Fig 1.
Kaplan-Meier curves from the ITT analysis including all patients (image A), with only treatment switchers in the placebo arm (image B), and with only non-switchers in the placebo arm (image C).
Sample size in the placebo arm is equal to 139, 111 and 28 patients in plots A, B and C, respectively. Sample size in the Everolimus arm is equal to 277 patients in all plots.
Fig 2.
Power with the log-rank (LR) for different total number of events under two different models: The proportional hazards model and the exponential progression switching model presented in Section 3.3 where patients switch treatment after disease progression with probability p.
Fig 3.
Markov chain states and transition rates.
Fig 4.
Plot A shows the survival probability for non-progressed patients (red), progressed and non-switched patients (green) and progressed and switched patients (blue) assuming p = 0.5, ,
,
.
In plot B, we show the resulting survival functions for the control (teal) and experimental (brown) arms, and also as a comparison, the control arm under proportional hazards (pink).
Fig 5.
Hazard ratio from the RECORD-1 trial and hazard ratio functions obtained with the model based on exponential progression switching (plot A), and corresponding weights functions from Eq (6) (plot B), for p = (0, 0.2, 0.4, 0.6, 0.8, 1), with
and
.
The weights using p = 0 are equivalent to those from the standard log-rank (LR) test.
Table 1.
Median OS, median PFS, sample size, and total number number of deaths based on the RECORD-1 trial.
Fig 6.
Power of the modified weighted log-rank (mWLR) test and the log-rank (LR) test (plot A) and efficiency of the mWLR test with respect to the LR test (plot B) assuming p = 1 and p′ = 1 for values of that range from 5 to 10 months and the corresponding hazard ratio η.
Fig 7.
Efficiency of the modified weighted log-rank (mWLR) test with respect to the log-rank (LR) test assuming matching values of p and p′ (i.e., (p = 0, p′ = 0), (p = 0.1, p′ = 0.1), …, (p = 1, p′ = 1)) for a fixed value of months.
Fig 8.
Power of the modified weighted log-rank (mWLR) test and the log-rank (LR) test (plot A) and efficiency of the mWLR test with respect to the LR test (plot B) assuming a fixed value of months, a fixed value of p′ = 1, and varying the value of p between 0 and 1.
Fig 9.
Efficiency between the modified weighted log-rank (mWLR) test and the log-rank (LR) test assuming a fixed value of months and varying both p and p′ between 0 and 1.
Values above 100% favor mWLR and values below 100% favor LR.
Fig 10.
Efficiency between the modified weighted log-rank test (mWLR) with respect to the test based on the restricted mean survival time (plot A), and efficiency between mWLR test with respect to the Max Combo test (plot B) assuming a fixed value of months varying p between 0 and 1, and p′ between 0.5 and 1.
Values of efficiency over 100% favor the mWLR test and values below 100% favor the test based on the restricted mean survival time or the Max Combo test.