McGill's Seminar Series in Quantitative Life Sciences and Medicine
Sponsored by CAMBAM, QLS, MiCM and the Ludmer Centre
Title: Learning from data
Speaker: Karen Kopciuk, University of Calgary
When: Tuesday, October 22, 12-1pm
Where: McIntyre Medical Building, room 1034
Data features can drive methodological ideas and inspiration. In this talk, I will provide an overview of three projects where data features drove the development of statistical methods that led to more learning from the data. For Lynch Syndrome families, many members experienced a second cancer so estimating the age-at-onset of a subsequent cancer was carried out in a progressive multi-state model. When genes or metabolites can be naturally grouped, selection of groups of variables or individual variables within a group is an important initial step in modelling the data. We developed an adaptive group bridge selection method in the semiparametric accelerated failure time model to address this limitation for variable selection for survival outcomes. The tissue of origin plays a critical role in the etiology and progression of cancer. With multiple measures of metabolites from cancer patients with different cancers from the same organ system, we can identify shared and unique metabolic pathways. We are developing mixed models that incorporate the not missing at random assumption and a Bayesian prior for the random effects.