Skip to main content
Biomedical Informatics and Medical Education > News and Events

News and Events

 

Chair’s Message

pth-use-this-oneWe are moving toward our vision with a number of activities across our various programs. We have updated our strategic plan in response to the 10-year academic program review that we recently completed. For our research-oriented MS and PhD programs, we have recently added a specialization in Data Science. We are completing a curriculum revision for our on line applied clinical informatics MS which will be effective Fall 2020. The work of our fellows in the clinical informatics fellowship program has received plaudits from clinical administrators and faculty, and we are currently recruiting a new faculty member in our department to assist with this program (view position description).  We are also recruiting a faculty member in medical education to start Summer 2020 (view position description). This is the beginning of a new cycle of admissions to our graduate programs, and we look forward to another productive year, and new growth in our department.

Cordially,

Peter Tarczy-Hornoch, MD
Chair and Professor, Department of Biomedical Informatics and Medical Education

Biomedical Informatics and Medical Education Newsletter

June 30, 2025 – July 4, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590 – See you in autumn quarter – 9/25/2025!

ANNOUNCEMENTS
On July 1st, Michelle Stoffel, MD, PhD (alumnae of the UW clinical informatics fellowship program) became Program Director of the Clinical Informatics Fellowship for the ACGME-accredited University of Minnesota-Hennepin HealthCare Clinical Informatics Fellowship Program. This multisite program combines the strengths of academic and community medical practice with research and didactic opportunities to prepare physicians for excellence in clinical informatics leadership roles.

PAPERS, PUBLICATIONS & PRESENTATIONS

  • Patricia S. Groves, Yelena Perkhounkova, Amany Farag, Maria Hein, Janice A. Sabin, Matthew J. Witry and Brad Wright, 2025, Concern and credibility: a factorial survey experiment on nurse judgments and intent to report patient-expressed safety events. 2025.  BMC Nursing, 24:798
  • LM-Merger: A workflow for merging logical models with an application to gene regulatory network models” by Luna Xingyu Li,  Boris Aguilar, John H Gennari, and Guangrong Qin

June 23, 2025 – June 27, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590 – See you in autumn quarter – 9/25/2025!

PAPERS, PUBLICATIONS & PRESENTATIONS

  • Deepthi Mohanraj, Raina Langevin, Libby Shah, Janice Sabin, Brian R. Wood, Wanda Pratt, Nadir Weibel, Andrea L. Hartzler. Leveraging Provocative Design Methods to Address Implicit Bias in Clinical Interactions through Technology. AMIA Annual Symposium 2025. Full paper
  • Julia C. Dunbar, Wanda Pratt, Lily Jeffs, Chelsea Ng, Sanaa Sayed, Jodi Smith, Ari H. Pollack. My Kidney T.R.E.K. – Thinking, Reflecting, and Empowering Kidney Transplant Patients, through technology. AMIA Annual Symposium 2025. Full paper
  • Hyeyoung Ryu, Sungha Kang, Wanda Pratt. Mitigating Stigma and Fostering Support: Improving AI-Generated Counterspeech for Microaggressions. AMIA Annual Symposium 2025. Full paper
  • Emma J. McDonnell, Wanda Pratt. From Chronic Health Condition to Disability Identity: Opportunities for Health Informatics Engagement. AMIA Annual Symposium 2025. Full paper
  • Lisa Dirks, Victoria BearBow, Wanda Pratt. Enhancing Health Research Results Dissemination for American Indian and Alaska Native Communities through Indigenous Community-Centered Design. AMIA Annual Symposium 2025. Full paper
  • Portuguese, Andrew J., Emily C. Liang, Jennifer J. Huang, Yein Jeon, Danai Dima, Rahul Banerjee, Mary Kwok, Kara I. Cicero, Alexandre V. Hirayama, Ryan Basom, Christy Khouderchah, Mazyar Shadman, Lawrence Fong, Andrew J. Cowan, and Jordan Gauthier. 2025. “Extramedullary Disease Is Associated with Severe Toxicities Following B‑Cell Maturation Antigen CAR T‑Cell Therapy in Multiple Myeloma.” Haematologica, June 19.
  • Ding X, Sheng Z, Hur B, Tauscher J, Ben-Zeev D, Yetişgen M, Pakhomov S, Cohen T. Tailoring task arithmetic to address bias in models trained on multi-institutional datasets. Journal of Biomedical Informatics. 2025 Jun 8:104858.
  • Sheng, Z., Ding, X., Hur, B., Li, C., Cohen, T., & Pakhomov, S. (2025). Mitigating confounding in speech-based dementia detection through weight masking. To appear in: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL), Vienna, Austria, 2025.
  • Saha, Aparajita;Molani, Sevda; Mease, Philip; Hadlock, Jennifer. Outcome-wide analysis of electronic health records data for identifying sequelae in Behçet’s disease. Accepted for podium abstract presentation at AMIA Annual Symposium 2025.
  • Oliver Bear Don’t Walk; Lauren W Yowelunh McLester-Davis; Susan Brown Trinidad.
    Rectifying Genocidal Data Stewardship: A Commentary on Ethical and Legal Obligations for Sharing Data With Tribal Entities
    https://www.jmir.org/2025/1/e77946
  • Xie SJ, Zhai S, Liang Y, Li J, Fan X, Cohen T,Yuwen W. Cultural Prompting Improves the Empathy and Cultural Responsiveness of GPT-Generated Therapy Responses. Accepted to AMIA Annu Symp Proc. 2025. Full paper.
  • Wang L, Carrington D, Filienko D, Jazmi C., Xie SJ, De Cock M, Iribarren S, Yuwen, W. (2025). Large language model-powered conversational agent delivering Problem-Solving Therapy (PST) for family caregivers: Enhancing empathy and therapeutic alliance using in-context learning. Accepted to AMIA Annu Symp Proc. 2025. Full paper. Also, In arXiv [cs.AI]. arXiv. http://arxiv.org/abs/2506.11376

