A joint latent class analysis for adjusting survival bias with application to a trauma transfusion study