Secondary use of clinical data for research requires a method to quickly process the data so that researchers can quickly extract cohorts. We present two advances in the High Throughput Phenotyping NLP system which support the aim of truly high throughput processing of clinical data, inspired by a characterization of the linguistic properties of such data. Semantic indexing to store and generalize partially-processed results and the use of compositional expressions for ungrammatical text are discussed, along with a set of initial timing results for the system.
Keywords: clinical NLP; compositional expressions; high throughput phenotyping.