June 16, 2025 – June 20, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590 – See you in autumn quarter – 9/25/2025!

ANNOUNCEMENTS
Dr. Neil Abernethy, who supported the Fred Hutchinson Cancer Research Center (FHCRC), CDC, HHS, and WA-DOH during the pandemic, is lead author of this Nature Digital Medicine paper describing the Coronavirus Prevention Network (CoVPN) Volunteer Screening Registry (VSR), which he co-designed with partners from FHCRC and Oracle Inc. The VSR collected over 650,000 volunteers and distributed them through hundreds of network sites to facilitate recruitment of diverse participants into multiple Phase 3 clinical trials. This effort helped accelerate COVID-19 vaccine development, underscoring the critical role informatics can play in rapidly addressing national health crises.

https://www.nature.com/articles/s41746-025-01666-3?utm_source=rct_congratemailt&utm_medium=email&utm_campaign=oa_20250506&utm_content=10.1038/s41746-025-01666-3

 

PAPERS, PUBLICATIONS & PRESENTATIONS

  • Feng Chen had poster presentation accepted by AMIA annual symposium: “Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models”: Feng Chen, Dror Ben-Zeev, Gillian Sparks, Arya Kadakia, Trevor Cohen. arXiv:2504.01216.
  • Good news from Anne Turner.  “Our abstract entitled, Digital Ageism and Informatics Research Involving Older Adults with Cognitive Impairment, has been accepted for panel presentation at the AMIA 2025 Annual Symposium this fall.  Two BIME-associated faculty,  Dr George Demiris (UPenn) and Dr. Amanda Lazar (Univ. of Maryland), along with Dr. Greg Alexander (Columbia) will be participating in the panel.”
  • Kuan-Ching Wu, Basia Belza, Donna Berry, Frances Lewis, Oleg Zaslavsky, and Andrea Hartzler. UTI risk factors in older people living with dementia: A conceptual framework and a scoping review. Dementia. https://journals.sagepub.com/eprint/E4MC5YDWAP3DDZFWIABI/full
  • Young D, Bartlett LE, Reid N, Hartzler AL, Bradford MC, Goss CH, Pilewski JM, Dunitz JM, Saavedra M, Berry DL, Kapnadak SG, Ramos KJ. Personal Narratives to Support Learning about Lung Transplant for People with Cystic Fibrosis. Patient Education and Counseling. 2025 May 5:108822.
  • Ji Z, Wedgeworth P, Mertens K, Jackson SL, Akinsoto NO, Klein J, Isaac M, Hartzler AL. EHR-Based Social Needs Screening and Referral in Primary Care: Clinician and Staff Perspectives on Practices, Barriers, and Benefits. AMIA Annual Symposium 2025, Full paper
  • Sarrieddine A, Lai C, Bear Don’t Walk O, Reid NFH, Sawicki G, Berlinski A, Rosenfeld M, Hartzler AL. Technology and Human Support Systems in Decentralized Clinical Trials: A Participant-Centered Case Study in Cystic Fibrosis. AMIA Annual Symposium 2025, Full paper
  • Lai C, Casanova-Perez R, Langevin R, Tsendenbal A, Saxena S, Pratt W, Sabin J, Wood BR, Weibel N, Hartzler AL. Automated Assessments of Clinical Encounters: Provider Perspectives. AMIA 2025 Annual Symposium, Poster.
  • Hartzler AL, Chung RY, Sarrieddine A, McNamara S, Lee M, Daines C, Davis J, Joseph L, Brown RF, Drake G, Revay D, Rushing S, Vanderbilt N, Rosenfeld M, Ong T. Are we ready for home spirometry? Patient and family perspectives on standardized implementation in pediatric clinical care NACFC 2025
  • Vasbinder, Knapp, Grassy, Harris, Basile, Young, Milinic, Bartlett, Hartzler, Kapnadak, Smith, Riekert, Berry, Ramos. Evaluating shared decision making in lung transplant discussions between clinicians and people with severe to advanced cystic fibrosis lung disease. NACFC 2025
  • Langevin R, Lai C, Mohanraj D, Shah L, Casanova-Perez R, Tsedenbal A, Saxena S, Sabin J, Wood BR, Pratt W, Weibel N, Hartzler AL. UnBIASED: Developing communication feedback technology to address implicit bias in patient-provider interactions . WPRN Annual meeting 2025.
  • Langevin R, Martin L, Bolivar KN, Keen S, Wedgeworth P, Lui PP, Hartzler AL. Using Large Language Models to Evaluate Patient-Centered Note Writing in Clinical Documentation. AMIA Clinical Informatics Conference, Anaheim CA, May 2025.

