They mostly hold data for reporting on the dashboard. Things like bandwidth used, storage used and payout info. They used to hold piece metadata, but that has been moved to the pieces themselves. So it’s all non-critical right now. I believe the only remaining db with a function to the operation of the node is the piece expiration db, which holds expiration dates for pieces. Losing that means your node doesn’t immediately remove expired pieces. However, those pieces will eventually be cleaned up by garbage collection as well.
So you’ll lose stats, mostly historic ones. But information like held amount might never be correct again, because the node is missing that information for previous months. It’s not ideal to lose the info, but it won’t break your node if you start over with clean db’s. I recommend following the info on the page @Alexey linked as that instructs you on how to recover only the corrupted db’s, limiting loss of information.