Well Praetorian2011 we are going to have to agree to disagree on...

Well Praetorian2011 we are going to have to agree to disagree on this one.

In a traditional analysis you are comparing the mean response on a rating scale between two groups. Lets say pain – on a scale of 1-10 with 10 being worse pain.

So your treatment group scores on average 5 and your control group 8 at follow-up. Depending on your variability and sample size lets say this is a statistically significant result.

But now the question is whether this 3 point improvement is clinically significant. And you would turn to papers that have derived what is considered a minimum clinically important difference for the scale you have used and in this patient population.

In a responder analysis the problem of what to set as the MCID is just pushed one step down the track. What counts as a responder has been determined – the 50% improvement idea.

But what is being tested now is the proportion of responder in each group. What difference in proportions would be considered to be clinically significant – in the sense of would lead to a change in practice.

Any difference can be statistically significant with a large enough sample. And so its not like you can have 10,000 people and think that a 1% improvement is clinically significant.

Dig deep Praetorian2011 – where else can you get your sample size from – all roads eventually lead back to clinical significance; and statistical significance is set from this - not the other way around.

In terms of the question of whether there is a lot of headroom between the 37% response rate in the P2 trial and my estimate of what is required.

Yep – but this is a very unsophisticated way of looking at the question.

The difference between the 37% and control group estimate was not statistically significant – indicating wide confidence intervals. The estimates arose from a post-hoc analysis where the endpoint was created after the trial had completed. There were numerous methodological weaknesses in the P2 trial. The endpoint in the P3 back pain trial was changed midstream. The results are running late. The landscape is one where MSB has failed to produce a convincing P3 efficacy signal for anything after numerous attempts

That is case for the prosecution. It may still be a very good risk / reward play. That’s for everyone to do their own little assessment of.