Biomedical Engineering Reference
In-Depth Information
of different patterns of censoring data, including interval-censored data with
PH model).
10.3
Monte Carlo Simulation Studies
10.3.1
Simulation Setup
We consider Monte Carlo simulation studies that mimic hypothetical oncology
Phase III clinical trials with two arms (C:control versus T: treatment) based
on 1:1 randomization, with a total sample size of 400 patients. We assume
every subject has a true progression time that satisfies the predefined criteria
for progression. The true progression time is assumed to follow an exponential
distribution with a median PFS of 12 weeks, or 24 weeks in arm C. The
hazard ratio (HR) between arms T and C is assumed to be either 0.67, or 1,
with a regression coecient of interest of log(0:67) = 0:40, or log(1) = 0. To
generate the simulated data sets, each exact progression time may be censored
by a prespecified time interval to result in a noninformative censoring . For
all simulation studies, we assume equal assessment schedules in both arms
that occur every 6 weeks, 8 weeks, or 12 weeks. Under each setting, 5,000
replicates are generated. For simplicity, we assume a constant patient accrual
rate and a common study closure date for all patients. We further assume that
no information is available after the last observed assessment. It is assumed
that we have an approximately 80% or 60% event rate, such that the trial
has about 95% or 85% power for detecting a 33% hazard reduction under 5%
Type I error (two-sided) if continuous event times are observed. In order to
keep the event rates at desired proportions, an additional assessment is added
correspondingly at the end of the study, if needed.
Throughout our simulation studies, we report results of hypothesis testing
and treatment effect estimates based on conventional methods and interval-
Search WWH ::
Custom Search