Construction and evaluation of an aspirin resistance risk prediction model for ischemic stroke.

BACKGROUND

Aspirin has become the drug of choice for the prevention and treatment of ischemic stroke (IS), but approximately a quarter of patients may be resistant to its effects and have an increased risk of recurrent ischemic events while also developing aspirin resistance. This study aimed to build a risk prediction model for aspirin resistance (AR) in IS patients, predicts the likelihood of IS patients developing AR.

METHODS

The retrospective research study included the clinical data of patients with ischemic stroke were retrospectively collected from January 2021 to January 2023 at the Affiliated Hospital of Beihua University in the Jilin Province. Univariate and logistic regression analyses were used to construct a risk prediction model. The Hosmer-Lemeshow χ test and a receiver operating characteristic (ROC) curve were used to check the differential validity and calibration of the risk prediction model. The AR risk assessment criteria for ischemic stroke were established based on the β values ​​of each risk factor and its variable types in the prediction model. The two evaluation criteria were compared and analyzed to determine the best criteria.

RESULTS

A total of 285 patients participated in this study, of whom 206 did not have AR, while 79 had AR. Seven risk factors were included in the prediction model. Sex (female), age (≥ 60 years), smoking, diabetes mellitus (DM), hyperlipidemia (HLP), platelets (PLT), > 350 × 10 g/L, and glycosylated hemoglobin (HbA1c) > 6.5% were independent influencing factors for the occurrence of AR in IS. The area under the ROC curve (AUC) of the risk score model in the training group was 0.834 (0.772-0.896, P < 0.001). The Hosmer-Lemeshow test predicted the model fit effect χ = 9.979, P = 0.267 > 0.05. In the validation group, the AUC was 0.819 (0.715-0.922, P < 0.001). The Base score model showed higher PPV (86.1%), the β × 4 model had better NPV (83.4%) with fewer false negatives (39), β × 4 showed slightly higher accuracy (82.8% vs 81.4%), its primary strength lies in enhanced AR detection sensitivity. Using the β value × 4 partial regression coefficient method, the scores and stratification of the AR risk prediction model were divided into three groups: no risk (0-3 points), low risk (4-15 points), and high risk (16-36 points).

CONCLUSIONS

Gender (female), age, smoking, DM, HLP, PLT and HbA1c are independent risk factors for AR in IS. The AR risk prediction model for IS demonstrates strong predictive and discriminative performance, enabling precise identification of high-risk patients.