Individualized prediction of stroke-associated pneumonia for patients with acute ischemic stroke

Abstract

Background: Stroke-associated pneumonia (SAP) remains a neglected area despite its high morbidity and mortality. We aimed to establish an easy-to-use model for predicting SAP. Methods: Two hundred seventy-five acute ischemic stroke (AIS) patients were enrolled, and 73 (26.55%) patients were diagnosed with SAP. T-test, Chi-square test and Fisher’s exact test were used to investigate the associations of patient characteristics with pneumonia and its severity, and multivariable logistic regression models were used to construct a prediction scale. Results: Three variables with the most significant associations, including age, NGT placement, and right cerebral hemisphere lesions combined with gender, were used to construct a stroke-associated pneumonia prediction scale with high accuracy (AUC = 0.93). Youden index of our SAP prediction model was 0.77. The sensitivity and specificity of our SAP prediction model were 0.89 and 0.88, respectively. Conclusion: We identified the best predictive model for SAP in AIS patients. Our study aimed to be as clinically relevant as possible, focusing on features that are routinely available. The contribution of selected variables is visually displayed through SHapley Additive exPlanations (SHAP). Our model can help to distinguish AIS patients of high-risk, provide specific management, reduce healthcare costs and prevent life-threatening complications and even death.