I would like to know why they used limiting factors on speech recognition for their trials. If you can't decipher one word in speech recognition, but after FX-322 you can, wouldn't that be a marker for improvement? Or if you can't decipher any words, are they saying FX-322 will not work?
I dug through the paper by Thornton and Raffin (attached) that they cited in their Phase 1B paper that explained their methodology for assessing clinical meaningfulness of word recognition (WR). Basically, here's the issue:
The tests consist of n=50 words and are randomized. When comparing average performances (percent correct) across tests, the tests are deemed "equivalent" difficulty. However, it's not true that for each individual person, the tests are exactly the same difficulty. The further the percentage is from 0 to 100 (extremes), the more this is true. In other words, if I score 50%, I'm more prone to variation than if I scored 100%, as I could have scored 100% so easily that it doesn't matter if you change the test on me.
In their assessment of being "clinically meaningful" WR, they dropped the patients with baseline scores below 90% because of the ceiling effect. This is because if someone has, say, a 98% baseline performance and they receive the drug and improve to 100%, it's basically impossible to know if that improvement is just from changing tests.
In the following picture, the dot lines create a 95% confidence interval around the dashed line. Thornton and Raffin used a scheme to show the effects of repeating tests. They established 95% confidence interval ranges that Frequency Therapeutics basically just referenced for their data set. The dashed line is the assumption of no improvement. Notice all of the clustering on the top right. It is too difficult to put these people in the running for clinical meaningful because of the variation.
What you have is the opposite, which is the floor effect. Say you take the test at baseline and score 0% and take FX-322 and improve to 2% (1 word). How do we know it wasn't just luck or that you got the right test? Moreover, what if you're at 0% initially, but are no where close to even 2%. Maybe FX-322 helps, but you're so bad off that you only jump to 2% or even stay at 0%. For this reason, the floor effect is a problem. They took care of eliminating the floor effect by working it into their exclusion criteria.