XGBoost identified as optimal model, achieving average test AUC of 0.90 before synthetic minority oversampling technique
By Elana Gotkine HealthDay Reporter
WEDNESDAY, April 10, 2024 (HealthDay News) — Use of advanced machine learning tools can help predict postinjury complications among patients with lower-extremity fractures, according to a study published online April 10 in the Journal of Orthopedic Research.
Mostafa Rezapour, Ph.D., from the Wake Forest University School of Medicine in Winston-Salem, North Carolina, and colleagues explored the potential of gait analysis coupled with supervised machine learning as a predictive tool for assessing postinjury complications among patients with lower-extremity fractures. Participants with lower-extremity fractures at a tertiary academic center were identified and underwent gait analysis. The raw data were processed, emphasizing 12 essential gait variables. Several machine learning models were trained, tested, and evaluated.
The researchers identified XGBoost as the optimal model before and after synthetic minority oversampling technique (SMOTE) application. The model achieved an average test area under the receiver operating characteristic curve of 0.90 before using SMOTE, with average test accuracy of 86 percent. A key role was attributed to the duration between injury occurrence and initial gait analysis. Early aggressive physiological compensations were identified, followed by stabilization phases. A significantly higher readmission rate was seen for patients with underlying medical conditions; the complication rate was also higher in this group, but not significantly so.
“Our study has provided significant insights into the complex dynamics of orthopedic recovery, especially in patients with lower extremity fractures,” the authors write.
Abstract/Full Text (subscription or payment may be required)
Copyright © 2024 HealthDay. All rights reserved.
