| Popis: |
Clinical trials are instrumental in translating outcomes of scientific research into medical practice. Enrollment of patients that meet requirements of a significant sample size, within a desired time span, is limited by the speed and efficiency of screening patients for these trials. The current process of eligibility screening involves repeated reading of clinical notes to evaluate patients against intricate eligibility criteria. Significant time, human effort and financial resources are consumed to accomplish this task. This dissertation analyzes possible reasons that limit the efficiency of the current clinical trial screening workflow and presents automated methods to facilitate and expedite the enrollment process. This includes extracting specific medical concepts from clinical notes, the study and reasoning of vague language used by healthcare professionals, and finally the identification of criteria-relevant text in clinical notes. We also show that active learning and semi-supervised learning techniques help in overcoming the challenge of limited and expensive training data in this domain. |
