Replies to post #354453 on NorthWest Biotherapeutics Inc (NWBO)

[exwannabe]

Welcome Onto, let me try answering this.

Alpha is the target for the p value that must be reached for a trial to be stat significant. It is not calculated from the data, it is defined in advance. For a single endpoint it is typically .05 (2 sided). At this level the chance of a false positive in 5%.

For multiple endpoints it gets a bit more complicated. You want the overall trial to only have that same 5% chance of a false positive. If each endpoint has an alpha of .05, then the chance of a false positive increases because of the multiple shots on goal.

So the purpose of alpha allocation is to insure the overall chance of a false positive remains the under 5%. There are numerous ways of doing this, but most all trials use one of two methods.

Splitting the alpha. Say we have two endpoints, A and B. Splitting the alpha might assign .01 to A and .04 to B. If either reaches that level it is stat sig. and that endpoint has been established. This is typically referred to as two primary endpoints (though the FDA notes they do not care about that terminology as the alpha math is more precise than words).

Sequential (Closed). In this case we allocate .05 to A. If that passes, we then evaluate B at .05. It is important to note in that that if A fails, then B can never be stat sig. That is what preserves the overall alpha. This is typically referred to as primary and secondary endpoints (though again terminology is imprecise)

[Flipper has correctly noted that there are alternatives to sequential testing described above. They might be of use in some cases, but have pros and cons and are rarely seen in clinical trials. There are also rare cases where the alpha can be split in a more complicated manner if the endpoints are dependent. This also is rarely done.]

The ".02" issue is a big unknown. There are multiple explanations.

On crossover, it has been disclosed that almost 90% of patients received DCVax-L. That is not a reason to assume the trial is a success, but could be a confounding factor in the OS analysis and is presumably why the want to use external controls.