Development and Validation of a Prediction Model for Respiratory Failure in Patients with Sepsis-Associated Acute Kidney Injury (SA-AKI) Within 48 Hours of Admission
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective: To identify patients with sepsis-associated acute kidney injury (SA-AKI) at high risk of respiratory failure within 48 hours of admission and enable timely intervention to improve patient prognosis. Methods: Data from SA-AKI patients admitted to Dongyang People’s Hospital between June 2012 and October 2024 were collected, including gender, age, and blood biochemical indicators at admission. Patients were randomly divided into training and validation groups. Independent risk factors for respiratory failure were identified in the training group, and a nomogram prediction model was developed. The model'sdiscriminative ability was assessed using the area under the receiver operating characteristic (ROC) curve (AUC), and its calibration was evaluated using the GiViTi calibration plot. Clinical effectiveness was examined using Decision Curve Analysis (DCA). The model was subsequently validated in the validation group. SOFA-based, NEWS-based, and various other machine learning models were also established and compared to the proposed model using DeLong’s test. Results: A total of 702 patients were included in the study. Independent risk factors for respiratory failure included D-dimer, lactate, pro-BNP, albumin, globulin, transcutaneous blood oxygen saturation, and pulmonary infection. The AUC values for the training and validation groups were 0.818 and 0.795, respectively, with calibration plot P-values of 0.973 and 0.864. The DCA curves for both groups indicated superior clinical utility compared to extreme scenarios. The SOFA model achieved AUC values of 0.583 (training group) and 0.763 (validation group), while the NEWS model had AUC values of 0.628 (training) and 0.618 (validation). DeLong’s test confirmed that the proposed model outperformed SOFA and NEWS models (P < 0.001). In the validation group, the AUC values for SVM, C5.0, XGBoost, and integrated models were 0.781, 0.757, 0.759, and 0.778, respectively, with comparable discriminative ability to the nomogram (P > 0.05). Conclusion: The nomogram developed in this study based on D-dimer, lactate, pro-BNP, albumin, globulin, transcutaneous blood oxygen saturation, and pulmonary infection was found to effectively predict respiratory failure risk in SA-AKI patients within 48 hours of admission.