Biotech Values

[DewDiligence]

Yet another comment on BIOD, including a Bayesian twist:

On the CC this morning, someone asked a great question: Since the type-1 and type-2 trials shared some of the trial sites in India, why was it reasonable to disaggregate the India data for the type-1 trial but not for the type-2 trial?

The answer given on the CC was that the India data was not statsig different from the non-India trial sites in the type-2 trial (the p-value on this test was about 0.50).

This struck me as a very odd and mathematically unjustified explanation. Given that the India data in the type-1 trial was statsig different from the non-India trial sites (p<0.05) and the two trials shared some trial sites in India, why would you insist on p<0.05 as the criterion to establish an “India effect” in the type-2 trial? This is clearly wrong from a Bayesian point of view.

Of course, the real reason BIOD did not disaggregate the data in the type-2 trial is that the type-2 trial hit the non-inferiority endpoint with the possibly bogus India data included. But this begs the question of whether removing the India data from the type-2 trial would have caused the non-inferiority endpoint to be missed. I have a hunch that the answer is Yes, and I’m surprised that no one on the CC asked.