The CLEF Chronicle: Transforming Patient Records into an E-Science Resource Jeremy Rogers, Colin Puleston, Alan Rector James Cunningham, Bill Wheeldin, Jay Kola Bio-Health Informatics Group Department of Computer Science University of Manchester

1
The CLEF Chronicle Transforming Patient Records
into an E-Science Resource Jeremy Rogers, Colin
Puleston, Alan RectorJames Cunningham, Bill
Wheeldin, Jay KolaBio-Health Informatics
GroupDepartment of Computer ScienceUniversity
of Manchester
2
CLEF Clinical E-Science Framework

• Improving the storage and processing of
  Electronic Health Records to enhance general
  clinical care
• Supporting clinical research via the creation of
  a clinical research repository, known as the CLEF
  Chronicle

3
Chronicle Query
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst remaining
in remission for the full extent of this period
THEN
4
Concepts from ExternalKnowledge Sources (EKS)
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst remaining
in remission for the full extent of this period
THEN
5
Properties from ExternalKnowledge Sources (EKS)
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst remaining
in remission for the full extent of this period
THEN
6
Implicit RelationshipsBetween EKS Concepts
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
shin part-of lower-leg part-of leg
Underwent surgical-intervention to remove all
tumours
THEN
mastectomy is-a surgical-intervention
ALSO
Survived for at least ten years whilst remaining
in remission for the full extent of this period
THEN
7
Temporal Information
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst
remaining in remission for the full extent of
this period
THEN
8
ARBITRARY TEMPORAL SEQUENCES
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst
remaining in remission for the full extent of
this period
THEN
9
Temporal Abstractions
WHAT PERCENTAGE OF PATIENTS WHO
Had cancer with stage of stage-2 located
somewhere in the leg with primary tumour that
doubled in size within a 3 month period
FIRST
that doubled in size within a 3 month period
Underwent surgical-intervention to remove all
tumours
THEN
ALSO
Survived for at least ten years whilst
remaining in remission for the full extent of
this period
THEN
whilst remaining in remission for the full
extent of this period
10
Chronicle System Overview
11
(1) Chronicle Representation
1
Chronicle Representation
12
(2) Chronicle Repository Query Engine
1
Chronicle Representation
Query Engine
2
Chronicle Repository
13
(3) Chroniclisation Process
3
1
Chronicle Representation
Chronicliser
Query Engine
Text Processor (Sheffield)
2
Chronicle Repository
EHR Repository (UCL)
14
(4) Chronicle Simulator
3
1
Chronicle Representation
Chronicliser
Chronicle Simulator
Query Engine
Text Processor (Sheffield)
4
2
Chronicle Repository
EHR Repository (UCL)
15
(5) Browser Query GUIs
5
Simple Browser Query Formulator
Query Formulator (Open University)
3
1
Chronicle Representation
Chronicliser
Chronicle Simulator
Query Engine
Text Processor (Sheffield)
4
2
Chronicle Repository
EHR Repository (UCL)
16
ChronicleRepresentation
17
Temporal Representation
SPAN Event
SNAP Event
start point
end point
occurrence point
Time
18
Temporal Representation
Note For the Patient Chronicle the atomic
time-unit equals one-day
hence, for example, Surgical-Operations and
Consultations are SNAP Events
SPAN Event
SNAP Event
end point
start point
occurrence point
Time
19
Temporal Representation
Example X-ray performed on specific day
with associated set of results
SPAN Event
SNAP Event
end point
start point
occurrence point
Time
20
Temporal Representation
Example Period of employment as Plumber,
spanning specific time-period
SPAN Event
SNAP Event
end point
start point
occurrence point
Time
21
Temporal Representation
SNAP
SNAP
SNAP
SNAP
Structured SPAN Event
end point
start point
Time
22
Temporal Representation
with set of snapshots representing same Tumour
at specific time-points
Example History of Tumour over specific
time-period
SNAP
SNAP
SNAP
SNAP
Structured SPAN Event
end point
start point
Time
23
Temporal Representation
whilst SPAN has set of temporal-abstractions
(e.g. max, min, etc.) summarising the
tumour-size attribute
Example cont. Each SNAP has associated value for
tumour-size attribute
SNAP
SNAP
SNAP
SNAP
Structured SPAN Event
end point
start point
Time
24
Chronicle Representation
Chronicle Model Java Object Model
Clinical Model
Generic Model
Clinical Knowledge Service
External Knowledge Sources (EKS) Ontologies,
Databases, etc.
EKS
EKS Related Inference
25
Chronicle Representation
Clinical Model
Generic Model
Chronicle Representation is embedded within a
generic Knowledge Driven Architecture
Clinical Knowledge Service
EKS
EKS Related Inference
26
Generic Model
Generic modelling classes
Clinical Model
Generic Model
Clinical Knowledge Service

