Causal inference with missing exposure information: Methods and applications to an obstetric study