Abstract Objective This study aimed to develop a predictive model for secondary infections in patients with severe or critical COVID-19 by a
Abstract Objective This study aimed to develop a predictive model for secondary infections in patients with severe or critical COVID-19 by analyzing clinical characteristics and laboratory indicators. Method A total of 307 patients with severe or critical COVID-19 admitted to Peking University Third Hospital from December 2022 to February 2023 were retrospectively analyzed, including 156 patients with secondary infection and 151 patients without secondary infection. The Boruta algorithm identified significant variables, and eight machine learning models were evaluated based on area under the curve (AUC) performance. The optimal model selected was further assessed, with model interpretability provided using SHapley Additive exPlanations (SHAP). Result Nine predictive factors were identified: Mechanical Ventilation, Procalcitonin (PCT), Interleukin-8 (IL-8), Interleukin-6 (IL-6), Blood Urea Nitrogen, Glucose, Creatine Kinase, Lactate Dehydrogenase, and Mean Platelet Volume (MPV). The random forest model demonstrated the best performance, with further evaluation showing an average AUC of 0.981 (CI 0.965–0.998) on the training set and 0.836 (CI 0.761–0.912) on the test set. SHAP analysis identified MPV, PCT, and IL-8 as the strongest predictors of secondary infections. Conclusion We developed an effective predictive model for secondary infection risk in severe COVID-19 patients using readily available clinical parameters, enabling early clinical intervention. This machine learning approach demonstrates potential for improving patient management. Clinical trial This study does not involve clinical trial interventions. Therefore, clinical trial registration was not applicable.