June 9, 2025 – June 13, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590 – See you in autumn quarter – 9/25/2025!

ANNOUNCEMENTS
You’re invited to help us celebrate 2024-25 BHI graduates, new emeritus faculty, and student/faculty awardees at the BHI End-of-Year Graduation Celebration! Here are the details:

When: Ceremony this Friday, June 13th from 2:00-3:00pm with reception immediately following from 3:00-5:00pm

Where:

  • SLU Orin Smith Auditorium for the 2:00-3:00pm ceremony or remotely (Zoom link below);
  • SLU 123 A & B for the 3:00-5:00pm reception (in person only)

Zoom link: https://washington.zoom.us/j/95887407122?pwd=8BbgfuyqBR2ISMEuthvMUV9P88qU5J.1

_______________________
Authors: Jimmy Phuong, MSPH, PhD; Adam B Wilcox, PhD, FAMIA; Shruti Sehgal, MD(Hom), MS; Anthony Solomonides, PhD, MSc, FAMIA, FACMI; Juan Espinoza, MD

Title: Maturity Models 101 – What does Roadmapping Institutional Growth in Informatics look like?. [Tutorial session]
Date: 11/15/2025
Time: 8:30 AM – 12:00 PM EST
Session Code: W04

_______________________
Please join us in congratulating Ehsan Alipour who successfully passed his PhD Defense!
Title: Evaluating Multi-Modal Data Fusion Approaches for Predictive Clinical Models Using Multiple Medical Data Domains

Abstract: Multimodal deep learning models have emerged as powerful tools in biomedical research, offering the ability to integrate diverse data sources such as clinical records, multi-omics data, imaging, survey responses, and wearable data to enhance predictive accuracy and deepen understanding of complex medical phenomena. Central to multimodal modeling is the process of data fusion, where information from different modalities is integrated in a unified model. Three primary fusion strategies exist in deep learning: early fusion (feature-level), intermediate fusion and late fusion (decision-level). While widely adopted in other domains, their comparative performance and implementation considerations remain underexplored in biomedical applications, where data heterogeneity, missingness, and varying dimensionality present additional challenges.

This dissertation aims to evaluate the implications of data fusion strategies in developing multimodal predictive models in medicine. Across three distinct aims, we assess the impact of early, intermediate, and late fusion techniques on predictive performance, implementation complexity, and generalizability using diverse combinations of data types, outcomes, and modeling strategies. These studies span multiple datasets and outcome types (binary vs continuous) providing a broad view of fusion strategy utility in real-world biomedical settings.

In Aim 1, we evaluated and compared early, intermediate, and late fusion strategies for integrating longitudinal EHR, genomic, and survey data to predict chronic kidney disease (CKD) progression in patients with type 2 diabetes using a novel transformer-based multimodal architecture. Using data from the NIH’s All of Us initiative, we trained models on a cohort of approximately 40,000 patients. While unimodal models—particularly those based on EHR data—achieved strong baseline performance with an AUROC of 0.73 (0.71 – 0.75), the inclusion of multimodal data offered only marginal improvement with an AUROC of 0.74 (0.72 – 0.76), with the benefit limited to the early fusion approach and lacking statistical significance. This aim highlighted the challenges of overfitting in complex fusion architectures and emphasized the role of modality-specific predictive strength.

