HIRU Health Information Research Unit for Wales

1
(No Transcript)
2
HIRUHealth Information Research Unit for Wales

• David Ford

3

• Centre for Health Information, Research
  Evaluation (CHIRAL)
• Institute of Life Sciences
• School of Medicine
• Swansea University

4
Centre for Health Information Research
Evaluation

• CHIRALs research areas
• Health Services Research
• Clinical Epidemiology Diabetes
• Social and Epidemiological Psychiatry
• Primary care
• Injury and Environment
• Health Information Management
• Mathematical Modelling
• Clinical Research Unit
• Qualitative Research

5
Cross cutting themes

• Patient-based trials
• Trials and evaluations of complex interventions
• Cohort studies
• Advanced methodologies (outcomes, qualitative,
  modelling, etc.)
• Use of routine data for research

6
Development of a health environment
information research platform Health
Information Research Unit

• Background and Methods

7
Simple questions which cannot be easily answered
in Wales

• What is the prevalence of any chronic disease in
  any area of Wales?
• Is disease x increasing or decreasing in Wales?
• What amount of money/resource is spent on
  different disease groups in Wales?
• Is the health of a population being adversely
  affected by living closer to
• If care is redesigned in a particular way what
  will be the likely impact on primary/secondary
  care and different populations?

8
More questions which cannot be easily answered in
Wales
2.

• What is the impact of poverty/socioeconomic
  status/deprivation on the demand for health
  services?
• Is this practices higher/lower referral
  rate/prescribing etc due to a different
  disease/illness burden?
• What additional workload is needed to meet
  national service frameworks (NSFs)/changes in
  contract etc?
• How many patients in Wales would benefit from a
  new treatment supported by NICE and how much
  would this cost?
• How does the physical environment influence
  health ?

9
HIRU Background and methods

• Vast amounts of electronically-stored information
  held in
• the NHS
• local government
• central government
• other organisations. . . . much of it held at
  an individual level
• These data, were they all to be available for
  simultaneous analysis, offer enormous potential
  to conduct and support research

10
The challenge

• Data collected for wide variety of purposes
• Generally only used for original purpose (if at
  all!)
• Data held in large number of organisations
• Held in variety of formats, variation in
  ontologies
• Known and unknown data quality issues
• Data extraction and transportation difficult
• Data volumes enormous
• Data protection / confidentiality hurdle
• Intelligent (thoughtful) analysis essential

11
The future looks good . . .

• National Programmes of NHS IT (Connecting for
  Health and Informing Healthcare in Wales) offer
  potentially excellent datasets for research.
• Current priority is delivering clinical benefits,
  although secondary uses of data is being
  considered.
• Still some time until research uses of data will
  be fully realised
• Full breadth of non-NHS data sources unlikely to
  be in scope in near future

12
Moving forward . . .

• Health Information Research Unit (HIRU) funded by
  research centre grant by Wales Office of Research
  Development (WORD)
• Three year grant (in the first instance long
  term research infrastructure development)
• Staff appointed Sept 06
• Formally launched by Minister of Health and
  Social Care, Welsh Assembly Government, Nov 06

13
HIRU built on a good pedigree of work in Swansea

• Clinical systems development and implementation
• RCP iLab project
• Public Health analysis
• Using routine data for primary research
• Blue C supercomputer at School of Medicine,
  Swansea University

14
Computing infrastructure

• Blue C supercomputer, one of the fastest
  computers in Europe dedicated to Life Science
  research
• Strategic partnership with IBM (through School of
  Medicines Institute of Life Sciences initiative)
• Advanced software toolset (database, data mining,
  GIS)
• Specialist support through Anix

15
Blue C Computer
16
HIRU programme aims

• Develop new methodologies for accessing and
  combining routine data in ways which do not
  breech data confidentiality rules and
  regulations, but which still permit the use of
  data for a wide range of research purposes.
• Explore how to use routinely collected and other
  data to support large scale multi-site
  intervention and cohort studies and policy
  relevant research.
• Develop innovative analyses of large and combined
  datasets
• Develop methods for data capture to common
  standards and definitions in multiple and remote
  locations.

