With the usual massive caveats of being a complete layman. However, I'll offer up my humble opinion on this specific paper anyway.
Firstly in the cases where CXL is recommended (i.e. progression in the last 6 months) then
not having it done isn't a zero risk option. Also CXL does have a significant history to back it up. That's a more general context for this to sit in.
Reviewing the conclusions:
Conclusions: CXL in human keratoconus can cause considerable morphologic corneal changes up to 30 months postoperatively[1]. Especially noteworthy is a long-lasting, maybe permanent[2], keratocyte loss in the anterior and middle corneal stroma involving the central and peripheral cornea. As long-term corneal damage after CXL is of genuine concern, particular care should be taken to perform this procedure only in accordance with investigational protocols.
[1] That's the point of doing it surely? Also only one cornea was 30 months post treatment.
[2] "long-lasting, maybe permanent" that's a
very considerable leap to jump to based on 6 corneas. So possibly only 3 patients one of whom was treated 5 months ago (so probably won't be stable yet) and the maximum was only 30 months.
It doesn't (in the abstract I can see anyway) detail what type/protocol of CXL was undertaken either.