The effect of labour induction on the risk of caesarean delivery: using propensity scores to control confounding by indication