A new survival status prediction system for severe trauma patients based on a multiple classifier system

Area of Science:

• Medical informatics
• Machine learning in healthcare
• Trauma care research

Background:

• Severe trauma necessitates accurate survival prediction for quality assessment.
• Current prediction systems rely on single models, limiting accuracy.
• Objective comparison of trauma center quality requires robust prediction tools.

Purpose of the Study:

• To develop and evaluate a multiple classifier system for severe trauma patient survival prediction.
• To improve upon the performance of existing single-model prediction systems.
• To enhance the trade-off between sensitivity and specificity in trauma outcome prediction.

Main Methods:

• Combined logistic regression and C4.5 decision tree models into a multiple classifier system.
• Utilized the Navarre major trauma registry (462 patients).
• Applied 10x10-fold cross-validation with sensitivity, specificity, and geometric mean as performance metrics.

Main Results:

• Achieved sensitivity of 0.8908, specificity of 0.6703, and geometric mean of 0.7661.
• Demonstrated increased specificity and geometric mean compared to single models.
• Statistical analysis (Mann-Whitney U test) confirmed significant improvements (p < 0.01).

Conclusions:

• Multiple classifier systems offer superior performance for severe trauma survival prediction.
• The proposed system provides an improved balance between sensitivity and specificity.
• This methodology enhances the objective comparison of trauma center emergency services.