Hi Marcel
I totally agree that deleting instances (or series) in a study is not normal, with the risk of ending up with 'dangling' references and an invalid hierarchy in the study. However, if the operation mentioned above can be performed 'legally', that is, the application exposes this functionality via its UI and cmdl interfaces, then it should also put the things right by updating the database. Otherwise, one can formally allow the system to be placed into an invalid state. It's not uncommon to have this functionality in Dicom systems, both Microsoft Azure and Google Cloud allow this (as shown by the links). But, please, do not resolve this by removing the individual deletions 
(as they are useful on occasions)!
Regards,
Virgiliu