Chiara Di Gravio dissertation defense – June 9
PhD candidate Chiara Di Gravio will defend her dissertation on Friday, June 9, at 8 a.m. Central Time, on-site and online. Her advisors are Jonathan Schildcrout and Ran Tao. All are invited and encouraged to attend.
The in-person event will be held in Room 11105 (11th floor) at 2525 West End Avenue.
For access to the event stream, please contact the department at biostatistics[at]vumc[dot]org.
Design and Analysis Methods for Modern Biomedical Studies with Longitudinal Outcomes
Electronic health records, existing cohort studies and clinical trials provide easily accessible data on outcome and covariates (e.g., disease status, vital signs) for most or all study subjects. However, many scientific questions require additional collection of costly variables that are the primary exposures of interest (e.g., DNA genotype, gut microbiome). When the exposure of interest is expensive to collect, resource constraints will limit the sample size. In these settings, two-phase outcome dependent sampling (ODS) studies are pragmatic solutions that allow researcher to identify the most informative subjects for expensive exposure ascertainment. In this dissertation, we aim to introduce novel two-phase ODS designs and inference procedures for settings where the outcome is collected longitudinally, and an expensive exposure needs to be ascertained retrospectively. In particular, we want to answer the questions on who we should sample for expensive exposure ascertainment, and how we can analyze the data after exposure collection.
First, we focus on settings with multivariate longitudinal continuous outcomes and introduce two designs and inference procedures that use different amounts of information to identify the most informative individuals and to estimate the parameters of interest. Importantly, we show how our approaches allow us to perform secondary analysis of data that have been previously collected in a two-phase ODS study. We demonstrate the advantages of the proposed methods through extensive simulations and an application to the Lung Health Study. Second, we focus on longitudinal binary outcomes and introduce a class of residual based designs that identifies informative individuals based on the available outcome and covariates data. We show how these designs can increase efficiency compared to existing two-phase ODS designs and introduce a semi-parametric approach to estimate the parameters of interest. Additionally, we introduce an R package that creates a framework for the design and analysis of two-phase ODS studies with longitudinal binary outcomes. Finally, we consider longitudinal ordinal outcomes and introduce and compare different analysis strategies to study the association between an expensive exposure and mortality in settings where patients who died were oversampled.
