Welcome

JCA welcomed the group. He said normally the meetings would take place on the fourth Monday of the month but the present meeting had been moved due to an IHTSDO meeting in Miami.

Declaration of Conflicts of Interest

None.

Approval of the Minutes from the Uruguay Meeting

GRE said there was one typo: he had mentioned that nesting expressions would require a couple of years for tooling from the modeling AG, it was minuted as requiring 20 years. It should be two years, he said.

With one correction, the AG approved the minutes from the Uruguay meeting.

Approval of the Minutes from the 30 November Meeting

The AG approved the minutes from the 30 November Meeting.

Laterality

JCA said that he had intended to have a revised document (see 2016-01-22 - Editorial Advisory Group Conference Call, Meeting Files, attachment: Arf223747_CTP Project Combined Inception-Elaboration Phase v01.docx) ready for review, but he had been too busy. However, based on comments from KCA, he had made some images of what the models would look like. (See 2016-01-22 - Editorial Advisory Group Conference Call, Meeting Files, attachment: lateralized concept examples.docx). JCA walked the audience through the examples. He said that after flattening out the representation, the addition of the inferred parent of “Disorder of left lower extremity” indicated that the classifier did not detect the equivalence of the pre-coordinated and flattened representations of laterality. He asked GRE for some insights into what they might do about it, because it would affect what they would do with existing concepts with precoordinated laterality in the role groups.

GRE said he had emailed a document to JCA, noting that he did not seem to have the permissions to edit the laterality page. The conclusions of GRE’s evaluations had been that the only option that retained complete semantic equivalency was option 1, with the nested body structure qualification by the laterality attribute. He said the result that JCA had gotten had been expected. Given that it was not possible to use nested expressions, he suggested discussing with the Modeling AG a proposal to support or change the way the classifier is loaded so that one could represent it in the role group in the authoring environment without supporting the nested expressions. The classifier would be loaded with the nested expression through a transformation that would be straightforward. Then the classification results would be correct. On the other hand, the computation of the distribution normal form might be affected, depending on the algorithm. GRE then said that the Modeling AG had med the previous Monday, and there had been two or three discussions related to OWL. They had discussed how it was not sufficient to represent nested relationships for current structure and anticipate future needs.  GRE noted that he would be holding a meeting with Robert Turnbull, Olivier Bodenreider and Harold Solbrig about how SNOMED is represented in OWL.

JCA asked GRE, in terms of technical changes to the classifier and the transforms that would be necessary to compute the semantic equivalent, was that something that was relatively straight forward and could happen in a release or two? Or would they have to go and revise the existing concepts to ensure that they had the proper inferences?

GRE replied that it only applied to particular areas. Procedures and findings would classify correctly. The problems only affected a small part and the tooling environment could not be changed so fast. Having it as a role group would be okay. Both KCA and he, GRE continued, had experience in feeding the classifier in the Workbench for particular assets.

In conclusion, GRE said, he and his team recognized that supporting anonymous concepts as the target of a relationship was the only semantically equivalent correct pure approach. But in the meantime, while there was no way to either author it that way or distribute it properly, the flattened model could only be transformed at classification time. The issue still had to be discussed by the Modeling AG, but essentially it would be about whether that approach could be done. It would take a few days to make the changes and a few weeks of testing, but it would be possible to do it for the next release. So his advice, he said, was to do it in a flattened model in the meantime.

KCA said it would be a good idea to further develop the model on nesting. The fact that they could not get the equivalencies that they wanted without resorting to nesting was an indicator that they needed to move towards nesting. They could begin considering what it should look like and they could develop guidance on how to nest anatomical structures when feeding them to the classifier. He noted that he and GRE had both written some things on the topic.

JCA asked if they needed to give the Modeling AG some use cases. GRE said that when it would be transformed into OWL, you would use a post-coordinated concept. That post-coordinated concept might have 2 parents or a nested relationship inside, so the Modeling AG recognized the complexity. In the meantime, flattening the model and making the guidance available would probably solve the authoring and distribution issue.  He said they could provide guidance to the internal technical team. In the long run, there was no way out of avoiding precoordination while maintaining semantic equivalence without supporting anonymous classes.

GRE then said that the work he and YGA had done had indicated that it would be too costly to precoordinate everything. Option 1 was still the best and the only option that was semantically equivalent.

• Action: ensure that AG members can update Confluence. (Note to AG members: if you experience any additional problems please write to Techsupport@ihtsdo.org for assistance.)

JCA said they should take GRE and YGA comments and widdle them down to something that could be presented to the Member Forum and/or the Management Board, because that would mean significant changes to the modeling. Not knowing how long the testing or technical changes would take, they might want to look at the concepts that were currently modeled with the precoordinated anatomy concepts and revise them so that there would be no differences in inferences until they had the ability to do the transformation. JCA estimated it was only a few hundred terms. GRE said it was about 1200.

JCA asked if the other AG members had any comments. BGO said he agreed with everything that had been said, and PAM said it sounded okay to him. 

Addition of "DURING" and "BEFORE" attributes

Next, BGO gave a presentation on the ECE evaluation of adding “DURING” and “BEFORE” attributes (see recording for presentation or the document "SNOMED CT Improvement Project: Combined Inception and Elaboration Phases, Topic: Adding new attributes to the associated with role hierarchy to represent additional temporal associations between procedures and their related disorders," dated "12/10/15," v0.01).

