Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Semantic Integration of Patient Data and Quality Indicators based on openEHR Archetypes
1. Semantic Integration of
Patient Data and Quality
Indicators based on
openEHR Archetypes
Kathrin Dentler, Annette ten Teije, Ronald
Cornet and Nicolette de Keizer
1 / 25
2. Patient Data
valuable, but semantic gaps
meaning-based integration
required
=> archetypes!
2 / 25
3. Quality Indicators
• Should be
well-formalised:
executable,
sharable &
comparable
results
• CLIF
• Research
question:
archetypes?
3 / 25
4. Outline
1) CLIF
2) Archetypes
3) Formalisation of indicator
4) “Archetyped” patient data
5) Case study & Lessons learned
6) Conclusions & Future work
4 / 25
5. Background: CLIF – Clinical
Indicator Formalisation Method
• Formalised indicator =
query / queries
• Required: standard
terminology for patient
data
5 / 25
6. 8 Steps of CLIF
1) Encode relevant concepts in terms of a
terminology
2) Define the information model <= standard
3) Formalise temporal constraints
4) Formalise numeric constraints
5) Formalise Boolean constraints
6) Group constraints by Boolean connectors
7) Formalise in- and exclusion criteria
8) Construct the denominator
6 / 25
12. Introducing Archetypes
• Computable specifications of clinical concepts.
• Constraints (e.g. occurrence, cardinality) &
ontological definitions.
• Used to record, exchange and integrate patient
data.
• openEHR archetypes: enthusiastic expert
community; publicly available.
12 / 25
13. Advantages of Archetypes
with respect to Indicators
1) Sharable, defined queries
2) Knowledge-level
3) Reality check
13 / 25
14. Sample Quality Indicator
Numerator: Number of
patients who had 10 or
more lymph nodes
examined after resection
of a primary colon
carcinoma.
Denominator: Number of
patients who had lymph
nodes examined after
resection of a primary
colon carcinoma.
- Exclusion criteria: Previous Reasons for this indicator: Evidence-‐‑based
radiotherapy and recurrent (correct staging leads to beYer outcome),
colon carcinomas requires data from several sources
14 / 25
15. Modelling Quality Indicators in
terms of openEHR Archetypes
1) Terminology <=> information model binding:
diagnosis codes <=> node “Diagnosis” of the
archetype “Diagnosis”
procedure codes <=> node “Procedure” of the
archetype “Procedure undertaken”
2) Inter-archetype relations between bound
concepts.
=> Bindings and relations are the backbone of
indicators (concept-level); used to build queries.
15 / 25
16. Sample Query
Patients with “Primary malignant neoplasm of colon”:
SELECT DISTINCT ?patient WHERE {
?patient a patient:at0000.1_Patient .
?patient schemarm:links ?diagnosis .
?diagnosis a diagnosis:at0000.1_Diagnosis .
?diagnosis schemarm:value_element ?diagcode.
?diagcode a diagnosis:at0002.1_Diagnosis .
?diagcode a sct:SCT_93761005 .
} ORDER BY ?patient
16 / 25
17. Patient Data
DWH
Entities
Codes
Mapped To
Patient
1,672,104
Diagnosis
2,925,156
ICD-‐‑9-‐‑CM
SNOMED CT
(ca. 50%)
(via crossmap)
Operation
144,860
Dutch SNOMED CT
classification
(manually, subset)
Admission
259,005
Pathology 92,870
-‐‑ (Dutch free text)
Reports
• DSCA dataset: e.g. radiotherapy & number of examined lymph
nodes.
• Matched based on based on sex, year of birth, operation,
discharge date and procedures => 192/229 patients.
17 / 25
18. Mapping between local Data
Structure and Archetypes
Table
Column
Archetype
Node
Patient
Identifier
Patient
Name
Admission
Admission Date
Patient Admission
Admission Date
Discharge Date
Discharge Date
Diagnosis
Code
Diagnosis
Diagnosis
Operation
Code
Procedure undertaken
Procedure
DSCA
Radiotherapy
Procedure undertaken
Procedure:
fixed SCT code
Multidisciplinary Procedure undertaken
Procedure:
meeting
fixed SCT code
Pathology
Procedure undertaken
Procedure:
fixed SCT code
Number of exam. Tumour-‐‑ Lymph node Number of nodes
lymph nodes
metastases
examined
18 / 25
19. Archetypes &
Patient Data in OWL 2
• Re-used archetype ontologizer.
• Transformed patient data into OWL based on
mapping.
• Loaded closure of SNOMED CT, archetypes &
patient data into OWLIM-SE 5.0
19 / 25
20. Sample Patient Graph
ihtsdo:SCT_50774009 procedure:at0002_Procedure ihtsdo:SCT_284427004
type type exactly_1 type type
data:SCT_50774009 procedure:at0000_Procedure_undertaken data:SCT_284427004
rm:DV_DATE_TIME
value_element type type value_element
type type
data:procedureTime_132_50774009 data:examinationTime_132
time time
data:procedure_132_50774009 hasTime hasTime data:lymphnodeexamination_132
2010_05_26T00:00:00
2010_05_27T00:00:00
links
links links
links
data:patient132
links
data:diagnosis_132_93761005
type 12 data:metastases_132
type value_element
patient:at0000.1_Patient hasNumber items
diagnosis:at0000.1_Diagnosis
data:SCT_93761005 data:nodeNumber_132
type
type
type
exactly_1 type ln_metastases:at0001_Number_of_nodes_examined
ihtsdo:SCT_93761005 max_1
diagnosis:at0002.1_Diagnosis ln_metastases:at0000_Tumour-_Lymph_node_metastases
20 / 25
21. Proof of Concept:
Calculating the Indicators
Indicator / Our Result
DSCA
Publicly Reported
Results
Lymph nodes
85,71% (42/49)
80,00% (43/54)
-‐‑
Meeting
91,66% (22/24)
100% (21/21)
-‐‑
Re-‐‑operation
1,66% (1/60)
9% (7/75)
8,33% (20/240)
One of the problems (meeting indicator):
DSCA: Colon sigmoideum <=> DWH: “Malignant neoplasm of
rectosigmoid junction” mapped to both colon and rectum via
crossmap…
21 / 25
22. Lessons Learned from Case Study
• High coverage of Clinical Knowledge Manager;
extending an archetype straightforward
• Intuitive mapping/modelling at knowledge-level
• Archetype Ontologizer useful, OWL easy to work
with
• Minor difficulties with datatypes; inter-archetype
relationships?
• High data quality required for re-use; problem-
oriented patient model
• UMLS mapping better
22 / 25
23. Conclusions
• Archetypes are suitable to bridge the gap between
clinical quality indicators and patient data.
23 / 25
24. Future Work
• Effect of data quality on reliability/validity of
indicator results
• Sharable queries: Who wants to run these or other
indicators on his/her archetyped data?
• New opportunities for automated reasoning at:
• patient-data level (infer implicit knowledge; validate data
based on archetypes; data-driven, bottom-up data entry),
• archetype-level (infer subsumption and equivalence
relationships between archetypes) and on the
• boundary between both: detect semantically equivalent
constructs!
• And: More bindings required => next presentation!
24 / 25