Measures of Dementia in Alzheimer’s and Their Distortion of Trial Results $AVXL $LLY Blarcamesine Donanemab

I have been asked about different measures of dementia and their impact on the trials. There are two basic aspects of these measures and dementia.

  • There is the innate to human cognition nature of deterioration resulting in greater or lesser progress to an observer.
  • There is the sensitivity of type and mix of measures build into a test tasks, or selected to create the composite score expressing the amount of deterioration

Mini-Mental State Examination on average has such plot for Alzheimer’s sufferes. In the pursuit of ideal measure DSRS has been created as it strives to produce straight line from MMSE 20 to 7. [3]

That is prospective decline from MMSE 17 points initial position to zero for an average patient.

Another illustration presents the respective scales relationship between 30 and 19 MMSE. [1]

Notice that MMSE scale is very insensitive at the almost healthy 30 points scale having a ceiling effect, that is early demantia, but is doing much better at 19 MMES vs. ADAS-Cog and CDR-SOB at this interval. The best to sense changes in early stages of dementia is ADAS-Cog and it is reasonable at 19 MMSE. The illustration of the relationship of these scales is in itself suspect of bias as the x-axis “Cognitive dysfunction” calls for its own criteria which is doubtful to be just entirely objective.

Some measures are questinnaires to be filled by care givers or doctors like CDR-SOB other are tests like ADAS-Cog or MMSE. At different points in patient deterioration they will have varing difficulties with relating the progress of the disease as the cognition is heterogeneous and each individual might follow a slightly differnt pattern. There are even problems with repeatibility as results taken few days apart might not be the same.

Depending on the movement of patients on these scales different values of Effect Size will be calculated. Effect Size Cohen’s d calculations need only the means and standard deviations. So it looks how the means are separated from each other and whether the distributions don’t overlap each other.

The p value in statistical parlance means that if you run the sample size experiment you have to get a p*100% chance to get it wrong, that is at least in the layman terms. Once this is below 5% chance of getting the difference between the drug and placebo as nonexistant due to random effects it is called statistically significant. The calculations take into consideration the distribution of data points and sample size.

Anavex posted the results for the 148 week OLE extension of the phase 2a trial. Assuming that the bars are standard deviations I calculated following Effect Sizes from the data.

Can these disparities between Effect Sizes be placed squarely on the shoulders of scale sensitivities? I do not know. One is certain that these things are real hard to assess as there is no completely objective measure of cognitive deterioration of people with Alzheimer’s or dementia. The large Effect Sizes are a bit concerning to me as to their authenticity but they follow the assumption of the bars being SD (standard deviations). I guess I must be doing somethig wrong or this is just insane efficacy. LOL

The trials are run against placebo. But if the sensitivity is varying along the path of progress or deterioration the picture is somewhat distorted. A solution could be to select few measures which complement each other. This can run foul of the accepted till now standards or the limited resourse a small biotech can expand on the trial. another solution could be to run all drugs on standard set of measures so that distortions are equivalent from trial to trial. But even this has it down side as different trials with cover varing parts of the scales. An example is trial by Anavex where MMSE hits ceiling effect with patients actually recovering and $LLY Donanemab trial where patients are deteriorating so close to placebo that the theory that one’s runs against placebo makes this to be the case where the distortions are equal in both arms is close to valid.

When Anavex run the phase 2 trial of PDD (Parkinson’s Disease Dementia) it used the CDR-CA (computerized assesment). As I said we have for most no objective measure of dererioration and all we can do is to compare between them and design ever better ones more inclusive but then they become extensive. Here are some results of comparison of MMSE, ADAS-Cog and components of CDR-CA. [2]

Ss you can see for the scores of MMSE 27-30 that the ADAS-Cog can actualy show improvement which is just another way of saying that it can detect improvement where MMSE can not find any. CDR-CA is correlated to AD but mostly used for PD as it measures also time delay in psychomotor skills and memory. The shortcomings of this measure vs. MMSE is exclusion of certain languge skills. One can see that it is most sensitive between 26 to 21 MMSE points. So follows ADAS-Cog. Both ADAS-Cog and CDR-CD lose their sensitivity below MMSE 20 points. I am speaking here of sensitivity but I am not able to distinguish it from the acceleration and deceleration in deterioration as percieved by humans. Nevertheless this graph confirms the information contained in the second plot in this post.

Here is a graph representing detrioration over some period of time in group of patients dosed with different SOC Alzheimer’s drugs which in general have a similar performance. The components of CDR-CA are enclosed in a box. They probably are going to be weighted to give a final score. In a sense all these measures are very close. Nevertheless ADAS-Cog is just 70% of MMSE score on the same deterioration but if the dosed arm does not move to different regime on the scale the distortion is minimal but the point is to have the drug to do precisely this thing. In case of Anavex2-73 phase 2a trial this was introducing uncertainty as MMSE scores lacked the sensitivity to pick up changes from the 4 patients who in general seemed to beat the odds and recovered from the disease.


[1] link

[2] link

[3] link


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

Leave a Reply