Biomedical Engineering Reference
In-Depth Information
Finkelstein's method, in terms of analyzing two well-known interval-censored
data sets: a retrospective study of time to cosmetic deterioration among breast
cancer patients, and an AIDS study of drug resistance to zidovudine. Sun and
Chen (2010) conducted Monte Carlo simulation studies to compare the per-
formance of Finkelstein's method with conventional methods when equal or
unequal assessment intervals and different shapes of hazard functions between
treatment arms are assumed, and recommended using bona fide interval-
censored methods, for example, Finkelstein's method, for interval-censored
time-to-event data whenever possible. These works illustrated the potential
usefulness of interval-censored methods in PFS analysis, but neither of them
were studied thoroughly in practical settings where potential bias might arise.
In this chapter, we consider several typical scenarios in oncology clinical
trials when potential biases may arise, and compare both conventional meth-
ods and interval-censored methods for PFS data analysis based on extensive
Monte Carlo simulation studies. As we are primarily concerned about the im-
plications of using interval-censored methods in phase III confirmatory oncol-
ogy trials, we limit our attention to treatment effect estimation and hypothesis
testing in studies with two randomized treatment arms. We do not aim to ex-
clusively cover all possible scenarios that may lead to potential bias in practice.
Instead, we try to illustrate the utility of conventional and interval-censored
methods in some common real-world scenarios for PFS data analysis, and try
to provide recommendations on additional analysis strategies when data ab-
normality arises. In Section 10.2 we briefly introduce the statistical methods
for PFS data analysis. The corresponding Monte Carlo simulation setup and
results are explained and shown in Section 10.3. In Section 10.4 we discuss
the advantages and disadvantages of different statistical methods based on the
simulation study results, as well as recommendations to avoid potential bias
in practice. All analyses are conducted with SAS 9.2, and all programs used
for simulations are available upon request.
Search WWH ::
Custom Search