Skew You Buddy

A recurring theme in these endeavors: data analysis is helpful only when there's a stable underlying data generation process. Among other things, this makes stat as a tool for macroeconomic policy analysis fraught with danger. The members of that branch of study have been arguing over the structure of analysis. The quants against the historicals. On the whole, historicals have been right. On the other hand, there's the old saw, "economists have accurately predicted 10 of the last 5 recessions!" As an example, here's an interesting post. I'll leave it to the reader to decide whether the effort is Type III error.

An even more egregious error happens when median/mean is relied on. It's also worth noting that one of the few shorts was explicit: they didn't believe an MoA existed. Ouch.

This time around, Celldex had comparison data that strongly suggested its lead candidate, glemba, could provide a survival benefit over chemo for patients with a difficult-to-treat form of breast cancer. Unfortunately, the data came from a subgroup contained within an otherwise failed trial. In retrospect, it looks like several outliers resulted in a lot of wasted resources.
[my bold]

Outliers, on one side of the median/mean, will skew you dead. Not to mention: subgroup post-hoc "analysis" is the Very Big Red Flag biostats. On often wonders why a legit math stat would get within a barge pole's length of such companies. But they do. I guess the money's not bad.

It is all too common for small bios to go all in, given that they have, in general, small pipelines. If I had a nickel for every one of these that generated traffic like, "but there's one guy still alive after 5 years taking WhizzBang250", I'd be sipping tequila on a Caribbean island. ImmunoOncology is one of the bigger deals these days, but there're even spontaneous remissions in cancer and always have been.

With PhARMA constantly braying that it costs $X billion to get "a drug to market" so drugs have to be really expensive to pay for all that, it's worth pondering whether the moolah is funnelled into R&D or mgt. pockets?