eLabs: Interoperable data, methods and people enabling populationbased research

1
e-Labs Interoperabledata, methods and
peopleenabling population-based research

• NCRI caBIG Informatics Conference, 13th July
  2007
• Iain Buchan (Director)
• Northwest Institute for BioHealth Informatics
• www.nibhi.org.uk

2
Current position

• Hard to access basic clinical data
• Research fails to harness care data
• Artificially slow expensive research
• Clinical creativity underused

3
Example of missing basics

• Hypothesis Tamoxifen ineffective due to
  interactions at CyP-26D,e.g. with paroxetine
• Request to cancer registry ?
• Breast cancer recurrence data missing
• Adjuvant therapy data patchy
• The data are in GP systems

4
Missing evidence

• Time-course
• Early natural histories
• Emergent risks
• Big picture
• True clinical effectiveness
• What-if/scenario planning

5
Treatment response evidence
Selection
Risks
Trial Population
Patient Population
GeneralPopulation
randomisation
loss
True response

• Predictors of treatment response
• Fixed risks
• Lifestyle factors
• Co-morbidities
• Co-treatments
• Dose delivery
• Treatment setting

Efficacy Safety
6
Givens for NHS data

• Increase in quantity structure of data
  (biological clinical population)
• Complex errors (miscoding, gaming, change of
  assay etc.) ? need meta-data
• Overloaded with irrelevant information cant
  find timely, relevant information

7
Opportunity of local collections

• There is a trend towards local clustering of
  academic NHS expertise around
• tissue banks
• omic facilities
• disease registers

8
Isolated research or care RD federation?
Evaluation
NHS
Extract
Research
Or
ResearchNetworks
E-Lab
Local NHS
Research
E-Lab
Local NHS
Local NHS
E-Lab
Service development
9
Building the Informatics Capacity

• The Northwest Institute for Bio-Health
  Informatics (NIBHI) was formed in 2004 to build
  informatics capacity to leverage the discovery
  potential of bio-health data
• The stakeholders are

10
What is Bio-Health Informatics?
inputs
intellectual space
outputs
Statistics

• Knowledge
• mechanisms
• interventions
• policies

Health
Computational Thinking
Data Information
omics
11
NIBHI objectives

• Create a critical mass of trans-disciplinary
  informatics expertise
• Build the core e-Labs programme between the NHS
  and academia
• Deliver proof of principle outputs
• Findings from novel uses of routine data
• New software and analytical methods

12
Discovery themes and platforms
Obesity
Inflammation lipids
Drug safety
New Cancer
Psychosis
Apply
Population-based E-Labs E-Epidemiology
Biostatistics Translational Mathematics
Engineer
Social BioHealth Informatics
13
Social Bio-Health e-Lab
Research
Service Development
E-Lab
Health Social Care Knowledge
Industry Partners
Academia Partners
Social Contract Governances
Local Information Systems
Population
14
e-Lab Processes
Research
Raw Data
Enhanced Data
Safety
GPs
Link records
Real effectiveness
Hospital(s)

Clean
Efficient trials
Biobanks
Organise
Exploratory
Surveys
Add metadata
LA/Council
Public health
Share algorithms
National
Service models
Depersonalise
15
Integrating Records ? Care in Salford
EMIS Vision
GP
Master Patient Index
Demographic Service
GP
XML (HL7v3)
Integrated Database (at PCT)
GP
Research Clinical Audit
XML Journal File
Optometrist eye screening Community nurses Podiatr
y

Pseudonymised Repository
Web Server
Analysis Tools
Web view of Patient Record
Web Forms
16
Chronicles of care
7
150
148
146
6
144
Total Cholesterol
Systolic BP
142
5
140
138
4
136
1993
1995
1997
1999
2001
2003
2005
2007
17
New diagnoses of type 2 diabetes
1600
1400
1200
1000
800
600
400
200
0
1997
1999
2001
2003
2005
18
(No Transcript)
19
Research Examples

• A) Iterative shaping of hypothesis between
  biology and medicine,plus enhanced data
  collection

20
African cows to Salford ICU
High cholesterol in African cattle identified as
a protective factor against death from
trypanosomiasis
Is high cholesterol a protective factor in humans
undergoing extreme inflammation?
ICU data and physicians in Salford E-Lab
accessible,and physiological clamping reduces
confounding
Data cleaning, meta-data capture, analysis
21
(No Transcript)
22
(No Transcript)
23
Cholesterol inflammation

• Discovery theme growing
• Shaping and testing hypotheses around causal vs.
  reverse causal models
• Speculative trials in mouse model
• Metabolomic proposal submitted
• Enhanced data collection and coding because it
  was of BENEFIT TO CARE data quality by-product
  of research