In Aim 2, we extended the fusion analysis to imaging data by combining a convolutional neural network (CNN) trained on longitudinal cross-sectional imaging with a shallow neural network trained on clinical and pathology variables to predict post-surgical margin status in patients with soft tissue sarcoma (n=202). Here, the intermediate fusion strategy significantly outperformed other approaches, achieving an AUROC of 0.80 (0.66–0.95), suggesting that cross-modal interactions between histologic features and imaging embeddings may be best captured through intermediate fusion. This result demonstrated the potential value of intermediate fusion when complementary signals exist across modalities.

In Aim 3, we explored fusion strategies for estimating continuous CT-derived body composition metrics (e.g., visceral, and subcutaneous fat volumes) using only chest radiographs and clinical variables in a dataset of 1,088 patients. A multitask multimodal model was developed and evaluated across early, intermediate, and late fusion strategies. Late fusion consistently delivered the best performance across most body composition metrics, closely followed by intermediate fusion. These results suggest that when individual modalities offer high independent predictive power, decision-level integration may be optimal for regression tasks.

Collectively, this work provides a comprehensive evaluation of data fusion strategies in multimodal biomedical modeling, highlighting their strengths, limitations, and practical considerations. Findings suggest that no single fusion strategy universally outperforms the others; rather, optimal fusion depends on data characteristics, model architecture, and task-specific objectives. This dissertation lays the groundwork for future research aimed at developing adaptive fusion strategies tailored to the complexities of real-world biomedical data.

PAPERS & PUBLICATIONS

 

June 2, 2025 – June 6, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590 – See you in autumn quarter – 9/25/2025!

ANNOUNCEMENTS
Andrea Hartzler was elected serve on the SOM Council on Appointments and Promotions for a 3-year term!

_______________________

Please join us in congratulating Velvin Fu who successfully passed her PhD Defense!

Title: Evaluating and Enhancing Large Language Models (LLMs) in the Clinical Domain

Abstract: Recent advancements in large language models (LLMs) have demonstrated human-level performance on many specialized medical tasks, even without annotated training data. However, three main challenges remain: (1) due to the sensitive and highly specialized nature of clinical narratives, as well as the high cost of human expert annotation, there is a lack of high-quality, well-structured, and clinically meaningful datasets for LLM training and evaluation; (2) current medical LLMs show limited generalization ability to interpret and extract complex clinical information on certain unseen natural language understanding (NLU) tasks; and (3) as LLMs are typically trained on vast amounts of data, there is a substantial risk of data contamination, where evaluation benchmarks unintentionally overlap with training data, leading to inflated test performance and potentially reduced performance on truly novel tasks.

In this work, we address these limitations through three core aims: (1) develop benchmark datasets for clinical information extraction (IE), a key NLU subtask, across two critical medical domains, and evaluate the performance of multiple state-of-the-art (SOTA) transformer-based language models (LMs), under both fine-tuning and in-context learning settings; (2) develop a more generalizable medical NLU model via instruction tuning, demonstrating enhanced performance on previously unseen clinical NLU datasets; and (3) systematically review existing detection approaches for data contamination and evaluate those approaches on datasets used during pre-training and fine-tuning LLMs, with our own and three other widely used open-source LLMs.

In summary, our work contributes to the development of both clinical benchmarks and robust LLMs, as well as highlighting the ongoing challenges in benchmarking LLMs’ generalizability.

UPCOMING EXAMS
Title: Evaluating Multi-Modal Data Fusion Approaches for Predictive Clinical Models Using Multiple Medical Data Domains
Student: Ehsan Alipour
Date/Time: Wednesday, June 11, 2025, 11am – 1 pm PT
In-person location: 850 Republican Street, Building C, SLU C259
Zoom: https://washington.zoom.us/my/peter.th

Abstract: Multimodal deep learning models have emerged as powerful tools in biomedical research, offering the ability to integrate diverse data sources such as clinical records, multi-omics data, imaging, survey responses, and wearable data to enhance predictive accuracy and deepen understanding of complex medical phenomena. Central to multimodal modeling is the process of data fusion, where information from different modalities is integrated in a unified model. Three primary fusion strategies exist in deep learning: early fusion (feature-level), intermediate fusion and late fusion (decision-level). While widely adopted in other domains, their comparative performance and implementation considerations remain underexplored in biomedical applications, where data heterogeneity, missingness, and varying dimensionality present additional challenges.

This dissertation aims to evaluate the implications of data fusion strategies in developing multimodal predictive models in medicine. Across three distinct aims, we assess the impact of early, intermediate, and late fusion techniques on predictive performance, implementation complexity, and generalizability using diverse combinations of data types, outcomes, and modeling strategies. These studies span multiple datasets and outcome types (binary vs continuous) providing a broad view of fusion strategy utility in real-world biomedical settings.

