Scottish School of Primary Care

1

New data from old linking primary and secondary
care data in research from WOSCOPS to DARTS and
 SHIP Frank SullivanPHCRED visiting Fellow Tri
State Seminar Flinders University
General Practitioner, Arthurstone Medical
Centre Professor of RD in GP, University of
Dundee Director, Scottish School of Primary Care
2
(No Transcript)
3
Perspectives

• 20 year research interest in PC Informatics
• Systematic Reviews and Evaluation
• Health Informatics Centre Dundee
• Wellcome and EU programme grants
• Advisory role
• Office for the Strategic Co-ordination of Health
  Research (OSCHR) eHealth Records Research Board
• External reference group of Research Capability
  Programme of Connecting for Health

4
The next 45 minutes

• Record Linkage in Scotland
• Deterministic and Probabilistic
• Access to primary care data
• Security and Confidentiality
• Some examples
• WOSCOPS, DARTS
• Planned Developments
• SHIP, TRANSFoRm
• Research Capability Program Connecting for Health
• Brief observations in Australia

5

• Each person in the world creates a Book of Life.
  This book starts with birth and ends with death.
  Its pages are made of records of the principal
  events in life. Record linkage is the name given
  to the process of assembling the pages of this
  book into a volume.

1978
Dunn H Record linkage. JAMA 1946, 36 1412-6
6
Deterministic linkage
CHNo
Lab Data
Screening
Hospital SMR
Dental
Investigations
Primary Care
Social Services
Pharmacy
7
Community Health Number
07 10 64 02 5 0
Sex
Date of Birth
Checksum
Sequence
8
Record-Linked DataCompleting the Jigsaw
Lab Data
Dental
GP
CHNo
Pharmacy
Hospital
Social Services
Screening
Investigations
9
Probabilistic Linkage by Information Services
Division (ISD) Scotland

• Bring together the pairs of records to be
  compared (Blocking)
• Quantify the relative probability that the two
  records belong to the same person (calculating
  weights)
• Make the linkage decision

10
Fragment of Soundex/New York State Identification
and Intelligence System Phonetic Code Algorithm

• char sout (char )malloc(strlen(s))int i
  0, i2 0if (s0 (char)NULL) return s
  while (isspace(si)) iwhile
  ((!isspace(si)) (si ! (char)NULL))
  iif (si (char)NULL) return ""while
  ((isspace(si)) (si ! (char)NULL))
  iwhile (si ! (char)NULL)

11
Dummy patient history as held in the linked
database
12
Information from cradle to grave...

• Mothers ante-natal records
• Maternity
• Neonatal record
• Register birth - NHS number
• Register with GP - CHI
• GP Appointments
• Dental Appointments
• Outpatients
• AE attendance
• General hospital admission
• Prescribing
• Cancer registration
• Cancer treatment
• Coronary heart disease
• Community care
• Death

13
Follow up A60K Over 15 years
Initial Trial A60M Over 5 years
14
Contents of GP RecordsRelational Databases

• Demographics
• Prescribing
• Diagnoses
• Tests and Procedures
• Free text

15
(No Transcript)
16
(No Transcript)
17
Programmable Automated Follow up of Read Coded
data
Data extracts are specified using a predefined
set of READ codes to define the patient
population, then for the patients identified
diagnostic, therapeutic and activity codes can
be extracted. These codes can be updated
nightly using an extract definition file.
Data are sent from the practice to the regional
Gateway server via eLinks and then loaded into
SQL Server.
18
The Walker Project
25,000
Parents of Index cases born 1927-1941 (55-75 yrs)
42,000
Index cases Babies born 1952-1966
73 data items
42,000
Pregnancies of Index cases 1952-1966 (30-50 yrs)
5,000
Grandchildren of Index cases (5-20 yrs)
19
(No Transcript)
20
DARTS
SCI-DC NETWORK
21
(No Transcript)
22
(No Transcript)
23
Patient Summary for clinicians
24
Summary for patients
25
(No Transcript)
26

Scottish Diabetes Survey 2002-2007 Recording of
Key Biomedical Markers n 220K
Percentage of Patients
Source Scottish Diabetes Survey
Data recorded within the previous 15 months
27
Alternatives to league tables
28
The Scottish Diabetes Research Network
unintrusive epidemiology
SCI-DC
clinical trials
With proactive consent for trial participation
29
Diabetes and Genetics

• Programme Grant
• United Kingdom Case Control Collection for Type 2
  diabetes
• 15,000 cases and controls
• In collaboration with Oxford, Exeter, Imperial,
  Cambridge, Mayo Clinic, Karolinska

