I expect you’ll be interested in these: UMLS::Similarity, UMLS::Interface, and a general search of the CPAN in the UMLS::* space. It's not trivial to set up but the instruction are good. Unfortunately there are no free medical stemming dictionaries I'm aware of. This really helps the kind of comparison task you're doing. If you know or learn of any, please do share. There are other approximate matching packages. Some like the metaphone stuff are probably too fuzzy for you and others like String::Approx are probably too slow.

You might want to index all your data first and use something like Lucy (to do stemming, tokenizing, and various normalization once instead of per search/match); or possibly resort to specialized data containers like Judy.