Risk stratification for in-hospital mortality in sepsis-associated acute kidney injury patients receiving continuous renal replacement therapy: an interpretable, externally validated machine learning study.

The gradient boosting machine model achieved an AUC of 0.890 for predicting in-hospital mortality in SA-AKI patients on CRRT, but its clinical utility may be limited by the study's retrospective design.

Summary and findings

This study developed and validated a prognostic model for patients with sepsis-associated acute kidney injury (SA-AKI) receiving continuous renal replacement therapy (CRRT). The model was based on data from 1,217 patients in the MIMIC-IV and eICU-CRD databases and validated in an independent cohort of 332 patients. The gradient boosting machine (GBM) model achieved an area under the curve (AUC) of 0.890 in the training cohort.

Abstract

Patients with sepsis-associated acute kidney injury (SA-AKI) requiring continuous renal replacement therapy (CRRT) have a high risk of in-hospital mortality, and early risk stratification may support timely clinical decision-making and efficient resource allocation. In this retrospective study, we developed and validated a prognostic model for SA-AKI patients receiving CRRT using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV version 3.1 United States, 2008-2022) and the eICU Collaborative Research Database (eICU-CRD United States, 2014-2015), with external validation in an independent cohort from the intensive care unit of the Second Affiliated Hospital of Anhui Medical University (AYEFY-ICU China, 2021-2024). Candidate variables were selected using the least absolute shrinkage and selection operator (LASSO) and Boruta algorithms, and eight machine learning models were constructed and compared. Model interpretability was assessed using SHapley Additive exPlanations (SHAP). A total of 1,217 patients from the MIMIC-IV and eICU-CRD databases and 332 patients from the AYEFY-ICU cohort were included, and ten predictors were ultimately identified. Among the evaluated models, the gradient boosting machine (GBM) showed strong performance, with AUCs of 0.890, 0.756, and 0.752 in the training, internal validation, and external validation cohorts, respectively. In the external cohort, its performance was comparable to XGBoost and LightGBM without significant differences, with overlapping confidence intervals, while exceeding conventional scores (SOFA, SAPS II). SHAP analysis identified urine output, serum creatinine, and age as key predictors. This multicenter-derived GBM model may support early risk stratification and clinical decision-making in SA-AKI patients receiving CRRT.

Elsewhere in the PT-141 (Bremelanotide) corpus