In Aim 1, we evaluated and compared early, intermediate, and late fusion strategies for integrating longitudinal EHR, genomic, and survey data to predict chronic kidney disease (CKD) progression in patients with type 2 diabetes using a novel transformer-based multimodal architecture. Using data from the NIH’s All of Us initiative, we trained models on a cohort of approximately 40,000 patients. While unimodal models—particularly those based on EHR data—achieved strong baseline performance with an AUROC of 0.73 (0.71 – 0.75), the inclusion of multimodal data offered only marginal improvement with an AUROC of 0.74 (0.72 – 0.76), with the benefit limited to the early fusion approach and lacking statistical significance. This aim highlighted the challenges of overfitting in complex fusion architectures and emphasized the role of modality-specific predictive strength.

In Aim 2, we extended the fusion analysis to imaging data by combining a convolutional neural network (CNN) trained on longitudinal cross-sectional imaging with a shallow neural network trained on clinical and pathology variables to predict post-surgical margin status in patients with soft tissue sarcoma (n=202). Here, the intermediate fusion strategy significantly outperformed other approaches, achieving an AUROC of 0.80 (0.66–0.95), suggesting that cross-modal interactions between histologic features and imaging embeddings may be best captured through intermediate fusion. This result demonstrated the potential value of intermediate fusion when complementary signals exist across modalities.

In Aim 3, we explored fusion strategies for estimating continuous CT-derived body composition metrics (e.g., visceral, and subcutaneous fat volumes) using only chest radiographs and clinical variables in a dataset of 1,088 patients. A multitask multimodal model was developed and evaluated across early, intermediate, and late fusion strategies. Late fusion consistently delivered the best performance across most body composition metrics, closely followed by intermediate fusion. These results suggest that when individual modalities offer high independent predictive power, decision-level integration may be optimal for regression tasks.

Collectively, this work provides a comprehensive evaluation of data fusion strategies in multimodal biomedical modeling, highlighting their strengths, limitations, and practical considerations. Findings suggest that no single fusion strategy universally outperforms the others; rather, optimal fusion depends on data characteristics, model architecture, and task-specific objectives. This dissertation lays the groundwork for future research aimed at developing adaptive fusion strategies tailored to the complexities of real-world biomedical data.

 

May 26, 2025 – May 30, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590
Presenter: Kari Stephens, PhD
Thursday, June 5th – 11-11:50 am
850 Republican Street, Building C, Room 123 A/B
Zoom Information: https://washington.zoom.us/my/bime590
Speaker will present via Zoom

Title:
Digital Behavioral Health Therapeutics: Is there an app for that and can AI be your next therapist?

Abstract:
Primary care treats the majority of mental and behavioral health conditions that are also on the rise, due to large care access issues. Integrating behavioral health providers into primary care is aiming to meet that need, but practices struggle with dissemination. Digital therapeutics may hold great promise to meet the growing need.

Speaker Bio:
Dr. Kari Stephens is Vice Chair of Research, Helen D. Cohen Endowed Professor, Director of Clinical Research Informatics and Professor in Family Medicine and Adjunct Professor in Biomedical Informatics and Medical Education at the UW School of Medicine. Dr. Stephens is a practicing clinical psychologist and biomedical informaticist conducting research focused on health equity, integrated behavioral health, chronic pain, posttraumatic stress disorder, anxiety, depression, substance use, cancer, long COVID, and informatics data sharing, particularly in primary care settings. Dr. Stephens currently conducts clinical research and leads informatics and innovations in data sharing as Director of Data QUEST, overseeing a regional electronic health record systems based primary care data sharing network, which has supported over $100M in grant funded projects, Associate Director with the National Alzheimer’s Coordinating Center, and executive faculty member and lead of regional informatics within the Institute of Translational Health Science’s Biomedical Informatics Core. Her work has been funded by NIH, PCORI, AHRQ, SAHMSA, DoD, CDC, private agencies, and local and state government agencies.

BIME 591
Wednesdays – 11:30-12:20 pm
Section B, HSEB 421
Zoom Information: https://washington.zoom.us/my/velvinfu?pwd=Y3dJbjNBeFpzTC9HTXV1UDFYYXlKQT09 

UPCOMING EXAMS
Title: Evaluating and Enhancing Large Language Models (LLMs) in the Clinical Domain
Student: Velvin Fu
Date/Time: Thursday, June 5, 2025,  11am – 1 pm PT
In-person location: 850 Republican Street, Building C, SLU C259
Zoom: https://washington.zoom.us/my/melihay

Abstract: Recent advancements in large language models (LLMs) have demonstrated human-level performance on many specialized medical tasks, even without annotated training data. However, three main challenges remain: (1) due to the sensitive and highly specialized nature of clinical narratives, as well as the high cost of human expert annotation, there is a lack of high-quality, well-structured, and clinically meaningful datasets for LLM training and evaluation; (2) current medical LLMs show limited generalization ability to interpret and extract complex clinical information on certain unseen natural language understanding (NLU) tasks; and (3) as LLMs are typically trained on vast amounts of data, there is a substantial risk of data contamination, where evaluation benchmarks unintentionally overlap with training data, leading to inflated test performance and potentially reduced performance on truly novel tasks.

