A prospective cohort study on the incidence and influencing factors of subsyndromal delirium in ICU patients.

Background: This study aims to develop and validate a machine learning-based risk prediction model for subsyndromal delirium (SSD) in ICU patients, while identifying key risk factors. Method: This study was a prospective study, selecting patients who were hospitalized in the ICU from October 2024 to May 2025. We compared seven machine learning algorithms: Random Forest (RF), Decision Tree (DT), K-Nearest Neighbor (KNN), Logistic Regression (LR), Elastic Network (EN), Extreme Gradient Enhancement (XGB), and Support Vector Machine (SVM). Result: In our study, the prevalence rate of SSD was 37.158%. The comparative analysis shows that XGB is the best predictive model (AUC = 0.84). Feature importance analysis identified four significant predictive factors: Use of vasoactive drugs (0.412), Monthly household income (0.306), Undergone surgery (0.191) and Number of Medications (0.036). Conclusion: The prediction model based on XGB has a good effect in identifying the risk of SSD in ICU patients. These findings enable clinicians to stratify high-risk groups and implement timely and targeted intervention measures, effectively reducing the risk of adverse consequences. Future multicenter studies should validate these results in larger cohorts.