• Including
• SNAP/SPAN temporal representation
• Temporal abstraction mechanisms
• EKS-concept handling

EKS
EKS Related Inference
27
Clinical Model
Extends generic model with clinical-specific
classes
Clinical Model
Generic Model
Clinical Knowledge Service
Examples ? SNAPS ProblemSnapshot,
SnapClinicalProcedure, etc. ? SPANS
ProblemHistory, ClinicalRegime, etc.
EKS
EKS Related Inference
28
External Knowledge Sources (EKS)
Clinical Model
Detailed (time-neutral) clinical knowledge sources
Generic Model
Clinical Knowledge Service
Currently Single OWL ontology Possibly Multiple
ontologies, databases, etc.
EKS
EKS Related Inference
29
External Knowledge Sources (EKS)
Clinical Model
Generic Model

• Provide
• ? Hierarchies of concepts
• ? Sets of inter-concept relationships
• ? Sets of instance-descriptor properties attached
  to concepts

Clinical Knowledge Service
EKS
EKS Related Inference
30
EKS-Related Inference
Arbitrarily complex inference mechanisms
Clinical Model
Generic Model
Drive ? Dynamic data creation ? Query formulation
Clinical Knowledge Service
EKS
Currently Description-Logic based
reasoner Possibly Rule-bases, procedural code,
etc.
EKS Related Inference
31
EKS-Related Inference
Clinical Model
Generic Model
Note Full EKS-related inference is neither
appropriate, nor required, for (time-critical)
execution of queries over thousands of patient
chronicles
Clinical Knowledge Service
EKS
EKS Related Inference
32
Clinical Knowledge Service

• Provides transparent access to
• External knowledge sources
• EKS-related inference

