Non-inferiority is an annoying phrase. Mostly because it’s got the feel of a double-negative about it, but also because it’s frequently quite a confusing thing to prove. There are an increasing number of clinical trials which are designed not to show SuperDrugOmab is better than Olde Elixir, but just that it’s not worse.
Now – how do you decide what is ‘as good’ as the current treatment? In an ideal situation you’d be able to say it had exactly the same efficacy, but we all know that with the wibbles of chance and the vagaries of life, they won’t come back with identical numbers. And even if they did, we’d not be convinced they always would do. So instead, we set a lower-bound for the difference, often to say it’s (95%) likely to be at least 90% as effective across a whole community or population. Now what would be even better would be if the 90% or more ‘guess’ was actually set as ‘no more lower effectiveness than the smallest difference that patients describe as important’; the minimal clinically important difference. For some conditions, scales or measures we know what this value is. For many things we don’t know this firmly, which is another call for excellent patient engagement in research from the beginning.