Speaker Schedule (Past and Future)

ML4H Seminar Google Calendar
for Outlook, copy and paste the following link exactly:
https://calendar.google.com/calendar/ical/c_550bf881ff28976aac7ddda130065136ab886eba48b8d25dc47baa54977d6d90%40group.calendar.google.com/public/basic.ics

Past Seminar Videos

Speaker Suggestion Form

Join the “ML4H Talks” Mailing List!


Speaker: Can Li, Ph.D. candidate (The University of Texas Health Science Center at Houston)

Title: Fairness and Equity in Healthcare Predictive Modeling

Abstract: The development of fair and equitable predictive models is crucial for preventing clinical prediction algorithms from amplifying existing disparities. As predictive models increasingly shape clinical decisions, longstanding disparities in healthcare outcomes remain, often driven by biases embedded in training data and modeling assumptions. These biases can contribute to unequal access to medical care and disparities in treatment quality across different demographic groups. One prominent example of such inequalities is liver transplantation, where limited organ availability and allocation policies influence who receives a liver transplant.
 
We developed a predictive modeling framework that simultaneously improves prediction accuracy and fairness for post-liver transplant complications. Our framework used transformer architectures to predict major post-transplant risks, including malignancy, diabetes, rejection, infection, and cardiovascular disease. The approach integrated task-balancing techniques and fairness-enhancing strategies within a multi-task learning framework to achieve equitable predictive performance across demographic groups.

Additionally, we introduced Fairness through Equitable Rate of Improvement (FERI), a method aimed at reducing demographic disparities in predictive modeling. FERI continuously adjusts learning rates across patient subgroups to ensure that improvements in model performance are more evenly distributed. This approach prevents certain groups from benefiting disproportionately while ensuring that underrepresented populations receive equitable improvements in predictive accuracy.

Our research extends to policy impact analysis through interrupted time series methods, examining how healthcare initiatives affect outcome disparities longitudinally. We further enhance equity through the fairness constraint optimization method and apply advanced survival analysis with counterfactual analysis to improve donor-recipient matching. These approaches contribute to a more equitable organ allocation system and better post-transplant outcomes across all demographic groups.

Bio: Can Li is a Ph.D. candidate majoring in Biostatistics and Data Science at the University of Texas Health Science Center at Houston and is currently in the final stage of her doctoral studies. She previously earned dual bachelor’s degrees in Biochemistry and Applied Mathematics from The Ohio State University, followed by a master’s degree in Biostatistics from Emory University.

Her research focuses on developing fair predictive modeling and addressing algorithmic bias in clinical research, particularly within healthcare applications such as liver transplantation. Her work integrates statistics, machine learning, and deep learning to improve equity in healthcare decision-making.

She has published several first-author papers in peer-reviewed journals and received a first-prize student paper award at a major conference. Through her research, she is committed to advancing fair and equitable predictive modeling that can improve healthcare delivery across diverse patient populations.

#science #talks #ml4hseminar 

Join Zoom Meeting:
https://broadinstitute.zoom.us/j/87629654270?pwd=UlxuK9eahEzP8xRaBchc2YU3rfKonb.1

(ID: 87629654270, passcode: 297879) Join by phone (US) +1 646-558-8656 (passcode: 297879) (US) +1 646-876-9923 (passcode: 297879)