17
Focus on Wales

• Small country - 3 million population
• Accessible leadership
• Desire to be smart
• 22 Local Health Boards ( PCT) coterminous with
  Local Authorities, 14 NHS Trusts
• Wide variation in health from near best to worst
• National Programme for IT (Informing Healthcare)
  working incrementally)

18
Progress through partnership

• Health Solutions Wales (HSW)
• Welsh Assembly Government Information Services
  Division (CHIP Programme PCIMT)
• Informing Health Care
• National Public Health Service Wales
• CRC Cymru organisations and other research groups
• And particularly . . .
• NHS and Local Authority organisations

19
Our approach

• Individual confidentiality is at the core of our
  approach (DAPP)
• No identifiable confidential information is
  passed to anyone involved
• The project is about information linkage across
  domains and organisations
• We act responsibly with the data. HIRU is not a
  data mart. Analysis is strictly protocol-driven.
  Disclosure control rules are applied to ensure
  confidentiality.
• We use data for health-related research and to
  provide useful information back to partner
  organisations that provide data
• Information used for knowledge not for
  performance management!

20
HIRU methodology (example)
Construct ALF
Validate
HIRU (Blue C)
Health Solutions Wales
Data Provider
Anonymisation process
Recombine
Encrypt and load
Operational system
HIRU (Blue C)
21
Matching Anonymisation Process
22
Matching

• Each record matched to a definitive population
  register for Wales - the NHS Administrative
  Register carefully maintained by the BSCs and
  HSW
• Exact match against date of birth, forename and
  surname
• Lexicon techniques common variations of
  forename, such as "Liz" and "Beth" for
  "Elizabeth"
• Then, if no match, use
• Soundex alternate phonetic spellings of the
  forename or surname, using Soundex functions.
• Fuzzy Matching Are there possible matches of
  either date of birth, address or name?
  Allowances for possible data entry errors.
  Possible matches are scored and weighted against
  frequencies of occurrence within the Welsh
  population. Matches are currently accepted if
  the probability is above 90.

23
Anonymisation

• NHS number encrypted using highly secure 256 bit
  encryption algorithm
• Unique surrogate value assigned against the
  encrypted value to become the Anonymous Linking
  Field (ALF)
• Encryption key held within HSW and only known by
  them
• Personal details deleted away before the file is
  returned to HIRU as File 3
• Second encryption of ALF occurs while data is
  flowed to HIRU (key known only to HIRU DBA) to
  create the ALF-E
• Under development R-ALFs (Anonymised
  households)

24
Geocoding

• Each address / postcode replaced by a geocode
• Lower Super Output Area standard small area
  geography of c. 1,500 people
• 1896 LSOAs in Wales

25
Data warehouse design

• Built in DB2 Data Warehouse Edition
• On IBM P Series supercomputer running AIX
• Supports the entire lifecycle of the data HIRU
  receives from
• loading of external data
• encryption and second level anonymisation
• base storage
• historisation of facts and dimensions
• data cleansing and
• presentation of the data to users

26
(No Transcript)
27
SAIL Pilot Project

• SAIL Secure Anonymous Information Linkage
• LHB/LA (PCT) area as early adopter pilot (pop.
  227k)
• Developing and refining the methods
• Testing quality
• Research question generation
• Sharing the benefits
• Reporting the learning
• Planning for scaling-up

28
SAIL Data Bank Pilot (Swansea)

• Access and load a range of datasets
• Inpatient day case admissions
• Outpatient appointments
• AE data
• Social Services data (older people, mental
  health, learning disability, children)
• Births and deaths
• GP morbidity prescriptions (35 of 36
  practices)
• Child health database
• Pathology results from NHS Trust
• NHS Direct Wales 0845 call centre contacts
• Many more in the pipeline . . . . .

29
Volumetrics (Swansea pilot)

• With national data for inpatients and outpatients
  and local (Swansea) data for primary care and
  social services, data volumes are large
• Information on 2.42 million people
• Detailed SAIL information on 227k people
• Success rate for ALFing NHS data c 99
• Success rate for social services data currently
  85 (using tight matching criteria)
• Number of health events recorded 34,700,000
• Number of records in the databank 105,million
  and rising!

