A proof-of-concept study on mortality prediction with machine learning algorithms using burn intensive care data.

Area of Science:

• Medical Informatics
• Computational Biology
• Trauma Surgery

Background:

• Burn injuries are a significant cause of mortality, necessitating accurate prediction of fatal outcomes.
• Traditional scoring systems like the Baux score (age, burn size) and revised Baux score (including inhalation injury) have limitations.
• Machine learning (ML) offers a data-driven approach to improve predictive accuracy by considering more variables.

Purpose of the Study:

• To evaluate the efficacy of various ML algorithms in predicting mortality for burn patients.
• To compare the predictive performance of ML models against the established Baux and revised Baux scores.

Main Methods:

• Collected admission data and mortality outcomes from 92 burn patients treated between 2002 and 2019.
• Selected 17 prognostic variables and trained multiple ML algorithms (Decision Tree, Extreme Boosting, Random Forest, SVM, GLM).
• Assessed model performance using Area Under the Receiver Operating Characteristic Curve (AUC) and compared with Baux scores via DeLong test.

Main Results:

• ML models demonstrated strong predictive performance, with AUCs ranging from 0.83 to 0.92.
• The Baux score and revised Baux score achieved AUCs of 0.85 and 0.84, respectively.
• No statistically significant difference was found between the predictive accuracy of ML algorithms and the Baux scores.

Conclusions:

• This study provides initial evidence for the utility of ML algorithms in predicting burn patient mortality.
• The findings suggest ML can achieve predictive performance comparable to existing clinical scores.
• Larger sample sizes and prospective validation are recommended for future research and clinical implementation.