Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV

Background Wound healing is a critical determinant of recovery and quality of life in patients with diabetes, particularly those admitted to intensive care units (ICUs). Identifying the key factors influencing wound healing and optimizing treatment strategies is essential for improving outcomes. Despite prior studies, there is limited comprehensive analysis that integrates multiple risk factors into predictive modeling frameworks. Objective This study aims to identify the significant factors affecting wound healing in diabetic ICU patients, evaluate the effects of different treatment approaches on healing outcomes, and develop a robust predictive model to assist clinicians in early risk identification and personalized treatment planning. Methods We utilized data from the MIMIC-IV database, encompassing 149,392 patient records. Key variables analyzed included demographic characteristics, chronic disease histories, wound-related factors, and treatment modalities. Descriptive and uni-variate analyses were performed to explore baseline characteristics and their associations with healing outcomes. Cox proportional hazards regression and logistic regression models were used for multi-factorial analyses, while machine learning models such as Random Forest and XGBoost were employed for predictive modeling. Models interpretability was enhanced through SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) analyses. Results Factors such as age, the presence of pressure ulcers, chronic kidney disease, and treatment modalities (e.g., insulin therapy, negative pressure therapy) emerged as significant predictors of wound healing outcomes. Random Forest achieved the highest performance among predictive models, with an area under the receiver operating characteristic curve (AUC) of 0.96. SHAP analysis identified age and death flags as critical determinants, while LIME provided patient-specific insights into model predictions. Conclusions This study underscores the importance of integrating multifactorial data to predict wound healing outcomes in diabetic ICU patients. The findings provide actionable insights for personalized treatment strategies and resource allocation in clinical settings. Future research should focus on validating these models in diverse datasets and exploring longitudinal impacts on patient recovery.