It's a really difficult question to answer in a general way. Without specifying a clinical domain you could only answer this for a datasset that covers all of healthcare.

But it is also an important question. Your information description has to span the domain you're interested in (scope) and with enough granular/detailed points to thoroughly describe it (completeness). Without considering this you will not end up with any useful insights from the data, and could end up with unsafe work (if important data points are not considered).

You also need to establish a meta-data framework that allows you to contextualise your data points (either patient centric or perhaps in youor case disease or condition centric?).