Carriers of ≥1 loss-of-function CYP2C19 allele had a higher risk of clopidogrel resistance, which was accurately predicted by an XGBoost machine learning model with an AUROC of 0.9925.
Observational (n=208)
No
Does a machine learning model incorporating CYP2C19 genotypes and clinical factors predict clopidogrel resistance in Chinese patients with ischemic stroke?
A machine learning model incorporating CYP2C19 genotypes, D-dimer, platelet count, and total bilirubin can accurately predict clopidogrel resistance in Chinese patients with ischemic stroke.
Absolute Event Rate: 9.85% vs 1.32%
p-value: p=0.038
Background: Clopidogrel resistance (CR) may diminish its antiplatelet effect, thereby increasing the risk of cardiovascular and cerebrovascular events. The cause of CR remains unclear, and it may be related to pharmacogenomics and coagulation markers. Machine learning is a novel approach to investigate the correlations among various factors. This study aimed to investigate the factors influencing CR in Chinese patients with ischemic stroke and to develop a precise and reliable predictive model for CR using machine learning. Methods: Thromboelastography (TEG), a standard technique for assessing platelet inhibition, was used to measure the adenosine diphosphate (ADP)-induced platelet inhibition rate. CR was defined as an ADP-induced platelet inhibition rate of less than 30%. Genotypes of CYP2C19 and PON1 were identified using fluorescence in situ hybridization. The relationships between genotypes, laboratory indicators, and ADP-induced platelet inhibition rates or CR were examined. An extreme gradient boosting (XGBoost) machine learning method was applied to predict the occurrence of CR. Adaptive Synthetic technique was used for reliable data augmentation and the predictive model was internally validated via nested cross-validation. Results: A total of 208 patients were enrolled in the study. Participants were categorized into the CR group (n=14) and the non-CR group (n=194). The CR group exhibited significantly lower activated partial thromboplastin time (APTT) levels compared with the non-CR group (P< 0.05). Carriers of at least one loss-of-function (LOF) allele of CYP2C19 had a significantly higher risk of CR than individuals without LOF alleles. Other risk factors for ischemic stroke, such as age, sex, and body weight, did not significantly affect platelet inhibition rates or CR. Based on the XGBoost model, CYP2C19 genotype, D-dimer levels, platelet count, and total bilirubin were major contributors to the prediction of CR in Chinese patients with ischemic stroke. The area under the receiver operating characteristic curve was 0.9925± 0.0067. The model’s accuracy and sensitivity was 97.44% and 91.82%, respectively. Conclusion: Genetic polymorphisms in CYP2C19 are the primary factors influencing CR. A machine learning model may be useful for early prediction of CR and for guiding the rational use of clopidogrel. Keywords: CYP2C19 , genetic polymorphism, clopidogrel resistance, ischemic stroke, machine learning
Xu et al. (Wed,) conducted a observational in Ischemic stroke (n=208). Clopidogrel vs. Patients without loss-of-function CYP2C19 alleles was evaluated on Clopidogrel resistance (ADP-induced platelet inhibition rate < 30%) (p=0.038). Carriers of ≥1 loss-of-function CYP2C19 allele had a higher risk of clopidogrel resistance, which was accurately predicted by an XGBoost machine learning model with an AUROC of 0.9925.