Prediction of Sleep-Disordered Breathing in Patients with Ischemic Stroke

doi:10.3969/j.issn.1004-5775.2025.07.001

Abstract：Objective: To evaluate the predictive value of a flexible statistical model based on baseline characteristics for sleep-disordered breathing (SDB) in patients after ischemic stroke. Methods: A total of 177 patients with ischemic stroke who were treated in the Department of Neurology at Qianjiang Central Hospital from March to December 2022 and had stable conditions for 7–10 days were selected as the study subjects. Sleep monitoring was performed on all 177 patients with ischemic stroke. Using R software, gender, age, hypertension, diabetes, heart failure, atrial fibrillation, smoking, coronary heart disease, body mass index (BMI), waist circumference, neck circumference, stroke scale score, history of previous stroke, alcohol consumption, and hypercholesterolemia were used as predictors. The random forest classification algorithm and stepwise regression were applied to assess the risk factors for SDB after stroke and to develop a predictive model. The predictive value of three questionnaires—the Sleep Apnea Clinical Score (SACS), Berlin Questionnaire (BQ), and Epworth Sleepiness Scale (ESS)—for SDB after stroke was evaluated. Results: Patients with ischemic stroke combined with obstructive sleep apnea (OSA) had higher neck circumference, waist circumference, BMI, and SACS scores compared to those with ischemic stroke alone, with statistically significant differences (t=-4.695, -4.357, -4.739, x2=-4.282; P<0.001). There were no statistically significant differences in ESS or BQ scores between the two groups (x2=-2.356, -2.660, P>0.05). The minimum oxygen saturation was lower in the ischemic stroke combined with OSA group compared to the ischemic stroke group, with a statistically significant difference (x2=-4.369, P<0.05). The incidence of Cheyne-Stokes respiration was higher in the ischemic stroke combined with OSA group than in the ischemic stroke group, with a statistically significant difference (x2=131.155, P<0.05). The areas under the ROC curve (AUC) for SACS, BQ, and ESS in predicting SDB in patients with stroke were 0.700 (95% CI: 0.627–0.766, P<0.001), 0.611 (95% CI: 0.535–0.684, P=0.007), and 0.607 (95% CI: 0.531–0.680, P=0.015), respectively. BMI, waist circumference, age, neck circumference, and the National Institutes of Health Stroke Scale (NIHSS) accounted for 77.34% of all features. The final random forest model achieved an accuracy of 83.33%, precision (combined) of 83.33%, recall (combined) of 83.33%, and F1-score (combined) of 0.83 on the test set. Conclusion: SACS has a certain predictive value for SDB after ischemic stroke. Neck circumference, BMI, and a history of previous stroke are the main risk factors for SDB after ischemic stroke.