PhD

dissertationThe electronic problem-oriented medical record was conceived to alleviate limitations of the paper-based medical record, and to improve its organization. The list of medical problems is at the heart of this problem-oriented record, and requires completeness, accuracy and timeliness to fulfill this central role. At Intermountain Health Care (IHC), a problem-oriented electronic medical record is being developed, and features a medical problem list at its core. This list is already in use in the outpatient setting, but is often incomplete, inaccurate and out-of-date. This issue is even more prominent for hospitalized patients. To help maintain a complete, accurate and timely problem list, I developed an Automated Problem List system using Natural Language Processing (NLP) to extract potential medical problems from the patient's electronic clinical documents. These problems are proposed to the user for inclusion in the "official" problem list, along with a link to allow viewing the documents the problem was extracted from. Two main applications compose this system. A background application does all documents processing and analysis using NLP, and the problem list management application allows viewing and editing these proposed problems. In the development of this system, the NLP module of the background application was evaluated first. This laboratory function study showed good recall and satisfying precision; accuracy was further improved by enhancing disambiguation and negation detection. A second study prospectively evaluated the whole Automated Problem List system in a clinical setting at the LDS Hospital. Patients benefiting from this system had more complete and timely problem lists. The sensitivity was higher, and the time between a medical problem's first mention in a clinical document and its addition to the list of problems was significantly reduced. In summary, this dissertation describes the planning, development, implementation and evaluation of a system using NLP to automatically extract medical problems from electronic clinical documents. This Automated Problem List system allowed better quality content of the problem list, opening doors to larger scale use of this system and contributing to possible answers to the challenge of making the problem list a cornerstone of our evolving clinical information system