. Author manuscript; available in PMC: 2018 Oct 1.

Published in final edited form as: Clin Trials. 2017 Feb 1;14(5):432–440. doi: 10.1177/1740774517692302

Table 1.

Simulation results for designs with a control arm in the null scenario with all p_k = 0.20, for N=250. n̄ = mean per-arm sample size. Each η_m = Pr(N_C > N_k + m), the probability that the number of patients randomized to arm C is at least m larger than the number randomized to arm E_k. Values in the row E₁ – E₄ are per-arm.

Method

Arm

Pr(Select)

Pr(Stop)

n̄(95% CI)

η₁₀, η₂₀, η₃₀

AR(1,0)

–

33 (10, 63)

–

E₁–E₄

0.01

0.67

42 (10, 135)

0.4, 0.27, 0.15

Total

202 (70, 250)

AR (\frac{1}{2}, 0)

–

40 (13, 69)

–

E₁ – E₄

0.01

0.74

40 (10, 109)

0.44, 0.32, 0.19

Total

200 (70, 250)

AR (\frac{n}{2 N}, 0)

–

38 (13, 65)

–

E₁ – E₄

0.01

0.73

40 (10, 116)

0.43, 0.3, 0.18

Total

199 (70, 250)

AR(1, 0.1)

–

38 (20, 63)

–

E₁ – E₄

0.01

0.74

41 (10, 124)

0.44, 0.3, 0.16

Total

200 (70, 250)

–

58 (17, 98)

–

E₁ – E₄

0.01

0.81

36 (10, 82)

0.6, 0.45, 0.33

Total

200 (70, 250)