Retrospective study develops a prognostic model for AKI patients on CRRT, supporting clinical decision-making.
Key Points
The aim is to develop a prognostic model for in-hospital mortality in sepsis-associated acute kidney injury patients receiving continuous renal replacement therapy.
Retrospective analysis of 1,217 patients from two databases and 332 from an independent cohort.
Variables were selected using least absolute shrinkage and selection operator (LASSO) and Boruta algorithms.
Eight machine learning models were constructed and compared, with gradient boosting machine (GBM) chosen for validation.
GBM achieved AUCs of 0.890, 0.756, and 0.752 in respective cohorts.
Key predictors included urine output, serum creatinine, and age identified through SHAP analysis.
Performance of GBM was comparable to other methods (XGBoost, LightGBM) and exceeded conventional scores like SOFA and SAPS II.