SEARCH

SEARCH BY CITATION

It has recently come to our attention that some of the results presented in Table II in ’A conditional error function approach for subgroup selection in adaptive clinical trials’ (Statistics in Medicine 2012; 31:4309–4320) are not correct. An R software coding issue resulted in numerical errors in the reported results for the conditional error function approach (CEF) and for the combination test approach by Spiessens and Debois (CT-SD). We have corrected these errors in the following table. We also, for reasons of consistency with the CEF approach, now present type I error rates for the CT-SD approach based purely on rejection of the intersection hypothesis inline image rather than as previously on rejection of both intersection and one or the other of the elementary hypotheses inline image and inline image. It is now clear that type I error rates are controlled at the nominal 2.5% level for both approaches, and our previous assertion that the CEF methodology was uniformly, although only marginally, more powerful than the CT-SD methodology is no longer supported by the simulation results for the selected scenarios. Correction of coding errors produced such small changes to data presented in Figures 1–3 as to be indistinguishable from normal simulation error; updated data underlying these plots are available on request from the authors. We apologize for any inconvenience this error has caused.

Table 1. Simulated probabilities to reject at least one null hypothesis under the null (inline image) or the alternative Δ{F} = 0 and Δ{S} = 0.3 for the conditional error function approach (CEF) or the test proposed by Spiessens and Debois (CT-SD) given various prevalence of the subpopulation τ, time points of interim analysis n1 ∕ (n1 + n2), and ε-selection rules (10,000 simulation replications per scenario for Δ{S} = 0.3 and 100,000 per scenario for Δ{S} = 0; a different random number seed is used for each cell of the table).
n1 ∕ (n1 + n2)ετCEFCT-SD
Δ{S}Δ{S}
00.300.3
0.5 ∞ 0.30.02550.48680.02540.4796
  0.50.02490.75060.02500.7507
  0.80.02510.94000.02520.9459
 00.30.02530.83800.02520.8347
  0.50.02510.92750.02500.9252
  0.80.02480.97320.02480.9747
 0.50.30.02460.81110.02450.8018
  0.50.02540.91710.02560.9220
  0.80.02550.97220.02560.9734
0.25 ∞ 0.30.02460.49730.02460.4966
  0.50.02510.75370.02500.7589
  0.80.02470.94370.02440.9482
 00.30.02480.85740.02480.8534
  0.50.02590.94290.02550.9431
  0.80.02560.98000.02560.9800
 0.50.30.02470.81240.02470.8067
  0.50.02450.93380.02460.9409
  0.80.02510.98110.02510.9801

Acknowledgements

  1. Top of page
  2. Acknowledgements

We are grateful to Gernot Wassmer who brought the problem to our attention. The results for the combination test approach can be reproduced using ADDPLAN 6.0 PE.