Prescribing Decision Support using SNOMED CT

1
Prescribing Decision Support using SNOMED CT

• Implementing SNOMED CT Conference
• Presentation by Malcolm Duncan, RSM 30th March
  2006

2
FDBE Core Product

• Multilex Drug Data File (Multilex DDF)
• Electronic Formulary including
• Clinical data supporting prescribing and
  dispensing
• Legal and reimbursement data
• Knowledge bases Decision support including
• Drug-drug interactions
• Active ingredient sensitivity checking
• Condition checking (proprietary and Read code
  driven)
• Codified side effects, uses, warnings etc

3
FDBE Strategy

• To provide decision support for EMR systems using
  SNOMED CT for clinical coding and the NHS
  dictionary of medicines and devices (dmd) for
  drug concepts
• Drug sensitivity checking (allergies)
• Condition checking (contraindications)
• Contraindications
• Precautions
• Future
• Side effects
• Indications

4
FAQ

• Doesnt SNOMED do decision support?

5
FAQ

• Why not create the FDB knowledge bases using a
  subset of SNOMED CT concepts directly?

6
A Mapped Solution?
7
Map Mismatch Grade 1
8
Map Mismatch Grade 2
9
Map Mismatch Grade 3
10
it depends what your definition of is is

• Bill Clinton

11
Same Time... Same Patient Same Disorder

• Gram positive diplococci
• Left lower lobe consolidation
• Bacteremia
• Community acquired pneumonia
• Penicillin sensitive streptococcus
• Medium dependency patient

12
Reasons for a Bespoke Ontology 1

• Represent concepts and their relationships /
  attributes useful to the medication decision
  support domain
• Contraindications
• Precautions
• Sensitivity checking allergies, adversogens
• Side effects
• Indications

13
Reasons for a Bespoke Ontology 2

• FDB already have the ontologies!
• FDB control their data model, hierarchy
  stability, concept addition and retirement etc
• Intellectual property / Licensing issues
• Need for a buffer layer.

14
Reasons for a Bespoke Ontology 3

• Terminology/classification independence
• ICD-9 (USA)
• Read Version 2 (UK)
• Others!

15
Reasons for a Bespoke Ontology 4

• SNOMED CT extension independence
• UK clinical
• dmd
• etc!

16
Enter DXIDs

• A bespoke ontology to represent medical
  diagnoses, disease states, and health-related
  conditions or procedures pertinent to medication
  decision support

17
DXIDs

• nlt4000 ?
• Semantic neighbourhoods can be used to represent
  useful non-is_a relationships
• Semantic neighbourhoods are defined separately
  for indications, side effects and
  contra-indications

18
Semantic Neighbourhood
19
FDBE - SNOMED Technical Solution

• SNOMED CT maps to DXID
• Related DXIDs recruited from semantic
  neighbourhood
• DXID linked to medications
• Not that simple ?

20
The Map Interface Fixing / Adapting to SNOMED
CT?

• Using SNOMED in the drug decision support world
• Drug decision support engines need their own
  ontological relationships.
• SNOMED has not been designed for this task
• FDBE harness the rich clinical content of
  SNOMED-CT to provide the interface to drug
  decision support
• Working around current problems
• Simple errors
• Deeper representational / ontological issues

21
Adapting to SNOMED

• Relating FBDE drug knowledge to SNOMED Concepts
• Creating decision support subsets
• Deciding on the meaning of terms in a drug
  decision support context e.g. Cardiovascular
  disorder

22
Simple errors

• Duplicate concepts
• False synonyms
• Type 1 hierarchy errors
• Type 2 hierarchy errors

23
Duplicate concepts

• Adrenaline
• 28372575003 Adrenaline (substance)
• 28387362001 Epinephrine (substance)
• Glycerol
• 80582002 Glycerol (substance)
• 259474004 Glycerol analyte (substance)
• 387330004 Glycerin (substance)