In this work, we address these limitations through three core aims: (1) develop benchmark datasets for clinical information extraction (IE), a key NLU subtask, across two critical medical domains, and evaluate the performance of multiple state-of-the-art (SOTA) transformer-based language models (LMs), under both fine-tuning and in-context learning settings; (2) develop a more generalizable medical NLU model via instruction tuning, demonstrating enhanced performance on previously unseen clinical NLU datasets; and (3) systematically review existing detection approaches for data contamination and evaluate those approaches on datasets used during pre-training and fine-tuning LLMs, with our own and three other widely used open-source LLMs.

In summary, our work contributes to the development of both clinical benchmarks and robust LLMs, as well as highlighting the ongoing challenges in benchmarking LLMs’ generalizability.

Title: Evaluating Multi-Modal Data Fusion Approaches for Predictive Clinical Models Using Multiple Medical Data Domains
Student: Ehsan Alipour
Date/Time: Wednesday, June 11, 2025, 11am – 1 pm PT
In-person location: 850 Republican Street, Building C, SLU C259
Zoom: https://washington.zoom.us/my/peter.th

 

May 19, 2025 – May 22, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590
Presenter: Lucy Lu Wang, PhD
Thursday, May 29th – 11-11:50 am
850 Republican Street, Building C, Room 123 A/B
Zoom Information: https://washington.zoom.us/my/bime590
Speaker will present in-person 

Title:
Evaluation and Meta-Evaluation of LLMs for Scientific and Medical Applications

Abstract:
Large Language Models (LLMs) are powering new opportunities to improve access to scientific and medical knowledge—especially for patients, caregivers, and other non-expert audiences. But these models do not always produce reliable or trustworthy information, and evaluating their performance remains a major challenge. Prior evaluation methods may not generalize to complex, retrieval-augmented, long-form generation tasks like literature review generation or plain language summarization. In this talk, I will introduce some settings in which LLMs are being used to support understanding of biomedical research, and present several recent studies of how to evaluate models more effectively. I will also discuss meta-evaluation strategies (how to evaluate the evaluation methods themselves) that can be used to probe model performance in these complex settings, and how our insights can help guide the use of LLMs in real-world healthcare and scientific communication.

Speaker Bio:
Lucy Lu Wang is an Assistant Professor at the University of Washington Information School, where she leads the Language Accessibility Research (LARCH) lab. She is also an affiliated Research Scientist at the Allen Institute for AI (Ai2), and holds adjunct appointments in UW Computer Science & Engineering, Biomedical Informatics & Medical Education, and Human Centered Design & Engineering. Her work spans scholarly document understanding, scientific evidence synthesis, and health communication. Specifically, she focuses on language technologies to improve access to and understanding of information in high-expertise domains like science and healthcare, with an emphasis on dataset development and evaluation practices. She is a core contributor to open science initiatives including Semantic Scholar and the COVID-19 Open Research Dataset, and public-facing systems focused on accessibility of science such as supp.ai and Paper to HTML. She received her PhD in Biomedical Informatics & Medical Education from the University of Washington.

BIME 591
Wednesdays – 11:30-12:20 pm
Section B, HSEB 421
Zoom Information: https://washington.zoom.us/my/velvinfu?pwd=Y3dJbjNBeFpzTC9HTXV1UDFYYXlKQT09 

ANNOUNCEMENTS
Please join us in congratulating Amber Chen who successfully passed her General Exam!

Title: Computational Approaches to Predict Health Outcomes Using Cytometry Data

Abstract: Flow cytometry plays a pivotal role in medicine, as demonstrated by the widespread use of the complete blood count (CBC) across medicine. Alongside summary markers such as cell counts, CBCs also generate rich single-cell datasets, which are drastically underused. To allow for enhanced clinical use of these single-cell data streams, there is a need for methods which can generate clinically relevant biomarkers, or which can directly use these data for clinical prediction. To tackle these issues, we propose three primary research aims: (1) algorithmically developing novel markers from clinical single-cell data using unsupervised techniques, (2) evaluate and develop novel single-cell data modeling methods for predicting health outcomes, and (3) applying these single-cell approaches to significant health conditions, including predicting pregnancy and cardiac surgery complications and predicting treatment efficacy for hematologic malignancies. Approaches will be validated using large-scale data archives from University of Washington Medical Center (UWMC), alongside multiple open-source single-cell datasets. Collectively, this project will lead to generation of novel computational methods which can create tangible benefit for multiple well-defined clinical challenges.

