We had a workshop on Tuesday that was an educational session and I specifically talked about phase I trial design and talked about some of the more modern methods of doing phase I trials from a statistical and design perspective. I was really trying to talk about moving forwards from traditional 3 3 designs which have been around for four years, probably more than that, to equally easy to do methods but better performance from a statistical perspective, more likely to select the correct dose to carry forward, able to monitor toxicity in expansion cohorts which we almost always do now and just more flexible and better able to be used to design future trials.
What are the advantages of newer designs?
3 3 designs put three patients on at a dose level, if there are no toxicities you go up to the next dose level and put another three patients on. If you get a toxicity you put six patients on and basically at that point you’re pretty much done so you never get to more than six patients on a dose level. Deciding the dose to carry forward into phase II trials and possibly even into phase III trials based on six patients doesn’t seem like a very informative way to make decisions. So the newer designs, some of the model assisted designs like a BOIN design, Baysian Optimal Interval design, or a keyboard design which is related to something called the Toxicity Probability Interval design, these allow you to add more patients on at any one dose level and to monitor toxicity going forward so that you get the expansion and you are then making decisions on perhaps fifteen patients or twenty patients. Sometimes that can even be the beginning of a phase II design so that you get a two-for.
Are these modern designs being used already?
Some of these designs are relatively recently published in the last three or four years. There’s an older style of design called model-based design that have some more performance, actually even better performance, but are more complicated to use called continual reassessment. Those designs have been around for a long time but they have not been taken up by the general research community, partly because they’re a little harder to use. Now people are starting to use these model assisted designs that are very easy to use but have similar performance and we’ll see whether they get to be used commonly. I’m using them all the time but I’m just one person, partly it’s getting the word out and getting people to understand that it’s just as safe, doesn’t take any longer but makes better decisions.
What resources are out there for people to find out more about this?
There are a couple of papers published recently in The Clinical Cancer Research Journal which is an AACR journal, one on the BOIN design and another one on the keyboard design. So that would be a good place to look to find an easy to read description for a clinician or for a researcher.
Do you think this is the future of trial design or is it just one way it can be done?
There’s only so much you can do in phase I because the primary goal is to look at toxicity and select the safe dose. There are always enhancements that can be made but these designs are really where we should be going and possibly some enhancements on top of what they’re already doing.
