Celebration Thread! Biogen is going to the FDA with Aducan.

[J11]

NF_52, thank you for your post.

Yes, I have become somewhat overly relaxed about considering mabs as the ADCOM rapidly approaches. My reference points for the APOE + /- argument about mab efficacy was related to the below figures:

APOE + - 2.gif

APOE 302 plus minus 2.png

[J11]

APOE + - 2.png

Subgroup 3.PNG

Subgroup 1.PNG

[J11]

APOE plus minus.PNG

APOE + - 3.PNG

302 Subgroups Edit 1a.png

[J11]

jamaneurol-e222793-g005.jpg

[J11]

"The tables in the report mention slow, typical, and fast progressors; they also mention race and baseline MMSE. It is still surprising to me that APOE genotype, gender etc. were not seen as relevant predictors -- from merely looking at the reported numbers APOE +/- did seem to be important. Was this also a covariate effect?" Page 70 of thread.

Final Model.JPG

[J11]

These were the thread reference points that I had in mind when thinking about the APOE +/- question. The first figure above (in the first of the figure posts) shows the phase 2 result for Lecanemab for ADCOMS. There was quite a large positive effect for the e4 positives and less so for the e4 negatives. There was however some confusion with this result because of the change in the plan while the trial was underway. This trial does not appear to be included in the study that you noted.

The next figure down was from the Aducanumab Briefing document. It also shows a substantial over-performance of the e4 positives. The first figure in the second post of the figure posts above showed my results of the regression by genotypes. The e4 negatives seemed to receive minimal benefit from treatment. However. the Clarity result that was added to the figure in purple hints that something went awry (more on this later). Note that the actual reported result for Clarity for the total carrier result was only 0.33. I do not find this plausible and it was largely driven by the e44 result. I do not see any great value in keeping results that for whatever reason are not that plausible so i think it is best to largely discard the e44 Clarity result. The one stable result that seems to be constantly present is from the e34s. We see this in the Lecanemab phase 2, Clarity, Aducanumab 302 high dose (likely), 302 low dose (likely), Donanemab phase 2 (likely) and probably also in the Donanemab phase 3 (TB-2).

OK it now is later. The next figure shows the Clarity result and we see how the epsilon 4 positives did very well, the 34s did as expected and the 44s appeared to underperform. The next figure shows us that when we add the 34s and 44s together it gives a luke warm result for the overall epsilon 4 positives. Next figure is the assessment of APOE epsilon genotypes and treatment response in the Aducanumab Briefing document. At that time it really seemed as though the epsilon 4s did better with treatment.

Then we see the figure from the currently reported result. Now it is felt that epsilon 4s appear to have an edge. However, I would suggest that some of these trials probably should be removed. For example, Solanezumab has a different mechanism of action with respect to amyloid that might distort the results. Probably would also want to edit out the Aducanumab 301 results and add in the Lecanemab phase 2 results. Of course the Trailblazer -2 outcome will add considerable additional input into this question. Graduate 1 and 2 would also be helpful. It is disappointing that more trials have not reported on specific APOE genotypes instead of only providing epsilon 4 positive and negative categories.

[J11]

In the first figure in the last of the figure posts above we see the expectation of the cognitive effect of Donanemab conditioned on epsilon 4 negative and 4 positive as a function of amyloid removal. This is very much similar to my original figure posted above. It really did not seem as though the epsilon 4 negatives benefited that much.

Finally I have included the FDA final model for Aducanumab. This model did not include a term for APOE.

It really does go to show that chasing after the statistical noise is not the optimal. I tried my best to stay more with the top line result and that worked out quite well ... chasing after the noise not so much. However, I did not think it plausible to merely ignore this question as there needs to be some consistency for the results to be credible.

[J11]

This is from the article:

"Pooled analysis of potentially efficacious antibodies lecanemab, aducanumab,
solanezumab, and donanemab shows slightly better efficacy in APOE ε4
carriers than in non-carriers. Carrier and non-carrier mean (95% confidence interval)
differences from placebo using Clinical Dementia Rating Scale–Sum of Boxes (CDR-SB)
were –0.30 (–0.478, –0.106) and –0.20 (–0.435, 0.042) and AD Assessment Scale–
Cognitive subscale (ADAS-Cog) values were –1.01 (–1.577, –0.456) and –0.80 (–1.627,
0.018), respectively. Decline in the APOE ε4 non-carrier placebo group was equal to
or greater than that in carriers across multiple scales. Probability of study success
increases as the representation of the carrier population increases.
DISCUSSION: We hypothesize that APOE ε4 carriers have same or better response
than non-carriers to amyloid-targeting therapies and similar or less disease progression
with placebo in amyloid-positive trials."

