Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 293

TABLE 10.8: Type I Error Rate (%) at = 5% (two-sided): 80% Event

Rate, 24 Weeks Median in C

Scenario Interval Right-point Mid-point Finkelstein's Sun's

6 4.9 4.9 5.3 4.9

I 8 4.8 4.8 5.3 5.1

12 4.3 4.3 4.8 4.8

6 5.0 4.8 5.1 4.9

II 8 4.9 4.8 5.0 5.0

12 5.2 5.0 4.9 5.1

6 4.9 4.6 5.2 5.0

III 8 4.8 4.6 5.1 4.8

12 4.3 4.3 4.7 4.8

6 8.0 8.0 5.5 5.3

IV 8 10.3 10.3 5.6 5.1

12 19.7 19.7 4.9 5.2

6 7.3 7.3 5.3 5.5

V 8 11.0 11.0 5.5 5.3

12 20.9 20.9 4.9 5.0

ExactlogranktestTypeIerror4.8%.

Meanwhile, under scenario IV, the severe underestimation (negative bias) from

conventional approaches naturally lead to extremely under-powered test re-

sults, for example, 48% for 80% event and 19% for 60% event when median

PFS in C arm is 12 weeks. On the contrary, we find over-powered test re-

sults from conventional methods under scenario V, for example, almost 100%

for both 80% and 60% event rates. We observe comparable empirical perfor-

mances of power rates between Finkelstein's score test and Sun's generalized
logrank test.