Forwarding this email thread along - of interest is the sample PCORI metadata on the i2b2.org demo site.
Hubert From: Murphy, Shawn N. [mailto:[email protected]] Sent: Friday, August 01, 2014 6:19 AM To: Parsa Mirhaji; Phillips, Lori C. Cc: Hong Luo; Campbell, James R; i2b2 Academic Forum ([email protected]); Hickman, Hubert B; Karthik Tumu; Munns, Michael B; Patrick Ryan Subject: RE: Does i2b2 Ontology support Multiple Inheritance of Concept ? Hi Parsa, I respect your use case of an interoperable database to support medication surveillance upon which OMOP is based, but that is different from why we built i2b2. The use cases for i2b2 were quite varied and it needs to adapt to many local requirements for ontologies of many different types, including clinical trials like are the thrust of transMart, and meaningful use ontologies in the NY Consortium as I'm sure you are aware. For i2b2 to be adaptable as a tool for researchers, there are no prescribed ontologies, but each research network that uses i2b2, of which PCORNet is the latest, will adopt an information model expressed in an i2b2 ontology as you can find at http://www.i2b2.org/webclient using "pcori" as the user name (same "demouser" password). This allows i2b2 to adapt to different research networks with different use cases. There is a difference between building ontologies that are understandable to human users and those that are understandable to computers. Most of the current i2b2 ontologies are really geared to have queries constructed directly by humans who do not have deep computer knowledge and so they try to stay understandable at that level when they are presented. Any acyclic graph can be represented by a vocabulary tree as I'm sure you are aware. To say this is confusing to users is crossing use cases. If OMOP tried to present its ontologies to users directly, I think you would indeed see a very confusing interface. This has been tried multiple times by projects such as BIRN and caBIG and they were not successful. So all is a compromise for the benefit of the users. Please note that for very confusing trees, when you want to present them in i2b2, you can "prune" them to the user and keep their functionality to the computer by hiding some of the terms in the tree. Thanks, Shawn. From: Parsa Mirhaji [mailto:[email protected]] Sent: Friday, August 01, 2014 2:40 AM To: Phillips, Lori C. Cc: Hong Luo; Murphy, Shawn N.; Campbell, James R; i2b2 Academic Forum ([email protected]<mailto:[email protected]>); Hickman, Hubert B; Karthik Tumu; Munns, Michael B; Patrick Ryan Subject: Re: Does i2b2 Ontology support Multiple Inheritance of Concept ? Lori Answering the highlighted green question,: 1. In order to answer this question I will have to give you some background: I am interested in this entire conversation from the perspective of our PCORI CDRN at NYC. I am the technical lead along with George Hripcsak at NYC-CDRN and working closely with OMOP team to align next release of the OMOP (v5) with PCORI requirements and CDM, since we have decided to OMOP as our primary CDM and vocabulary. However, as I am sure you all know, there is an interest in also presenting our resources as an i2b2/SHRINE infrastructure. The question then immediately becomes ontology harmonization and automation. While OMOP has made investments and systematic effort to consolidate their vocabularies under a documented and controlled framework aligned with other biomedical standards (including SNOMED), it seems to me that i2b2 community has not (or has it?). So the dilemma is: we have a point A which is well documented and controlled (OMOP ontologies) and a point B that is heterogeneous and fragmented (i2b2 ontology). This makes mapping between A and B a bit challenging to say the least. Maybe SHRINE is or will eventually become that controlled/standardized ontology for i2b2 consortium? I just don't know so I don't want to make assumptions. But until there is one such reference ontology with controlled terms and alignment with standard biomedical vocabularies (SNOMED and RXNorm for example), I would think that OMOP is going to be our point of reference for ontologies and ontology mappings. We would want to map everything else to OMOP and to other standards (including PCORI) using OMOP mappings. 2. This brings us to my initial question in this correspondence: what is the best practice to support multiple hierarchy in i2b2? We are interested in any work around simply because OMOP ontology is an acyclic graph allowing multiple ancestors for a concept and deep recursions, as does SNOMED and some others. This doesn't sit well with i2b2 technical infrastructure, even if there were standard i2b2 ontologies. I am of course interested in work arounds and suggestions that would allow us to move forward. 