Re: Celebration Thread! Biogen is going to the FDA with Aducan.
Posted: Tue Jul 06, 2021 9:42 pm
I drew in the "U"s for the blue and red curves to show the broad trend of the data near the middle. In the two figures at the top of the previous post I then applied 20 point moving averages (medians) for the 302 high treatment arm (after removing those points which were above 4 CDR-sb change -- this was the cognitive decline level that was posted recently that was found to be a rapid progression outcome from maintaining amyloid positivity through time. A few patients in the 302 high dose exhibited this behavior and I felt it best to remove these data points because they distort the more typical pattern.)
What these figures show is a strong parabolic trend for both the average and the median. It was especially strong for the moving average with a correlation of ~0.90. I had been very unclear earlier about what this pattern could mean. The post from yesterday apparently provides an explanation. The strong non-linear trend is likely a result of the covariate differences that were noted. The patients on the ends had higher cognitive decline risk and so the parabola is exhibiting more this increased risk than any SUVR effect.
The effect is extraordinarily high and is completely unobservable by simply looking at the blob of dots.
For me the big lesson here is that data analysis is really all about showing these obscure patterns in the data. Whoever can construct the biggest jigsaw puzzle with the most order and meaning wins -- er, here, that would be $100 billion and the gratitude of tens of millions of people coping with a progressive neurodegenerative illness. Creating meaning through constructing the data into the higher order of the correlation plots is part of this meaningfulness construction. As we can see in the above figures this creation of meaning also might be possible even closer to the individual level (or at least ~ the level of 20 patient moving averages). If we could get the overall regression from the other large anti-amyloid trials to fit well into this figure, then we would have very powerful evidence on all scales of anti-amyloids.
Also wonder with this style of thinking whether a randomized trail for aducan would actually be needed. What might be done instead is to simply treat everyone with high dose. Different patients would clear different SUVR amounts and then they would fall along the regression line. In this way all the patients would contribute meaningfully to the result; with placebo we know that they will have an average of ~ 0 SUVR and and average of ~ 1.5 on CDR-sb; this does not seem overly helpful. Having more patients defining the regression line would be much more relevant.
What these figures show is a strong parabolic trend for both the average and the median. It was especially strong for the moving average with a correlation of ~0.90. I had been very unclear earlier about what this pattern could mean. The post from yesterday apparently provides an explanation. The strong non-linear trend is likely a result of the covariate differences that were noted. The patients on the ends had higher cognitive decline risk and so the parabola is exhibiting more this increased risk than any SUVR effect.
The effect is extraordinarily high and is completely unobservable by simply looking at the blob of dots.
For me the big lesson here is that data analysis is really all about showing these obscure patterns in the data. Whoever can construct the biggest jigsaw puzzle with the most order and meaning wins -- er, here, that would be $100 billion and the gratitude of tens of millions of people coping with a progressive neurodegenerative illness. Creating meaning through constructing the data into the higher order of the correlation plots is part of this meaningfulness construction. As we can see in the above figures this creation of meaning also might be possible even closer to the individual level (or at least ~ the level of 20 patient moving averages). If we could get the overall regression from the other large anti-amyloid trials to fit well into this figure, then we would have very powerful evidence on all scales of anti-amyloids.
Also wonder with this style of thinking whether a randomized trail for aducan would actually be needed. What might be done instead is to simply treat everyone with high dose. Different patients would clear different SUVR amounts and then they would fall along the regression line. In this way all the patients would contribute meaningfully to the result; with placebo we know that they will have an average of ~ 0 SUVR and and average of ~ 1.5 on CDR-sb; this does not seem overly helpful. Having more patients defining the regression line would be much more relevant.