Evaluation of a multi-arm multi-stage Bayesian design for phase II drug selection trials – an example in hemato-oncology

Table 2 Simulation results in terms of absolute bias based on a fixed sample size for increasing benefit of the experimental arm– all priors on p_k (k=A,B,C) are non-informative Be (1,1) priors; p₀=0.30; n=40 or 100 patients per arm

Sample size	True benefit	Posterior mean estimate biases		Mean square errors		Decision criterion 1		Criterion 2	Criterion 3
	dB	p_A	p_B	p_A	p_B	A	B	B	B
40	0.00	0.0086	0.0090	0.0048	0.0048	0.4850	0.4835	0.5012	0.1414
	0.05	0.0082	0.0078	0.0048	0.0052	0.4854	0.2982	0.6358	0.2399
	0.10	0.0100	0.0042	0.0050	0.0054	0.4798	0.1632	0.7440	0.3518
	0.15	0.0097	0.0014	0.0049	0.0055	0.4798	0.0753	0.8352	0.4803
	0.20	0.0098	−0.0001	0.0049	0.0057	0.4795	0.0297	0.9028	0.6132
	0.25	0.0096	−0.0013	0.0049	0.0056	0.4800	0.0093	0.9477	0.7322
	0.30	0.0095	−0.0049	0.0049	0.0055	0.4804	0.0027	0.9731	0.8271
	0.35	0.0088	−0.0070	0.0048	0.0051	0.4841	0.0005	0.9886	0.8999
	0.40	0.0109	−0.0087	0.0048	0.0047	0.4749	0.0001	0.9958	0.9477
	0.45	0.0092	−0.0116	0.0049	0.0043	0.4808	0.0000	0.9984	0.9755
100	0.00	0.0042	0.0032	0.0020	0.0020	0.4863	0.4925	0.4960	0.0460
	0.05	0.0037	0.0030	0.0020	0.0022	0.4885	0.2156	0.7042	0.1363
	0.10	0.0040	0.0014	0.0020	0.0023	0.4866	0.0661	0.8503	0.2896
	0.15	0.0038	0.0012	0.0020	0.0024	0.4881	0.0128	0.9404	0.4925
	0.20	0.0040	−0.0003	0.0020	0.0024	0.4858	0.0015	0.9807	0.6875
	0.25	0.0027	−0.0009	0.0020	0.0023	0.4947	0.0001	0.9950	0.8473
	0.30	0.0046	−0.0017	0.0020	0.0023	0.4836	0.0000	0.9989	0.9362
	0.35	0.0042	−0.0032	0.0020	0.0022	0.4856	0.0000	0.9998	0.9792
	0.40	0.0032	−0.0037	0.0020	0.0021	0.4922	0.0000	1.0000	0.9952
	0.45	0.0036	−0.0047	0.0020	0.0018	0.4897	0.0000	1.0000	0.9992

N is the sample size; p_A, p_B and p_C refer to the posterior means of response probability in arms A, B and C, respectively; dB refers to the true benefit of B over A in terms of response probability. Bold data refer to the null hypothesis of absence of any treatment difference (dB=dC=0), or to the alternative hypothesis of an expected true 0.15 benefit of treatment B (dB=0.15)

ISSN: 1471-2288