Typically we would plug in a procedure details cluster, specific to each speciality/ procedure.
This is from the openOutcomes Hip arthroplasty template
and is based on a form used for outcomes reporting.
If nothing else, this acts as a conversation space where we might try to rationalise the datapoints and related coded valuesets, captured very variably even within the orthopaedic knee surgery community.
Is this hard, yes, does it keep changing yes. but I don’t think NLP shortcuts the standardisation/rationalisation process, though it may be a great way to enter the data. You still need to get some degree of clinical consensus on what are the questions and the related acceptable answers and other data constraints