关键词: font-family:Inter, -apple-system, BlinkMacSystemFont, ", font-size:16px, background-color:#FFFFFF, ">脓毒症、糖尿病、老年患者、急性肾损伤、预测模型

Abstract: Objective: The objective is to develop a machine learning model that predicts acute kidney injury (AKI) in elderly diabetic patients with sepsis, aiding clinicians in more effectively identifying high-risk patients. Methodology: Based on the MIMIC-Ⅳ database, elderly diabetic patients with sepsis were selected, with acute kidney injury (AKI) as the outcome defining the patients’ condition. The Boruta algorithm, Logistic Regression, and Lasso Regression were used to identify model variables. Clinical prediction models were built using 10 machine learning algorithms, including Logistic Regression, Gradient Boosting Machine, Random Forest, K-Nearest Neighbors, Neural Networks, and Extreme Gradient Boosting (XGBoost). After hyperparameter optimization, the best model was selected for SHAP analysis. Results: The study ultimately included 5984 elderly diabetic patients with sepsis. Through the application of four variable selection algorithms, ten indicators were identified for modeling: APSⅢ score, SOFA score, mechanical ventilation use, weight, lymphocyte count, serum lactate, systolic blood pressure, pH, arterial oxygen pressure, and the use of vasopressors. The XGBoost model achieved areas under the curve (AUC) of 0.838 and 0.804 for the training and validation sets, respectively. Decision curve analysis and calibration curves confirmed that the model provided significant net clinical benefit and exhibited good prediction stability. SHAP analysis revealed that the APSⅢ score made the highest contribution to the model. Given the strong clinical accessibility of the identified indicators, the XGBoost model can quickly identify the probability of AKI risk in elderly diabetic sepsis patients. Conclusion: A risk prediction model for AKI in elderly diabetic patients with sepsis was developed and validated based on the XGBoost algorithm. The model demonstrated strong predictive performance and stability, providing valuable support in assessing risk factors and stratifying treatment for these patients, thereby optimizing resource allocation and improving patient outcomes.

Key words: font-family:Inter, -apple-system, BlinkMacSystemFont, ", font-size:16px, background-color:#FFFFFF, ">sepsis, diabetes mellitus, elderly patients, acute kidney injury, prediction model