Field methods in medical record abstraction: assessing the properties of comparative effectiveness estimates