Development and Validation of a Machine Learning-Based Risk Prediction Model for Acute Kidney Disease After Cardiac Surgery

Renal injury after cardiac surgery is one of the common complications with high incidence rate, high risk of death and progression to chronic kidney disease (CKD). Previous evaluations of perioperative renal function mainly focused on acute kidney injury (AKI) related to cardiac surgery within seven days after surgery. The newly proposed concept of acute kidney disease (AKD) in recent years refers to acute or subacute kidney injury lasting seven to ninety days. Research has found that AKD can occur after AKI or in patients without AKI, and the two are both related and independent of each other, possibly indicating different subtypes of kidney injury. AKD is not uncommon and is a more significant predictor of mortality and end-stage kidney disease (ESKD). Therefore, AKD may be an important window for identifying and managing high-risk patients after cardiac surgery. Due to limited research on AKD after cardiac surgery, the risk factors for AKD are currently unclear, and there are no clinically practical and effective risk stratification tools available. This study aims to establish a multimodal perioperative data platform through a retrospective cohort, and use machine learning methods to construct a risk prediction model for AKD after cardiac surgery. The accuracy and stability of the model will be validated in a prospective study cohort, and an online risk prediction and clinical decision-making tool will be developed to help clinicians quickly conduct personalized risk assessments and optimize diagnosis and treatment strategies, thereby improving patient prognosis and reducing medical costs.