Metabolic control may alter antithrombin III activity but not its plasma concentration in diabetes: a possible role for nonenzymatic glycosylation.