Effect of changing the stillbirth definition on evaluation of perinatal mortality rates