Constructing risk adjustment models for percutaneous coronary intervention: implications for quality assessment