Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 273

10.1.2 Potential Bias in PFS Analysis

While the planned response assessment schedules are designed to minimize any

potential biases that may be introduced by periodic assessments, deviations

from scheduled assessments and variations in determining progression status

are common in practice, and thus lead to different types of bias and vari-

ability. Dancey et al. (2009) and Bhattacharya et al. (2009) summarized three

types of biases that may occur in evaluating PFS endpoint: (i) evaluation-time

bias|introduced by the random or systematic dierences in evaluation times

according to treatment arms; (ii) evaluation bias| introduced by investiga-

tor's knowledge of adverse events leading to unmasking of treatment arms;

(iii) attrition bias |introduced by the imbalance between treatment arms in

missing assessments that impact assessment of progression and/or differen-

tial patients drop-out rates. Evaluation-time bias represents a major source

of potential bias for PFS analysis. It may occur when the differences between

the actual progression event and documented progression date are systematic

between randomized arms. In practice, several scenarios may lead to potential
evaluation-time bias:

1. Unscheduled response assessments may occur more likely in one arm

than the other arm. In open-label studies, early reevaluation of patients

may be more likely in the control arm than in the experimental arm be-

cause of a concern regarding lack of ecacy. Alternatively, unscheduled

examination may also occur more frequently in the experimental arm

than in the control arm because of worsening symptoms and concerns

regarding toxicity.

2. Imperfect protocol adherence may exist in practice. While the sched-

uled response assessments are designed to be equal and symmetric be-

tween treatment arms in many cases, various random or systematic de-

viations from scheduled assessments occur frequently in the real world.

One such example is a randomized open-label phase III study of com-