Abstract
BACKGROUND: Predicting the prognosis of patients with acute cerebral infarction (ACI) is crucial for clinical decision-making and personalized treatment. However, existing models often lack the comprehensive integration of clinical and biological indicators necessary for accurate and interpretable predictions. This study aims to develop and validate a predictive model using a combination of clinical assessments and inflammatory biomarkers to improve the prognostication of ACI patients. METHODS: This real-world, retrospective cohort study was conducted at Luhe Hospital, Beijing, and included 1,017 ACI patients admitted within 24 h of symptom onset. The dataset was randomly split into a training set (80%) and a validation set (20%). Twelve machine learning models were developed and evaluated, with the optimal model and feature set selected based on comprehensive performance metrics. To enhance interpretability, the Shapley Additive exPlanations (SHAP) method was employed to quantify and visualize the contribution of each feature to the model's predictions. RESULTS: The final model, utilizing the Logistic Regression (LR) algorithm, incorporated six key features: NIHSS at 24 h (NIHSS_24 h), NIHSS_change, D-dimer, neutrophil count (N), lymphocyte percentage at 24 h (L_pct_24 h), and length of stay (LOS). NIHSS_24 h emerged as a critical early prognostic indicator, effectively predicting three-month outcomes post-discharge. Inflammatory markers, including D-dimer, N, and L_pct_24 h, significantly enhanced the model's predictive performance. The SHAP method provided both global and local interpretability, elucidating the relative importance of each feature in the model's predictions. To facilitate clinical decision-making, a web-based application was developed for real-time prognostic assessment. CONCLUSION: This study developed a robust and interpretable predictive model for ACI prognosis by integrating clinical and inflammatory biomarkers. The model underscores the prognostic significance of NIHSS_24 h and inflammatory markers, highlighting the critical role of early assessment and personalized treatment strategies. Future research should focus on multi-center validation and the incorporation of additional predictive variables to further enhance the model's accuracy and generalizability.