3. With that said, here is my work plan that I would like to be subjected to your critical analysis. And of course any help is greatly appreciated. * At Montefiore medical center we have created a terminology management infrastructure that sits outside of any system (i2b2, OMOP, our EMRs and several other clinical applications). Everything else is a client to this terminology service. The system is based on SKOS that is a W3C standard for representing ontologies and terminology systems, taxonomies, etc using RDF/OWL. * This service allows us to interlink between UMLS, OMOP, IMO, etc * I would wish I had access to a version of SHRINE or an i2b2 ontology that is considered by the consortium as best practice or defacto standard, somehow (explicitly) mapped to any of the existing standards. May be with the follwoing process: * Demographics mapped to HL7/CDC * Diagnoses mapped to SNOMED or ICD9 or ICD10 * Medications mapped to RXNorm or NDF or others * Labs mapped to LOINC or CPT * Procedures mapped to CPT or SNOMED or ... * The format of the maps would be such that every leaf node in the SHRINE or the i2b2 ontology would be assigned to one or more paths, but hopefully not combinatorially (which is the problem with crude algorithms dealing with multiple hierarchy) * We would then use OMOP mappings to cross-walk from i2b2 paths to standard terms, and from there to OMOP terms, and annotate OMOP terms with their corresponding i2b2 Paths. * The format of our terminology service allows versioning and tracking the provenance of maps, therefore for every given OMOP term, we can maintain several Paths associated with SHRINE, and different instances of i2b2 implementation * The format of our terminology service would also allow automated cross walks between paths, using OMOP terms as their reference map. * This would enable us to submit a query to our terminology service such that would extract all OMOP codes that have at least one Path associated with them (as leaf nodes of any given i2b2 or SHRINE implementation) * We would then automatically compute an entire i2b2 ontology by traversing upward from leaf nodes, based on their paths. * This assumes that leaf nodes themselves don't participate in any other ancestorship relationship with each other anywhere else in the i2b2 hierarchy. That is, ancestorship is driven merely through paths and not imported from OMOP or SNOMED ancestorship relations. * We would use references to OMOP concepts as concept codes in i2b2 ontology to enable linkage between i2b2 ontology and the mapping framework * The i2b2 instance would be populated using OMOP codes that are instantiated in its ontology (ETL considerations can be automatically computed by OMOP ontology) to back and forth between the two models Comments are welcome. Best regards Parsa Parsa Mirhaji MD, PhD Director, Clinical Research Informatics Associate Professor, Systems and Computational Biology Albert Einstein College of Medicine, Montefiore Medical Center Institute for Clinical and Translational Research (ICTR) Administrator: Ms. Jeannette Lindquist [email protected]<mailto:[email protected]> Tel: (718) 430-2856 or (914) 457-6775 Fax: (718) 430-2521 Jeannette: (718) 430-2916 or (914) 457-6801 From: <Phillips>, "Lori C." <[email protected]<mailto:[email protected]>> Date: Wednesday, July 30, 2014 at 8:45 AM To: Parsa Mirhaji <[email protected]<mailto:[email protected]>> Cc: Hong Luo <[email protected]<mailto:[email protected]>>, "Murphy, Shawn N." <[email protected]<mailto:[email protected]>>, "Campbell, James R" <[email protected]<mailto:[email protected]>>, "i2b2 Academic Forum ([email protected]<mailto:[email protected]>)" <[email protected]<mailto:[email protected]>>, "Hickman, Hubert B" <[email protected]<mailto:[email protected]>>, Karthik Tumu <[email protected]<mailto:[email protected]>>, "Munns, Michael B" <[email protected]<mailto:[email protected]>> Subject: RE: Does i2b2 Ontology support Multiple Inheritance of Concept ? I2b2 does not have a universal ontology that others are mapped to. In essence, the Diagnoses part of the i2b2 ontology is ICD9. However if your site used SNOMED, it could easily be SNOMED. Having said that though, we performed an exercise at Partners in preparation for ICD10. In that exercise we mapped ICD9 to ICD10 in the manner I described. This merged ICD10/ICD9 ontology is provided in i2b2's general 1.7 release. My question to you is: are you attempting to create a universal ontology that includes SNOMED? What are you using as the target hierarchy that everything would be mapped to? Lori From: Parsa Mirhaji [mailto:[email protected]] Sent: Tuesday, July 29, 2014 11:11 PM To: Phillips, Lori C. Cc: Hong Luo; Murphy, Shawn N.; Campbell, James R; i2b2 Academic Forum ([email protected]<mailto:[email protected]>); Hickman, Hubert B; Karthik Tumu; Munns, Michael B Subject: Re: Does i2b2 Ontology support Multiple Inheritance of Concept ? I think the below statement somehow answers my question! The problem is that there are some concepts in OMOP vocabulary for example that have multiple parents, and multiple children that in their own turn, have multiple parents and this recursively happens several levels deep. I can get you a list of those examples later but the situation is more severe on medications where drugs are characterized under several classification schemes and they all have their child concepts that are in turn classified under multiple classification schemes. Since OMOP ontology is basically a thesauri of several sources, chances are a medication and its children inherits classifications that originate from SNOMED, NDF, RXNorm, etc and essentially forms an acyclic graph and not necessarily a hierarchy. Therefore a crude algorithm to compute all valid paths to the leaf nodes creates too many unnecessary nodes, and we don't have resources or processes that can create/assign a specific map/path only to those leaf nodes of interest rather than computing all paths... The problem is not as big in SNOMED only or ICD9 only representations. With