Biomedical Engineering Reference
In-Depth Information
here to study the Freidlin et al. (2007) method and the other methods under
ATD and CIA, as well as under informative censoring, which was not studied
in Fay and Shih (2012).
Here we do simulations to examine the size and power of the different
tests under different conditions. We consider eight different two-sample lo-
grank tests. The tests are the permutation test using the permutational cen-
tral limit theorem (PCLT), the score test (score), the exact permutation test
by Monte Carlo simulation (ex), three different tests based on multiple im-
putation (within subject resampling): the method proposed by Huang et al.
(2008) (hly), the modification using the permutational central limit theorem
(pclt), and the modification using Monte Carlo simulation (mc), the Freidlin
et al. (2007) method (2pt), and the naive right endpoint imputation (rei).
The simulation conditions are designed to mimic a study where the end-
point is cancer progression. For this study there are n = 100 in each treatment
arm. In the simulation, true progression times are generated that follow an
exponential distribution with different median values. Patients are assumed
to be entered uniformly into the study over a 24-month period and will have
a minimum of 24 weeks of follow-up. For each patient, regularly scheduled
doctors visits will occur every 4 weeks with a 1-week window (that is, they
are uniformly distributed on all the days between 1 week earlier and 1 week
later than the target date); scheduled event assessments occur every 8 or 12
weeks (every second or third doctors visit). We assume that the scheduled and
unscheduled assessments will occur at the doctors visits.
For each visit there is an underlying probability that an event assessment
will occur. For scheduled assessment times, this probability is 1. For other vis-
its we modify the probability of assessment according to the assessment prob-
ability distributions described below. We examine thirty-two different simu-
lation conditions, which are described by eight scenarios (labeled 1 through
8), each examined under four different assessment probability distributions
(labeled a through d). The eight scenarios describe different survival distribu-
Search WWH ::
Custom Search