Presenter: George Stefan
Supervisory Committee: Eleanor Pullenayegum (Supervisor), Olli Saarela, Aya Mitani
Date and Time: Thursday, May 9, 2024, 12-2pm EST
Location: 155 College Street, Health Science Building, Room 734
Abstract: Longitudinal data can be helpful in estimating disease trajectories based on potential prognostic factors, either at the population or the individual level. However, observation times or dropout may be associated with the health outcome, which if left unaddressed may introduce bias in the estimation. Methods have been developed by Lin, Scharfstein and Rosenheck, Lin and Ying, and Bůžková and Lumley to analyze irregularly observed longitudinal data under various assumptions. Lin et al. use an approach based on inverse-intensity weighted generalized estimating equations (IIW-GEEs) and Lin-Ying employ more traditional estimating equations in which a quasi-residual term containing the intensity model estimates is incorporated to yield unbiased longitudinal trajectories. However, the Lin-Ying methods are unable to deal with outcome-dependent follow-up and informative dropout at the same time and IIW-GEEs are unable to account for informative dropout. Methods have since been proposed to address these two issues simultaneously, but only via shared latent variables, or under a relatively strict set of covariate assumptions. We will propose methods which extend the Lin et al. and Lin-Ying methods to incorporate outcome-dependent follow-up and informative dropout. We will demonstrate asymptotic unbiasedness of our estimates through mathematical derivations, show how our statistical method can be implemented via computational software and verify estimates’ bias and variability via simulation studies. We will derive efficient standard errors and compare these with naive standard errors which do not take into account the uncertainty in the weights.
All are welcome!