It found a -0.30 CDR-sb for the carriers and -0.20 for the non-carriers in its evaluation.
That just seems much too low an estimate to me. I do not think the FDA would consider mab approval with such small benefit. 0.4 seemed to near the edge of approvability. Averaged together 0.3 and 0.2 would only give you in the highish mid-20s.

I would probably stay away from this approach of finding the point estimate without thinking in the CDR-sb-SUVR space. The regressions that I have done on thread worked out magically well when I took such a perspective.
When you take the totality of the results and then look at the dose response relationship then you wind up being within a few pips from the parabola. However, it is then best not to look under the hood and ask too many questions -- you might not like what you find. specifically here the APOE epsilon e4 negatives do not seem to be that well behaved. Or at least this is what was found only in Clarity. The e4 negatives dramatically outperformed my prior in Clarity.

However, we now also have the top line for Trailblazer-2 and it was quite startling at nearly 0.70 CDR-sb. Here the parabolic nature of the regression is of potential explanatory power -- Donanemab removes a lot of amyloid as we know so this should magnify the treatment effect. In addition there was selection involved to remove those with low tau. This is going to select for those who will be responders -- it is not entirely clear though whether APOE e4 positives or negatives would relatively gain more by such selection. It is exciting to realize that with the averaging of the epsilon 4 genotypes there will be one genotype that logically will be above 0.70 and one that is below 0.70. However, in the above results from the paper we have the 95% upper confidence interval of 0.47 and 0.43 for epsilon 4 positives and negatives.

(0.478 - 0.106)/ 2 standard Deviations = 0.362 CDR-sb / 2 Standard Deviations = 0.181 / 1 Standard Deviation


Probably best not to continue to peer into the data noise and try and make sense of it, though considering the Trailblazer-2 result and Clarity my current best estimate is still along the regression shown above (though updated into a parabola). So, for Trailblazer, 0.6 for the total epsilon 4s might be somewhat conservative. This would almost certainly imply that the e44s did much better than in Clarity. In fact, if the quoted research is to be believed, then it should be more likely that the e4 positives were the subgroup that reached over the 0.70 mark on CDR-sb.

The basic number crunch for Trailblazer-2 is that if it were the e33s who were the subgroup over 0.7 and the e4 positives averaged out at 0.6, then with 30% e33s and 70% e4s you would have 0.7*0.6 =0.42 subtract 0.7 = 0.28 and you now need this from the 30% of e33s. You would need almost 1 CDR-sb for the e33s. That is a lot! This would even be called clinically meaningful. It almost has to be the other way round: it has to be the epsilon 4s who were above 0.70. Clearly the numbers are starting to move upwards. Will be very excited to see where TB-2 reports out on APOE genotype.

[J11]

In terms of why we might be seeing these discordant results, one aspect of note is that epsilon 4 carriers are more likely to be actual AD patients. Somewhere or other I encountered the number that ~95% of e44s with amyloid and cognitive impairment will be true AD patients, while only 90% of e34s will be true AD patients and only ~75% (? not sure about this number) of e33s. So epsilon 4 genotype is a very powerful tool in selecting the patients who will most likely respond to amyloid lowering. I think the problem that arises though is that e44s than have so much more ARIA. The mab trials might then be somewhat heterogeneous in terms of how they cope with these side effects and how they profile the risk. For example, Emerge and Engage did not allow anti-coagulant use, while Clarity did. Indeed the miss with Engage might have been about how the clinical sites dealt with these worrisome ARIA events. Even in the Lecanemab phase 2, treatment for epsilon 4 positives was disrupted as the risk of ARIA was reconsidered mid-trial.


https://www.desmos.com/calculator/smmhariuuq
https://www.desmos.com/calculator/cuesvabfrs

Might be too much if we subdivided the Carriers into e34s and e44s and take this into 3D. Would require everyone to locate their 3D glasses. What the url is showing is that with the reported 0.696 CDR-sb from TB-2, there will almost have to be a fairly large result in one of the subgroups.


3D Oh, yeah!
https://c3d.libretexts.org/CalcPlot3D/i ... ;zoom=1.08


https://c3d.libretexts.org/CalcPlot3D/i ... ;zoom=1.21



https://www.desmos.com/calculator/uulaum5esy
https://www.desmos.com/calculator/kiml2v9pwk

https://www.desmos.com/calculator/csgktmottn


https://www.desmos.com/calculator/vauay1yye5
https://www.desmos.com/calculator/2dhd09xggu

[J11]

APOE + - 3 Edit 3.png