Update to rct_minim

[Philip Ryan]

Thanks as ever to Kit Baum an updated version of rct_minim is available on SSC. rct_minim allocates treatments to subjects in a randomised controlled trial using the covariate adaptive method of Pocock and Simon (1975; full ref in help file). This update corrects a bug in the previous version that affected users of Stata Releases 10 through 13. No incorrect output was produced, the program simply would not run. I thank Brendan Scott for alerting me to the problem.

[Philip Ryan]

Thanks as always to Kit Baum an update to rct_minim is now available on SSC. rct_minim implements a covariate adaptive randomisation method following Pocock and Simon (1975); full reference in the help file. The new version (2.3.0) provides a sleep() option that introduces a pause in execution to allow repeated rapid calls to the program (during a simulation) enough time for a new seed to be set. I thank Karla Hemming at the University of Birmingham, UK, for suggesting this.

[Jeewaka Mendis]

Thanks Philip for creating this module. I am wondering if the user could change the seed so that a second statistician can verify the randomisation using the same seed. Thanks

[Philip Ryan]

The short answer, for the end user, is no, at least as rct_minim is currently programmed.

The longer answer is that rct_minim, as with any other open source ado file, can be tweaked by anyone, but I won't be doing it. I haven't done any serious Stata programming for over a decade and I have only Release 14 available to me. I am happy, when notified, to address any bugs reported in my programs, but that's it. Much more would be unsafe for the end user.

With a fixed (non-adaptive) trial design (such as a random permuted block design, supported by ralloc.ado (SSC)) you just set the random number generator via a seed once and the allocation sequence for the entire trial proceeds then and there. When and how each subjects is recruited and randomised is not an issue, it is all preordained. But with an adaptive design, you will likely have subjects available over time and you must know their covariate status prior to randomisation. You may well have turned off your computer at various times between randomisations. This affects Stata's RNG status for the next randomisation. I addressed this by setting a new seed for each randomisation (ie each call to rct_minim) based on the date and time, to the second. If I were to program rct_minim in 2022 (and I am not) then I would probably use some of the tools that have become available (eg the rngstate system parameter) to save the state of the RNG with each randomisation in the data set to avoid problems with too-frequent setting of seeds and also to facilitate the sort of functionality you wish.

Sorry I cannot be of more help.

[Jeewaka Mendis]

Thanks Philip for your response. Really appreciate it.