30
Patient Journey Analysis- Health and Social Care
31
Post-pilot developments (in progress)

• Data extraction utility for all 497 GP practices
  (weekly extracts)
• Automated incremental feeds from national
  datasets (monthly)
• National data transportation fabric (for NHS and
  non-NHS) with full file handling, checking,
  receipting etc)
• Fully automate processes for matching,
  anonymisation and database loading
• Support national clinical data collections for
  research
• Extend depth and breadth of health data suppliers
  to cover all Wales
• Focus on non-health suppliers (social care,
  education and housing)
• Building on pathology data to integrate imaging
  and biomedical testing data
• Part of Phase 1 of ONIX with NCRI Informatics
  Initiative (live meta-data index)

32
Data transportation fabric
33
Building clinical (2 care) datasets

• Working with Wales-wide groupings of clinical
  specialities
• Agree a basic national research dataset
• Developed a generic data architecture to support
  data collection, for all conditions
• Make available web tool for data capture and
  local analysis
• Anonymised version fed into SAIL Data Bank
• Starting with stroke, ankylosing spondylitis, and
  soon cancer

34
Individual-level data acquisition strategy
AE Attendances
Social Services
Other national datasets
Clinical data collections
Child Health
GP Out of Ours
Education attainment
Pathology
Radiology
NHS Direct
Housing
Cancer
Screening
Anomalies
Out patients
Inpatients
GP Data
Geographical coverage
Ecological / environmental data
35
Some datasets individual and ecological

• Clinical databases
• Cancer
• Screening (multiple conditions)
• Congenital Anomalies
• Myocardial Infarction
• Diabetes
• Stroke
• Arthropathies
• Etc.
• Ecological datasets (many are GIS)
• Census
• Ordnance Survey - Mastermap
• Social Housing
• Transport
• Environmental Health
• Planning
• Leisure
• Government departments and agencies

• Individual level - health
• Population (NHSAR)
• Inpatients -PEDW (HES)
• Births
• Deaths
• Outpatients
• AE
• GP Data
• Laboratory systems
• Child Health Database Wales
• NHS Direct Wales
• Radiology- Reporting and Imaging
• Individual level non health
• Social Services
• Educational Attainment
• Housing

36
Assuring confidentiality

• Data Anonymisation Process approved by wide range
  of Information Governance officials and bodies
  (Caldicott etc)
• Large scale and very rich data (all anonymised)
• Subsets of data not sent externally
• Access controls necessary
• Research protocol driven access
• Analysis plans crafted to ensure no disclosure
  (avoiding low cell counts)
• Secure analysis laboratory with secure
  workstations
• Portals in development with no data transfer

37
Types of research supported by data linkage

• Demonstrating and explaining variation in health
  service utilisation with a view to developing
  testable hypotheses
• Improving capacity for (and efficiency of)
  clinical trials
• Improving drug and device safety (long term
  follow-up)
• Population health studies and evaluation of
  policy initiatives
• Hybrid cohort studies using traditional
  recruitment and e-approaches
• Health system and biomedical Modelling

38
Thank you
39
HIRU programme aims

• Develop new methodologies for accessing and
  combining routine data in ways which do not
  breech data confidentiality rules and
  regulations, but which still permit the use of
  data for a wide range of research purposes.
• Explore how to use routinely collected and other
  data to support large scale multi-site
  intervention and cohort studies and policy
  relevant research.
• Develop innovative analyses of large and combined
  datasets
• Develop methods for data capture to common
  standards and definitions in multiple and remote
  locations.

