Prediction model for seat position identification using patient profiles, bone fracture, and dislocation characteristics in a collision and non-collision motorcycle accident in Thailand

Abstract

Motorcycle accidents accounted for the most common prevalence of Road traffic collision (RTC). Therefore, identifying the rider or passenger is crucial for ensuring fairness. However, patients who suffered from RTC frequently could not provide any information due to loss of consciousness, memory loss, or death. We aim to develop two separately multivariable prediction models based on the differences in the facture pattern and demographic data between the rider and passenger in collision with another vehicle and non-collision motorcycle accident. A total of 1,816 patients had fractures from motorcycle accidents. 1,583 and 233 were riders and passengers, respectively. After a multivariable logistic regression with stepwise backward elimination, six final predictors, including Age, sex, femur fracture, wrist and hand fracture, leg including ankle fracture, and pelvis and lumbar spine fracture, were required for the final models. The prediction models had an acceptable level of discrimination (auROCs of 0.79 and 0.77 for collision and non-collision accidents, respectively) and appeared well-calibrated. Overall, the prediction model is potentially useful as an assisting tool for identifying seat positions.