Acute kidney injury (AKI) is a heterogeneous clinical syndrome associated with substantial mortality risk. Conventional prognostic models may have limited ability to provide individualized risk estimation because AKI patients differ widely in baseline characteristics, illness severity, laboratory profiles, and treatment exposures. This study aimed to develop a personalized survival prediction model for AKI patients using an evolutionary learning-based Cox proportional hazards approach, named EL-AKICox, to enable individualized survival estimation, risk stratification, and visualized survival curves for clinical decision support.

Data were obtained from the retrospective data repository, which included 291,945 records collected between 2017 and 2022. From this database, 30,271 AKI samples were identified for model development and evaluation. The original dataset contained 45 variables, including demographic characteristics, laboratory measurements, and medication-related information. After feature engineering, 89 candidate variables were considered, including 44 derived features. The EL-AKICox algorithm was applied to select a compact but informative feature set and construct a survival prediction model with optimized prognostic performance.

The EL-AKICox model selected 29 prognostic factors and achieved C-index values of 0.732 in the training set, 0.720 in the test set, and 0.736 in the imputed test set. Based on model-derived risk scores, AKI patients were stratified into high-risk and low-risk groups, allowing individualized estimation of survival probability over time. Among the selected predictors, five clinically relevant factors were identified as important contributors to risk estimation. By integrating these variables with individualized risk scores and survival curves, the model may help clinicians recognize high-risk AKI patients and support more targeted monitoring and intervention strategies.

The EL-AKICox model provides individualized survival prediction for AKI patients using a reduced set of informative clinical features. Its ability to generate personalized risk scores, survival curves, and risk-group classification may enhance prognostic assessment and support clinical decision-making in AKI care. Further prospective validation is warranted to confirm its generalizability and clinical utility. Part of these results was previously presented as poster at the Taiwan Society of Nephrology meeting in 2025.