A polytomous conditional likelihood approach for combining matched and unmatched case-control studies.