Call to order and role call

Jim Case (JCA) welcomed everyone. Attendees are listed at 2016-08-22 Editorial Advisory Group Conference Call. Keith Campbell was unable to attend but sent comments in advance of the meeting.

Approval of prior minutes

Approved.

Disjunctive components (LOINC)

Daniel Karlsson (DKA) from the Observables project gave a presentation using slides previously posted on the agenda page. He said there were a number of LOINC terms in the cooperative area in the current agreement to be mapped to SCT expressions. There were 814 LOINC terms, but some of those used a plus sign in a different way and should not be counted. He estimated that there were 700-800 LOINC terms in scope. The problematic ones, he said, were where there was a component and a union of different substances. It could be any or all of those substances, meaning it would be a disjunctive statement.

There were several possible solutions, DKA said. A disjunction was not really an alternative, so that could be ruled out pretty quickly, he said. Another alternative would be to continue to manually maintain the observables. However, that strategy had been shown to introduce some inconsistencies in the substance hierarchy, so it was not a perfect solution either. Another alternative was to extract the subset, classify the terms using a more expressive classifier, and export the result back as primitives. Within that solution were two alternatives, which he described. 

JCA asked for clarification: one of the options would be to represent them in a more robust environment like Protege and then integrate them back in? DKA said yes, one alternative would be to introduce the concepts as new substance concepts. JCA asked whether there had been an testing on the impact that would have on classification. DKA replied that he did not think the impact would be large since the concepts would mainly be used as observables. However, they would show up, looking strange as substances in the middle of the hierarchy. 

Toni Morrison (TMO) said she would be uncomfortable introducing such concepts into a foundational hierarchy and she was unsure of what it would mean elsewhere. The concepts could not be used to model products, and she was not sure how they would model through when modeling allergies. She thought some testing would be needed. DKA said they would behave like intermediate primitive groupers. That was not a great solution, he continued, but they would be generated using a more expressive logic, so it would not be a manual curation of the groupers. TMO asked how someone using SNOMED CT would know that they were not supposed to model with them. DKA replied that he did not think they would be forbidden from being used, but you could not use them to postcoordinate new concepts. He added that the list of 814 would seem pretty random out of context.

Guillermo Reynoso (GRE) said it was difficult enough to clean up the substance hierarchy and adding this kind of contractors into the hierarchy would be difficult to deal with. He suggested having an extension without putting them into the core. Making it more useable for this particular use case, he said, made it less useable for everyone else. 

DKA said he assumed that the solution would be to make the observables primitive. One could classify those using more expressive logic behind the scenes.

Yongsheng Gao (YGA) said it would be difficult to maintain in the core, and substances did not even have a model so it would be hard to maintain regardless of the solution. He said he preferred to look at the EL use of GCI to achieve the goal.

Paul Amos (PAM) said that if he found something like that in his patient's notes, it would be really unhelpful and not give him confidence that he knew the patient's problem. In the laboratory it would be more narrowed down and then would go in the patient record well enough, and he could accept some ambiguity in the laboratory record, but he would prefer it to be in an extension. PAM added that they had spent a lot ot time trying to get rid of and/or in ICD, and if it were used now, it could open the floodgates. He suggested working with Regenstrief to adjust its model rather than bending to Regenstrief's model.

BGO thought the best approach involved making them primitive and inserting the individual components manually, and he would keep them in the core.

DKA said he was not sure what to keep in the core - they are all expressions and not SNOMED CT concepts, so the full observables would never be according to the agreement in the core. JCA replied that BGO's point had been that the disjunctive substances would be in the core. 

Responding to a question, DKA said the hierarchies that would be impacted were substances, cells and organisms.

JCA said he wanted a summary of the currently supported options for the next call. If they could not finish the topic there, they could discuss it in Wellington. He asked how critical the answer was to the LOINC modeling group's work. Farzaneh Ashrafi (FAS) said the candidate baseline was scheduled for February.

JCA summarized the main ideas: (1) just model them as primitive in the core and assign the subtypes manually (in substances they would have to do that anyway because there was no model); (2) create an extnsion that could be used in conjunction with the core. DKA added another option: (3) make the 814 observables primitive - there would not be a component in there, it would be part of a primitive component.

JCA said those three options would be discussed next time. He asked about the impact of going ahead and making them primitive in the initial release and revising them at a later date. FAS said the team could do that and had done that before, but she wondered what DKA thought of that solution. DKA agreed that that was the least work-intensive solution.

• LOINC modeling team to take the 814 disjunctive concepts and make them primitive in the initial release, then revise them as necessary based on further modeling discussions.