I do have a few critical questions that came to mind about the methodology and interpretation, though.
1. My biggest concern is about medication. Your patients were on various psychotropics, and you used a “medication load index” as a covariate. But isn’t it possible that the drugs themselves, not just the disease, are causing both the enlarged choroid plexus and the cognitive issues? How can you really separate the disease effect from the long-term medication effect here?
2. Regarding your ChP segmentation: you used a deep learning model and then manually checked the results. Were the people doing the manual checking blind to the diagnosis (BD, MDD, or HC)? If not, couldn’t that have introduced bias, especially since you expected to find differences?
3. In the BD group, the correlation between ChP volume and logical memory disappeared when you controlled for illness duration. But the correlation with visual reproduction stayed significant. That’s weird, right? If the effect is driven by the disease progressing over time, why would one memory test be affected and not the other? Doesn’t that inconsistency make the finding a bit shaky?
4. Also, your correlation coefficients are really small (like -0.15). With your huge sample size, you can get statistically significant results with tiny effects. How confident are you that these small correlations are actually clinically meaningful or relevant for an individual patient?
5. I noticed you used FDR correction for some results but not all (like the correlations with SDMT and VFT in MDD). Why the inconsistency? Shouldn’t all those multiple comparisons be strictly corrected?