Reanalyzing MERIT design using Julia + recent numerical algorithms

Here’s an interesting case of a paper that made it through peer review at Statistics in Medicine despite using Monte Carlo estimation without regard to its inherent error. An extensive tabulation and simulation study were presented, with absolutely no mention of MCSE’s! (As one would expect, this vitiates the reported results.)

Two interesting aspects of the above critique are:

  • Relatively recent numerical work [1] played a crucial role
  • Julia proved remarkably effective in rendering the reanalysis clearly

  1. Frey J. An algorithm for computing rectangular multinomial probabilities. Journal of Statistical Computation and Simulation. 2009;79(12):1483-1489. doi:10.1080/00949650802286753
2 Likes

Thanks for sharing this feedback. I had read the paper, but these details go over my head.

However, generally speaking, there seems to be a need (following Project Optimus as well) to randomise in dose-optimization oncology studies. Yet, bec this typically requires large N, I don’t think this has happened in oncology (no? except when requested following later confusion in the development pipeline), whereas these dose-ranging studies are pretty typical outside of this domain.

There seems to be a plethora of designs, but most prominent of which are MERIT and DROID (from Ying Yuan’s team, I presume). The former, which you are discussing here, seems to be comparing each arm to predefined thresholds of toxicity & efficacy, rather than actually comparing arms to each other, resulting in an OBD admissable set (Stage 1). Do you find this as another valid point against MERIT?

2 Likes

This is a keen observation, Eslam! It had not occurred to me to criticize MERIT on such grounds, since one [valid?] perspective on early-phase trials is that they are to inform the grossest kind of considerations, taking existing therapies as the comparator. (“Is this new drug in a reasonable ballpark to trial further, considering what we already have available? If so, which of several doses should we study?”) But I now think you have indeed identified yet another vulnerability here.

1 Like