Abstract: Health text data contain rich and valuable information that can improve clinical care, patient outcomes, and population health. For example, social media posts generated by patients and the public can be analyzed to understand patient experiences and public opinions toward health-related events and policies such as COVID-19 mask and vaccination mandates. Unstructured clinical notes authored by clinicians document rich information about patients’ medical history, conditions, medications, and social determinants of health. Health text data, however, are often of huge volume and heterogeneous, which present challenges for developing and applying appropriate computational methods that can analyze them at scale with accuracy. In this talk, I will discuss my work that empirically applied computational analysis such as Natural Language Processing (NLP) methods on health text data including large-scale Twitter data and online physician reviews, to generate real-world insights for public health authorities and clinicians. I will also present my work that developed methodological innovations so that the computational methods are more appropriate and tailored for the contexts of health text data. Last, I will share my most recent work that evaluated emerging large language models (LLMs) on extracting clinical information from notes of veterans with lymphoid malignancies, and our efforts in developing a tailored NLP pipeline for reliable information extraction. I discuss how health text as sociotechnical products require careful study design when developing and applying computational analysis, which is significantly lacking in health informatics.

Rigor and Reproducibility Seminar Series - UF Interdisciplinary T32 in Movement Disorders and Neurorestoration

This seminar series is jointly hosted by the UF Movement Disorders and Neurorestoration Program and the UF Health Science Center Libraries. Talks focus on rigor and reproducibility topics for the pre-doctoral training program.

This program and seminar series is funded by the National Institute of Neurological Disorders and Stroke (NINDS) through T32 NS082128, awarded to PIs Dawn Bowers (College of PHHP) and David E. Vaillancourt (College of HHP).

The seminar will be recorded, with the recording distributed to registered attendees after the event.