Biomedical Engineering Reference
In-Depth Information
true survival time. But this conclusion does not hold for the case of less frequent
visits (i.e., Month=3) because the intervals are getting larger and the midpoint ap-
proximation would be further away from the true survival time. In the case of less
frequent visits, the bias from \Cox.Right" could be 40% to 50% and \Cox.Mid" could
be 20% to 30%.
T TABLE 11.1: Bias (%) for Treatment Compariso n
p Month
IntCox
Cox.Right
Cox.Mid
0.1
1
2.17
2:93
0.02
0.2
1
0:21 8:73
3:09
0.3
1
-0.86 13:14 5:78
0.4
1
2.14 13:18 3:98
0.5
1
2.00 14:72 3:89
0.6
1
0.52 21:67 9:39
0.7
1
0.16 15:82 2:55
0.8
1
0.40 19:68 3:90
0.1
3
6.42 18:48 12:21
0.2
3
0.65 26:65 16:81
0.3
3
6:58 30:51 17:91
0.4
3
8:41 37:36 21:33
0.5
3 11:67 39:37 23:02
0.6
3 11:81 44:75 27:21
0.7
3 12:96 51:84 32:73
0.8
3 10:71 39:13 18:39
Similar conclusions hold for the other parameters. Figure 11.1 illustrates the
bias as a function of the probability of censoring for all four parameters of 1 ; 2 ; 3 ,
and 4 for the two visit schedules. In this gure, the solid lines denote \IntCox,"
the dashed middle lines denote \Cox.Mid" and the bottom dashed lines with points
denote \Cox.Right." We can see that the \IntCox" is always best among the three
methods and \Cox.Right" is the worst. For the case of more frequent visits, \IntCox"
Search WWH ::




Custom Search