See also: IRC log
<matthias_samwald> people report that they are in this IRC chatroom, but nobody else is there
<matthias_samwald> strange...
mibbit probs
use irc.w3.org intstead
<matthias_samwald> Example of current modelling approach: http://samwald.info/res/pharmacogenomic_CDS_precoordinated_4_test.owl
<matthias_samwald> Example of current modelling approach: http://samwald.info/res/pharmacogenomic_CDS_precoordinated_4_test.owl
<matthias_samwald> http://trowl.eu/
<scribe> scribenick: ericP
matthias_samwald: OWL working out
well
... finding inconsistencies in input data
... sometimes have to e.g. exchange C/G or A/T in the PharmGKB
polymorphisms
ericP: null hypothesis, can you use SPARQL?
matthias_samwald: not trivial to
use SPARQL alone -- started doing that
... because of performance or completeness issues, i may
migrate some of the DL inference to SPARQL rules
... working in OWL, you can continue to use OWL for
design-time
... but for CDS algorithms, implement as SPARQL rules.
... (where you need results in e.g. 5 seconds)
<mscottm> why 5 seconds?
matthias_samwald: CDS messages
are aournd 5s
... there are studies about how physicians use decision support
when e.g. placing order
... results showed that 20s was too late
<mscottm> 1+ to what BobF just said
BobF: given time-scale, it might
not be possible to use real-time reasoning if it takes more
than .5 sec (Mayo cutoff)
... may need to compile overnight
... .. when using the reasoner to go from genotype to called
alleles (the long part of the job)
matthias_samwald: spectrum:
... .. enter raw SNP data and get results AFAP
... .. everything is pre-computed and e.g. physician enters a
medication and gets the pre-compiled inferences
... .
... scenario for the former: in medicine safety codes, raw SNP
data is scannable from a 2D barcode which physician scans and
uses when prescribing
BobP: this might be a business case, where a service compiles SNPs into alleles overnight, available when the bearer of the SNPs shows up in the hospital
matthias_samwald: working with @@1 who's active in Hl7 genomics
<mscottm> IHE = Integrating the Healthcare Enterprise
matthias_samwald: he's working
with a consortium to create an IHE
... the draft from @@1 is heavily focused on HL7 and CDA
... but there is a mention of ontologies, etc.
<mscottm> IHE has indeed focused on XML rather than RDF until now.
<mscottm> Cecil Lynch?
<mscottm> Lloyd MacKenzie??
rafael richards: anaesthesia group is using ontologies
scribe: we still ballot in spreadsheets
<mscottm> Raphael Richards, anesthesiologist, prof at Hopkins, anesthesiology group working on data dictionary (DD?) at HL7
scribe: the idea is that a
patient brings a key with their context
... i'm working on a CCR CCD extension
... (HL7 is trying to tack functionality onto a 35-year-old
broadcast-only protocol)
<mscottm> Richard: 1 to 5 physician groups are the market not big hospitals, with patients moving between them.
<mscottm> ..Big hospitals know how to take care of themselves and run HL7, etc.
matthias_samwald: experience with pharmacogenomics in HL7?
bobF: coming from data
integration for research purposes
... not necessarily for drugs [and CDS]
... in our standard EHRs, we don't have genetics available
<BobF> (bobF isn't speaking right now)
bobF: at the transition of care,
i just want problem lists, medications, history,
allergies
... all pubmed publications show this is where the prob is
<mscottm> Richard: Same problems in transfer all the time: reliable transfer of medication, allergy, problem lists, history, ..
Rafael Richards: stale data is useless in a
hospital where the half-life of data is 24hrs
... almost valueless for decision-making
... we need to inject the word "real-time"
... we need to federate and follow the patient
... big EHRs in big hospitals are growing
... but we need to integrate outside the hospital
... CCR/CCD in RDF would be a great starting point
... want to integrate two small hospitals
... i've been keeping track of the HL7 clinical genomics.
haven't been integrated with the OWL work
connected to HL7, eg chats with Mollie Ullman
scribe: Genetic Test Report Group
has been working with LOINC to code value sets as responses to
"LOINC questions".
... e.g. was the sampled tissue normal or ...
... or "is your predicated enzyme level going to be
{high,medium,low} activity?"
... but no SNPs
... so they have LOINC data about the front end (tissue
samples), and the back end (predicated results), but missing
the middle
<mscottm> So, they are missing the provenance information that would help you evaluate their interpretation (such as the predicted enzyme level).
matthias_samwald: so it would be
good to use LOINC
... ... for e.g. the phenotype
... we don't really care about the test procedures
<mscottm> Richard Raphael works with Conor Dowling and David Booth. (!)
<BobF> bobf needs to drop off - bye all
BobF, tx kindly
all, i'd be happy to get feedback/edits to http://www.w3.org/2012/Talks/0506-egp-EHRs/
<mscottm> http://www.smartplatforms.org/