Bias Assessment in Progression-Free Survival Analysis Using Interval-Censored Methods - Interval-Censored-Time-to-Event Data: Methods and Applications - page 294

TABLE 10.9: Type I Error Rate (%) at = 5% (two-sided): 60% Event

Rate, 24 Weeks Median in C

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

6 5.5 5.5 5.7 5.4

I 8 5.7 5.7 5.7 5.4

12 5.5 5.5 5.7 5.6

6 5.4 5.2 5.6 5.4

II 8 5.4 5.3 5.5 5.6

12 5.7 5.4 5.5 5.1

6 5.4 5.3 5.4 5.1

III 8 5.6 4.9 5.4 5.0

12 5.4 5.1 5.4 5.3

6 8.1 8.1 5.6 5.0

IV 8 10.6 10.6 5.9 5.3

12 18.5 18.5 5.8 5.4

6 7.3 7.3 5.6 5.1

V 8 10.0 10.0 5.8 5.3

12 17.3 17.3 5.7 5.4

ExactlogranktestTypeIerror5.3%.

10.3.4 Summary of Simulation Results

Under all the settings and scenarios considered, interval-censored methods

(e.g., Finkelstein's method and Sun's generalized logrank test) are robust and

reasonably accurate in terms of point estimation and hypothesis testing.

Results from conventional approaches, on the contrary, are highly depen-

dent on the overall information retained in the observed PFS data because they

treat observed data as exact event times. Factors such as overall assessment
frequency and event proportion would moderately influence point estimation

in well-conducted studies, for example, slightly underestimated treatment ef-

fect (negative bias in log HR, toward null) when assessment is infrequent or