Extreme restriction design as a method for reducing confounding by indication in pharmacoepidemiologic research