Upload
nuffield-trust
View
293
Download
3
Tags:
Embed Size (px)
DESCRIPTION
In this slideshow, Liam Smeeth, Deputy Director and Head of Department of Non-Communicable Disease Epidemiology of the London School of Hygiene and Tropical Medicine discusses big data, e-health and the Farr Institute. Liam Smeeth spoke at the Nuffield Trust event: The future of the hospital, in June 2014.
Citation preview
Big data, e health and the Farr Institute
Liam Smeeth
London School of Hygiene and Tropical Medicine
Thanks to: Harry Hemingway, Emily Herrett, Harriet Forbes, Ian Douglas, Krishnan Bhaskaran, Tjeerd van Staa, Ben Goldacre, Iain Chalmers and many others
Funding: Wellcome Trust, MRC, BHF, HTA
Plan
• Big data and e health
• Examples of research
• Data quality
• The Farr Institute
UK Government: big data
Universities and Science Minister
Chancellor of the Exchequer
Prime Minister
Health Minister
Big data: is it something new?
Two answers:
• No
• Yes
Big data: is it something new?
Yes
Computers mean that more health related data are available and can be linked together
Genomic and metabolomic data are available at a new scale and new level of detail
The computerisation of health related data and the -omic revolution extraordinary opportunities for research
• Better research
• More efficient research
• Research that couldn’t otherwise be done
Examples
Measles mumps rubella vaccination and autism
• 1998 Lancet paper: MMR vaccination might cause autism
• MMR vaccine coverage fell internationally
• Measles outbreaks occurred
MMR coverage by time of 2nd birthday, England NHS Immunisation Statistics, HSCIC
Study raises concerns
Measles mumps rubella vaccination and autism
• United Kingdom Medical Research Council funded case-control study
• Similar large studies in USA and Denmark
• Only possible because of electronic health records (big data)
Effect.5 .75 1 1.25 1.5 2
Combined
Current study
DeStefano et al
Madsen et al ASD
Madsen et al autism
Effect size (95% CI)
0.92 (0.68 – 1.24)
0.83 (0.65 – 1.07)
0.93 (0.66 – 1.30)
0.86 (0.68 – 1.09)
0.87 (0.76 to 1.001)
Decreased risk Increased risk
Smeeth et al, Lancet 2004;354;963-9
MRC study
72.0
74.0
76.0
78.0
80.0
82.0
84.0
86.0
88.0
90.0
92.0
94.0%
MM
R c
ove
rage
Autism risk published
MMR coverage by time of 2nd birthday, England NHS Immunisation Statistics, HSCIC
Our study published
• Cohort study within the Clinical Practice Research Datalink (CPRD)
• 5.2 million people
• 33.9 million person-years of follow-up included
• 184,594 people (3.5%) experienced one of the 21 commonest cancers
Body mass index and cancer
1980 1984 1988 1992 1996 2000 2004 2008 2012 2013
Age-standardised prevalence of overweight and obesity ages ≥20 years, by sex, 1980–2013
Ng M et al Lancet 2014
Bhaskaran K et al Lancet in press
Body mass index and cancer: a cohort study of 5.2 million people
Bhaskaran K Lancet in press
Different causes
Bhaskaran K et al Lancet in press
Data quality: myocardial infarction as an example
Capture of acute myocardial infarction events in primary care, hospital admission, disease registry and national mortality records
Emily Herrett, Anoop Dinesh Shah, Rachael Boggon, Spiros Denaxas, Liam Smeeth, Tjeerd van Staa, Adam Timmis, Harry
Hemingway
BMJ 2013; 346; f2350
Herrett E et al. BMJ 2013;346:bmj.f2350
Incidence
Incidence
Herrett E et al. BMJ 2013;346:bmj.f2350
Diagnostic validity • Around 90% of patients with an ST elevation
myocardial infarction recorded in the national registry (MINAP) had raised cardiac enzymes or characteristic EKG findings, but….
