Classifying Encounter Notes in the Primary Care Patient Record (original) (raw)

The ability to automate the assignment of primary care medical diagnoses from free-text holds many interesting possibilities. We have collected a dataset of free-text clinical encounter notes and their corresponding manually coded diagnoses and used it to built a document classifier. Classifying a test set of 2,000 random encounter notes yielded a coding accuracy rate of 49.7 %. Automated coding of primary care encounter notes is a novel application area, and though imperfect our method proves interesting enough to warrant further research.