Gravar-mail: Estimating bias from loss to follow-up in a prospective cohort study of bicycle crash injuries