GRE said BGO should probably create a small model, as KCA and YGA had done with laterality, to find out which was more efficient, semantically better, more expressive, and creating the fewest number of complications. He said it needed to be tested for a while. BGO agreed and said he would work on it using the 40 or so examples. JCA said they would record that as an action but not put a deadline on it. He added that it would be very useful to work on that because it would address a number of tracker items at the same time. GRE suggested using either the Workbench or Protege, then testing it in the other environment as well. GRE added that he would be happy to help, so BGO should not hesitate to contact him.

• ACTION: BGO to test out the "DURING" and "BEFORE" attributes using the 40 or so examples in the combined inception and elaboration document, using Workbench and Protege.

Extending meetings to 90 minutes

The AG members agreed to extend their monthly meetings from 60 to 90 minutes.

Is there a need for a description that matches the FSN?

JCA said a question had arisen on whether there was now a need in RF2 to mandate a description that matches the FSN without the semantic tag. That had been the rule in the Workbench, but the new FSN naming conventions that the team had adopted made some of the FSNs less user-friendly. 

KCA said he supported the change because he had always found it redundant to have two almost or completely identical descriptions, with the only difference being the semantic tag. BGO agreed.

GRE, however, disagreed. He said it had been the rule for the last 10 years, and the benefits depended on the use case. He said he recognized that recently the unfriendly FSNs might not be an advantage and sometimes the difference was very small, but most FSNs in earlier releases were glorified preferred terms. There are probably several thousand FSNs that are ambiguous and will need to be replaced, requiring more extended versions and clarification of meaning. In some use cases, like reporting, it was not easy to retain the context of the original record in secondary usage, GRE continued. Sometimes it could be done using FSNs in the display or the preferred term. The idea of the expanded version of the description, he said, was to enable the user to choose the description in the localization or to be more explicit. 

GRE then said he would abstain on the issue because it would just affect the English version, and other language editions would be able to decide on what is best depending on the use case. He suggested limiting the decision to the English representation.

KCA said in descriptions the use of the semantic tag in the FSN was over-burdensome, but hat could be a separate consideration. He said he had never liked the approach of duplicating 2 descriptions and having a semantic tag difference. One solution was possibly to have the semantic tag go away.

GRE said most FSNs were ambiguous - at least 10 percent, he had found - without the semantic tag. 

JCA said that even with the semantic tag, currently there was a common misuse of terms based on the FSN and ignoring the semantic tag, which provided a marker to reveal whether the user was in the right area of the terminology. GRE agreed that there were entire hierarchies with similar terms that could not be distinguished without the semantic tag. He said he would probably want to retain the practice in the Spanish edition; if the change were limited to the English edition, then the native English speakers would know better.

JCA said the proposal provided some challenges to the editors and to those requesting new content. Many requests for new content involve creation of a preferred term without the semantic tag. If there is no other description, then the editors do not know if the submitter wants that as the preferred term or not. He said it was something the editorial team would need to discuss because it had an impact on modeling. From a tooling aspect, JCA added, the authoring tool has a description that matches the FSN without the semantic tag. 

PAM said he thought it required some more thought. Also relevant, he said, was the issue of natural language processing, which would be impacted by not having something that was deemed to be fairly unambiguous. By keeping the semantic tag in, it might mess up the NLP and have an impact on translations.

JCA said if requestors ask for a more user-friendly term as the preferred term, it often is not a semantic match, which meant having to explain to them why they could not have their preferred term. So he agreed that it was not as straight forward as one might think. He said the editorial team would discuss it.

Use of URLs in Concept Descriptions

JCA asked for opinions on whether to allow or disallow URLs in text descriptions for concepts. BGO said that given the dynamic nature of URLs, it would not be a good idea. KCA said he supported URLs in definitions but not descriptions. JCA clarified that he was talking specifically about definitions, and that he was talking about plain text.

PAM said that there needed to be a database of articles read at the time that a decision was made so that the articles do not disappear.

KCA said a URL did not have to be a web location. Having the ability to use, for example, file URLs to provide a richer way of conveying information was a step in the right direction. There could be policies around how to do it, like having an automated way to check that it is still there, or one could make the web content part of the International Edition using the file URLs or other things. At some point, he said, they needed to get SNOMED into the modern world and think about modernizing how they represent subscripts, superscripts and text definitions, and having URLs in definitions was a first step that could open up opportunities.

PAM said he sort of agreed but it was important to deal with the underlying issue of not being able to afford to lose the basis upon which you based your decision. BGO said that if an article is referenced, one could just add the reference to the article instead of the URL, but if it were referencing just text on a website, that could present problems.

GRE said the first 500 text definitions were about the meaning of the concept in SNOMED CT, but more recently text definitions are trying to define a concept in a more general, generic way.

JCA said it would require more evaluation because it was about more than just URLs. There were a lot of issues around text definitions that they may want to take on. He said further that they had been using a definition that had come from SKMT, the Standards Knowledge Management Tool that inherits definitions from a number of standards and puts them into a metathesaurus.

• JCA said he would go in and create new discussion pages for each of these topics.

Any other business

GRE said it would be important to define the dividing line between the responsibilities of the Modeling AG and the Editorial AG. JCA agreed.

Next call

22 Feb. 2016.