Table 2.
The effect of surrogate distributional discrepancy and delay in obtaining primary outcomes on power and expected treatment failures (standard deviation) for the S-P replacement algorithm.
| S-P Replacement algorithm | |||||||
|---|---|---|---|---|---|---|---|
|
pPA = pSA pPB = pSB |
pSA ↑ Δ pSB ↑ Δ |
pSA ↓ Δ pSB ↓ Δ |
pSA ↑ Δ pSB ↓ Δ |
pSA ↓ Δ pSB ↑ Δ |
|||
| Δ = 10% | Trt A surrogate success probability (pSA) = | 0.70 | 0.80 | 0.60 | 0.80 | 0.60 | |
| Trt B surrogate success probability (pSB) = | 0.30 | 0.40 | 0.20 | 0.20 | 0.40 | ||
| Optimal allocation Trt A target = | 0.60 | 0.59 | 0.63 | 0.67 | 0.55 | ||
| Delay = 25% | Power | 91 | 90 | 90 | 90 | 90 | |
| Failures (std dev) | 28.4 (3.6) | 28.3 (3.6) | 28.5 (3.7) | 28.2 (3.7) | 28.8 (3.6) | ||
| Delay = 50% | Power | 91 | 90 | 89 | 89 | 91 | |
| Failures (std dev) | 28.4 (3.6) | 28.5 (3.6) | 28.2 (3.6) | 27.7 (3.6) | 29.2 (3.6) | ||
| Delay = 75% | Power | 90 | 91 | 90 | 90 | 90 | |
| Failures (std dev) | 28.3 (3.6) | 28.6 (3.5) | 27.8 (3.7) | 27.2 (3.6) | 29.5 (3.7) | ||
| Δ = 20% | Trt A surrogate success probability (pSA) = | 0.70 | 0.90 | 0.50 | 0.90 | 0.50 | |
| Trt B surrogate success probability (pSB) = | 0.30 | 0.50 | 0.10 | 0.10 | 0.50 | ||
| Optimal allocation Trt A target = | 0.60 | 0.57 | 0.69 | 0.75 | 0.50 | ||
| Delay = 25% | Power | 91 | 90 | 90 | 90 | 91 | |
| Failures (std dev) | 28.4 (3.6) | 28.3 (3.6) | 28.7 (3.6) | 28.1 (3.7) | 29.1 (3.6) | ||
| Delay = 50% | Power | 91 | 90 | 90 | 88 | 90 | |
| Failures (std dev) | 28.4 (3.6) | 28.5 (3.8) | 28.5 (3.9) | 27.6 (3.9) | 29.8 (3.8) | ||
| Delay = 75% | Power | 90 | 90 | 89 | 88 | 90 | |
| Failures (std dev) | 28.3 (3.6) | 28.8 (3.7) | 27.8 (3.9) | 26.7 (3.9) | 30.4 (3.7) | ||
| Δ = 30% | Trt A surrogate success probability (pSA) = | 0.70 | 0.99 | 0.40 | 0.99 | 0.40 | |
| Trt B surrogate success probability (pSB) = | 0.30 | 0.60 | 0.01 | 0.01 | 0.60 | ||
| Optimal allocation Trt A target = | 0.60 | 0.56 | 0.86 | 0.91 | 0.45 | ||
| Delay = 25% | Power | 91 | 90 | 90 | 89 | 91 | |
| Failures (std dev) | 28.4 (3.6) | 28.1 (3.6) | 29.5 (3.7) | 28.2 (3.8) | 29.4 (3.6) | ||
| Delay = 50% | Power | 91 | 89 | 89 | 88 | 90 | |
| Failures (std dev) | 28.4 (3.6) | 27.6 (4.2) | 30.7 (4.6) | 28.2 (4.3) | 30.6 (3.8) | ||
| Delay = 75% | Power | 90 | 87 | 86 | 84 | 90 | |
| Failures (std dev) | 28.3 (3.6) | 27.4 (4.6) | 30.7 (6.0) | 28.5 (5.6) | 31.2 (3.8) | ||
The sample size (N = 62) was selected that yielded simulated power of approximately 90% under complete randomization. S-P replacement algorithm method implemented doubly-adaptive biased coin design (γ = 2). Simulations = 5000 replications (α = 0.05, two-sided). Surrogate weight ws = 0.5.