麻豆AV

Jessica Gronsbell, PhD

Assistant Professor | Department of Statistical Sciences |
University of Toronto

WHEN: Wednesday, November 29, 2023, from 3:30 to 4:30 p.m.

WHERE: Hybrid | 2001 麻豆AV College, Rm 1140 | &

Note: Dr. Gronsbell will present in-person

Abstract

In spite of the enormous increase in the volume and diversity of clinical data in the last decade, the use of machine learning to improve patient care remains a largely unfilled opportunity. A critical bottleneck is the lack of methods that can properly address statistical inference questions that arise in 鈥渓ife after machine learning鈥�. Time permitting, I will consider two such questions. First, I will show how to reliably evaluate a model鈥檚 performance and whether it is fair in the semi-supervised setting when an extremely small proportion of testing data is labeled. Then, I will discuss our recent method for regression modeling when the outcome of interest is predominantly derived from a machine learning model due to the time or expense of ascertainment. The practical utility of my proposals will be illustrated with analyses of electronic health record data from Mass General Brigham healthcare system and population biobank data from the UK Biobank.

Speaker Bio

Jesse Gronsbell is an Assistant Professor in the Department of Statistical Sciences with cross-appointments in the Departments of Family and Community Medicine and Computer Science at the University of Toronto. She is interested in the development of statistical learning and inference methods that address key challenges of analyzing modern observational health data, including extreme missing data, complex measurement error, data heterogeneity, and bias and fairness. Jesse鈥檚 work is primarily supported by NSERC, CIHR, and the Ontario Ministry of Health. Website: