I admire your persistence re the ERP/EEG “scent” in tracking its benefit in making our trials, as well as others’ hereout, shorter, more cost effective, higher efficiency and greater accuracy compared to trials of old. Work smart, not hard, and with better results.
I made an extra effort to dig into the Cognision site, approval trials, design and function. I studied why my machine learning investment program likes the “slow, steady” enrollment of our AD trial and assigned it a higher probability of success. I reread your thesis, several times.
Try this on, buddy:
ERP/EEG data can, in fact, shorten our trial for AD with greater precision! Here’s the high level summary of how:
Assume we have fewer sites in Australia conduct the trial. I’m assuming 11 would be ideal. My thought process is that Dr M and team, having found success in the ph2a, success defined as the vast majority of patients having slower decline than historical models (and slightly over 20% having reversal of symptoms 7/32), will now reconstruct 11 similar trials with the same “n”. This time, the trials will include a placebo set within each group to take the place of historical data - but should return the same results (*/-).
It then follows that if the dosed patients (2/3 at each site) are able to be very accurately measured as to mental decline or increase or stability in cognitive brain signals across a battery of tests, by the ERP - that’s what it detects very accurately and gained approval for, it can be demonstrated that the results of 2/3 of the patients, n less (non placebo), outperform the placebo group (1/3) or do not. No human bias. Measured brain signals. This rules out RWE and stands on hardcore science alone. After the fact or in addition, RWE can be included as well, of course. And, doses should correlate as well. ERP will be our truest ally in validating these results.
“An ERP system records electrical signals at the scalp that are produced by the brain when performing cognitive tasks. By doing this study, we hope to evaluate various performance parameters of the COGNISION(TM) system.”
This is from the 2010 FDA clinical trial which approved Neuronetrix’s ERP.
Now, why “less is more”:
If we use 70 sites, we may enroll patients faster, however, the ERP data will be for each site which will contain approximately 6 patients. 2 of these will be placebo. 2 will be low dose. 2 will be high dose. 80% will carry the wild type gene. 20% will carry the less responsive gene. Once this is all mixed into one single outcome, there will be a bunch of small “n”’s which may not show the overall picture especially during a rolling enrollment.
If we go with higher “n”’s and fewer sites, the 1/3, 2/3 and gene breakouts will be able to be compared across 11 sites to demonstrate via ERP that there is efficacy above placebo.
If that’s the case and the approved machine is able to show consistency across the board at several 35 n sites, we have a strong enough case to be made at 6 mos. My opinion. Along with safety, RWE, the extension data and any other indications which may have received approval at that point.
I think this is how the ERP should be our battle ax - reproducing similar results across blocks of decent sized n’s against placebo with discernible outcomes, not as an overall mix.
This is like running 11 small trials and getting a “pass” and same grade range on each. Not one pass/fail exam.
Also, when patients complete the trial (if so) and are switched from placebo to active drug since justification for equipoise would be inhumane at that point, we present stronger evidence if there is efficacy signal.
As you said in one of your previous posts, We have the proof and we will show them.
This is how I would proceed to make the case in a short, accurate trial using ERP/EEG REAL TIME. 11 stat sig models in same range measured without bias and calibrated across sites, will blow a single result of 450 pts out of the water, IMO.
That’s why my program liked it...Xena was right - it IS biased! But, there’s wisdom in designing it in this fashion. Why “cut for high card” for all the stakes, when we can demonstrate that we can beat the house, 11 “hands” in a row?
Open for thoughts and rebuttals.
And, even though the outcome may be the same (should be) as with the whole trial overall, this allows us to show that no matter how you slice it (by CRO, by subjective personnel, by rolling enrollment - time of year - it’s winter down there now! by geographical location) you will always end up with the same data. Missling said we would be allowed to check on interim data, I believe. This would put that to good use.
Signing off to give you time to digest and think - but still in earshot.
I have to admire your persistence. I salute you! (It motivated me to hatch this).
Great posting, Nidan! I respect those who do not go quietly away without answers! :) thumbs up