Toronto, Ontario

Methods of Quantifying Within-Subject Variability for Longitudinal Sata with Irregular and Informative Observation (Proposal Presentation)

Read Post

Presenter: Fatema Johara

Supervisory Committee: Eleanor Pullenayegum (Supervisor), Linda Hiraki, Andrew Paterson

Date and Time: Friday, May 10, 2024, 1-3pm EST

Location: 155 College Street, Health Science Building, Room 734

Abstract: Variability in longitudinal outcomes is often perceived as a nuisance parameter in statistical models and is not usually estimated. However, within-subject variability may in fact be informative. For example, childhood-onset systemic lupus erythematosus (cSLE) is a chronic autoimmune disease, which is commonly affected by kidney disease or lupus nephritis (LN). Lupus nephritis (LN) is one of the most severe manifestations of cSLE where genetics play a key role in cSLE risk. Since SLE exhibits a relapsing-remitting nature due to disease activity, it may be clinically interesting to model within-subject variability of kidney function to genetic features of interest. Other examples where variability is informative include: menstrual cycle variability may inform infertility, and lack of variability in fetal heart rates may indicate compromised heart failure. While there are statistical methods available for estimating means and medians, there is no method suitable for modeling variability when the observations are irregular and informative. This gap in existing methods has motivated me to develop new methods for modeling within-subject variability under semi-parametric framework as part of my PhD thesis. After establishing a model and estimation procedure between predictor variables and an outcome, it can be challenging to determine which covariates should be included in the model. For instance, in a genetic study with numerous genetic predictors, variable selection becomes essential. Thus, further plan is to devise a technique to develop estimating equations with penalty functions to perform variable selection and parameter estimation. Finally, an R package will be created to apply the proposed methods to any real-life data application where the objective is modeling variability.

All are welcome!