Destined to fail, page-37

Cafa, I don't know whether to thank you or curse you for making me research into this further and relive the horrors of uni-level maths. But in my research I have found out a couple of references of interest.

I like this one: http://www.ncbi.nlm.nih.gov/pubmed/25487708?dopt=Abstract. And the key sentence within this abstract is this: Assuming an increase in the median PFS from 9.4 months to 12.5 months with the addition of SIRT to mFOLFOX6, recruiting >=450 patients will be sufficient for 80% power and 95% confidence.

So the key question for overall PFS is whether the MEDIAN PFS increases by 3.1 months. Not the mean. Before I read this abstract I did not realise this.

So in a study of 600 patients, of which half are control, what is the PFS increase of the 151st patient in the list of trial patients, in ascending order of PFS improvement (i.e. actual PFS - expected PFS)? If we assume that 120 patients do not improve PFS at all (or minimally), we are looking at the 31st patient in the list of 180 liver-only patients, sorted in order of increased PFS.

So unless I have mis-interpreted this, we now know for a fact that the 31st patient in the list did not experience PFS improval of 3.1 months. Otherwise the primary outcome would be a success.

But because the liver only group Secondary Outcome WAS a success, we now also know that the median of this subgroup - i.e. the 91st liver-only patient - did experience PFS improval of 3.1 months.

So in summary (assuming minimal improvement on the patients who had tumours outside the liver), of the liver-cancer only patients, We have 30 who definitely had PFS improvement of less than 3.1 months, and 90 who definitely had PFS improvement of greater than 3.1 months. The other 60 we have to wait on until we see the detail.

To me this is like a light bulb going off and it explains - perhaps - why this trial was not destined to fail. Because if ALL the liver-only patients showed more than 3.1 months PFS improvement, the overall PFS for the group would still be >3.1 if measured by the improvement in the MEDIAN patient.

And of course all the above analysis may be wrong, as I am learning as I go along