Anavex Life Sciences Corp (AVXL)

[williamssc]

Reply from amstock82 posted on another board(investorvillage)- Updated - Re: Primary Endpoints Claims Trial Endpoints Not Met
The Primary endpoint of the ADAS Cog is first given this is the first thing analyzed by Anavex on the slide. 
 
1) ADAS-Cog score change of or better -0.50 @ 48 weeks is defined as a clinically significant change. Therefore, patients with an improvement of -0.50 were defined as improved. (A later slide shows the improvement of -1.85 overall for all patients treated)
 
2) Among patients that improved with Anavex 2-73, the mean ADAS-Cog score improved -4.03. 
 
They give the odds ratio for ITT or Intent-to-Treat population as 1.839. Basically, if an odds ratio is 1, then it is 50/50. 1.839 is 83.9% change of something. Or as they give it, people who got Anavex had an 83.9 or 84% more likely to improve compared to the placebo group. 
 
The treatment group includes both the 30mg group and 50mg group as well as some of the patients who have the 20% chance of the altered gene (the gene that causes Anavex 2-73 to not work as well). And if the patient is already fairly far gone, they won't be able to measure improvements in the ADAS-Cog. They might see changes from MRI and other tools. 
 
But given the -4.03 number, it means that the group of patients that improved - improved tremendously. 
 
The p value is very good at 0.015. (the great p value is attributed to the large improvement and is unlikely to be random). 
 
Next slide ADCS-ADL
 
This slide is analyzed the same way as the first ADAS-Cog.
 
The treatment group was 167% more likely to improve when compared to the placebo group with this test. This is very significant, and they also have a good p value of 0.0255.
 
Next slide. ANAVEX®2-73-AD-004 Primary Endpoint – ADAS-Cog
 
The raw data shows patients who took the placebo began with ADAS Cog of 29.18 (standard error of 0.61). For any one sample, the standard deviation can be large. (Standard deviation is a measure of the variability of single samples, while Standard Error is how much the population mean may differ from the sample mean. 
 
The slide indicates that patients taking the placebo's scores in 48 weeks declined by 4.11 overall (SE=0.86). 
 
The slide indicates that patients taking the 30 mg and 50 mg drug declined overall by 2.26 (SE=0.51).  
 
At 48 weeks, they ended up with 161 patients for placebo and 301 in the treatment group. Inferred from slide 25 safety population of n=161 and n=301. 
 
To do the t-test, use the following formula.  
 
\begin{equation*}t=\dfrac{\bar{x}_{1}-\bar{x}_{2}}{\sqrt{(s^2(\frac{1}{n_{1}}+\frac{1}{n_{2}}))}}}\end{equation*}
 
There are a few potential ways to calculate this. That is because the baseline and placebo both have baseline measures and SE and then the SE changes at 48 weeks. 
 
But to do it the easy way, go to the web tool that what's his name used and put in correct values and do the test for sample mean is greater than placebo/test mean. 
 
 
 
(Thanks to the Evan Miller organization website test tool for t-test): https://www.evanmiller.org/ab-testing/t-test.html
 
(SE = Standard deviation / Square root of number of samples). Since we have SE and number of samples, it is easy to derive standard deviation to put into the tool from Evan Miller. Have to click in the Sample summary box to change the data. I.e., click in the round box next to the (Sample 1 Summary) and on the other side in the (Sample 2 Summary) to input data. Anavex also indicated on one of their slides the change in numbers due to deaths or dropouts in testing. 
 
The p=0.0327 (Anavex has it at on their slide at p=0.033).
 
I also calculated the same thing using a more complex calculation which I won't go into, getting the t value, degrees of freedom and comparing it to the chart: https://cdn.scribbr.com/wp-content/uploads/2022/06/Students-t-table-one-tailed-two-tailed-L-Scribbr.pdf
 
It comes up to approximately the same.
 
The trick short traders played is mostly to not use the correct comparison with the tool. (That is why the person who posted a picture from the Evan Miller Organization cut off part of the results from his picture). What we want to determine is if there is a difference in the 'means' that is statistically significant? In other words, could the change in means be random chance or is it real? Second, he didn't put in the right values but that would only have had a minor impact. 
 
Note, some may say that the company didn't point out what type of t-test that they are using. They don't have to do so because to anyone who has knowledge of statistics, the right methodology is apparent based on what they are testing for. In other words, all they have to do is know that they are trying to prove that the use of Anavex 2-73 improves cognitive decline. Based on what they are trying to prove, that determines what test to use. The slide indicates:
 
"ANAVEX®2-73 treatment slowed cognitive decline by 45% compared to placebo at 48 weeks" 
 
The test is to show whether or not the above claim is true. It is determined to be true if the test results in p < 0.05. I did the calculation 3 ways one of which was using Evan Miller's tool. Sample 1 using his tool is the placebo data. Sample 2 is Anavex 2-73. The Evan Miller tool shows that Sample 1 mean is greater with p value better than 0.05 and an actual p value of p=0.0327. Since Sample 1 mean is greater, this indicates that Anavex 2-73 slowed cognitive decline with a certainty of p=0.0327. The 45% can be calculated as given on the slide. 
 
The other two ways I calculated t values came out to approximately the same conclusion with ~ p=0.03. The reason I calculated it manually is to verify the Evan Miller organization tool and the overall methodology. 
 
NOTE: Some say that the FDA indicates a t-test two tailed test must be used and a one tail test is not the norm. This is BS. The statistical test used is based on what they are trying to show. Based on what is being proved by Anavex, the test is determined to be a t-test two sample test.