Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 282

Therefore, we consider the following scenarios for illustration purposes:
1. Scenario I: Per-protocol compliance for response assessments. This is
the ideal scenario in the PFS data collection process, where assessments
are equal between randomized arms and there are no deviations from
scheduled assessments. This scenario may help us understand the actual
performance of competing methods considered given observed interval-
censored data.
2. Scenario II: Random deviations from scheduled assessments. In both
arms, we assume that the actual assessments may uniformly deviate
from schedules either 1 or 2 weeks at the left-point and right-point
of the true progression time. Among all patients, 40% have scheduled
assessments, 20% have such random deviations on both left-point and
right-point, 20% have deviations on left-point only, and 20% have devia-
tions on right-point only. Such imperfect protocol adherence is virtually
unavoidable in practice, and this scenario helps us understand whether
random deviations from scheduled evaluations may be of a concern for
potential bias.
3. Scenario III: Randomly missing scheduled assessment. In both arms,
we assume that 20% of subjects may randomly miss one scheduled as-
sessment that is closest to true progression time. Such data abnormali-
ties may arise when patient noncompliance occurs randomly in blinded
studies, and may help us understand the impact of randomly missing
scheduled assessments.
4. Scenario IV: Early determination of progression in experimental arm.
We assume that 40% of subjects in experimental arm have progression
disease (PD) as soon as they occur. This often occurs in open-label
studies, when worsening of symptoms or concerns of toxicity arise in
experimental arms.