30
Wellcome Trust Functional Genomics Programme
Intranet project management

• gt16,000 recruited
• 20 people/day
• DNA distributed nationally
• Serum locally

31
The FTO gene is the second most important gene
for Type 2 diabetes risk
P 10-12
TCF7L2 gene
FTO gene
-logp value
P 10-8
P 10-5
P 10-3
Chromosome
32
Pharmacogenetics of the response to
Sulphonylureas (SU)
SU response N540
5
-5
0
Variation of response Absolute HbA1c reduction
Data from DARTS, Tayside, Scotland
33
Defining Drug Response
Treatment HbA1c
Other measures Lipids, Biochemistry, BP, Other
treatment
SU treatment in 1200 patients Metformin treatment
in 1400 patients (with DNA)
34
Using GP phenotypic data in DARTS to look at
genetic determinants of response in type 2
diabetes

• TCF7L2
• Large population attributable risk
• Likely effect on beta-cell function (direct or
  indirect)
• Hypothesis
• Variants in TCF7L2 (rs7903146) will affect
  response to sulphonylureas but not metformin

Pearson et al Diabetes 2007
35
TCF7L2 genotype modulates SU response.
Outcome HbA1c lt7
SU response
HR 1.56, p0.03
Metformin response
P0.82
Pearson et al Diabetes,Aug 2007
36
Acute Recruitment Tool
37
Current SOP for access to patients with
pre-existing conditions in Scotland

• SPCRN staff undertake to search practice records
  for potentially eligible patients on behalf of
  individual practices and working under practice
  staff supervision.
• There is a current generic non-disclosure
  agreement with each practice.
• SPCRN staff must have a current NHS substantive
  or honorary contract.
• Each practice should have formally agreed to
  collaborate in a study
• Normally all searches and mail outs will be done
  from within the practices, and no identifiable
  data will be removed from the practice without
  explicit patient consent. These circumstances
  will have been specigied to the ethica cttee and
  Caldicott
• In exceptional circumstances, an encrypted
  electronic file containing patient identifiable
  data can be taken from the premises provided that
  it is kept secure by password protection, and
  destroyed as soon as invitations to take part in
  a study have been issued. These circumstances
  will have been specified to the ethics cttee and
  Caldicott

38
SPCRN Patient Recruitment Using GP records
ISD/HIC data
GP data
Option 1 Practice sends letters
Practice / SPCRN prepare send invitations
Info about study. Request to participate
List of eligible patients
Practice agrees to participate
Practice screens data
ISD/HIC aggregates, by practice
SPCRN
Option 2 Central Mailing
Blinded mailing list / study identity
Practice screens data
As option 1 until...
Blinded mail-merge
Blinded print mail
Invitation letter
39
Categories of confidentiality in Scotland

• Clinical care, governance
• Epidemiology
• Patient specific

• Inform and allow opt-out
• Ethics and Caldicott approval
• Signed consent from practices and patients

40
Key responsibilities of a Caldicott Guardian
include

• a management audit of current practice and
  procedures
• annual plans for improvement, monitored through
  the clinical governance framework and
• development of protocols to govern the disclosure
  of patient information to other organisations.

41
PACs role is to advise ISD and the Registrar
General on the protection of the privacy of
patient information while at the same time
recognising the need for legitimate access to
records by research workers, and those involved
in health administration for well-defined and
bona fide purposes, subject to appropriate
safeguards to maintain confidentiality.
42

• Staff may use some of your personal health
  information in research and staff training.
  Healthcare staff may use information from their
  patients to help them find the causes of disease
  and the effects of treatment and for planning new
  treatments. If the research involves you
  personally you will be contacted and asked for
  your consent.

43
(No Transcript)
44
Recent UK Wide Recommendations

• 1. Mandate common patient Identifier
• 2. Communicate the relevance of research to
  healthcare
• 3. Federate existing databases
• 4. Improve data quality
• 5. Initiate governance discussions
• 6. Engage key stakeholders

45
UK-Wide data extraction Tools
46
OSCHR e-health funders strategy

• Coordination required for
• Infrastructure, access management, quality
  standards and governance
• Capacity building and funding novel research
• Stakeholder Engagement
• Benefits realisation Evaluation and monitoring
• Optimising the interface with Industry

47

A7.9M over 4 years MRC ESRC EPSRC Chief
Scientist Office
48
Multi Institution Linkage Authorisation

• No copies of datasets each data owner sends an
  extract to user on request
• User browses dataset descriptions, requests
  extracts they need
• Approval given by MILA Executive, in consultation
  with dataset owners linkage assessment panel
• Separation of powers no data supplied without
  agreement of 2 independent agencies data owners
• Explicit record of all requests and data supplied

