I am struggling to finalise a statistical framework/methodology that would be best suited to analyse a 3 arm clinical trail data.

The problem is as follows, I have two clinical trails, parallel designs, 3 arms (placebo, drug 100mg, drug 200mg. It has a primary efficacy point is proportion of responders at week 12 and secondary endpoints at week 12, 28, and 52.The data also includes several demographic and baseline characteristics. I need to have a framework that would better characterize profile of those patients that respond better to 200mg dosage Vs 100mg.

I can either analyse these datasets separately or pool them together but I am not sure what will be the benefits or drawbacks. In addition, does anyone knows a package in R that could perform this kind of test.

I am new to the world of clinical trails so please guide me or direct me to a resource that would be a good starting place.


1 Answer 1


There's a whole literature about "best practices". Effectively your problem is one of multiple hypotheses. The overarching objective is to propose a testing framework that conserves the nominal 0.05 (or other) alpha testing level. Probably the best and most authoritative reference on the subject comes from the American FDA (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/multiple-endpoints-clinical-trials-guidance-industry).

Some studies run analyses like the above in a formal phase II study. You can design and power and the study in several ways. One possibility is to sequentially test the hypothesis that either 100mg or 200mg is superior to placebo - if you believe that 100mg is a minimally effective dose, then using a Holm procedure, test whether 200mg is superior to 100mg. Or you can do dose adaptation, to overenroll 100mg and 200mg dose levels and use an interim analysis to drop the less effective arm. You can test 100mg vs placebo and 200mg vs placebo as coprimary or separate hypotheses, so it's patient or physician's choice of dose level.

The software is far from the most important thing here. What's important is that you formally translate the investigator's question of interest into a formal hypothesis.

  • $\begingroup$ Thank you for your reply. In your opinion, will Dunnett's test for multiple comparisons, which compares each treatment with a control (Placebo) or with another established drug will work in this case? $\endgroup$
    – Angelo
    Mar 1, 2022 at 22:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.