Prediction of Prognosis in Patients with Trauma by Using Machine Learning.

Area of Science:

• Medical Informatics
• Computational Biology
• Trauma Surgery

Background:

• Trauma patient outcomes require accurate prediction for effective resource allocation.
• Machine learning offers potential for analyzing complex clinical data to predict patient trajectories.
• Existing predictive models may not fully capture the nuances of trauma recovery and long-term needs.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting mortality and chronic care needs in trauma patients.
• To identify the most accurate machine learning model for trauma outcome prediction using historical clinical data.

Main Methods:

• Collected clinical data from 5871 trauma patients admitted in 2015-2016.
• Applied various machine learning techniques, including eXtreme Gradient Boosting (xGBT), to the patient dataset.
• Utilized cross-validation to select the model with the highest predictive accuracy.

Main Results:

• Developed two xGBT models: a complete model and a short-term emergency department (ED) model.
• The complete model achieved 86% recall for recovery, 30% for chronic care, and 67% for mortality.
• The short-term ED model showed 89% recall for recovery but only 41% for mortality and 25% for chronic care.

Conclusions:

• The developed machine learning algorithm demonstrates strong predictive capability for healthy recovery in trauma patients.
• The algorithm's performance is currently unsatisfactory for predicting chronic care requirements and mortality risk.
• Future improvements may involve incorporating age classification, severity scoring, and variable calibration to enhance predictive power.