Size Matters - Part The Second

Generally speaking, one sees drug developers announcing that some clinical trial is going so well that they've decided to increase sample size. That's baloney, of course, since trial success is intimately tied to p-value, and that's (for a given delta) inversely proportional to n. So, what they really mean is, "this shit ain't making it, so let's drown them with data points!" May be, just may be, the trial will squeak into stat sig (even if it's some post-hoc subset!) before the budget runs out. The delta might not even be clinically meaningful, but what the hell; FDA could still approve.

Today brings something of an exception: a revision (if I read this correctly) to lower sample size:

6:06 am Intec Pharma announces that the protocol for the Accordance Study has been amended to reflect input from opinion leaders and biostatisticians specializing in PD (NTEC) : Co states, "The amended protocol's sample size estimation uses a predicted standard deviation that is based on recent evidence from similar Parkinson's disease clinical trials. Using this estimated standard deviation and the same predicted difference between the active intervention and control, we achieve a substantial savings in sample size, without altering the objectives, endpoints or statistical power of the Accordance Study."
[briefing.com 7 November 2016]

Something of a roll of the dice, given that neuro drugs are notoriously difficult to measure in the first place.