Look at the standard error, page-65

1. Using standard error overlap is overly conservative. Really I should have convert the SE to CI multiplied by 83% and then correct for multiple hypothesis testing. If I were to do this the US 5% BID may not actually be an outlier anymore. I should mention that it really doesn't matter too much as the US 5% BID performing much worse than all the other treatments including the control which is hardly a positive.

2. I think the SE observed do suggest wide variability in the patient data that really do need to be explained by BOT. There may be a treatment effect, but if there is one it is being swamped by the variability.

The issue with trial variability is not confined to the USA data. Both the Australian and USA cohorts have approximately the same SE. The variability issue is more than just with the US batches.

There is no "proponent" of the null hypothesis - this is just what all trials have. What the data suggests is that the null hypothesis of no treatment effect can't be rejected.