"Nonparametric Multi-state Analysis of Longitudinal and Survival Data "
Liang Li, Ph.D., Cleveland Clinic
November 30, 2010 @ 3:30 pm - 4:30 pm
Location: Blockley Hall - Room 701
Biostatistics
TITLE: Nonparametric Multi-state Analysis of Longitudinal and Survival
Data
ABSTRACT: Most methodology for joint modeling of longitudinal and
survival data approach the problem from a subject-specific perspective
via shared parameter model or its variants. This paper proposes a new
nonparametric procedure that produces population averaged inference,
leading to multi-state probability curves with the states defined
jointly by longitudinal and survival data. To account for informative
censoring and flexible nonlinear shapes of the longitudinal profiles, a
bias corrected penalized spline regression is applied to estimate the
longitudinal profile for each subject. The probability curves are then
estimated sequentially based on the observed survival data and the
smoothed longitudinal profiles. Simulation Extrapolation (SIMEX) method
is employed to correct the deconvolution bias caused by the measurement
error of the longitudinal data. A bootstrap test procedure is developed
to compare multi-state probability curves between different groups. We
present theoretical justification of the procedure along with simulation
results. The procedure was applied to data from the African American
Studies of Kidney Disease and Hypotension (AASK), a multi-center
clinical trial. This is joint work with Bo Hu (Cleveland Clinic) and Tom
Greene (University of Utah).
