3. An ongoing problem…
“In attempting to arrive at the truth, I have applied
everywhere for information but in scarcely an instance have I
been able to obtain hospital records fit for any purpose of
comparison.”
“If they could be obtained, they would enable us to decide
many other questions besides the one alluded to. They would
show subscribers how their money was being spent, what
amount of good was really being done with it or whether the
money was not doing mischief rather than good.”
- Florence Nightingale, 1863
Credit: Heather Leslie
4. Why is health IT so hard?
•The banks are doing it; why can’t health?
•Complex and dynamic domain
•Lifelong records
•Clinical diversity
•Confidentiality
•Mobile population
Credit: Heather Leslie
5. Complexity
•Both the number of concepts and the rate of
change is high
•Health is big, and continually growing…
–In breadth
–In depth
–In complexity
•Clinical knowledge is continually changing
Credit: Heather Leslie
6. How have we been dealing?
•Free text(Specialist and administrative systems have more
structured data, but generic electronic health records
are still mainly free text)
7. So what do we need structure for?
•Avoid repetition and
shadow records
•Retrieval and overview
•Reuse of record info
•Clinical decision support
•Quality indicators
•Management data
8. Longitudinal information access
•How long are you planning to live?
•Do you expect your health record to survive
that long?
•Even if it does survive, will
it be readable for future
systems and users?
Credit: Ricardo Correia
9. Celsius
Aural measurement
IR aural thermometer
Environment: 5° C
Wet clothing Space blanket
temperatureBody
Structuring health is hard
Credit: Bjørn Næss
10. Structuring identically is even harder
Example: Smoking status in
national registers:
• 9 different variations on
“Smoking status” in 26
different forms
• Additionally: number of cigs
per day, month quit smoking,
number of months since
quitting date, etc.
Brandt, Linn (2016). Report from REGmap February 2016 – Complete mapped register set - Preliminary analysis.
11. Semantic interoperability
•A (or The?)
Holy Grail of Health Informatics
•[…] the ability of computer systems to
exchange data with unambiguous,
shared meaning
•Requires (among other things)
shared information models
and terminologies
1
12. «Information model»?
•A definition of the structure and content of the
information that should be collected or shared
– A "minimal dataset"
– A message or interface definition
•Internally all applications have some sort of
information model
•Sharing information requires developing shared
information models
Credit: Ian McNicoll
13. How have we been doing infomodelling?
•Locked into each product
•In ways that clinicians don’t understand
•Few clinicians participating
•Technicians are left to interpret
•New requirements?
15. Semantic interoperability* requires
identical data models
Clinical information modelling is difficult and
expensive, and should be done once
⇒ Information models should be
shared and strictly governed
* Level 4 semantic interoperability; Walker et al. (2005); http://www.ncbi.nlm.nih.gov/pubmed/15659453
16.
17. openEHR
• Specification for structured health
records
• openEHR Foundation (openehr.org)
• Free (as in beer AND speech)
• International community
• Two level modelling
– Reference model
– Content models (archetypes)
& data sets (templates)
18. openEHR reference model
• EHR structure
• Security
• Versioning
• Participants, dates/times,
data types
NO CONTENT
Credit: Heather Leslie
20. Archetypes
• Implementable specification for one clinical concept
• Comprehensible for non-techies
• Maximum datasets (aspirational)
• Reusable
THE STANDARDISED CONTENT
Credit: Heather Leslie
22. Templates
• Combinations of constrained archetypes
• Data sets for forms, messages, interfaces, etc
• For specific use cases
• NOT user interfaces
THE USECASE SPECIFIC CONTENT
Credit: Heather Leslie
24. National governance
•January 2014: Archetype governance started
– Goal: quality and interoperability
– Four people
•Online collaboration
– arketyper.no / wiki.arketyper.no
– 350 registrered users
– Vendors and registries participate
0
50
100
150
200
250
300
350
400
jan.14
apr.14
jul.14
okt.14
jan.15
apr.15
jul.15
okt.15
jan.16
apr.16
Number of registrered users
25. Processes
• Formalised processes for
development, review and
publishing
– Archetype development is
a “do-ocracy”; not centrally
prioritised
– Re-use of international
models
0
5
10
15
20
25
30
35
40
45
jan.14
mar.14
mai.14
jul.14
sep.14
nov.14
jan.15
mar.15
mai.15
jul.15
sep.15
nov.15
jan.16
mar.16
Number of published archetypes
26. Reviews and approval
• National editorial committee defines review requirements
• Review rounds last one week, several rounds per archetype
• Until clinicians reach consensus…
• If requirements are met, the archetype is «Published» on
arketyper.no
• Unpublished archetypes are unstable and should not be used
27. Are information models enough?
•Sure, if we want to make 100k archetypes for
each diagnosis, lab result, symptom, …
•Sure, if we never want to query for all the
patients who had pneumonia caused by viruses
•We need something more: Terminologies
2
28. Things that are always
true:
Terminologies
Vocabularies,
classifications,
ontologies; ICD-10,
SNOMED CT, ICF
Things that are true for a
specific individual:
Information
models
Information structure;
openEHR archetypes, FHIR
resources
Things that are true given a
specific set of conditions:
Inference models
Rules and knowledge bases
used in decision support
and alert systems.
Rector AL, Rogers J, Taweel A. Models and inference methods for clinical systems: a principled approach. Stud
Health Technol Inform. 2004;107(Pt 1):79-83
Some overlap
29. Terminologies
•Controlled vocabularies (NCMP, NCSP)
–Flat lists of coded concepts
•Classifications (ICD-10, ATC)
–Hierarchies
•Ontological thesauri (SNOMED CT, ICNP)
–Polyhierarchies with associated attributes
–Some are combinatorial
3
30. Terminologies vs information models
Information models can be said to
describe the "questions"
Terminologies can give (some of) the
"answers"
Complementary concepts
ICD_10::L40.0::Psoriasis vulgaris
and
SCT_2015::74757004::Skin structure of elbow
SCT_2015::6736007::Moderate
???
31. Where terminologies rule
•(100s of) Thousands of concepts
–Diagnoses, symptoms, lab results, body
structures, organisms, procedures, …
•Inference based
on relations
between concepts
3
33. Context
•"Let’s just chuck the codes in here so we can bill for
this cancer treatment!"
•15 years later, from the brand new Dr. Google:
–"Ma’am, I’m sorry to tell you you have ovarian
cancer."
–"What!? They were taken out 15 years ago!"
3
More: http://e-patients.net/archives/2009/04/imagine-if-someone-had-
been-managing-your-data-and-then-you-looked.html
34. Quantitative data types
•"It’d be really nice to just have a code for the
number of the pregnancy the woman is in…"
•"Yeah. 10 ought to be
enough for anybody."
Famous last words…
3
35. Complex concepts
•Combinatorial explosion
–"Every kind of rash for every skin area"
–Every combination of oral glucose challenge
⇒ the 601 LOINC "glucose" codes:
•Postcoordination may
mitigate, but beware…
3
36. Grey area
•Small value sets
•Some contextual information
–Actual diagnosis vs. tentative vs. risk vs. exclusion
vs. family history
•Consistent use is hard, and not always appropriate
–Different use cases will have different requirements
3
37. Summary
• Clinical information models must be free and shared
• Clinicians must drive clinical modelling
• Terminologies are necessary additions to information models
• Grey areas -> pragmatic choice based on requirements
More info:
• One day seminar/two day course in Oslo May 23/24: http://goo.gl/UW7et5
• Videos of one day seminar in Sweden 2015: http://goo.gl/6Ibbkf