24
False synonymy

• glacial acetic acid dichloroacetic acid
• 111142004 dichloroacetic acid (substance)
• 111142004 dichloroacetic acid
• 111142004 glacial acetic acid

25
Adapting to SNOMED CT?

• Machine and human readability?
• Expressivity and cohesion?
• Data standards versus over-granularity
• The non-disjoint sibling issue
• The non-enumerated sibling issue
• The hidden assumption
• Trouble with groupers

26
A SNOMED CT Hierarchy
27
Type 1 (False Positive) Hierarchy Error
28
Fixing a Type 1 Error
29
Fixing a Type 1 Error (Zoomed In)
30
Type 2 (False Negative) Hierarchy Error
31
Asthma

• Asthma
• Acute asthmaChronic asthma
• Mild asthmaModerate asthmaSevere asthma

32
Asthma The Expectation

• Parent
• Asthma
• Children
• Acute asthma
• Chronic asthma
• Mild asthma
• Moderate asthma
• Severe asthma

33
Asthma SNOMED CT Jan 2006

• asthma
• acute asthma
• chronic asthma
• asthma finding
• mild asthma
• moderate asthma
• severe asthma
• Most proximal common parent is Respiratory
  Finding!

34
Asthma SNOMED CT Jan 2006
35
The Promoted Sibling
36
Fixing Type 2 Errors
37
Fixing Type 2 Errors (Zoomed In)
38
Is this a Type 2 Error?
39
Is Crohns Disease Missing Children?
40
Crohns Disease
41
Would this be an Oxymoron?
42
Type 2 Error?
43
Definitions Standards?

• 6 different SNOMED CT disorder concepts
• acute asthma
• asthma with status asthmaticus
• asthma attack
• exacerbation of asthma
• severe asthma
• asthma - currently active
• Useful or reproducible distinctions?

44
Troubles with groupers.
45
The I know what I mean by Grouper Concepts

• Connective tissue disease
• Cardiovascular disease
• Fully defined in SNOMED CT

46
49601007disorder of cardiovascular
systemsynonym cardiovascular disease
47
105969002disorder of connective tissuesynonym
connective tissue disease
48
Non-disjoint siblings
49
ICD-9
50
The Concept Which Isnt There 1
51
The Concept Which Isnt There 2
52
The Concept Which Isnt there 3
53
The Hidden Assumption?
54
ICD-9
55
10304006Mycteroperca jordani
56
227393006curry leaf
57
Amphetamine
58
Grouper or Molecular Entity?
59
SNOMED CT Supports the Encoding of an EMR BUT

• SNOMED CT does not provide decision support
  knowledge bases out of the tin
• SNOMED CT cannot represent, much less actually
  include, all decision support metadata as
  attribute values etc

60
SNOMED CT Can Assist Decision Support Ontology
Development

• SNOMED CT makes helpful suggestions about
  relationships that might be used to draw
  inferences about concepts
• One has to know which suggestions to ignore!
• The SNOMED CT model will not suggest all (or
  even most) inferences for a given decision
  support scenario

61
Implementing SNOMED-CT Driven Decision Support

• SNOMED-CT can be interfaced to decision support
  ontologies
• This requires
• Sophisticated software tools
• A large investment of clinical knowledge worker
  time
• over and above the maintenance of the decision
  support ontology itself

62
Caveats

• Because something is SNOMED CT encoded, it does
  not automatically follow that decision support
  can run informatively against that artefact
• Implementation Issues
• Constrained value sets
• Provenance
• Currency
• Clinical data standards
• Technical data standards
• A maintained problem list?
• How reusable are clinical data?

63
FDBE Decision Support in the Future

• SNOMED CT post-coordination
• More than equivalence / subsumption checking
• Decision support axis modificationltgtSNOMED CT
  axis modification!
• Disorder has course acute, chronic, unstable,
  non-recurrent, etc
• Sell by date e.g. surgery within last 30 days
• Lab Results
• Beyond the HL7 V3 CD data type