Guest Lecture: Nicholas Mitsakakis, MSc, PhD, P.Stat.
Date: Thursday, February 27
Time: 1:30 pm - 3:00 pm
Location: Rm. 101, Health Sciences Addition
Probabilistic Graphical Models and its use in Health Sciences
Nicholas Mitsakakis, MSc, PhD, P.Stat.
Biostatistician of THETA
Assistant Professor
Institute of Health Policy, Management and Evaluation
University of Toronto
Abstract
Probabilistic graphical models are tools in Statistics and Computer Science that use the representation of a graph in order to represent and examine dependencies and probabilistic relationships between a given set of random variables. Methodological developments on graphical models are increasingly active in both fields of statistics and machine learning, mainly motivated by the need of analyzing high-dimensional data, often generated in high-throughput biology. As such, graphical models have been and continue to be used for tackling complex research questions in the biological sciences.
In this talk I aim to discuss the applicability and utility of probabilistic graphical models in the health sciences. First, I will discuss some preliminary results of a scoping review of clinical epidemiology published studies that used graphical models. Afterwards, I will present two studies illustrating the use of graphical models in health research. In the first study, undirected graphical models are used for modeling the relationship and dependence between the attributes of a disease-specific preference based utility instrument (PORPUS), using data from prostate cancer patients. In the second study, Bayesian Networks, a type of directed graphical models, are used for investigating the nature and risk factors of sleep disturbance in liver transplant patients, and its effect on health related quality of life.
