Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 275

disease progression between treatment arms. The former one may be intro-

duced by protocol noncompliance that may or may not depend on treatment
assignment. Meanwhile, the reasons that lead to different withdrawal rates

between treatment arms are often complex but may lead to informative cen-

soring, thereby violating the independence censoring assumptions, as discussed

by Bhattacharya et al. (2009).

Previous research has shown that ignoring the evaluation-time bias and

evaluation bias may lead to incorrect and misleading PFS analysis results.

For example, Panageas et al. (2007) explored the situations when overesti-

mation of median PFS is possible if failing to recognize the interval-censored

data structure of PFS; Sun and Chen (2010) explored the possibility of in-

troducing bias if unequal assessment intervals are scheduled in different arms;

Freidlin et al. (2007) discussed that unscheduled tumor assessments may er-

roneously conclude the existence of treatment effect even when there is none;

Bhattacharya et al. (2009) considered that differences in assessment time in-

tervals between treatment arms may account for nonexisting treatment effect,
and suggested various sensitivity analysis approaches to assess the robustness

of analysis results.

10.1.3 Conventional PFS Analysis and Interval-Censored

Methods

In the pharmaceutical industry, the standard methods for right-censored time-

to-event data analysis are (i) Kaplan{Meier estimates for median event time

and related distributional quantities; (ii) logrank tests for hypothesis test-

ing of treatment effect; and (iii) semiparametric Cox proportional hazard

(PH) model for treatment effect estimation (Kalbfleisch and Prentice (2002)).

For interval-censored data such as PFS, there is no widely accepted interval-

censored method. After reviewing FDA briefing documents for new oncologic

drug applications, as well as \Guidance for Industry: Clinical Trial Endpoints

for the Approval of Cancer Drugs and Biologics" (FDA (2007)), we found that