THESIS DEFENSE ABSTRACT
The Analysis
of Informatively Coarsened Discrete Time-to-Event Data
Michelle
Shardell, PhD Candidate, Johns Hopkins Department of Biostatistics
In many prospective studies, participants are evaluated for the
occurrence of an absorbing event of interest (e.g., HIV seroconversion) at
baseline and a common set of pre-specified visit times after enrollment. Since
participants often miss scheduled visits, the underlying visit of first
detection may be interval censored, or more generally, coarsened.
Interval-censored data are usually analyzed assuming non-informative censoring,
a special case of coarsening at random (CAR).
We posit a class of models for the event-time distribution that loosen the CAR
assumption and use the EM algorithm for estimation. To perform inference on the
estimated survivor functions, we propose an extension of the logrank test
utilizing the EM-based estimates. We extend this methodology to estimate
regression parameters for a discrete-time proportional hazards model with a
low-dimensional covariate. Since CAR is usually not testable and often
scientifically implausible, estimation is performed by incorporating elicited
expert opinion about the relationship between event times and visit compliance
into the model. The procedures are illustrated using data from the AIDS Link to
the Intravenous Experience (ALIVE) study, an observational study of HIV
infection among injection drug users in Baltimore, Maryland. Performance of our
method is assessed via simulation studies.
Return to Upcoming
Events List
|
Return
to Home Page
|