Screened selection design for randomised phase II oncology trials: an example in chronic lymphocytic leukaemia

Table 1 Simulation study to evaluate and compare performance of SSD, Modified SSD, SWE and Bayesian selection strategy

Scenario	True CR (pA,pB)	Simulated overall selection probabilities
		SSD			Modified SSD			SWE			Bayesian selection strategy
		Arm A	Arm B	No Arm	Arm A	Arm B	No Arm⁺	Arm A	Arm B	No Arm	Arm A	Arm B	No Arm
1	(0.01,0.01)	0.025	0.025	0.949	0.025	0.025	0.950 (0.001)	0.500	0.500	0	0.013	0.013	0.974
2	(0.1,0.1)	0.455	0.454	0.091	0.311	0.311	0.379 (0.287)	0.500	0.500	0	0.383	0.383	0.234
3	(0.2,0.2)	0.500	0.498	0.002	0.320	0.320	0.359 (0.357)	0.500	0.500	0	0.490	0.490	0.019
4	(0.3,0.3)	0.500	0.500	0.000	0.334	0.335	0.331 (0.331)	0.500	0.500	0	0.500	0.500	0.001
5	(0.01,0.03)	0.023	0.167	0.810	0.021	0.164	0.815 (0.004)	0.315	0.685	0	0.012	0.094	0.894
6	(0.01,0.2)	0.002	0.950	0.048	0.001	0.947	0.051 (0.003)	0.003	0.997	0	0.002	0.864	0.134
7	(0.20,0.35)	0.100	0.900	0.000	0.042	0.805	0.154 (0.154)	0.099	0.901	0	0.104	0.894	0.002
8	(0.2,0.4)	0.047	0.953	0.000	0.017	0.897	0.086 (0.086)	0.046	0.954	0	0.049	0.950	0.001

The table displays a simulation study to evaluate the operating characteristics of Screened Selection Design (SSD) and Modified Screened Selection Design (Modified SSD), and compare it to Simon-Wittes-Ellenberg Selection Strategy (SWE) and Bayesian Selection Strategy, based on 1 million replications. The overall probabilities of selecting Arm A, selecting Arm B and selecting neither arm (No Arm) for each design under various scenarios are presented. Scenarios 1–4 denote situations when the rates at both arms are the same, whereas Scenarios 5–8 denote situations when the rates are different. The values in bold indicate the correct selection probabilities under specific scenario. CR: Complete Response rate, pA: true CR rate for Arm A, pB: true CR rate for Arm B. ⁺The values in parenthesis for Modified SSD indicate the probability that neither arm is selected based on the primary endpoint when the observed difference is less than the clinically-relevant value of 5%. This then directs the decision making process to other additional clinical factors to choose between the two active arms.

ISSN: 1471-2288