SLIDE 1
Background The Model Catalogue Future Work
Outline
◮ Clinical Data for Research ◮ Model Catalogue, Case studies ◮ Future work, Conclusions
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 1/21
Domain Specific Modelling for Clinical Research Jim Davies, Jeremy - - PowerPoint PPT Presentation
Domain Specific Modelling for Clinical Research Jim Davies, Jeremy - - PowerPoint PPT Presentation
Domain Specific Modelling for Clinical Research Jim Davies, Jeremy Gibbons, Adam Milward, David Milward, Seyyed Shah , Monika Solanki, James Welch Department of Computer Science The University of Oxford October 2015 Background The Model
SLIDE 2
SLIDE 3
Background The Model Catalogue Future Work
Clinical Research Data: motivation for DSLs
◮ What motivates our work:
◮ Interoperability ◮ Meta-analysis ◮ Standards ◮ Portability of datasets ◮ Capturing meaning Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 2/21
SLIDE 4
Background The Model Catalogue Future Work
Clinical Data : Producers and Consumers
◮ Not just clinicians and patients ◮ Data Managers ◮ Programmers ◮ Researchers ◮ Pharmaceutical companies ◮ NHS/Department for Health divisions ◮ Auditing
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 3/21
SLIDE 5
Background The Model Catalogue Future Work
Clinical Data for Research : meta analyses
‘Analyse findings from a collection of previous studies, as a whole, to find patterns or results not always obvious from the individual studies.’
- 1. Formulate hypotheses
- 2. Identify candidate studies
- 3. Set incorporation criteria (quality, data availability,
reproducibility etc)
◮ Determine interoperability of variables
- 4. Select dependent variables or summary measures tone included
- 5. Selection of a statistical model
- 6. Carry out analyses, determine correlation, avoid pub bias
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 4/21
SLIDE 6
Background The Model Catalogue Future Work
Hypothetical data experiment
“histological type of tumour” is an common data point collected in breast cancer studies.
◮ determine whether two data sets using this data point are
compatible
◮ share the data point from one organisation to the next
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 5/21
SLIDE 7
Background The Model Catalogue Future Work
Meta data structural interoperability
Two clinical trials collect the data point “histological type of tumour”
◮ Study one:
in-situ ductal only | tubular/cribriform | ductal grade unknown | mixed
◮ Study two:
invasive ductal or no specific type | tubular | mucinous | invasive cribriform
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 6/21
SLIDE 8
Background The Model Catalogue Future Work
Meta data pathway: context of data collection
E3 E3 GP Referral A & E Step 3 USS/CT, Biopsy,Cytology, CA125 if required Step 5 Suspect Cancer? Yes Step 6 End No More tests required or review in x months Could be in primary or secondary care Step 36 Refer to other MDT including Palliative Step 36 Refer to other MDT including Palliative Step 9 Treatment recommendations Step 9 Treatment recommendations Step 11 Primary or Neoadjuvant Neoadjuvan t Primary Step 10 Discuss with patient Curative Step 37 Get Pre-Treatment Samples Step 37 Get Pre-Treatment Samples Step 35 Pre-Treatment tests (Scan,CA125) Step 35 Pre-Treatment tests (Scan,CA125) Step 38 Pre-Treatment Assessment Step 38 Pre-Treatment Assessment Step 39 Clinic appointment (GONC) OPA Step 39 Clinic appointment (GONC) OPA Step 15 Get Pre-Cycle Samples Step 15 Get Pre-Cycle Samples Step 18 On-Treatment Clinic Response Assessment Step 18 On-Treatment Clinic Response Assessment Step 16 Chemo Cycle Step 16 Chemo Cycle Yes Yes Respons e x3 additional cycles Step 26 Get Surgery Samples Step 26 Get Surgery Samples Step 40 Primary Surgery Step 40 Primary Surgery Step 42 Get Surgery Samples Step 42 Get Surgery Samples Step 43 Pathology Diagnosis Step 43 Pathology Diagnosis Step 45 End of Treatment @ GONC Step 45 End of Treatment @ GONC Step 44 MDT Review (Pathology Diagnosis/ Staging) Step 44 MDT Review (Pathology Diagnosis/ Staging) Step 41 Chemo Required? No Yes Adjuvant No Other MDT Step 1 Initial Contact Step 1 Initial Contact Step 46 USS/CT, Biopsy,Cytology, CA125 if required Step 47 Test Results Step 47 Test Results Step 48 Confirm Diagnosis and Site ID Step 48 Confirm Diagnosis and Site ID Emergency Surgery Step 21 Neoadjuvant or Adjuvant? No Response Adjuvant x6 cycles Neoadjuvan t x3 cycles No Relapse 1 2 1 2 Step 33 End Discharg e Completion Chemo x2 additional cycles Step 17 Complete Cycle? No Death No Treatment Step 2 Initial Contact Step 2 Initial Contact E5 E2 E2 E1 Line 1 Line 2 Line 3 E6 E6 E7 E7 E8 E8 E9 E9 E1 E1 E1 3 E1 3 E1 2 E1 2 E11,E14,E1 6,E17,E19, E24,E26,E2 7,E29,E31, E41,E43,E4 5,E46 E11,E14,E1 6,E17,E19, E24,E26,E2 7,E29,E31, E41,E43,E4 5,E46 E1 5 E1 5 E2 1 E2 E2 E2 2 E2 3 E2 5 E2 8 E3 E3 E1 8 E3 6 E3 3 E3 9 E3 7 E3 4 E3 8 E4 E4 2 E4 4 E4 4 E4 7 E4 9 E4 9 E4 8 E5 1 E5 2 E5 E5 4 E5 4 E5 5 E5 7 E58,E61,E62,E6 4, E66,E68,E69,E7 0, E72,E73,E76 E5 9 E6 E6 3 E6 5 E6 7 E7 1 E7 4 E7 8 E7 7 E8 1 E8 2 E8 E10 9 E7 5 Yes Treatment for Relapse + Line E10 4 E10 6 E10 8 E11 E11 2 E11 3 E11 4 E11 6 E11 8 E12 Step 4 Test Results Step 4 Test Results Step 7 MDT Review (Pre-Treatment Diagnosis) Step 7 MDT Review (Pre-Treatment Diagnosis) Step 8 Confirm MDT Diagnosis and Site ID Step 8 Confirm MDT Diagnosis and Site ID Step 12 Pre-Treatment tests (Scan,CA125) Step 12 Pre-Treatment tests (Scan,CA125) Step 13 Pre-Treatment Assessment @ Gynmo/HEC Step 13 Pre-Treatment Assessment @ Gynmo/HEC Step 22 CT Response after #6/Completion Chemo? Step 22 CT Response after #6/Completion Chemo? Step 34 End Step 25 IDS Step 25 IDS Step 27 Pathology diagnosis Step 27 Pathology diagnosis Step 28 MDT Review (Definitive Diag/ Staging) Step 28 MDT Review (Definitive Diag/ Staging) Step 20 CT Response after #3 at MDT? Step 20 CT Response after #3 at MDT? Step 29 End of Treatment @ Gynmo Step 29 End of Treatment @ Gynmo Step 30 Off-treatment Follow up/- Mainten. Therapy
