Enhancing clinical decision-making in closed pelvic fractures with machine learning models

Area of Science:

• Orthopedic Surgery
• Trauma Management
• Medical Informatics

Background:

• Closed pelvic fractures pose significant risks, including hemodynamic instability and mortality.
• Accurate prediction of these complications is vital for effective clinical management.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting hemodynamic instability and mortality in patients with closed pelvic fractures.
• To identify key clinical risk factors associated with these adverse outcomes.

Main Methods:

• A retrospective study of 133 patients with closed pelvic fractures.
• Utilized machine learning algorithms including Logistic Regression, Random Forest, and others.
• Analyzed 40 clinical variables and performed factor analysis.

Main Results:

• Random Forest (RF) and Logistic Regression (LR) models demonstrated superior predictive performance compared to traditional methods.
• RF model achieved an AUC of 0.92, accuracy 0.86, precision 0.81, and F1 score 0.87.
• Identified TILE grade, heart rate, creatinine, WBC, fibrinogen, lactic acid, and ISS as significant risk factors.

Conclusions:

• RF and LR algorithms are effective tools for predicting hemodynamic instability and mortality in closed pelvic fractures.
• These models can enhance clinical decision-making and improve patient outcomes.
• Lactic acid levels >3.7 and ISS >13 are significant predictors of adverse events.