Removing tables or columns is more straightforward than changing them. Once certain properties of the domain model are not required anymore, their usages can be removed from the application.

The first deployment changes the application code to stop reading from the database column but still to write to it. This is required to ensure that the old version can still read values other than null.

The next deployment will remove a potential not null constraint from the database column. The application code stops writing to column. In this step, occurrences of the model property can already be removed from the code base.

The final deployment step will drop the column. As mentioned before, it highly depends on the business use case whether column data should actually be dropped. Rollback scripts would need to recreate removed columns, which implies that the previous data is gone.