Abstract: FR-PO039
Using Machine Learning to Predict AKI in Trauma Patients: A Single Trauma Center Study with Temporal Validation
Session Information
- AKI: Epidemiology, Risk Factors, and Prevention - 2
October 25, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 101 AKI: Epidemiology, Risk Factors, and Prevention
Authors
- Lim, Sunggyul, Dankook University Hospital, Cheonan, Chungcheongnam-do, Korea (the Republic of)
- Yi, Yongjin, Dankook University Hospital, Cheonan, Chungcheongnam-do, Korea (the Republic of)
Background
Acute kidney injury (AKI) is a common and serious complication among trauma patients, associated with increased mortality and longer hospital stays. Early identification of AKI in this population is difficult due to its complex causes and the unique characteristics of trauma patients.
Methods
We conducted a retrospective study using electronic health records (EHRs) from a trauma center to develop and validate with extreme gradient boosting (XGBoost) predictive model for predicting AKI. The model was trained on data from Jan 2015 to Jul 2021 and validated on data from Aug 2021 to Jul 2023, internally. Model performance was assessed using the AUROC, and feature importance was evaluated using SHapley Additive exPlanations values.
Results
A total of 11,687 patients (9,063 in the development set and 2,624 in the validation set) were included. The incidence of AKI was 6.6% in the development and 5.4% in the validation group. The models showed AUROCs of 0.864 and 0.886 for predicting AKI stages 1-3 and stages 2-3 at 48 hours, and AUROCs of 0.904 and 0.903 for predicting AKI stages 1-3 and stages 2-3 at 24 hours. Important features influencing model predictions included in-hospital creatinine, age, and laboratory markers such as lactate dehydrogenase.
Conclusion
The machine learning models effectively predict AKI in trauma patients up to 48 hours in advance using readily available EHR data. The performance of this approach highlights the potential for machine learning to enhance predictive capabilities in trauma care.