Sepsis-associated acute kidney injury (SA-AKI) is a common, high-mortality complication among critically ill patients. Traditional diagnostic criteria for AKI rely on functional markers that often lag behind the onset of renal injury. Because immune dysregulation is central to the pathogenesis of SA-AKI, incorporating immune phenotypes into early risk stratification may improve prediction and facilitate earlier intervention. The present study aimed to identify immuno-inflammatory signatures associated with new-onset SA-AKI and to develop a machine learning–based prediction model that integrates immune biomarkers with routine clinical variables.
The analysis used data from a multicenter prospective cohort of 1,551 patients with sepsis. Enrollment occurred across five tertiary hospitals in Beijing. Clinical information and an array of immuno-inflammatory biomarkers were collected within 24 hours of sepsis diagnosis. Biomarkers assessed included humoral measures, complement components, and T lymphocyte subset profiles. The investigators then applied machine learning methods to derive predictive models for SA-AKI using these early clinical and immunologic data.
In this cohort, new-onset SA-AKI developed in 44.8% of enrolled septic patients. The occurrence of SA-AKI was associated with increased mortality relative to patients who did not develop AKI. The study used these outcomes to evaluate the discriminative value of immuno-inflammatory markers and combined clinical-immune prediction approaches.
A distinct high-risk immuno-inflammatory endotype associated with SA-AKI was identified. Key features of this endotype included:
These immune alterations indicate simultaneous humoral/complement changes and divergent T-cell subset dynamics in patients at high risk for SA-AKI. The characterization suggests that both innate and adaptive immune derangements contribute to renal vulnerability in sepsis.
The investigators developed an Extreme Gradient Boosting (XGBoost) model to predict new-onset SA-AKI. The integrated clinical-immune model combined the identified immune biomarkers with routine clinical variables. The clinical variables explicitly reported as included in the model were APACHE II, the SOFA score, procalcitonin (PCT), prothrombin time (PT), and sex. The model leveraged the complementary information from early immune phenotyping and established severity scores to improve risk stratification for SA-AKI.
Model discrimination favored the integrated clinical-immune approach. The XGBoost-based clinical-immune model achieved an area under the receiver operating characteristic curve (AUC) of 0.914. By comparison, the clinical-only model had an AUC of 0.788. Decision curve analysis demonstrated greater net clinical benefit for the integrated model relative to the clinical model, supporting potential utility in clinical decision making. The source reports these comparative performance metrics but does not provide additional internal validation statistics, calibration measures, or external validation details in the abstract.
The study's findings emphasize the role of immune dysregulation in the development of AKI among patients with sepsis. An immuno-inflammatory endotype characterized by elevated IgA and PCT, C3 depletion, and divergent T-cell subset changes identified a subgroup at high risk for SA-AKI. Integrating immune biomarkers with routine clinical severity measures in an XGBoost model substantially improved discrimination for new-onset SA-AKI (AUC 0.914) versus a clinical-only model (AUC 0.788), and decision curve analysis indicated higher net benefit for the integrated model.
The authors conclude that early immune phenotyping can meaningfully augment early identification of patients at elevated risk for SA-AKI. Details on model calibration, handling of missing data, feature selection procedures beyond the listed predictors, and external validation were not reported in the abstract. Further publication of the full methods and results would be required to assess implementation feasibility, generalizability, and potential integration into clinical workflows.