Dezon K. Finch, PhD, MA
My education was fortified with additional coursework in programming, statistics, data base systems and machine learning. My experience over the past 15 years in the VHA includes several studies that utilize my education and skills. Specifically, in a five-year study funded by NIH and HSR I developed the NLP to identify and quantify empirically derived, key dimensions of Pain Care Quality in Veterans with musculoskeletal pain. I extracted pain management data from structured fields (coded) and unstructured fields (clinician text notes) in the electronic health record to assess factors associated with Pain Care Quality in a nationally representative sample of veterans with musculoskeletal pain and to determine whether VHA facilities that have adopted the SCM-PM provide higher quality pain care. I developed a rules-based NLP system and applied it to over two million notes to assess the quality of pain care. I am still developing this effort through the PMOP project to compare pain care quality across various clinic types. The current version of the vocabulary includes over 18,000 entries and includes coverage for CIH treatments and their results. In other work I have led NLP development efforts including two on which I was PI. Both are studies that evaluate adherence to treatment guidelines. Treatments and symptoms are extracted from medical progress notes and compared to treatment guidelines. I achieved an F-measure of .92 across all concepts using a custom parser with section detector. I am skilled at uncovering new features for the use of machine learning for a wide variety of classification tasks and choosing the best algorithm for the task. I recently developed a sentence classification model using deep learning to identify communication in secure messages. Previously, I have derived and tested over 100 features that describe lines of text in medical progress notes which has proven useful in detecting sections in documents, detecting templates and extracting information from them and other types of semi-structured text elements. I collaborated across VA facilities starting with the Consortium for Healthcare Informatics and currently with investigators from Vanderbilt, Yale, and University of Utah.
Contact: Dezon.Finch@va.gov