PAPERS & PUBLICATIONS

  • Jie Fu, Sharon Pai, Daniel Hippe, Faisal Yaseen, John Kang, Ramesh Rengan, Jing Zeng, Stephen Bowen, Sunan Cui. “Predicting Voxel-Level Local Progression for Locally Advanced NSCLC following Chemoradiation Using Multi-Task Deep Learning” has been accepted for presentation in ORAL scientific session for the 67th annual meeting of American Society for Radiation Oncology (ASTRO)
  • W Sami, F Abbas, M Sabir, A Razzaq, H Nazeer, MA Hayat, M Ali, MI Ali, F Yaseen, U Nazir “Data-Driven Prediction of Malaria Outbreaks in Nigeria” was accepted in IEEE International Geoscience and Remote Sensing Symposium (IGARSS).

UPCOMING EXAMS
Title: Evaluating and Enhancing Large Language Models (LLMs) in the Clinical Domain
Student: Velvin Fu
Date/Time: Thursday, June 5, 2025,  11am – 1 pm PT
In-person location: 850 Republican Street, Building C, SLU C259
Zoom: https://washington.zoom.us/my/melihay

Abstract: Recent advancements in large language models (LLMs) have demonstrated human-level performance on many specialized medical tasks, even without annotated training data. However, three main challenges remain: (1) due to the sensitive and highly specialized nature of clinical narratives, as well as the high cost of human expert annotation, there is a lack of high-quality, well-structured, and clinically meaningful datasets for LLM training and evaluation; (2) current medical LLMs show limited generalization ability to interpret and extract complex clinical information on certain unseen natural language understanding (NLU) tasks; and (3) as LLMs are typically trained on vast amounts of data, there is a substantial risk of data contamination, where evaluation benchmarks unintentionally overlap with training data, leading to inflated test performance and potentially reduced performance on truly novel tasks.

In this work, we address these limitations through three core aims: (1) develop benchmark datasets for clinical information extraction (IE), a key NLU subtask, across two critical medical domains, and evaluate the performance of multiple state-of-the-art (SOTA) transformer-based language models (LMs), under both fine-tuning and in-context learning settings; (2) develop a more generalizable medical NLU model via instruction tuning, demonstrating enhanced performance on previously unseen clinical NLU datasets; and (3) systematically review existing detection approaches for data contamination and evaluate those approaches on datasets used during pre-training and fine-tuning LLMs, with our own and three other widely used open-source LLMs.

In summary, our work contributes to the development of both clinical benchmarks and robust LLMs, as well as highlighting the ongoing challenges in benchmarking LLMs’ generalizability.

 

May 12, 2025 – May 16, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590
Presenter: Elizabeth Putnam, PhD
Thursday, May 22nd – 11-11:50 am
850 Republican Street, Building C, Room 123 A/B
Zoom Information: https://washington.zoom.us/my/bime590
Speaker will present Via Zoom

Title:
Joys and challenges of research with small, rural communities

Abstract:
The establishment of trust relationships has been critical to research progress with small rural and tribal communities and thus has become a central theme of community focused research. Understanding the contexts presented by these communities has educated investigators seeking to connect with them to address local concerns about research and potential resulting discrimination. The challenges of recruiting participants from small populations and difficulties in reaching valid conclusions in these studies will be addressed.

Speaker Bio:
Elizabeth Putnam is Professor of Molecular Genetics and Toxicology at the University of Montana, serving as Chair of the Department of Biomedical and Pharmaceutical Sciences in the Skaggs School of Pharmacy since 2015. After completing undergraduate work in Biochemistry at Rutgers College in New Brunswick, NJ, she earned a PhD in Biomedical Sciences from the University of Texas-Houston Graduate School of Biomedical Sciences. Following postdoctoral fellowships at M.D. Anderson Cancer Center and the University of Texas Medical School, she moved to the University of Montana in 2000. Her research efforts include outreach to rural and underserved communities including projects to provide comprehensive, collaborative and participatory initiatives that specifically increased community understanding of the interplay between genes and environment. Since 2007, this has included CBPR with the Confederated Salish and Kootenai Tribes to develop community understanding of precision medicine while learning from these communities of the appropriate, culturally respectful ways to disseminate information about precision medicine opportunities.

BIME 591
Wednesdays – 11:30-12:20 pm
Section B, HSEB 421
Zoom Information: https://washington.zoom.us/my/velvinfu?pwd=Y3dJbjNBeFpzTC9HTXV1UDFYYXlKQT09 

ANNOUNCEMENTS
Chris Lewis was selected as a 2025 Innovators & Influencer Honoree for the American Academy of Physical Medicine & Rehabilitation (AAPM&R)!
https://www.aapmr.org/members-publications/member-recognition/innovatorsandinfluencers

_______________________
Have a new idea that you think will transform medicine? If so, be sure to review the details in our 2025 RFA and submit a proposal by May 23rd, 2025.