Step 30 Off-treatment Follow up/
- Mainten. Therapy
Step 31 Recurrence of Disease? Step 32 Hospital Admission Step 32 Hospital Admission Step 23 MDT Review (Response) Step 23 MDT Review (Response) Step 24 Clinic appointment @ GONC Unit Referral E83,E88,E90,E92, E93,E96,E98,E100,E10 3,E105,E107,E108,E110 Step 14 Pre-Cycle tests Step 14 Pre-Cycle tests
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 7/21
SLIDE 9
Background The Model Catalogue Future Work
Expected benefits for Clinical Data researchers
◮ Compatible/comparable datasets ◮ Combine datasets ◮ Interlinking separate metadata sets ◮ Traceability/provenance links ◮ Data/metadata sharing across departments ◮ ‘Grok’: Comprehension/communication
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 8/21
SLIDE 10
Background The Model Catalogue Future Work
ISO11179: Metadata Registries
◮ Design for metadata registries ◮ Registration, versioning, publishing, user roles ◮ Generic, structure-less ◮ Assumes a single conceptual model
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 9/21
SLIDE 11
Background The Model Catalogue Future Work
Model Catalogue: Data Modelling Language
Model Type Element Class EnumValue Reference Primitive Enumeration Contents newVersionOf refines contains 1 1
* *
0..1
* * * * * * * ◮ ISO11179-like - registration versioning, publishing support ◮ GRAILS MVC implementation ◮ Concept domain, value domain tagging ◮ ‘Discourse’ plugin for collaborative development
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 10/21
SLIDE 12
Background The Model Catalogue Future Work
Model Catalogue Modelling
Form Implementation Language Form Model Language Data Model Language Model Metalanguage Form Implementation Form Model Data Model instanceOf implements refines instanceOf instanceOf instanceOf instanceOf Form Implementation Language Form Model Language Data Model Language Form Implementation Form Model Data Model instanceOf implements refines instanceOf instanceOf instanceOf instanceOf
◮ Common meta-language vs. model refinement ◮ Parameterised transforms vs derived/composed(?) models ◮ Languages and tools limit implementation
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 11/21
SLIDE 13
Background The Model Catalogue Future Work
Case study I: Overview
◮ Mapping of metadata from clinical information systems
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 12/21
SLIDE 14
Background The Model Catalogue Future Work
Case study I: Reuse
◮ Common measurement units, automatic reuse ◮ Factorising models
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 13/21
SLIDE 15
Background The Model Catalogue Future Work
Case study I: Reporting
◮ Producing reports from data models
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 14/21
SLIDE 16
Background The Model Catalogue Future Work
Case study I: lessons learnt
◮ Metadata was an afterthought in data reuse/sharing ◮ Pathway context large affect on meaning ◮ Spreadsheets, XML, Models
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 15/21
SLIDE 17
Background The Model Catalogue Future Work
Case study II: Overview
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 16/21
SLIDE 18
Background The Model Catalogue Future Work
Case study II: Search Tools
◮ Lookup for metadata
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 17/21
SLIDE 19
Background The Model Catalogue Future Work
Case study II: Collecting Metadata
◮ Metadata sharing mechanisms
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 18/21
SLIDE 20
Background The Model Catalogue Future Work
Case study II: lessons learnt
◮ Single authoritative models cumbersome ◮ Separate models with interlinks ◮ Complexity of metadata models (genome sequencing)
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 19/21
SLIDE 21
Background The Model Catalogue Future Work
Future work: ALIGNED
ALIGNED:
◮ Alignment between data and software engineering ◮ Seshat: Researcher-sourced Archaeological and
Anthropological data
◮ Centralised model with two perspectives ◮ Co-evolution of data and schemas ◮ Generation of data manipulation artefacts from schemas ◮ Generalise model catalogue approach ◮ Different domains Jurion / DBpedia / Poolparty ◮ Model Driven implementation of Model Catalogue
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 20/21
SLIDE 22
Background The Model Catalogue Future Work
Clinical Research Data: lessons learnt
◮ Flat data dictionaries are inadequate ◮ Catalogues of models must be inter-linked ◮ Domain experts must be able to use the models ◮ Data models are constantly evolving artefacts ◮ Automation is key to taming complexity
Jim Davies et. al. Domain Specific Modelling for Clinical Research DSM2015 21/21