It's there to handle situations where the interrupt source is lost before the vector is handled. See page 791 in the manual for a description. (An example of how this can occur is also given in the PIT section page 697).
It has the highest priority and is always handled by the CPU so you must always assign a vector and interrupt service routine to it. The situation is typically very rare and how you handle it depends on how deterministic your system needs to be. If you find this interrupt occurring often then you probably have some systematic defect somewhere. If you are unable to tolerate any lost interrupts then you may need to reset. Alternatively, you may wish to keep a record of these events for diagnostic analysis.