In V2, OBX-33 is repeating, because we needed to be able for one observation specify multiple specimens - for example, creatinine clearance or Amylase/Creatinine (https://loinc.org/1811-9/, https://loinc.org/54355-3/, https://loinc.org/70268-8/, https://loinc.org/70264-7/, https://loinc.org/14117-6/).

In some cases, a single observation is made and result returned based on more than one specimen.  An example is testing for oligoclonal bands in CSF and serum:

https://www.labcorp.com/tests/019216/oligoclonal-banding-serum-and-cerebrospinal-fluid

In order to address this, there are several possible approaches:

1. The ideal solution would be to increase the cardinality of Observation.specimen from 0..1 to 0..*, to match the V2 capability, but this breaks strict backward compatibility since it changes the wire format (for JSON), and since Observation is a normative resource and very widely implemented, this may likely be a problem (at least it likely requires quite significant implementer consultation and reaching agreement).
2. Add a sibling Observation.additionalSpecimen element, with cardinality 0..*.  In cases where multiple specimens are required, both the 'specimen' and 'additionalSpecimen' elements would be used.
3. Since there is a real requirement for specifying multiple specimens, but the need for that is expected to be rather infrequent (probably considerably less than 5% of lab tests?), similar to #2 this could instead be done using an 'additionalSpecimen' standard extension.
4. This is also to some degree a variation on #2: Add an Observation.specimen2 element (whatever the actual name), which going forward would be expected to be used instead of the 'specimen' element (all specimens would be represented in one place).  The use of the original 'specimen' element would be discouraged or formally deprecated (however we go about doing that).

There may be additional alternatives (or combinations and variations on the ones above) which could be proposed and added to this list for consideration.