40
Clinical trials linked data

• Quicker and more accurate feasibility studies,
  searching multiple databases for
  inclusion/exclusion criteria
• Development of methods for electronic enhanced
  recruitment (EER) of participants
• Testing efficacy of EER an RCT within an RCT
• Provision of long term outcomes

41
Identifying suitable trial participants Site
screening
Sample population (n6474)

• Excluded (n 6409)
• Does not have type 2 diabetes (n 6222)
• Is lt18 years (n 0)
• Not receiving 1 or 2 OADs for 2 months (n
  171)
• The subjects most recent HbA1c is not 7.5 and
  lt 11 (if on mono therapy) or 7 and lt 10
  (if on combo therapy) and do they have a BMI lt 40
  kg/m2 (n 6)
• Smoked within the last 6 months (n 3)
• The subject has clinically significant active
  pulmonary disease (excluding asthma) or
  cardiovascular disease (unstable angina within
  the last 6 months or MI within the last 12 months
  and/or heart failure NYHA class I to IV) (n 6)
• The subject has proliferative retinopathy or
  maculopathy requiring acute treatment (n 0)
• The subject has uncontrolled / untreated
  hypertension ( 180/100 mmHg) (n 1)
• The subject has had treatment with systemic
  steroids within the past 2 months prior to
  screening (n 0)

Potential participants identified (n65)
42
An RCT within an RCT

• Test the hypothesis that electronic enhanced
  recruitment (EER) works
• More patients entered
• Patients recruited more quickly
• Cost effective
• Cluster RCT design
• Research practices randomised into I and C
• EER instituted in I Group

43
Example of HSR using linked HIRU data (with
Cannings, Butler, Dunstan Cardiff)

• Are patients who are not prescribed an antibiotic
  for an acute Respiratory Tract Infection (RTI) at
  a higher risk of developing a complication than
  those who are prescribed an antibiotic?
• Cohort Patients diagnosed with a first
  episode of an acute RTI in 24 general
  practices
• Exposure Antibiotics prescribed at first
  presentation of RTI
• Outcome Complications diagnosed in Primary and
  Secondary Care

44
(No Transcript)
45
(No Transcript)
46
Can routinely collected, electronically stored
data be used for health technology assessment by
randomised controlled trial?

• We repeated the analysis of four exemplar RCTs
  using data extracted from local electronic data
  systems
• Studies were small, multicentre HTAs in South
  Wales addressing four different technologies
  (open access to outpatients investigation of
  sleep apnoea autologous blood transfusion
  surgery for incontinence)
• Funded by the HTA Programme

47
Summary of exemplar studies
48
Data sources

• Central returns PEDW
• Hospital information systems PAS Pathology
  Radiology
• Clinical system GeneCIS (symptoms, signs,
  diagnoses, interventions)

49
Conclusions

• Routinely collected data can support RCTs if
  clinically rich, and held in electronic form
• Patient and professional preference would still
  need to be collected
• Data availability, validity and standardisation
  must be improved
• Costs would be less, and larger trials could be
  run
• Williams JG et al The value of routine data in
  health technology assessment can randomised
  trials rely on existing electronic data? Health
  Technology Assessment 2003vol 7no 6
• Cohen et al Estimating the marginal value of
  better research output Designed vs routine
  data in randomised controlled trials. Health
  Economics 200312959-74

50
Can outcomes be monitored using clinical data as
a proxy?

• Yes - if clinical data is captured in structured
  form in sufficient detail
• Symptoms and signs can be used as proxy measures
  for generic and disease specific HRQL measures
• Assessment of the usefulness and cost in routine
  practice needed
• Hutchings H, Cheung WY, Williams JG et al
  International Journal of Health Technology
  Assessment 2005

51
Longitudinal Tracking Patient Journey Analysis

• Being able to track anonymised individuals across
  multiple datasets longitudinally has huge
  potential benefits
• Understanding complex NHS and SS care packages
• Economic modelling
• Effects of interventions in one sector on others
• More comprehensive and long term follow up

52
Population health studies

• Electronic cohort studies eCohort
• Suitable when exposures and outcomes are
  routinely collected
• Electronically enhanced cohorts a hybrid
  between a traditional and eCohort with reduced
  costs
• Case series, case control studies, etc
• Evaluation of policy initiatives interrupted
  time series

53
Future Research plans

• Work collaboratively with thematic research
  networks in Wales
• Develop/support major research platforms
• Public Health Centre of Excellence
• MRC Centre for Ageing Research
• NIHR Injury Prevention
• Wellcome Trust LADA Cohort
• WORD/RCs - Environments for Healthy Living a
  Family Cohort Study on the impact of the changing
  social, physical and technological environment on
  health
• Cost effectiveness of conducting national
  clinical audits using routine data (Health
  Fundation)

