NOTE: BIS 525 students are required to attend in person. Others are invited to attend in person but may also attend via Zoom.
SPEAKER: Kevin Lybarger, PhD, Assistant Professor, Department of Information Sciences and Technology, George Mason University
TITLE: “Extracting Social Determinants of Health: Annotation, Extraction, Information Gain, and Ethical Considerations"
ABSTRACT: Social determinants of health (SDOH) are social, behavioral, and environmental factors that impact health and inform patient care. Within the Electronic Health Record (EHR), the clinical narrative contains nuanced and detailed descriptions of many SDOH. The utilization of this text-encoded information in secondary use applications necessitates extraction via natural language processing (NLP). This talk explores the extraction of SDOH information, including: i) the creation of annotated clinical data sets, ii) the development of large language model (LLM) extractors, iii) the exploration of a sizable EHR data set, and iv) providing commentary on ethical considerations. We will discuss a range of LLM extraction approaches, spanning model types (encoder vs. generative models), learning strategies (supervised vs. in-context), and prompting strategies (structured text generation vs. question answering). We introduce an annotated dataset integrating granular, relation-style annotations with SDOH concept normalization to create a multi-faceted representation optimized for secondary use applications. We empirically validate the efficacy of this representation for SDOH extraction, achieving high performance using LLMs in both supervised and in-context learning paradigms.
YSPH values inclusion and access for all participants. If you have questions about accessibility or would like to request an accommodation, please contact Charmila Fernandes at Charmila.firstname.lastname@example.org. We will try to provide accommodations requested by November 21, 2023.
George Mason UniversityKevin Lybarger, PhDAssistant Professor