Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 301

performs slightly better than right-point imputation. Based on our experi-
ence, we agree with Sun and Chen (2010) and recommend Efron's methods
for handling ties for the Cox model in conventional approaches.
When using PFS as a primary or secondary endpoint in phase III oncology
clinical trials, based on our simulation studies, it is important to regard PFS
as interval-censored data rather than assume it to be right-censored data.
In the study design stage, we strongly recommend adopting consistent and
symmetric assessment intervals across treatment arms whenever possible. The
length of the assessment interval should be narrow enough to capture the
change of disease natural history under different therapeutic interventions,
and the frequency of assessments is crucial to determine whether the planned
number of events is sucient to achieve the desired power at the required
Type I error level. In the data analysis stage, we highly recommend carefully
reviewing the potential sources of biases from evaluation, time, evaluation
and attrition. Along with the additional sensitivity analysis currently adopted,
interval-censored methods, such as Finkelstein's method and Sun's generalized
logrank test, among many others, should also be considered for sensitivity
analysis to reassure the validity and robustness of analysis results.
Our current simulation study is far from perfect, and other different set-
tings and scenarios remain for further evaluations. The practical performance
of many competing interval-censored methods under different sample sizes,
treatment effects, assessment frequency, and imbalanced evaluation should be
further explored.
Acknowledgments
The authors would like to thank Xing Sun and Cong Chen at Merck Research
Laboratories for kindly providing the SAS macro of Finkelstein's method