Biomedical Engineering Reference
In-Depth Information
TABLE 10.11: Empirical Power (%) at = 5% (two-sided): HR = 0.67,
60% Event Rate, 12 Weeks Median in C
Scenario
Interval
Right-point
Mid-point
Finkelstein's
Sun's
6
86.3
86.3
86.8
86.3
I
8
86.8
86.8
86.2
86.1
12
84.7
84.7
84.9
84.8
6
85.0
85.9
86.8
85.7
II
8
84.2
85.7
86.4
85.3
12
78.3
82.2
84.1
83.6
6
86.3
86.7
86.1
86.0
III
8
85.2
85.5
85.5
85.1
12
84.1
83.4
84.7
84.6
6
63.5
63.5
86.5
86.1
IV
8
47.4
47.4
85.5
84.7
12
18.5
18.5
85.1
84.7
6
96.4
96.4
86.6
86.1
V
8
98.5
98.5
86.0
85.8
12
99.6
99.6
85.6
84.9
Exactlogranktestpower86.8%.
event rate is low. Moderately random protocol noncompliance, such as devia-
tions from scheduled assessment (scenario II) and missing assessment (scenario
III), seem to have reasonable impacts on point estimation. However, when a
systematic difference in progression evaluation between randomized arms oc-
curs, the point estimate tends to be severely biased, and the direction of bias
depends on the treatment arm (scenarios IV and V).
Hypothesis testing results from conventional approaches are generally con-
sistent with the findings in point estimation. Empirical Type I error and power
rates approximately remain at nominal levels when assessment frequency and
event proportion are high. When low assessment frequency and event pro-
Search WWH ::
Custom Search