Biomedical Engineering Reference
In-Depth Information
to ensure no systematic difference in evaluation time or progression evaluation
between both arms. Depending on whether sucient information is retained
in the observed interval-censored data, the additional information loss may
or may not have significant impact on estimation and hypothesis testing. For
example, in our simulation studies, more severely biased results of scenario
II than scenario I are only obtained when median PFS in C arm is 12 weeks
and event rate is 60%, for which the information retained in observed PFS
data is intuitively the least among all simulation settings. Our limited sim-
ulation studies also suggest that, while virtually no study can guarantee all
assessments at scheduled intervals exactly, as long as noncompliance is ran-
dom and there is no substantial information loss through interval-censored
sampling, the impact on estimation and hypothesis testing may be insignifi-
cant. Recently, Kay et al. (2011) expressed a similar consideration based on a
re-analysis and additional simulation studies of a completed phase III oncology
trial to address the FDA's concern on evaluation-time bias.
The latter source of information is more relevant to the potential bias
issue discussed in Section 10.1.2, as systematic differences in progression de-
termination or evaluation time always raise suspicions of potential bias in
practice. Our simulation studies confirm that the bias could be substantial if
there are systematic differences in progression assessment between treatment
arms (scenarios IV and V). Furthermore, conventional approaches may erro-
neously conclude the existence of treatment effect even if there is none, which
is consistent with findings by Freidlin et al. (2007).
Given the robustness of interval-censored methods under different scenar-
ios in our limited simulation studies, we would also recommend using interval-
censored methods to reassure the analysis results for confirmatory oncology
clinical trials. If we have to use conventional approaches in practice due to the
absence of commercial software implementation for interval-censored meth-
ods, it is advised to interpret the results from conventional approaches with
caution. Similar to Sun and Chen (2010), we find that mid-point imputation
Search WWH ::
Custom Search