• Registry (an audit) incomplete
• Hospital Episode Statistics more complete
• Primary care clinical record much more complete: but all three together best
• Cross validation suggested primary care diagnosis had a high validity
Electronic health data and evaluation
• Generalisability or external validity
– adherence to intervention
– clinical care received
– co-morbidities
– co-prescriptions
– selected groups of participants
– absolute risks and benefits different
Poor guides to clinical practice and policy
Challenges for randomised trials 1
• Recruitment: inadequate sample size
– review of all 114 multicentre trials from two major UK public funders over seven years
– only 31% of trials achieved their recruitment target
– over half had to be awarded an extension Campbell MK et al Health Technol Assess 2007
• Loss to follow-up: leading to bias
• Costs: up to $10,000 per participant not unusual
Challenges for randomised trials 2
Can electronic health records help with randomised trials?
• recruitment
• generalisable
• outcomes
• costs
incorporate evaluation into everyday care?
Electronic health data and evaluation
What to do in the absence of evidence?
What to do in the absence of evidence?
randomise
Are the patient and the doctor or the policy maker and manager
happy to randomise?
Option A Option B
100% follow-up: totally electronic records based
Is there an absence of clear evidence?
Results included in the evidence base
Text messaging reminders for influenza vaccine in primary care (TXT4FLUJAB)
A randomised controlled trial using electronic health records
Emily Herrett, Tjeerd van Staa, Liam Smeeth
• Targets for the elderly are reached
• Targets for patients under 65 at risk are missed
• Last year 51.6% of eligible patients were vaccinated compared to a 75% target
Influenza vaccine uptake Vaccine uptake, 2011/12
0
10
20
30
40
50
60
70
80
% vaccinated
UK government target: 75%
SMS text message reminders
• Widely used by practices
• Effective for appointment reminders
• High mobile phone usage (93% for age <60, 70% for age 60+)
TXT4FLUJAB methods
• Design: cluster randomised trial using English primary care electronic health records
• Intervention: text message vaccine reminder to patients under 65 in risk groups:
– “Hello Fernanda, to reduce your risk of serious health problems from flu we recommend vaccination. Call 0207 927 2837 to book. The London medical practice”
Consenting practices
randomised
Text messaging group:
60 practices
≈ 600,000 people
SMS reminder to patients
under 65 at risk
Standard care group:
60 practices
≈ 600,000 people
Seasonal flu campaign as
planned
Practices invited to
trial
Researchers ascertain exposure and outcome data
remotely from practice records
TXT4FLUJAB costs
• Total costs to date: £50,000
• Cost per clinic: £200
• Average 1400 patients per clinic receive intervention or control: about 200,000 patients
• Likely total cost: £100,000
Cost per patient: £2 per patient included
In 2012, four Health Informatics Research Centres were
awarded by a consortium of 10 United Kingdom
funders led by the Medical Research Council
Our Story
Farr London
Farr Scotland
Farr at Swansea, Wales
Farr N8 Manchester
Strengthening health informatics research
• MRC coordinated 10-partner £19m call for e-health informatics research centres across the UK
Cutting edge research using data linkage
capacity building
• Additional £20m capital to create Farr Institute
• UK Health Informatics Research Network
Coordinate training, share good practice and
develop methodologies
Engage with the public, collaborate with industry
and the NHS
“To harness health data for patient and public benefit
by setting the international standard in the use of
electronic patient records and related data for large-
scale research.”
Our Vision
Basic discoverie
s
Proof of concept
(Experimental medicine)
Clinical Trials
Quality and
outcomes research
Health gain
What are the aims of Farr London?
= research along the translational pathway
+
Farr Tools: Informatics methods
Farr People: Capacity development
Farr Curated data: Research-ready cohorts with 10m person years now
Reverse translation
Bringing together people
Inter-disciplinary: genomics, biostatistics, epidemiology, bioinformatics,
health informatics, computer science, social science etc.
Inter-institutional
William Farr
“Diseases are more
easily prevented than
cured and the first
step to their
prevention is the
discovery of their
existing causes”
•
• Photo, quote
• And a
William Farr’s grand challenge
Health records ‘An arsenal
that the genius of English
healers cannot fail to turn to
account’
William Farr 1874
supplement to 35th annual report
of the Registrar General,
What is needed?
• Expertise
• Novel methods and approaches
• Ensuring high data quality
• Confidentiality and security of data
An expectation by patients/citizens, clinicians and policy makers that research and evaluation is a normal - in fact a necessary - part of health care and policy