Biomedical Engineering Reference
In-Depth Information
5. Scenario V: Early determination of progression in control arm. Contrary
to scenario IV, we assume that 40% of subjects have progression disease
(PD) in control arm as soon as they occur. This often occurs in open-
label studies because of a concern regarding the lack of ecacy in the
control arm. Both scenarios IV and V aim to mimic situations where
a systematic difference in evaluation time exists, or assessment of pro-
gression is impacted by the investigator's knowledge of adverse events
leading to unmasking of treatment arms in open-label or blinded studies.
The above scenarios cover some typical situations that may introduce potential
bias in PFS analysis, as discussed in Section 10.1.2. Some other scenarios,
such as unbalanced drop-out and missing data in PFS, may involve dependent
censoring issues in interval-censored data analysis that are beyond the scope of
this chapter. Situations involving unequal assessment intervals without further
data abnormality, which occur less often in practice than equal assessment
intervals for apparent concerns of bias, have been considered by Sun and Chen
(2010). Hence, these scenarios are not included here.
10.3.2
Results on Point Estimation
Table 10.2 summarizes point estimates under scenarios I through V with equal
scheduled assessment intervals (every 6, 8, or 12 weeks) in both arms, when
the true hazard ratio between T and C is 0.67 ( = 0:4), the median pro-
gression time in C is 12 weeks, and the event rate is 80%. Results from the Cox
regression of exact failure times served as benchmarks. We use relative bias,
which is defined as the percentage of bias compared with the true parameter
value, throughout our analysis. The asymptotic standard deviation (ASD),
Monte Carlo empirical standard error (ESE), and 95% coverage probability
(CP), and root mean square errors (RMSE) are also reported.
Search WWH ::
Custom Search