That is actually something you will find little documentation on publicly. There are classes/workshops that ESRI charges for or that you can attend at the ESRIUC but less in the public space.
A couple of the points you mention come out of good DBA practices; but some really dont and really depend on your sytems and needs.
For integrity, for sure a stable hardware platform (I always go clustered systems) is key; and using the right DB backend. I am from a strong state and federal background so I usually stick to big-iron DBs like MSSQL or Oracle. MSSQL2008 spatial gives you a good toolkit for the cost; versus if you really want bullet-proof with 3rd party access you want to consider Oracle; where lots of apps connect to the spatial side.
Security; well there you have a strong DBA need to make sure you have accounts configured properly; but then you also have a need to use the built-in ESRI controls for who can read/write etc from your DBs. This also brings into question the structure of your data and your systems using it; do you need to create higher security silos where you have more tighter user level security; or is just a huge single DB ok with you maintaining the user by user level permissions across the features.
Things you also want to think about is using internal objects like domains versus reference tables with those values good for you; domains may not be accessible from 3rd party apps looking at your data; where a reference table can be joined and you have those coded values.
Data normalization is a big topic to consider; some people are happy doing a solid normalized design with geometry only SDE features and then all the tabular stuff in business tables that you then have a lot of indexes and views to optomize performance for searching by other apps.
So if you have particular areas you are curious about; feel free to add more details to your questions but much of what you looking to really depends on your platform and your functional needs. A high-speed web-app maynot want SDE versus just fGDB; versus dynamic data would demand SDE.