Detection of practice pattern trends through Natural Language Processing of clinical narratives and biomedical literature.

Clinical knowledge, best evidence, and practice patterns evolve over time. The ability to track these changes and study practice trends may be valuable for performance measurement and quality improvement efforts. The goal of this study was to assess the feasibility and validity of methods to generate and compare trends in biomedical literature and clinical narrative. We focused on the challenge of detecting trends in medication usage over time for two diseases: HIV/AIDS and asthma. Information about disease-specific medications in published randomized control trials and discharge summaries at NewYork-Presbyterian Hospital over a ten-year period were extracted using Natural Language Processing. This paper reports on the ability of our semi-automated process to discover disease-drug practice pattern trends and interpretation of findings across the biomedical and clinical text sources.