<Mustafa> OMOP project
Mustafa: OMOP: ICD9/SNOMED
<Mustafa> CrossMap project
Mustafa: Crossmap: (ISO/WHO) ICD10/SNOMED
... ICD[9,10] to MeDRA
... mapping info + hierarchy (e.g. SNOMED-CT)
... e.g. specific MCI codes in SNOMED to generic in MeDRA
... using this for query expansion
... e.g. retriving diabetes patients with a single code for diabetes gets expanded to type I, II, gestational
emory: cognative medical systems (mine) plus subs to caregraph (Connor's) creating as much open source infrastructure for health care as possible
... structural and semantic normalization for enhanced decision support
... using a series of terminology [capabilities?] that can be queries in Jena or CTS2 from Mayo
... CTS2 has ~130 methods, ~40 implemented by bio-ont
... we expanded that to ~57 methods for a demo to VA
... convert REST queries to SPARQL queries against the vocabs in the Jena triplestore
conor: we tried using OWL and autoclassification
... insufficient ROI so we used SKOS
... SKOS takes you from flat enumerations of medications
... e.g. RxNorm can't support much inference
... hierarchical schemes for specialties
... you'll never get codes *just* in SNOMED
... so we wanted SKOS 'cause we didn't wnat an enumeration to look like classification
... once we embraced SKOS, we could go to town on importing
... e.g. RxNorm is a monthly dump
... UMLS is 6 months behind LOINC
... some schemes are amalgams of multiple sources (when necessary)
<conor> www.caregraf.info -- specifically schemes.caregraf.info … the triple store is 4store
conor: SALUS uses Virtuoso, caregraf uses 4store, emory's uses jena
... some relations back to RxNorm are broader, e.g. @@1 which includes packagers
... branded drugs are reducabable to generics in RxNorm
... so we record both, which means you don't need do query expansion
mustafa: most of the type we don't have the expansion
... on the system side, we don't have to targets to match the source system
... e.g. a clinical researcher might use third-level queries and get fourth level responses [scribe didn't understand]
conor: the EHRs' data is much more expansive than the normalized representation
ericP would say "sounds like a bit of forward chaining"
if he could turn off his mute
emory: we should focus on something implementable
... we want the implementation to survive the group
<conor> Disorder/Diagnosis … very crude: cataracts -- making sense of it in SNOMED (this is over a year old) -- http://schemes.caregraf.info/analytics/lc_snomedCataractsClassification.html
emory: we should be really focused, and the area of greatest value to us and apparently you [SALUS] is diagnoses and allergens
... these drive a lot of the decision-making in health care
... is there an op to share ideas, code, mappings?
... can we create a reference that's more up-to-date than bioontologies?
... can we have high enough quality to be used in a production environment
<conor> snomed disorders - facet'ed by specialist ie/ cancer, ophthalmologist ...
color: some doctor's have said that SNOMED is to "busy" for their specialties
... we need to facet this by specialty and see if we can draw in experts
... we can look at the hierarchy and see how much pre/post coordination can be done
... we can look at NCI ontologies to compare detail level
... how much do our tools manage expressiveness and how can we push that further?
<conor> too busy and too little! you can't please everyone. Points to "lite" and "full"? … cleaner anyhow
I'd like something where the maintenance could survive the group
emory: i looked at ontowiki and wikipedia extensions for ontology mappings
... what you do is tag a wikipedia article with a concept code and then map the concept codes
... if the govornence becomes a proprietary product....
... if we can do that in a public way, we potentially give access to countries/intsittutions that don't ahve the $$$
... ericP and I can list the topics that need development
... e.g. what are the rules for ontology development?
... should we adopt using post-coord as much as possible?
conor: when we were dealing with alergens in SNOMED, we said "it's arbitrary" and they said "we'll pull it out"
... there are different levels of detail
emory: can we release code, even ontologies (if you promise that you're a UMLS licensee)?
... maybe not publicly editable
... how do we conduct this in a way that doesn't look like yet another ontology
conor: we can boot with SNOMED-CT disorders (and morphology), maybe MeDRA CT, NCI cancer ontologies
looking at FHIR as ontologies
8PDT/11EDT/18 wednesday