SNOMED Clinical Terms

Download Report

Transcript SNOMED Clinical Terms

SNOMED Clinical Terms (Snomed CT)
– the language for healthcare
A presentation for the KIDDM Mashup,
17th September 2007
Ian Herbert
Vice Chair
BCS Health Informatics Forum
With grateful acknowledgements to Dr David Gain of NHS CFH
Why do we need Snomed-CT?
To enable consistent representation & retrieval of clinical info:
• about individual patients
• in knowledge sources, e.g. drug formularies & guidelines
To avoid a combinatorial explosion of the terms needed
To provide a flexible set of classifications of terms
Need a terminology that can be extended quickly & indefinitely
This necessary (but may not be sufficient) for:
• analysing patient information, e.g. for mgmnt & research
• automated decision support, e.g. for safe prescribing
• semantic interoperability between care providers & systems □
What is SNOMED CT?
A conceptual
classification
A controlled
clinical
vocabulary
Dictionary
of Clinical
Concepts
SNOMED CT is a terminological resource that can be
implemented in software applications to represent clinically
relevant information reliably and reproducibly □
Why not use ICD10 or OPCS4?
Not not rich enough (or intended) for patient records
No facility to combine expressions to clarify meaning
• ‘emergency’ + ‘thoracotomy’
• ‘recurrent’ + ‘IGTN’ + ‘left’ + ‘great toenail’
Updates too slow (every 10 years for ICD)
SNOMED CT can respond reasonably quickly to
• changes in the wider field of medicine
• changes in local policy
• individual requests for additions
and will never be complete □
What’s wrong with free text?
Free text is an extremely valuable and flexible way of
recording details about individual circumstances, but…
• The meaning may be ambiguous, & open to
misinterpretation
• Its meaning is not available for computation, e.g.
• it can’t automatically be analysed for audit or payment
• it can’t direct care pathways
• it can’t trigger automatic warnings about allergic reactions or
interactions □
How is the information used?
Documentation in
electronic records
Decision
support
Clinical audit
Reporting
Direct
care
Administrative /
management
information
Epidemiology
Research
Summaries
Billing &
reimbursement
Indirect
Care
Aggregation
functionality
Resource
management
One concept, many names
Some of the descriptions associated with ConceptID
22298006:
• Fully Specified Name: Myocardial infarction (disorder)
DescriptionID 751689013
• Preferred term: Myocardial infarction
DescriptionID 37436014
• Synonym: Cardiac infarction
DescriptionID 37442013
• Synonym: Heart attack
DescriptionID 37443015
• Synonym: Infarction of heart
DescriptionID 37441018
Avoiding ambiguity
To a neurologist
Cord compression means Spinal cord compression
To a midwife
Cord compression means Umbilical cord compression
Transmission and sharing of information requires
consistency of terminology – and its use □
Will the computer limit what I can say?
More concepts
• 400,000 health care concepts
More descriptions
• 1,000,000 clinical terms
More information
• 1,500,000 semantic relationships
Contextual modification of expressions
• possible, Family history of, planned, refused,
aborted etc. □
Depth of clinical expression
peripheral angiography
special peripheral angiography procedures
peripheral graft arteriogram
femoral-femoral crossover arteriogram □
How is it organised?
Multiple top level concepts, e.g:
- body structure
Each with a hierarchy of concepts beneath
Strictly organised by ‘IS A’ relationships
- index finger ‘is a kind of’ finger
- finger ‘is a kind of’ hand part, etc
Each concept may have permitted qualifiers, e.g.
- pain ‘has qualifier’ severity □
Hierarchies
Examples
Clinical Finding:
Contains the sub-hierarchies of Finding and Disease
Important for documenting clinical disorders and
examination findings
Finding: Swelling of arm
Disease: Pneumonia
Procedure/intervention:
Concepts that represent the purposeful activities
performed in the provision of health care
Biopsy of lung
Diagnostic endoscopy
Foetal manipulation
Observable entity
Concepts represent a question or procedure which,
when combined with a result, constitute a finding
Gender
Tumour size
Ability to balance
Body structure
Concepts include both normal and abnormal
