Matching for several sparse nominal variables in a case-control study of readmission following surgery