Cohen’s d Effect Size And Time In The Trials for Alzheimer’s Drug Blarcamesine And Why Results Come In The Second Half of 2024 $AVXL $SAVA

This Blog Is Only For Educational Purposes. Do Not Trade On This Blog Alone. Do Your Due Diligence or Consult a Professional.

The Data from Phase 2a Tells a Story…

The graph on the left shows the plot of changes in the MMSE scores (cognition) over 148 weeks of Phase 2a and the Open-Label Extension. Because the graph values include MMSE scores, Standard Error, and n, we can calculate Standard Deviation to arrive at Cohen’s d Effect Size at 57 and 148 weeks.

The definition of Standard Error is SE = Standard Deviation / Square root ( n )

Once we obtain Standard Deviation, we can calculate Pooled STANDARD DEVIATION used in calculating Cohen’s d.

The definition of Pooled Standard Deviation is SDpooled = Square root(( SDdosed^2 + SDplacebo^2)/2)

To calculate Cohen’s d, we divide the difference between the cognition scores of the Placebo and Dosed arms by the pooled Standard Deviation. The result is the Cohen’s d Effect size.

Of course, since we do not have the accurate Placebo arm in Phase 2a, we assume that the Low/Medium Concentration Arm is not showing any therapeutic effect of Blarcamesine in Alzheimer’s disease and, therefore, can serve as a Placebo arm.

After going into calculations, we obtained the following results

The Effect Size for Blarcamesine after 57 weeks of dosing was ES57= 0.38

The Effect Size for Blarcamesine after 148 weeks of dosing was ES148 = 0 96

Can We Repeat the Same Excersize for Phase 2b/3 Results?

I cannot provide you with a quick answer to this question. Only partially, since we do not know the mean cognition and Standard Deviation results for the 50mg Cohort in Phase 2b/3. Let’s see what we know at this moment.

We are given Standard Error for the Placebo Arm and The Mean. We have the combined arms for the Anavex2- 73; 30mg and 50 mg. 30mg corresponds to the Medium Concentration Cohort from Phase 2a, so we assume there is no therapeutic effect in that cohort. The Odds ratio leads us to believe we can have a 70% / 30% distribution between those scoring on average -4.03 ADAS-Cog13 and those scoring +4.11 (Placebo-like) ADAS-Cog13. We can also assume that the maximum effect distribution might be 80% / 20%. We also have to assume the values of Standard Errors (by extension of Standard Deviations) to be the same as given in this slide of the December 1, 2021 Presentation. These assumptions introduce unknown errors in the Cohen’s d Effect Size estimation for Phase 2b/3. The results can only be used as general guidance. I will not present here all the calculations and will limit myself to results.

The 80% / 20% distribution of results attained Cohen’s Effect Size of 0.57 after 48 weeks of dosing

The 70% / 30% distribution of results attained Cohen’s Effect Size of 0.50 at 48 weeks of dosing

Similar results were calculated for 48 + 96 weeks per Attention-AD trial (, assuming 2% deterioration in dosed arm with Placebo Arm deterioration of the same rate as in the 48 weeks.

The 80% / 20% distribution of results attained Cohen’s Effect Size of 1.15 after 144 weeks of dosing

The 70% / 30% distribution of results attained Cohen’s Effect Size of 1.10 after 144 weeks of dosing

Two observations can be readily made. The error in my assumptions is not leading to a significant divergence between the results from Phase 2a and Phase 2b/3, and with time the Effect Size has roughly doubled. Also, Phase 2b/3 validates Phase 2a even against arguments to dismiss it due to its small n.

To put these numbers in perspective, I will calculate the Effect Size of Cohen’s d for the 12 months (52 weeks) for $SAVA Simufilam. We have almost all the data available on January 24, 2023, Press Release from Cassava Sciences except for Placebo arm deterioration at cognition score of 19.1 ADAS-Cog11. Previously done analysis led to this plot of baseline cognition numbers related to expected Placebo deterioration. link

When converted from ADAS-Cog13, the baseline cognition for Blarcamesine is ~24 ADAS-Cog11 points. Still, for Simufilam, with a higher level of baseline cognition at 19.1, the rate of Placebo arm deterioration seems to be about 2.25 or 2.5. Plugging all the numbers into the EFeect Size Cohen’s d formula gives the following results for Simufilam.

Cohen’s d Effect Size for Simufilam at 52 weeks of dosing at a baseline of 19.1 ADAS-Cog11 is ES = 0.18

Cohen’s d Effect Size for Simufilam at 78 weeks of dosing at a baseline of 19.1 ADAS-Cog11 is ES = 0.22

From the (link, we can read that the Attention-AD trials add another 96 weeks of dosing to the already accrued 48 weeks giving altogether 144 weeks of dosing which is about 2.77 years. A trial that long, at least for Alzheimer’s disease time, eliminates all who would spontaneously test cognitively higher than the baseline, leaving no doubts about the drug’s efficacy. This was the conclusion of the paper I gave a link to (link. The trial is to be completed by 7/31/2024.