We are seeking applications for one-year pilot projects in the area of medical data science, for potentially up to $45K, with $25K guaranteed direct costs + the possibility of adding up to $20K in cloud credits, pending agreement with cloud service providers.

Please note that only UW faculty are eligible to apply. Learn more here.
_______________________
2025 UW Wellness Symposium
Enhance your well-being at the 2025 UW Wellness Symposium, Wednesday, May 21, and Thursday, May 22. Join expert-led virtual sessions on mental, physical and financial health with The Whole U.

PAPERS & PUBLICATIONS

  • Riley BC, Phuong J, Hasan RA, Stansbury LG, Hess JR, Roubik DJ. Cold-stored versus room-temperature platelets for hemostasis in combat trauma patients. Transfusion. (Accepted on 30 Apr 2025).
  • S. Zeng, N.C. Jani, A.M. Sotolongo, G. Luo, M. Arjomandi, and M.J. Falvo. Clinical Utility of Pulmonary Function Testing in Assessing Longitudinal Outcomes of Deployed Veterans with Preserved Spirometry. Annals of the American Thoracic Society (AnnalsATS), 2025.

 

May 5, 2025 – May 9, 2025

UPCOMING LECTURES AND SEMINARS
BIME 590
Presenter: Jared Erwin, PhD
Thursday, May 15th – 11-11:50 am
850 Republican Street, Building C, Room 123 A/B
Zoom Information: https://washington.zoom.us/my/bime590
Speaker will present In-Person

Title:
Scaling standards based health care data

Abstract:
Healthcare data standards have been successful in improving interoperability, both for clinical use cases as well as research. The accessibility of data has expanded its use cases beyond the originally intended scope of the standards. This talk explores the challenges and opportunities of scaling standards-based healthcare software to support large-scale data storage, analytics, and secure access. Drawing from real-world implementations, we will examine patterns for scaling FHIR and OMOP, integrating with data lakes and distributed computing, and evolving approaches to security and identity management, including the roles of TEFCA and QHINs.

Speaker Bio:
Jared Erwin has a 20 year history developing healthcare software, from real-time embedded device to cloud based service.  Currently he develops the Microsoft Azure FHIR service.  He is also a lecturer at the University of Washington, teaching research methods.

BIME 591
Wednesdays – 11:30-12:20 pm
Section B, HSEB 421
Zoom Information: https://washington.zoom.us/my/velvinfu?pwd=Y3dJbjNBeFpzTC9HTXV1UDFYYXlKQT09 

ANNOUNCEMENTS
­­­­­Have a new idea that you think will transform medicine? If so, be sure to review the details in our 2025 RFA and submit a proposal by May 23rd, 2025.

We are seeking applications for one-year pilot projects in the area of medical data science, for potentially up to $45K, with $25K guaranteed direct costs + the possibility of adding up to $20K in cloud credits, pending agreement with cloud service providers.

Please note that only UW faculty are eligible to apply. Learn more here.
_______________________
2025 UW Wellness Symposium
Enhance your well-being at the 2025 UW Wellness Symposium, Wednesday, May 21, and Thursday, May 22. Join expert-led virtual sessions on mental, physical and financial health with The Whole U.

UPCOMING EXAMS
Title: Computational Approaches to Predict Health Outcomes Using Cytometry Data
Student: Ya-Lin, Chen
Date/Time: Tuesday, May 13, 2025, 11 AM – 1 PM (Pacific Time)
Location: 850 Republican Street, Building C, Room 359
Zoom:  https://washington.zoom.us/j/92178550574

Abstract: Flow cytometry plays a pivotal role in medicine, as demonstrated by the widespread use of the complete blood count (CBC) across medicine. Alongside summary markers such as cell counts, CBCs also generate rich single-cell datasets, which are drastically underused. To allow for enhanced clinical use of these single-cell data streams, there is a need for methods which can generate clinically relevant biomarkers, or which can directly use these data for clinical prediction. To tackle these issues, we propose three primary research aims: (1) algorithmically developing novel markers from clinical single-cell data using unsupervised techniques, (2) evaluate and develop novel single-cell data modeling methods for predicting health outcomes, and (3) applying these single-cell approaches to significant health conditions, including predicting pregnancy and cardiac surgery complications and predicting treatment efficacy for hematologic malignancies. Approaches will be validated using large-scale data archives from University of Washington Medical Center (UWMC), alongside multiple open-source single-cell datasets. Collectively, this project will lead to generation of novel computational methods which can create tangible benefit for multiple well-defined clinical challenges.