anatomical structures
Abnormal structures are represented in a subhierarchy as morphologic abnormalities
Lingual thyroid ( body
structure)
Neoplasm (morphologic
abnormality)
Hierarchies
Examples
Organism
Coverage includes animals, fungi, bacteria and plants
Necessary for public health reporting and used in
evidence-based infectious disease protocols
Hepatitis C virus
Streptococcus pyogenes
Acer rubrum (Red maple)
Felis silvestris (Cat)
Substance
Covers a wide range of biological and chemical
substances
Includes foods, nutrients, allergens and materials
Used to record the active chemical constituents of all
drug products
Dust
Oestrogen
Haemoglobin antibody
Methane
Codeine phosphate
Physical object
Concepts include natural and man-made objects
Focus on concepts required for medical injuries
Prosthesis
Artificial organs
Vena cava filter
Colostomy bag
Physical force
Includes motion, friction, electricity, sound, radiation,
thermal forces and air pressure
Other categories are directed at mechanisms of injury
Fire
Gravity
Pressure change
How is it constructed?
Defining and qualifying characteristics used to construct &
refine a terminological model of healthcare
Concepts combined with Attribute-Value pairs
• Procedure with:
- method
– excision
- site
– both tonsils
- using
– laser device
(the post-coordinated representation)
= Bilateral laser tonsillectomy
(the pre-coordinated equivalent & a ‘kind of’ tonsillectomy)
‘Method’, ‘site’ & ‘using’ are defining characteristics An
additional ‘success’ attribute would be a qualifier □
Getting the right Snomed CT term
•
•
•
•
Search for term if you think it’s in there
Search the term hierarchies to find the term
Use a combination of the two
Info. can always be entered in post-coordinated form, q.v.
the bilateral laser tonsillectomy example, but equivalent
pre-coordinated term may be available
• Where system constrains context, a data entry template
can have possible terms in manageable drop-down lists
(including post-coordination qualifiers)
• Automatic encoding of entered text
• highly desirable, but far from reliable at present
• generated codes must be approved by user before commital □
Snomed-specific issues
• Detecting equivalence of same thing said in pre-coordinated &
various post-coordinated representations
• Expressing negation - this comes in many forms, e.g:
• diabetes excluded
• appendectomy not done
• no pain in right leg
• NAD - nothing abnormal detected
• Consistent authoring of the terminology
• Enabling accurate speedy use in unconstrained situations, e.g.
when taking a patient history □
- & non-Snomed-CT specific issues
• Human beings are lazy & good at inference
• So patient records full of short cuts, e.g:
• BP 140/80 means ‘blood pressure taken and systolic pressure
observed to be 140 mm hg, and diastolic pressure 80 mm hg’.
Assumed to be of patient whose record it’s in, & taken during
the encounter it lies within
• Computers are pedantic & pernickety. So is Snomed.
It has xx codes for a blood pressure
• Users want biggest bang per keystroke buck, so
unconstrained searching for terms & post-coordination
not popular □
Why is Snomed CT ‘not sufficient’?
•
•
•
•
Snomed CT consists of concepts, i.e. types
Doesn’t deal with numeric values, e.g. weight 70 kg
Doesn’t identify individual objects, e.g. people
So needs to be used within an external syntax to bind
instances of Snomed concepts to their context, e.g.:
•
•
•
•
who it’s about - the subject (typically a patient)
when action / event occurred or observation made
who performed action / made observation
where action / event occurred or observation made □
Where are we now?
•
•
•
•
•
Snomed CT adopted by the NHS
Now in the hands of an independent international body
Adopted by several countries, more coming
Has no significant global rivals
But not much practical experience in patient record
keeping with it yet, virtually none in real-time □
Are we winning?
“We will know we have succeeded when clinical
terminologies in software are used and re-used,
and when multiple independently developed medical
records, decision support, and clinical information
retrieval systems sharing the same information using
the same terminology are in routine use.”
Alan Rector 2000
“Clinical Terminology: Why is it so hard?”