  • Phase 2b/3 improves over the Phase 2a results. We must remember that Phase 2a had over 70% APOE4 carriers, while the statistical Alzheimer’s population holds only 43%. They deteriorate 22% faster than other Alzheimer’s patients.
  • With time, Cohen’s d Effect Size increases in both Phases. Releasing the results later has the advantage of erasing doubts about the drug’s efficacy as 2.77 years trial duration purges any patients who could spontaneously score better than at baseline.
  • Simufilam is not competing with Blarcamesine. If it reported its Effect Size at 3 years at present deterioration rates, its ES would be only about 0.34 versus Blarcamesines ~1.00.
  • Any values present here are approximate, as the respective Standard Deviations may change with time. They only give insight into general trends.
  • Stocktwits community members speculate regarding the release of full Alzheimer’s trial data. It seems that the latest the data can be released is after 7/31/2024, as the accrued Effect Size will be most remarkable. This generally agrees with Dr. Missling’s words of surprising the market. In this case, data would show unquestionable Blarcamesine’s efficacy.
  • The baseline of the Blarcamesine trial in cognitive terms is so low that compared superficially to other drugs, it suffers from a comparative disadvantage in the eyes of investors. The lower the cognitive baseline, the faster the deterioration of patients, and the stronger must be the drug’s therapeutic effect to overcome deterioration.
  • This phenomenon is not lost on the people in the field, but investors are not accurately aware of it. That is why KOL are enthusiastic about Blarcamesine’s prospects.
  • As Investor, I realize that the full release of data on Alzheimer’s trials will likely happen the following year (2024). I suggest forbearance and focusing on Rett’s results.

Please, donate. I just bought grammar software to improve my writing


Make a one-time donation

Make a monthly donation

Make a yearly donation

Choose an amount


Or enter a custom amount


Your contribution is appreciated.

Your contribution is appreciated.

Your contribution is appreciated.

DonateDonate monthlyDonate yearly

6 thoughts on “Cohen’s d Effect Size And Time In The Trials for Alzheimer’s Drug Blarcamesine And Why Results Come In The Second Half of 2024 $AVXL $SAVA

  1. I very much appreciate your hard work, expertise and professionalism. I made another contribution to help you find your efforts.


  2. I appreciate all the good work you do on this blog, but I can see here from your analysis that you have misunderstood the ADAS-Cog Odds Ratio result in the Anavex AD 2b/3 trial. Per the Anavex Dec 1 2022 press release, the OR were only applied to the top cognitive scorers in the trial, both dosed and placebo, and not across all participants: “Patients treated with ANAVEX®2-73 were 84% more likely, to have improved cognition by ADAS-Cog score change of -0.50 points or better from baseline to end of treatment than patients on placebo, Odds Ratio = 1.84 (p = 0.015).” The numbers of patients in this group were not disclosed, but they are definitely a subset of the whole trial, and likely a small subset, albeit clinically significant.

    Further, you state that APOE4 carriers represent 43% of AD patients. My research (and your early posts) concluded more like 70% of AD patients are carriers, with ~ 10% APOE4/4 in that group. This may not affect your results.

    Finally, your analysis of the Anavex 2a results suggests that the distribution curve in the 2b/3 is also highly eccentric and does not resemble standard distribution. This could well explain why Anavex is claiming (per a MIssling comment at a conference this year) 2b/3 trial success just based on Odds Ratios. I don’t know whether that might affect your Cohen’s d SD analysis, but I expect it would.

    1. The situation with the Odds Ratio is very simple. Some 30% of patients spontaneously score better than at baseline (at this trial baseline)(calculated base on Standard distribution and an information in a paper) so 1.84*.30=.552. From Phase 2a we know that Low and Medium Concentration cohorts do not exhibit any therapeutic effect and decline like placebo so we allocate .552 population to 50mg cohort which is the impossible 110% of the 50mg cohort then we alocate 30% background scoring better than baseline to 30mg cohort and we end up with 70%/30% or 80%/20% split in 50mg cohort. At the baseline of 24 ADAS-Cog11 there is about 30% patinets who naturally score above their baseline score on cognitive tests after one year into the trial. Also the mean and the standard deviation support this number as it is easy to calculate – my old engineering book. Since therapeutic effect only takes place (this might not be entirly right) in 50mg cohort as Phase 2a suggests it all that matters. it is easy to increase the dosage. Had I have not red papers I would not believe it either. After some 3 years the number those who spontaneously score better than at baseline dwindle to almost zero so for any Alzheimer’s drug real trial should be at least 3 years duration as this distortution is eliminated by time and therapetice effect is all what is left to see. I hope that at least answers first part of your question.

  3. While I accept that 30% of the placebo group can be expected to naturally score above baseline under the conditions of the 2b/3 trial, the question for purposes of the Odds Ratio has to be, What percent of the placebo arm scored at least 0.50 points *better* than baseline (as described in the Anavex PR)? This will be the denominator for the raw numbers in the Odds Ratio. Less than 30%, surely, but can it be estimated using a standard curve for placebo response (which should be the case)?

    Based on the 4.03 point average improvement for the dosed arm, a clue to the numerator in the Odds Ratio, and the 1.84 OR itself, I have calculated a 35/19 ratio between the dosed and undosed n’s above 0.50 points improved (assuming whole number ADAS-Cog results). That includes the n possibilities of 35/19, 70/38, etc. — if that helps.

  4. Answering my own question, assuming standard distribution and the 30% positive deviance from a 4.11 points average deterioration, that would equate to a ~ .59 SD for placebo baseline performance or higher. Moving along the curve another half point yields a ~ .66 SD or higher. And that equates to a textbook 24 to 25 percent subset of the placebo group of what, about 160 n?

    In the end, it looks like 70/38 fits best as the raw numbers of the ADAS-Cog Odds Ratio. Seventy dosed improvers, i.e., 32 dosed improvers over the placebo rate. Quite impressive if true.

Leave a Reply