Meeting Record biosurveillance meeting: Friday March 30

http://www.w3.org/2012/03/30-hcls-minutes

                            Surveillance follow-up 1

06 Apr 2012

Attendees

   Scribe
          ericP

Contents

     * [2]Topics
     * [3]Summary of Action Items
     __________________________________________________________________

   <scribe> scribe: ericP

   <mscottm2> Background: Msgs came in through msg queueing system, HL7
   2.X msg was parsed, distilled into components necessary for analysis
   and RDF created

   <mscottm2> Eric: <checking> validating reports real-time?

   <mscottm2> who is talking?

   <mscottm2> tx

   <mscottm2> eschrip: coded representation for topic, such as particular
   antibiotic for TB and another - whether there was a response.

   <mscottm2> ericp: how were they tied together?

   <mscottm2> eschrip: each reporting body (State?) created its own
   software

   JoshMandel: did all the states have a common hl7 guideline?

   <mscottm2> JoshMandel (?): When does validation happen, before or after
   RDF?

   <mscottm2> eschrip: we apply some OWL reasoning and also use Jess to
   the msg

   cecyl: initial submission is paper-based
   ... first electronic view is when the state gets a copy of all the
   paper forms

   <mscottm2> ericp: trying to ascertain if the assertion of efficacy is
   tied to the event (msg).

   eschrip: [biosynth demo]
   ... working from collection of different kinds of messages, those that
   appear in a given clinic or lab
   ... system sends them all to a processing point
   ... processing does:
   ... .. categorize events for reported disease, e.g. upper-respiratory
   issues
   ... .. identify syndromes, trends, etc.
   ... reported e.g. when a doctor did a test for anthrax, regardless of
   disposition
   ... parsing clinical data, different 'cause e.g. TB forms are would be
   normalized
   ... called for a rete engine
   ... unified patients based on hospital IDs, no inter-clinic links
   ... we were able to establish liklyhoods of different diseases
   ... rete used to asses symptoms, followup tests, etc.
   ... we classified into Cecil's disease ontology

   ericP: if you were able to push the processing down, is there value to
   reporting in RDF?

   <eschrip> That was Craig

   <eschrip> (Craig dialed in from Salt Lake City)

   cecil: drawing from e.g. prescriptions, facebook, etc. folks were able
   to beat CDC's influenza predictions by two days

   <mscottm2> Google predicted an outbreak in influenza 2 days earlier
   than CDC could do it by looking at search patterns

   eschrip: using RDF for messaging isn't as interesting as RDF for
   querying
   ... biosynth data lifetime is very short
   ... for TB, we keep it around longer to eliminate duplicates
   ... for a Quality of Care project, we kept data for the stay of a
   patient
   ... need validation

   ericP: what if we used SPARQL for validation?

   eschrip: we used OWL, and SPIN a little
   ... we used lots of SPARQL to examine the data, even in OWL format
   ... yes, a story should be told that novel JSON is the same as novel
   RDF for self-discovery requirements

   JoshMandel: there are a couple JSON schema language, and i expect
   validators
   ... in RDF/SPARQL land, you'd have a zillion little validating queries?

   ericP: maybe monolithic

   cecil: how do a write a standard query for population health when
   there's not single query point
   ... GELLO provides a canonical model to which you'd map via d2r, etc.
   ... talking to chris chute, stan huff, ken madel, the SHARP grants are
   trying to create standard APIs

   JoshMandel: at Harvard, we're creating canonical models
   ... has the same flavor of pushing the mapping back to the source
   system

   ericP: when you have combinations of models, can one message meet
   multiple models?

   JoshMandel: yes
   ... we have a payload validator which:
   ... .. is there a code?
   ... .. is it RxNorm?
   ... .. is there a dose?

   cecil: i'm leading the clinical model CIMI task force
   ... FHIR use cases are similar to the narrow use cases
   ... you don't need the monolithic model when you know your target
   ... HL7 has Common Message Element Types for e.g. the lab domain
   ... aggregates e.g. SMART models

   <scribe> scribenick: ericP

   Craig: we have an org which is trying to move into a managed care model
   ... (instead of payment for test types, etc.)
   ... we're building towards being able to do decision support and risk
   analysis

   mscottm2: when you've got finer granularity than state-level, you can
   do interesting mashups of, say, epidemic data
   ... is your data public? can we watch a virus traverse the states?

   eschrip: the format depends on the organization
   ... infectious disease may be different from ...

   Craig: they want to control and vet the data before publication, way
   into the episode of an outbreak
   ... there are many bits and pieces of information which preceed a
   diagnosed case

   <Bosse> I need o leave, thanks for an interesting discussion. /Bosse

   cecil: EpiInfo, if you can demonstrate a need to know of edidemiology
   info
   ... most states have a weekly fact sheet of what's happening, what
   needs attention
   ... even though i get that data, i can't really use it beyond figuring
   out how to use it late

Summary of Action Items

   [End of minutes]
     __________________________________________________________________


    Minutes formatted by David Booth's [4]scribe.perl version 1.136
    ([5]CVS log)
    $Date: 2012/04/06 13:01:22 $

References

   1. http://www.w3.org/
   2. http://www.w3.org/2012/03/30-hcls-minutes#agenda
   3. http://www.w3.org/2012/03/30-hcls-minutes#ActionSummary
   4. http://dev.w3.org/cvsweb/~checkout~/2002/scribe/scribedoc.htm
   5. http://dev.w3.org/cvsweb/2002/scribe/

-- 
-ericP

Received on Friday, 6 April 2012 13:07:51 UTC