Thank you

I was referring to this one: onlinelibrary.wiley.com/doi/abs/10.1...

Thanks a lot for replying! When do you think is a single-arm trial appropriate? Estimation?

...Novel designs using randomisation, like MERIT, seem to compare each arm against against an assumed fixed population rates too IN EACH ARM.

I would love to hear what you think of single arms in the phase 1 dose finding/optimization setting. Which is virtually 100% of onc trials. Optimus explicitly encourages randomisation.

I am tempted to try and simulate a single arm sampled from a true response rate of 0.75, test against a null response rate 0.4, and compare that design to a 2-arm RCT sampled from 0.75 and 0.4. Surely someone has done it, no?

I don’t like ‘undershot’ as a criterion. I want non-coverage in each tail to be very close to alpha/2. And Wilson works fine for large n so not need to be switching. R’s Hmisc::binconf is an old function that computes Wilson’s interval. Also there is no excuse for continuing use of symmetric CIs

I use Wilson for small n trials. Clopper-Pearson has been pretty much standard for a lot of trials in my experience bec it seems to guarantee that the nominal coverage is not undershot.

I think Wald's is so widespread although it seems the literature agrees it shouldn't be used.