Predicting acute kidney injury in trauma using an extreme gradient boosting model

Area of Science:

• Medical Informatics
• Nephrology
• Trauma Surgery

Background:

• Acute kidney injury (AKI) is a frequent and challenging complication in trauma patients.
• Early identification of AKI is crucial for timely intervention and improved patient prognosis.

Purpose of the Study:

• To develop and validate an extreme gradient boosting (XGBoost) model for predicting AKI in trauma patients.
• To forecast AKI occurrence 24 and 48 hours in advance.

Main Methods:

• Analysis of 17,859 trauma patient records from a regional trauma center (January 2015 - July 2023).
• Development and validation of XGBoost models using demographic, clinical, and laboratory data.
• Performance evaluation via Area Under the Receiver Operating Characteristic Curve (AUROC) and feature importance analysis (SHapley Additive exPlanations).

Main Results:

• The XGBoost models achieved high predictive performance, with AUROCs ranging from 0.864 to 0.904.
• Accurate prediction of AKI stages 1-3 and stages 2-3 at both 24 and 48-hour intervals.
• Key predictors included serum creatinine, injury severity score, age, lactate dehydrogenase, D-dimer, platelets, albumin, and C-reactive protein.

Conclusions:

• XGBoost models demonstrate significant potential for early AKI prediction in trauma patients.
• Clinical data can be effectively utilized to forecast AKI up to 48 hours prior to onset.
• This predictive capability can facilitate proactive management strategies for trauma-induced AKI.