体、平均血小板体积、凝血酶原时间比值、尿酸、糖化血红蛋白和载脂蛋白B。随机森林分类器机器学习方法得出的氯吡格雷抵抗风险预测模型的曲线下面积（AUC）=0.8730，准确率=0.8033，表明该模型在临床使用氯吡格雷后识别氯吡格雷抵抗风险方面具有良好的预测能力。本研究开发了一种具有高度预测性的氯吡格雷抵抗风险预测模型，可协助临床决策以制定更好的治疗策略。