49
Virtual Microdata Laboratories in Safe Havens
Including 1y Care
50
Translational Medicine and Patient Safety in
Europe (TRANSFoRm)

• A13.2 M European FP 7 Integrated project
• 15 partners from 8 EU countries
• Aims to integrate clinical and research
  activities in primary care via
• Rich capture of clinical data via a generic
  dynamic interface
• Distributed interoperability
• Controlled vocabulary and standardised data
  elements

51
Demonstration and dissemination
WP8 DEMONSTRATION INDUSTRY CONTRACT
RESEARCH ORG ACADEMIA
User requirements Development and Evaluation
ICT
WP 1 RESEARCH USE CASES
WP7 SYSTEMS AND SERVICES FOR DATA INTEGRATION
WP9 DISSEMINATION WEBSITE PUBLICATION WORKSHOPS P
ROTOTYPES COLLABORATION INFRASTRUCTURE TECHNOLOGY
PLATFORMS
WP 2 PATIENT SAFETY USE CASE
WP10 MANAGEMENT
52
Interoperability

• Based on models
• HL7 RIM
• CEN 13606 openEHR
• HL7 Genotype model
• Implementation of models within systems
• May limit the richness of semantic
  interoperability
• Controlled vocabulary may help

53
Controlled vocabulary
54
TRANSFoRm
WIDER ADOPTION OF eHRS PROMOTING PATIENT SAFETY.
(WP2) -More accurate diagnosis -Richer coding of
data -Integration of personal risk support
(genetic or clinical data)
DYNAMIC CLINICAL INTERFACE (WP5)
IMPACT PARTNER -EGPRN -UoC -UoB (WP2,9)
Archetypes Rich ontology (WP4)
SYSTEM INTEGRATION (WP7)
RESEARCH SEMANTIC WORKBENCH (WP5)
INTEROPERABLE CLINICAL TRIAL DATA MANAGEMENT
SYSTEMS (WP1) -Wider adoption of
ISO1179 -Services for metadata archiving and
re-use -Larger, faster, more economic RCTs
GENOTYPE-PHENOTYPE STUDIES WITH POWER (WP1)
WP3 Security WP5 Provenance Data Mining
tools WP6 Interoperability Models Vocabulary
service WP7 Meta-data repository
IMPACT PARTNER -ECRIN -QUINTILES -EGPRN (WP8,9)
IMPACT PARTNER -ECRIN -QUINTILES -EGPRN (WP8,9)
55
Record Linkage in Australia

• WA Data Linkage System since 1995
• QoF data in Mt. Gambier
• GHRANITE

56

• Generic interfacing / communication with GP and
  other databases (Current standards e.g. HL7 not
  sufficient for research)
• Comprehensive patient consent management
• Scheduled extraction of data
• Data provider control of data flows (data
  inspection prior to transmission)
• Secure information exchange
• Ethical Record Linkage
• Scalable support mechanisms

57
Consent management
Different types of consent can be managed
58
Typical GRHANITE Data
Patients
Consultation dates
Chlamydia test results
59
GRHANITE current installations
60
The cycle of patient information
Higher quality care
Patient
Better informed decisions
Patient data
Retrieval
Patient records
Relevant patient data
Comparison with targets
Quality improvement
Storage System
Local insights
Grouped analysis
Research
New knowledge
Sullivan and Wyatt ABC of Health Informatics 2006
61
(No Transcript)
62
Safe Havens

• A Safe Haven is a secure environment in which
  person-level data are managed. A Safe Haven
• may contain identifiable data
• must have strict controls for physical and
  electronic access
• must be managed by a TTP or Honest Broker
• may be accessed by other users for particular
  purposes, following the necessary approvals and
  process
• Must comply with certain physical and
  organisational security arrangements

63
Honest Broker

• A trusted custodian of patient-identifiable data
  who has responsibility to implement systems of
  access according to legislation and policy,
  responsible for
• ensuring that pseudonymisation and anonymisation
  processes are correctly specified and
  implemented
• carrying out any permitted linkage of different
  sources of identifiable health and social care
  data and
• Carrying out data quality checks that are not
  possible for researchers and other data users to
  do themselves for reasons of confidentiality

64
Trusted Third Parties

• A TTP is an organisation that has been
  authorised to manage (usually) identifiable data
  for specific purposes. These purposes include
  the supporting and/or carrying out of medical and
  health-related research and potentially other NHS
  business purposes. A TTP should manage these data
  using a Safe Haven and operate physical and
  electronic access controls, which comply with the
  standards set out for Safe Havens. The TTP is
  the Data Controller for the data which it is
  authorised to manage.