The horizontal and vertical nature of patient phenotype retrieval: new directions for clinical text processing

The author reviews the historical problem of identifying appropriate patients for retrieval from a clinical repository of patient records, compares the competing features of document classification and natural language processing, and proposes an alternative approach. The alternative approach 1) codes inquiries in an ontology to lend a vertical axis to retrieval knowledge instead of coding the target body of notes, 2) invokes natural language indexing and lexical normalizations on the corpus of notes that is scalable and tractable, and 3) leverages thesauri of word-level synonyms and near-synonyms to expand term searches "horizontally" around the concept spaces drawn from the ontology in which the queries were "coded."