EPSRC

1
EPSRC Pathways to Impact Grant. Training and
education for the developers of databases in
research and clinical practice.

• Rick Jones, Yorkshire Centre of Health
  Informatics, University of Leeds

2
Context

• The UK can significantly enhance its clinical
  research capability by using, strictly within the
  bounds of patient confidentiality, the electronic
  patient data that the UKs National Programmes
  for IT in the NHS have the potential to allow.
  This will have enormous benefits for all types of
  clinical, public health and health services
  research and for many aspects of patient care.

3
Context Simulations

• 4.1 Surveillance (Pharmacovigilance)
• The vision for an ideal surveillance system is of
  a nationwide active system for tracking
  patients responses to medical interventions
  (POMS, immunisations and OTC drugs as well as
  Devices) and of disease and other incidents
  requiring reporting.
• 4.2 Clinical Trials
• Within the range of activities involved in
  running a successful clinical trial in the
  future, there will be a need to access and
  process data from electronic records at a number
  of stages before, during and after the trial.
• 4.3 Prospective tracking of a known cohort
• This simulation team concluded that in order for
  UK Biobank (which is a resource for prospective
  studies) to be able to provide benefits access to
  data will have to be at patient level data
  (identifiable) both coded and textual.
• 4.4 Observational epidemiology
• The construction of retrospective observational
  epidemiological studies based on routine data
  sources requires access to data from a very wide
  range of electronic records, both within and
  without the health services.

4
Context The NHS Research Capability Programme

• Six work streams
• Technical Architecture
• Functional Scope
• Data Quality, Standards and Linkage
• Information Governance and Threat Assessment
• Infrastructure
• Communications and Stakeholder Engagement

5
Harvesting research outcomes from clinical
databases - demonstrating the potential of TPP
SystmOne.

• Richard Gillott, Cardiovascular Database
  Developer, LTHT
• Rick Jones, Yorkshire Centre of Health
  Informatics, University of Leeds

6
Aims objectives

• Based on the work of the RCP could a practical
  trial be carried out as proof of concept to
• Prove the feasibility of extracting identifiable
  patient data from GP systems for use in research
• Construct an architecture to enable the rapid,
  repeatable, and secure query collection of data.

7
Pilot trial

• A small trial was planned with the aid of 2
  research groups based in Leeds
• The pilot aimed to demonstrate the value of the
  information contained in the patient record, and
  prove whether the data was sufficient in its
  coverage of the population and its completeness

8
Methods
9
Results Yields of Records
81 Cardiovascular 66 Oncology
10
For linked records

• We need
• Granulated consent all or part of record
• A model to allow selective searching by
• Clinical relevance
• Administrative components
• Therapeutics
• Diagnostics
• An understanding of when and how frequently to
  conduct searches
• A business model to reimburse GP systems
  suppliers / data guardians for their search time

11
Workshop 1. Headlines

• Specificity of Consent Statements
• Baseline Knowledge of Electronic Record Systems
• Videos of Consent Process
• Consent to approach relatives
• Trial Designs for Long-term outcomes
• Knowledge of Accountable Parties
• Specimen Consent Forms and Information Sheets

12
Today Architecture
13
Background

• Building upon the work undertaken by the RCP
• Proposed a number of high level architectures
• Centralized vs federated data
• Concentrates on Secondary Care
• Honest broker
• The RCPs work and its database of documents was
  reviewed.
• The aim was not to provide a conclusive model for
  implementation, but to discuss some of the
  important considerations that need to be made.

14
About the model

• Uses the Service Orientated Architecture (SOA)
  paradigm, in a format that ties in with the work
  being conducted by the Research Capability
  Program to link data from Primary, Secondary and
  Tertiary sites across the NHS.
• Adopts the hybrid model of centralized
  federated data storage and processing
• Recognises the mechanisms necessary for the
  querying of data are already available in
  some systems, such as SystmOne, thanks to the
  GPES initiative. However, this scheme will need
  to be extended, to support both the push and the
  pull methods required by a scalable solution.

15
Centralized vs distributed
16
Overview diagram
17
Deployment diagram
18
In this workshop

• Can we explore and determine an architecture?
• Can we consider the metrics of use of a linked
  search service?
• Numbers of patients
• Frequency of searches
• At what scale should this be developed?
• Local pilot
• Employ National infrastructure

19
Exercise 1

• What are the main use-cases of a system?
• Use simple UML notation
• High level view
• Identify as many actors as possible

20
UML Use Cases

• Keep it simple
• Identify actors
• Patient
• Researcher
• Clinician
• Sys Admin
• Key tasks
• Dont worry about syntax

21
Intermission Food for thought

• Tony Solomonides et al,
• Privacy compliance and enforcement on
  Europeanhealthgrids an approach through ontology
• Phil. Trans. R. Soc. A 2010 368, 4057-4072

22
Exercise 2

• Can we identify the components needed to meet the
  requirement?
• Systems
• Processes
• Connections
• Well sort out the UML later!

23
It doesnt need to be pretty
24
Next steps

• Workshop output will be synthesised and
  circulated for validation
• Overview of work to date will be prepared for
  31st March Workshop
• Concept is that the overview will be presented
  and critiqued by participants and prepared for a
  final white paper report
• We are expecting a number of patients at the
  final session

25
At 1pm..

• Free Text in GP Patient Records How much extra
  information is there and how can we extract it?
  Rob Koeling, University of Sussex.
• The UK General Practice Research Database
  provides a valuable source of information for
  health services research. Coded data is
  supplemented by free text (physicians' notes or
  letters). However, due to the difficulty of
  extracting information, and the cost of
  anonymisation, free text is seldom utilised in
  research. One of the goals of the PREP (Patient
  Record Enhancement Project) project at the
  University of Sussex and the Brighton and Sussex
  Medical School, is to explore how much extra
  information (i.e. on top of the coded data) is
  concealed in the free text fields of GP patient
  records. I will present some results of an
  annotation effort in which a corpus containing
  text records of 344 women in the year prior to a
  diagnosis of ovarian cancer were marked up with 5
  commonly presenting symptoms. I will also report
  on a simple method we developed for automatically
  extracting these symptoms using string matching.
  I will talk about what this means for the
  estimates of the incidence of these symptoms and
  what we can do to extract this information. How
  far can simple, readily available, methods take
  us and what is the scope for more complex
  information extraction techniques.

26
Yorkshire Centre for Health Informatics

• Director Dr Susan Clamp
• 44 (0)113 343 4960
• s.clamp_at_leeds.ac.uk

www.ychi.leeds.ac.uk