The use of propensity score stratification and synthetic data to address allocation bias when assessing bicycle helmet effectiveness

Abstract

Case-control studies have found bicycle helmet use significantly mitigates the risk and severity of head injury in a motor vehicle collision. However, critics argue the decision to wear a helmet is confounded with other factors related to cycling safety such as cycling speed. If such an allocation bias exists, results from casecontrol studies may be invalid if confounding factors are ignored. Although allocation bias and bicycle helmet effectiveness is frequently mentioned in the literature, there is a paucity of research that has explored this relationship. This study aims to examine bicycle helmet effectiveness in a motor vehicle collision using the propensity score stratification method, which removes allocation bias from case (head injury) and control (no head injury) groups to allow for direct comparison of helmet effectiveness in reducing head injury. Due to privacy and data accessibility issues, synthetic data was created from a recently published Australian study of linked hospital and police data over a nine-year period. In a motor vehicle collision, helmet use was associated with factors that have been argued to influence estimates of helmet effectiveness; however, using propensity score stratification, there is no evidence these confounding factors influence estimates of helmet effectiveness.

Publication
2016 IRCOBI Conference Proceedings - International Research Council on the Biomechanics of Injury