Clinical Model
Generic Model
Clinical Knowledge Service
Simple interface Takes Instance of concept X,
including set of descriptor values Returns
Updated descriptor-set for X (including updated
constraints)
EKS
EKS Related Inference
33
Problem-Types
Bodily-Locations
Chronicle Representation Example Representation
of the history of a specific clinical problem as
displayed by a particular patient A problem
is either a pathology (e.g. cancer) or some
manifestation of a pathology (e.g. a specific
tumour)
location
type
Problem History
snapshots
Problem Snapshot
Problem Snapshot
Problem Snapshot
34
Problem-Types
Bodily-Locations
location
type
Problem History
Chronicle Model Objects
snapshots
Problem Snapshot
Problem Snapshot
Problem Snapshot
35
Problem-Types
Bodily-Locations
SPAN Event
location
type
Problem History
SNAP Events
snapshots
Problem Snapshot
Problem Snapshot
Problem Snapshot
36
Problem-Types
Bodily-Locations
External Knowledge Sources (EKS)
location
type
Problem History
snapshots
Problem Snapshot
Problem Snapshot
Problem Snapshot
37
Bodily-Locations
type concept selected from EKS
Tumour
location
type
Problem History
snapshots
Problem Snapshot
Problem Snapshot
Problem Snapshot
38
descriptor variables derived from type concept
Bodily-Locations
Tumour
Integer History
location
type
tumour-size
Problem History
snapshots
Integer Snapshot
Integer Snapshot
Integer Snapshot
tumour-size
Problem Snapshot
Problem Snapshot
Problem Snapshot
39
Values allocated to snapshot descriptors
Bodily-Locations
Tumour
Integer History
location
type
tumour-size
Problem History
snapshots
Integer Snapshot
Integer Snapshot
Integer Snapshot
time-point
4/3/98
tumour-size
value
7
Problem Snapshot
Problem Snapshot
Problem Snapshot
40
History descriptor values derived automatically
Bodily-Locations
Tumour
start-point
4/3/98
Integer History
end-point
7/2/02
location
type
tumour-size
start-value
7
end-value
43
Problem History
minimum
Temporal Abstractions
7
maximum
82
range
75
snapshots
Integer Snapshot
increase-rate
0.051
Integer Snapshot
Integer Snapshot
tumour-size
Problem Snapshot
Problem Snapshot
Problem Snapshot
41
location concept selected from EKS
Breast
Tumour
Integer History
location
type
tumour-size
Problem History
snapshots
Integer Snapshot
Integer Snapshot
Integer Snapshot
tumour-size
Problem Snapshot
Problem Snapshot
Problem Snapshot
42
Additional descriptor variables inferred via
EKS-related reasoning
Breast
Tumour
Integer History
location
type
tumour-size
Problem History
Boolean History
her2-receptor
snapshots
Integer Snapshot
Integer Snapshot
Integer Snapshot
tumour-size
Problem Snapshot
Problem Snapshot
Problem Snapshot
Boolean Snapshot
Boolean Snapshot
Boolean Snapshot
her2-receptor
43
Values allocated/derived for new descriptors
Breast
Tumour
start-point
4/3/98
Integer History
end-point
7/2/02
location
type
tumour-size
start-value
false
end-value
true
Problem History
Boolean History
always-true
false
her2-receptor
always-false
false
percent-true
63.72
snapshots
Integer Snapshot
percent-false
36.28
Integer Snapshot
Integer Snapshot
tumour-size
Problem Snapshot
Problem Snapshot
Problem Snapshot
Boolean Snapshot
Boolean Snapshot
time-point
4/3/98
Boolean Snapshot
her2-receptor
value
false
44
Chronicle RepositoryandQuery Engine
45
Chronicle Query Engine Requirements

• Querying over Large Numbers of patient chronicles
• Basic RDF/RDFS-Style Reasoning, involving
• Hierarchical relationships (is-a)
• Property relationships (part-of, has-location,
  etc.)
• Transitivity
• Temporal Reasoning, including
• Reasoning about temporal sequences
• On-the-fly temporal abstraction

46
Chronicle Repository

• An RDF/RDFS-based repository (currently using
  Sesame RDF-store)
• RDF/RDFS representation to facilitate
• Querying over Large Numbers of patient chronicles
• Basic RDF/RDFS Reasoning (must incorporate
  transitivity)
• Additional Temporal Reasoning mechanisms will be
  required (including on-the-fly temporal
  abstraction)

47
ChroniclisationProcess
48
Electronic Health Records (EHR)

• Document based
• One document per clinical procedure
• Minimally structured
• No inter-concept references
• No inter-document references
• Mainly free-form text
• For human consumption
• Incomplete information
• Many implicit assumptions

49
Chroniclisation

• Complex heuristic process
• Input Largely unstructured EHR data
• Output Highly structured chronicle data
• Process will involve
• Text processing
• Co-reference resolution
• Temporal reference resolution
• Inference of implicit information

50
CLEF Chronicle Summary

• Chronicle Representation
• Temporal Representation
• External Knowledge Sources (OWL, etc.)
• Complex EKS-related reasoning (DL, etc.)
• Chronicle Repository Query Engine
• Querying large numbers of patient records
• Simple EKS-related reasoning (RDF/RDFS)
• Temporal Reasoning
• Chroniclisation Process
• Input Largely unstructured EHR data
• Output Highly structured Chronicle data