|
Spring 2007
STATISTICS
COLLOQUIUM
Wednesday, March 21, 2007
3:30-4:00—Refreshments
4:00-5:00—Talk
Yost Hall, Room 101
Mark Schluchter, PhD
Department of Epidemiology and Biostatistics, CASE
Joint Modeling of longitudinal and Time-to-event Data - Applications to Studies of Progressive Renal Disease and Cystic Fibrosis
Abstract
This talk will illustrate an approach to joint modeling of a continuous longitudinal variable Y, measured longitudinally, and a time-to-event outcome, possibly right censored. The model used assumes that Y follows a linear mixed model with random slope and intercept, where the patient-specific slope and intercept, along with some transformation of the time-to-event, follow a trivariate normal distribution. Two examples will be used to illustrate use of this modeling approach for different purposes. The first example is from a randomized trial of patients with progressive renal disease, where the primary outcome is rate of change in Y=glomerular filtration rate (GFR), and the time-to-event variable is time to dropout from the study due to end stage renal disease. Here, the goal of the analysis is to estimate and compare mean rates of GFR decline among treatment groups, while accounting for biases caused by nonignorable dropout, also called informative dropout or informative censoring. The second example focuses on estimating the relationship between longitudinal changes in Y= FEV1 % predicted and survival in patients with cystic fibrosis. Here, the focus is on using the longitudinal FEV1 data to predict the time-to-event outcome (survival, or age at death), and on obtaining predictions of individual patients' quantities such as their FEV1 % predicted at a specified age, or their survival time (age at death).
|
