Brainiacs

Before the lecanemab show discussed in this morning's post, there was this sitting in the queue. Time to hit Enter, I guess.

There's much ado about the lecanemab results. Here's a thread from Feuerstein and a comment, to wit:

... highly statistically significant loss of brain volume across all Rx groups in their phase-2 study?
-- MadhavThambiset

Hmm, one might say. Moreover, the SAP is Bayesian. Color me suspicious. While it may not happen soon, this drug will fail in real-world practice, same as every other anti-amyloid compound. Mammy Yokum has spoken. Hokum is hokum.

Here's a lengthy report (the great Derek Lowe).

The antibody was dosed at 10mg/kg twice a week, and the [PII] Bayesian trial was set to look for an 80% chance of a 25% or greater decline in disease progression, as measured by ADCOMS, the Alzheimer's Disease Composite Score. It missed that endpoint at the specified 12-month evaluation point, and at the 18-month point it was actually showing weaker effects than it did at 12 months.

Now, the entire point of whispering Bayes is that stat sig will be easier to find since you're only attempting to improve on 'prior knowledge', not starting from scratch

The similarity to some cancer therapies is worth keeping in mind: in oncology, if you die at the about the same time you would have otherwise, but with smaller tumors (and that only confirmed by imaging, not by quality of life), no one cares.

Ouchy!!

And talk about Ouchy!! This is an even more pointed report; may haps a takedown?

Clarity AD had a sole primary endpoint of an effect on CDR-SB versus placebo. CDR-SB is a subjective tool used by doctors to make an initial assessment of a patient's condition, but became an acceptable endpoint when in 2013 the FDA relaxed guidelines, conceding that co-primary cognitive/functional measures might prove impractical in early Alzheimer's.
...
One [question] concerns the extent to which Aria-E occurrence disclosed which patients were on lecanemab and called Clarity AD's result into question. [unblinding the arms]

And, not to ignore the obvious, this is this author's take on the PII. He is not a Bayes fan, either.

Here's the skinny on 'adaptive Bayesian' trials.