Where Quant Matters

There's only one field where quant really, really matters and it's not tech and it's not finance (666). It's biopharma. And the reason, naturally, is that, past the discovery process, it's all about the numbers. Do the numbers from the trials demonstrate that CompoundX is really better than some placebo or Standard of Care?? Bayesians, may you rot in hell if you get your hands on clinical trials.

If you follow biopharma more than casually, you know of Derek Lowe. Today's musing is mandatory.

How many of the bullish investors in that company [Axovant] realized (or really understood) that the same exact compound had failed a well-run Phase II trial in Alzheimer's, and that since then two other drugs with the exact same mechanism of action had done the same?

I've used the epithet more than once: "as any math stat will tell you, give me a large enough sample size and I'll get you stat sig for a ham sandwich curing cancer". Well, here's a snip from a comment to Lowe's piece

I went back to look at the REVEAL results, and no wonder they are not filing it. The CV event rates were 10.8% for anacetrapib and 11.8% for placebo. That is a microscopic effect size. True, the p-value was 0.004, but that is probably due to the "black hole" effect common to many outcome trials - small difference trend toward significance when the sample size becomes large. No payer would reimburse for this level of efficacy, p-value or not.
-- Emjeff

Read through the comments. As of when I type this, they're at least as acerbic as Lowe. Well, a whole lot more. Good on 'em.