that said, did I understand correctly that you are saying ICD9 and SNOMEDCT are already mapped to i2b2 ontology? Is it explicitly represented somewhere or some level of parsing of paths/concept codes is required to extract it? Best regards Parsa From: <Phillips>, "Lori C." <[email protected]<mailto:[email protected]>> ... We don't merge in branches of trees intact. Each term is handled individually and placed as a leaf. ... From: <Phillips>, "Lori C." <[email protected]<mailto:[email protected]>> Date: Tuesday, July 29, 2014 at 10:14 PM To: Parsa Mirhaji <[email protected]<mailto:[email protected]>> Cc: Hong Luo <[email protected]<mailto:[email protected]>>, "Murphy, Shawn N." <[email protected]<mailto:[email protected]>>, "Campbell, James R" <[email protected]<mailto:[email protected]>>, "i2b2 Academic Forum ([email protected]<mailto:[email protected]>)" <[email protected]<mailto:[email protected]>>, "Hickman, Hubert B" <[email protected]<mailto:[email protected]>>, Karthik Tumu <[email protected]<mailto:[email protected]>>, "Munns, Michael B" <[email protected]<mailto:[email protected]>> Subject: Re: Does i2b2 Ontology support Multiple Inheritance of Concept ? Parsa, I realize now that I don't understand the context you are referring to. When you spoke of concept A was it respect to mapping or merging ontologies? ICD9, SNOMED, etc already have their hierarchies, so I guess I am not following with respect to the creation of your concept A and its children. Isn't Concept A already in SNOMED, etc.? If you are describing the merge or mapping of concept A to SNOMED, etc. We don't merge in branches of trees intact. Each term is handled individually and placed as a leaf. In your concept A example, you would map concept A , then map each of its children. The problem is O(n) not exponential. I hope this helps; It might be better if you give me a specific example. Lori On Jul 29, 2014, at 9:31 PM, Parsa Mirhaji wrote: Lori The problem is not just having to have 25 vs 5 records in the ontology cell to reflect on the multiple hierarchy. It exponentially explodes if each of the children would have 5 more subclasses. It will multiply 25 * 5 and in ontologies like SNOMED or OMOP representations that support deep nesting with multiple hierarchy it immediately explodes to a size that is not tractable with ordinary measures... The question is, are there any other workarounds that you know of to enable us to represent those graph-based ontologies in hierarchic model that i2b2 supports? Best Parsa From: <Phillips>, "Lori C." <[email protected]<mailto:[email protected]>> Date: Tuesday, July 29, 2014 at 4:37 PM To: Hong Luo <[email protected]<mailto:[email protected]>>, "Murphy, Shawn N." <[email protected]<mailto:[email protected]>>, "Campbell, James R" <[email protected]<mailto:[email protected]>>, "i2b2 Academic Forum ([email protected]<mailto:[email protected]>)" <[email protected]<mailto:[email protected]>>, "Hickman, Hubert B" <[email protected]<mailto:[email protected]>>, Karthik Tumu <[email protected]<mailto:[email protected]>>, "Munns, Michael B" <[email protected]<mailto:[email protected]>> Cc: Parsa Mirhaji <[email protected]<mailto:[email protected]>> Subject: RE: Does i2b2 Ontology support Multiple Inheritance of Concept ? Hong, In your example, Concept A and its children would have to be placed 5 times for (25) total records in the ontology table. This tends to occur in medications where something like ibuprofen may show up under Pain reliever and Anti-inflammatory folders. To take my example further, this does not mean that to query for patients taking ibuprofen you have to select both Pain reliever and anti-inflammatory concepts . Only one is necessary; either folder will pick up all patients taking ibuprofen. Lori From: Hong Luo [mailto:[email protected]] Sent: Tuesday, July 29, 2014 4:05 PM To: Murphy, Shawn N.; Campbell, James R; i2b2 Academic Forum ([email protected]<mailto:[email protected]>); Hickman, Hubert B; Karthik Tumu; Munns, Michael B Cc: Parsa Mirhaji <[email protected]<mailto:[email protected]>> ([email protected]<mailto:[email protected]>) Subject: Does i2b2 Ontology support Multiple Inheritance of Concept ? Hi Shawn and All, Can you please let me know if the latest release of i2b2 Ontology support Multiple Inheritance of Concept? For example, if Concept_A has 5 parents, we have to create 5 concept records in Ontology table since their C_fullname or concept_path are different. The challenge is that if concept_A has 5 parents and 5 children, then 5*5 concept records will be created. The number of concepts that needs to be created will increase exponentially if we have many multi-hierarchy concepts in our vocabularies. Your response is greatly appreciated, and will certainly help us implement the logic to transfer data into our Ontology cell. Regards, Hong The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail. ________________________________ The Nebraska Medical Center E-mail Confidentiality Disclaimer The information in this e-mail may be privileged and confidential, intended only for the use of the addressee(s) above. Any unauthorized use or disclosure of this information is prohibited. If you have received this e-mail by mistake, please delete it and immediately contact the sender.
_______________________________________________ Gpc-dev mailing list [email protected] http://listserv.kumc.edu/mailman/listinfo/gpc-dev
