Extracting Patient Profiles from Patient Records and Online Literature

We present a representation model for the content of medical documents (journal articles and the patient’s record) that allows the extraction of critical relationship information from online texts and tabular data. Our model relies on a list of attributes with associated values that are dynamically determined using an efficient finite-state grammar and an automatic term verifier. Extracted relationships are partitioned in different subsets according to the particular group of patients they refer to, thus enabling the retrieval of multitopic documents and the targeted selection and presentation of a portion of a document’s information. We present results from a system implementing this representation model and contrast our approach with traditional information retrieval and information extraction techniques.