I have worked with this data for a while now and I noticed a distinct lack of correspondence between the time-indicating variables between the different files and a probably very problematic coding with the month indicators of the questionnaire files:
There are three ways, in which time is coded: Visit numbered (1,2,3,4,5,...), month abbreviated (m00, m03,m06...), and month as variable name (Baseline, Month03, Month06,... or some variation thereof).
Unfortunately, the visit variable (e.g. in syn4984917) does not correspond coherently to a time point. This can be seen in the "Age_At_Visit" variable. Sometimes a person ages one year during Visit 1-5 (as expected according to the paper) and other times the person ages 5 years during Visit 1-5. This would be ok, as it it somewhat easy to detect and correct. Unfortunately there are incoherences which can not be solved.
I will make an example for one participant ("LNDCTL002"):
According to syn4984917, that person visited 6 times, and aged 1 year in-between visits. So the time-frame should be 6 years. This files contains MMSE data for that person.
According to syn16807689, that person had Scans at time point m00, m03, m12, m24. This does not work out for me, as m03 is not listed in syn4984917 (except Visit 1-6 stands for m00,m03,m12,m24,m36,m48, and that persons birthday fell between m00 and m03, possible, but I think unlikely).
According to syn4013024, that person visited Baseline, Month03,Month06,Month09,Month12,Month15 and has no otherwise, especially not Month24. The worst problem is the following: The MMSE data listed in syn4984917 is identical to the MMSE data in syn4013024, despite the first file listing those measurements as yearly and the second one as Baseline, Month03,Month06,Month09,Month12,Month15. This can not be solved by just declaring the "Age_At_Visit" variable in syn4984917 wrong, as the MMSE data in syn4013024 suggest, that there was no Visit at Month24. However syn4984917 and syn16807689 suggest otherwise.
Normally I would declare the time point from one of the three datasets as invalid and go on if the other two datasets give corresponding values (which would be a bit problematic). In this case however, all three datasets indicate different time points (which is really problematic).
This was just an example for one participant, but my I would guess, that roughly ~50% of the data has some form of incoherence with the timestamps. Not every one is as bad as this one above. None the less, this severely disturbs my trust in the dataset and the conclusions published from them. I would very much welcome any suggestion on what the underlying error is and whether there is a possibility to fix it.
Kind regards
Jannis Denecke
Edit: I need to work with the raw MRI data, so ANMerge is probably not a working alternative (I am not sure, as i don't have access to the data). And I currently have no confidence whether this error has been corrected or just masked in ANMerge.
Edit2: Don't trust the age column in the proteomics data (syn5367753), as nobody aged a single year between visit 1 and 5, although visit actually means time point here (1 = baseline, 5 = m12, regardless of diagnostic group)
Created by Jannis Denecke jannis.denecke Hi @jannis.denecke,
There is an issue with the visit number and the time stamp in the original data and what you outlined above is one of the reasons for this (different protocol for when patients were invited according to diagnosis). I fixed those in the ANMerge data, so I would highly recommend using this version.
For some MRI images (the 3month ones you referred to) there is no corresponding other data for CTL and MCI. This is due to the study design and unfortunately there is no solution for this.
The raw MRI data for AddNeuroMed can be used with ANMerge data. The images themselves did not change during the rework of the data. You might run into issues with patient IDs but syn22023005 should help with this. It might also be the case that you find less raw images in the AddNeuroMed data than there is data in the processed ANMerge data but this is since some people were reluctant to share the raw data when I tracked it down.
Best
Colin @jannis.denecke please note that there is an update of this study available. See link to 'ANMerge': https://www.synapse.org/#!Synapse:syn2790911/wiki/235387 I think, I may have pinned the error down: Control and MCI participants have visited only yearly, except for those who had MRI Scans. Those have a time point at m03, but without behavioural data. Therefore the visit variable is not consistent, as described in Birkenbihl et al., 2020. On top, the month indication in the variable names of the behavioural data is wrong and each column represents the respective number of visit (Baseline = Visit 1, Month03 = Visit 2, ...). This would be the only explanation i could come up with right now. However, that would also mean, that the answer in https://www.synapse.org/#!Synapse:syn2790911/discussion/threadId=5355 is incorrect.
I think, this is supported by the fact, that some behavioural data has the month indication as variable and some has visit as variable indication (compare syn4013024 and syn4013001).
As for the lack of behavioural data at time point m03 for MCI and CTL, but the availability of MRI scans: Can somebody confirm, that this is actually the case, or if behavioural data should be available?