@MollyR you could use a database, like a local H2 DB and a primary key to identify cases you already have inserted. The question then would be what would constitute such an identifier.
The reference row filter also could accomplish that if you would define what constitutes a row you want to exclude if you would do that in KNIME.
Maybe best you come up with sample data and construct a workflow that demonstrates the whole of your challenge without giving away secret information.