24
Example 2) Discovery from routine datain a long
seriesSurveillance of obesity in Wirral
children whilst the epidemic took hold
25
Child Deprivation (2001 Census)
Fifths of IDAC 2004 Proportion of households with
childrenclaiming benefits Red (light) most
deprived Red (dark) Purple Blue (dark) Blue
(light) most affluent
26
Adiposity of 3 yr olds 1988 - 1989
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
27
Adiposity of 3 yr olds 1990 - 1991
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
28
Adiposity of 3 yr olds 1992 - 1993
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
29
Adiposity of 3 yr olds 1994 - 1995
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
30
Adiposity of 3 yr olds 1996 - 1997
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
31
Adiposity of 3 yr olds 1998 - 1999
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
32
Adiposity of 3 yr olds 2000 2001
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
33
Adiposity of 3 yr olds 2002 - 2003
Fifths of adiposity SDS BMI fifth Red (light)
fattest Red (dark) Purple Blue (dark) Blue
(light) thinnest
34
Taller, faster-growing children have carried more
of the obesity epidemic
35
Faster infant growth and slower early child growth
6 weeks to 8 months
8 months to 2 years
17
16.5
16
15.5
Mean (95 CI) length/height gain (cm)
15
14.5
14
13.5
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
Birth Year
36
Obesity and linear growth

• Direct public health / policy use
• New quest to identify the links between growth
  and obesity - ? targets within individuals or
  just shared environment
• Mouse equivalent of tall-fat child phenomenon
  identified with knock out of gene on putative
  pathway - ? target

37
Discipline-based E-Labs

• e.g. PsyGrid

38
(No Transcript)
39
PsyGrid

• Aim to e-enable the worlds largest cohort
  study of first-episode psychosis
• 2 years in
• Cohort study running well
• Two e-enabled trials mounted on cohort
• Extreme care with security
• Focus on rapid, systematic data capture
• Adopted by Mental Health Research Network
• Translating to other networks

40
LinkageData-dataResearcher-dataResearcher-comp
utation

• Engineer a factory,not just a warehouse...

41
Parallelization for Genome-wide Screening of
Common Complex Human Diseases
University of Manchester (NIBHI, ARC), Microsoft
Technology Centre (Thames Valley, UK) Melandra
Ltd.
The scientific problem
The solution
42
Shared Genomics with Microsoft

• Proof of concept goals achieved
• enabled more powerful SNP interaction studies
• made the research tools more accessible
• Speed
• Test set 100K SNP, 600 samples, 10K permutations
• 1 way
• PLINK 4-5hrs NIBHI 65 sec
• 2 way
• PLINK 38hrs NIBHI 11 mins
• Full project from Oct 2007
• Web-delivered genetic epidemiology tools
• Fundamental mathematical and software work
• Employ more powerful computers

43
Future Big, Fast Exploration
Problem Question (Hypothesis)
Factors
Model
Population Individual Molecular
Drill down
Findings
Mine
44
Research Object
Question(s)
Protocol
Publications
Data/Material/Pointers
Reflections
Results
Analysis
Making research more transparent, sharable,
reproducible, and easily preserved
45
Clear Public Good
Unclear Public Good
Research Objects
Research
Depersonalise
e-Lab Population
Health Records
Health Records
Local Ownership
Asset Enrichment
46
E-Lab Local Sustainability
Research Development

• Standards
• Governance
• Technical
• Easy, open tools
• Local use ownership
• Compelling findings

Care e-Records
Citizen-driven e-Health
Interoperable local e-Labs confederation
47
Northwest e-Lab Confederation
Technology Experience Local ownership
Federatede-LABs
Regional adoption 3.5-5.0M
e-LAB
e-Lab model adopted within key NHS Trusts 1.5M

• ETHOS

Salford e-Lab pilot 230k
0 1 3

Time (years)
48
Learning from e-Lab pilots

• RD pipelines from/for healthcare
• Data
• Local expertise ? meta-data
• Easy collaborative research platforms
• Federate local ownership enrichment

49
Cancer e-Lab recommendations

• Co-invest to create a critical mass of Social
  Bio-Health Informatics
• Embed spokes of e-Labs in cancer centres
• Set cancer interoperability tests for wider
  health intelligence systems dont isolate
  cancer

50
Future cancer intelligence tests

• Tamoxifen question answered within 3h
• Kaplan-Meier update within 14d of event
• Routine pre-diagnosis signals detected
• Phase IV natural control groups identified
• Real-time feasibility analysis
• Real-time workflow-based recruitment
• Long-term content-feeds on trial participants
• gt90 of tissue samples registered in e-Lab
• ...

51
Thanks for your attention

• www.nibhi.org.uk