LAUSR

Objective This study aims to investigate novel clinical risk factors for cognitive impairment (CI) in elderly. Methods A total of 3221 patients (259 patients with CI and 2,962 subjects without CI) were recruited into this nested case-control study who underwent cerebral magnetic resonance angiography (MRA) from 2007 to 2021. All of the clinical data with MRA imaging were recorded followed by standardization processing blindly. The maximum stenosis score of the posterior circulatory artery, including the basilar artery, and bilateral posterior cerebral artery (PCA), was calculated by the cerebral MRA automatic quantitative analysis method. Logistic regression (LR) analysis was used to evaluate the relationship between risk factors and CI. Four machine learning approaches, including LR, decision tree (DT), random forest (RF), and support vector machine (SVM), employing 5-fold cross-validation were used to establish CI predictive models. Results After matching with age and gender, 208 CI patients and 208 control subjects were finalized the follow-up (3.46 ± 3.19 years) with mean age at 84.47 ± 6.50 years old. Pulse pressure (PP) in first tertile (<58 mmHg) (OR 0.588, 95% confidence interval (CI): 0.362–0.955) was associated with a decreased risk for CI, and ≥50% stenosis of the left PCA (OR 2.854, 95% CI: 1.387–5.872) was associated with an increased risk for CI after adjusting for body mass index, myocardial infarction, and stroke history. Based on the means of various blood pressure (BP) parameters, the performance of the LR, DT, RF and SVM models accurately predicted CI (AUC 0.740, 0.786, 0.762, and 0.753, respectively) after adding the stenosis score of posterior circulatory artery. Conclusion Elderly with low pulse differential pressure may have lower risk for cognitive impairment. The hybrid model combined with the stenosis score of posterior circulatory artery, clinical indicators, and the means of various BP parameters can effectively predict the risk of CI in elderly individuals.