Citation

Somboon S, Phunghassaporn N, Tansawet A, Lolak S. Asian J. Surg. 2021; ePub(ePub): ePub.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.asjsur.2021.09.010

PMID

34657787

Abstract

Accidents on the road are problems affecting many countries worldwide. There are up to 12,000 casualties from traffic accidents per year in Thailand, or about 1-2 people dying per hour.1 Therefore, it is important to triage and prioritize which patients to treat first, and the ability to predict the outcome and survivability rate of each patient will be beneficial. Unfortunately, out of the most commonly used scores in Thailand, only Trauma and Injury Severity Score (TRISS) has been properly validated.2 This study explored the possibility of using machine learning (ML) to better predict and compute the mortality rate among road traffic injury patients in Thailand.

Data were obtained from Vajira's road traffic injury cohort resulting in 1033 subjects from 1st January 2015-31st December 2018. The outcome was in-hospital mortality within 30 days of hospitalization. The model, named Death from Road Injury Prediction (DRIP), was constructed based on logistic regression (LR) using the scikit-learn library in python. First, data were split into a training set and test set in an 80:20 fraction. Since the dataset was highly imbalanced, a Synthetic Minority Oversampling Technique (SMOTE) algorithm was applied to up-sample a 'death' group in the training set to 30% while maintaining the original fraction in the test set. Next, each numerical variable was standardized, then recursive features elimination (RFE) was used for variable selection. The C hyperparameter was optimized by searching for the best hyperparameter with grid-search, and evaluated by five-fold cross-validation. The predictive power of DRIP model was assessed using the area under the ROC curve (AUC) with a 95% confidence interval (95% CI), sensitivity, specificity, accuracy, and f1-score.

Sex, SBP, DBP, HR, RR, and GCS were selected as significant factors and used in...

Language: en

Keywords

Traffic accidents; Machine learning; In-hospital mortality