Speaker: John Boscardin, PhD, Professor, Departments of Medicine and Epidemiology & Biostatistics, University of California, San Francisco
Title: Building better prognostic models: balancing accuracy, interpretability, and usability in the care of older adults
Abstract:
Prognostic models can be powerful tools to help clinicians, patients, and caregivers make individualized decisions—particularly in the complex care of older adults. I will share our group’s experience developing and refining prognostic models tailored to the needs of older populations with a focus on several recurring methodological issues: (1) the tradeoffs between traditional regression and more complex machine learning approaches; (2) model development strategies including feature engineering and variable selection, and internal validation of the full process; (3) balancing the demand for high predictive performance with the need for interpretability of individual features; (4) assessing the stability of individual predictions across a range of potential models; (5) practical approaches for handling of missing data during model implementation.
Bio:
John Boscardin, PhD is Professor in the Departments of Medicine and Epidemiology & Biostatistics at the University of California, San Francisco. He is Director of the Statistical Laboratory for Aging Research, Co-leader of the UCSF Pepper Center (NIH/NIA P30) Data and Analysis Core, and Co-leader of the Analytics Core for a joint NIH/NIA P01 program project between UCSF and Mt. Sinai. The laboratory is comprised of ten full-time data and statistical scientists and provides deep analytic support across the full lifecycle of research projects. He is a long-time core faculty member for the CTSI K-Scholars Program, serving as statistical mentor to numerous early-stage investigators. Dr. Boscardin received his PhD in Statistics from the University of California, Berkeley and was a faculty member at UCLA in the Departments of Biostatistics and Medicine before joining UCSF in 2008. His areas of methodological expertise include analysis of longitudinal and repeated measures and time to event data, treatment of missing data, Bayesian statistical modeling, and computational methods.
Susan Holt is inviting you to a scheduled Zoom meeting.
Join Zoom Meeting
Https://umaryland.zoom.us/j/92035628613?pwd=xPKCao6RQS359x9fAhQaOajnBZuRna.1&from=addon
Meeting ID: 920 3562 8613
Passcode: 913766