Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 281

censored methods, namely (i) logrank tests and Cox PH model for the con-

ventional approach; (ii) Finkelstein's score test and Sun's generalized logrank
test, as well as Finkelstein's method for interval-censored Cox PH model, un-

der different simulation scenarios. When equal assessment intervals are applied

on both arms and no assessment deviation exists, mid-point and right-point

imputations of the conventional approach are identical, and only right-point

imputation results are reported. For benchmarking purposes, results from the

logrank test and Cox regression based on exact event times are also reported

to gauge the error from Monte Carlo simulations.

Our simulation study design aims to cover some common scenarios in PFS

analysis. For example, the ratio between median PFS and assessment interval

in control ranges from 1 to 4, and the maximum number of assessments ranges

from 3 to 35. Table 10.1 provides overall trial duration and the corresponding

maximum number of assessments under different assessment schedules (every

6, 8, and 12 weeks). The primary focus of this chapter is to assess the potential

bias in PFS analysis when assessment deviations and potential bias might
exist.

TABLE 10.1: Overall Duration (in weeks) and Maximum Number of Assess-

ments
Median Hazard Event Overall Max Number of Assessments

in C Ratio Proportion Duration Every 6 Every 8 Every 12

1 80% 88 15 11 8

1 60% 40 7 5 4

12 weeks 0.67 80% 108 18 14 9

0.67 60% 48 8 6 4

1 80% 172 29 22 15

1 60% 80 14 10 7

24 weeks 0.67 80% 206 35 26 18

0.67 60% 92 16 12 8