54
I could go on and on . . .But I wont!Thank
you!
55
Demonstrating and explaining variation in health
service utilisation examples using record
linkage

• Radiology rates in AE
• Hospital admissions from ankle fractures
• Co-morbidity adjustment for outcomes
• MI admissions
• Length of stay Stroke admissions

56
(No Transcript)
57
(No Transcript)
58
2 way linked data admission rates for ankle
fractures by hospital
59
Helping to understand trends in emergency
attendances and admissions
60
(No Transcript)
61
(No Transcript)
62
Emergency Hospital Admission rates for Swansea
(LSOA)
Demand on NHS for Emergency Admissions for
2004-2005
63
Wider area comparisons LOS variability
64
Outcome Measurement

• WAG is moving to measuring/commissioning on
  outcomes
• Important that measures used are robust and fair
• HIRU is assisting with methodological
  developments in outcome measurement (SLIM)
  project
• Reconstructing QoL measures from Routine Data

65
Co-morbidity adjustment for outcomes

• Co-morbidity affects outcomes
• Are co-morbidities reliably recorded on hospital
  discharge data ?
• What is the variation in frequency of
  co-morbidity recording by hospitals?
• What is the frequency of COPD in subsequent
  admissions where listed as co-morbid factor in
  primary admission with an MI?

66
(No Transcript)
67
Frequency of co-morbidity recording in hospital
discharges

• Variation in recording between trusts no
  logical reasons for this data quality issue
• 80 likelihood of subsequent mention of COPD
• Could be improved by primary/secondary care data
  linkage
• import co-morbidities from primary care
• severity and diagnostic validation from pathology

68
Majority Use NHS Trustsin Wales Admissions
Outpatients
69
An example of collaborative operational research

• Clinicians/NHS would like to know the likely
  impact of bowel screening on workload
• Collaboration between Screening Services/ WCISU /
  HIRU
• Plan to link anonymised cancer registry data with
  utilisation datasets in primary and secondary
  care
• Determine service utilisation by stage of cancer
• Model likely impact on service workload
• Plan services to anticipate needs

70
3 way linkage Stroke Survival Rates
71
Mapping and GIS applications

• The physical and built environment influences
  health
• An area where local government can strongly
  influence health
• An area where there is tremendous opportunity for
  collaborative work and research between health
  and local government
• Development of mapping, GIS and increasing
  numbers of ecological datasets offers huge
  potential

72
The Built Environment and Health

• How does the design of the built environment,
  planning and land use policies influence health
  through?
• Providing equitable access to services
• Supporting physical activity through walking and
  cycling
• Providing safe environments for pedestrians
• Providing safe and exciting play areas for
  children
• And many more.

73
Mastermap topographic layer
74
Mastermap premise types from address layer
75
Example of traffic calming distribution
76
Social equity in the provision of traffic calming
77
Primary School Entry Children Posterior mean
relative risks for overweight children
Non-Spatial Model
78
Percentage of overweight children by 5th of
Deprivation, LSOA, WIMD 2005, n13,416
79
Collaborative research with City and County of
Swansea

• Development and testing of automatic pedestrian
  activity devices
• Studies in pedestrian safety
• Evaluating interventions aimed at increasing
  physical activity
• Use of GIS datasets to support research into
  environmental influences on health
• Playground distribution
• Others.

80
(No Transcript)
81
(No Transcript)
82
Distance to Play Areas in Swansea
(m)
2.3km
83
Example of research question using linked data

• Are patients who are not prescribed an antibiotic
  for an acute Respiratory Tract Infection (RTI) at
  a higher risk of developing a complication than
  those who are prescribed an antibiotic?
• Cohort Patients diagnosed with a first
  episode of an acute RTI in 24 general
  practices within Swansea LHB, in 2005.
• Exposure Antibiotics prescribed at first
  presentation of RTI
• Outcome Complications (quinsy, pneumonia etc)
  diagnosed in Primary and Secondary Care

84
Linkage
Hospital data
85
(No Transcript)