A comparative analysis of multi-level computer-assisted decision making systems for traumatic injuries

Area of Science:

• Medical Informatics
• Machine Learning Applications
• Clinical Decision Support Systems

Background:

• Trauma experts face critical decisions with numerous patient variables under time constraints.
• Developing reliable computer-assisted decision-making systems is crucial for optimizing treatment in trauma care.
• This research focuses on traumatic brain injury (TBI) and traumatic pelvic injuries.

Purpose of the Study:

• To compare existing machine learning methods for medical informatics.
• To develop reliable, rule-based computer-assisted decision-making systems for trauma treatment recommendations.
• To utilize previously recorded trauma case data for predictive modeling.

Main Methods:

• Machine learning techniques, specifically Classification and Regression Trees (CART) and C4.5, were used to extract decision rules.
• Rule sets were tested against other methods like AdaBoost and Support Vector Machines (SVM).
• Logistic regression was employed for variable filtering prior to rule discovery.

Main Results:

• CART demonstrated superior performance for survival prediction when all variables were considered.
• Neural networks performed best when using only significant variables.
• A combined C4.5/CART system achieved high accuracy (89.7%) in predicting patient outcomes and (93.1%) ICU length of stay for TBI patients.

Conclusions:

• An efficient computer-aided, rule-based system for TBI decision-making was developed.
• Combining CART and C4.5 with logistic regression enhances rule performance and quality.
• The developed rule-bases outperform systems using all variables for TBI outcome prediction.