harnessing health data for patient and public benefit professor tjeerd van staa herc...
TRANSCRIPT
Harnessing health data for patient and public benefit
Professor Tjeerd van Staa HeRC ([email protected])
van Staa TP, Dyson L, McCann G, Padmanabhan S, Belatri R, Goldacre B, Cassell J, Pirmohamed M, Torgerson D, Ronaldson S, Adamson J, Taweel A, Delaney B, Mahmood S, Baracaia S, Round T, Fox R, Hunter T, Gulliford M, Smeeth L. The opportunities and challenges of pragmatic point-of-care randomised trials using routinely collected electronic records: evaluations of two exemplar trials. Health Technol Assess 2014 Jul;18(43):1-146. doi: 10.3310/hta18430.
Phases of drug development
=>
Explanatory and pragmatic trials (Schwartz and Lellouch 1967)
• Explanatory trials => to verify a biological hypothesis – study population well adapted to the problem at hand, homogeneous and low withdrawal rate
• Pragmatic trials => to choose between two treatment. Usually complex interventions; no external validity beyond class of patients studied; trial should be representative of class of patients
• Schwartz and Lellouch did not focus on the current beliefs that explanatory and pragmatic trials are different with respect to internal and external validity!
Exemplar trials
• -RETRO-PRO: the effectiveness of simvastatin compared to atorvastatin—a feasibility study (ISRCTN33113202)
• -’cold’ recruitment
• -funded by Wellcome Trust
• -eLUNG: the effectiveness of antibiotics compared to no antibiotics for exacerbations of chronic obstructive pulmonary disease: a feasibility study (ISRCTN72035428)
• -’hot’ recruitment
• -funded by the HTA NIHR
• => Primary objective: to test feasibility
• Qualitative research on GP and patient perspectives
Electronic Health Records:Clinical Practice Research
Datalink
• Central repository of anonymised EHRs
• EHR records of General Practitioners across the UK = central healthcare provider; EHR for record keeping
• About 5% of the population included
• Pseudo-anonymised records (using opt-out system)
• Linked to other datasets using NHS number (e.g. hospital data, death certificates, registries)
• Regular transmission of data from practice to CPRD (monthly update of research database; daily for trial practices)
• ResearchOne alternative to CPRD
Drop-out rates of practices at each step in the recruitment and approval procedure
Step in site recruitment
Retropro eLung
N practices % N practices %
Practices contacted 459 100% 459 100% Practices returned letter of interest 377 82.1% 377 82.1% Interested 270 58.8% 270 58.8% Declined 107 23.3% 107 23.3% Site agreement received 63 13.7% 53 11.5% Site-specific submission 50 10.9% 45 9.8% NHS permission granted 48 10.5% 43 9.4% GCP and protocol training complete
35 7.6% 32 7.0%
Approved to recruit by study team 30 6.5% 31 6.8% Practices remaining following subsequent withdrawals
25 5.4% 24 5.2%
Were seeking to recruit patients 17 3.7% 8 1.7% Recruited patients 17 3.7% 6 1.3%
IT system in EHR trials
Pop-up boxes - flags for eLung and Retropro
eLung Retropro
Selection of eligible patients
• Monthly research database: creation of list of eligible patients
patients with >15% CVD risk according to Framingham or QRisk2 + no exclusions
COPD patients + no exclusions• Daily processing: further exclusions
Statin Rx => removal from list Antibiotic => removal for 2 weeks
• GP confirms eligibility (on website): only patients on the list can be recruited!
• First eligibility assessment done centrally => complex searches can be done
Recruitment of patients
• Three different models:
List of eligible patients: screening by GP => directly to study website + log-in + entry of patient number (Retropro)
Data entry of specific Read code during consultation for patient on eligibility list => LEPIS activated => link to study website (Retropro)
Start consultation for patient on eligibility list => LEPIS activated => link to study website (eLung/ Retropro)
Monitoring of LEPIS
Trial_ID
Clinic_ID
Patient_ID Patient_Status Staff_Actio
n Timestamp
eLung e9f9635dafc72e9742d8993aaa2adee4f806afa
P P 2013-03-14 11:53:56
eLung e9f9635dafc72e9742d8993aaa2adee4f806afa
FG N/A 2013-03-14 11:52:57
FLU-CATs
N/A N/A TRU 2013-03-14 11:51:46
RCT_Test
e61a31b5b182c5d37383fd978b6ec947411eaa
FG N/A 2013-03-14 11:48:49
Follow-up in trials
• Major clinical outcomes
Linked datasets GP to confirm: prospective, randomised, open, blinded endpoint
evaluation (PROBE) design Blinded review by assessors
• Persistence to treatment
• Additional data may be collected
QoL+FEV1 with eDiary in eLung Blood test for genetic analyses at month 3
Characteristics of Retropro and non-trial participants
Retropro participants
(N=301)
Other statin starters in
recruiting practices#
(N=1163)
Statin starters in non-
recruiting practices#
(N=26898)
Gender woman 105 (34.9%) 525 (45.1%) 12691 (47%)
Mean age (years) 67.0 63.1 62.5
Region England 203 (67.4%) 905 (77.8%) 23846 (88.4%)
Scotland 98 (32.6%) 258 (22.2%) 3143 (11.6%)
QRISK2 10-year CVD risk < 5% 1 (0.3%) 62 (5.3%) 2121 (7.9%)
5-10% 5 (1.7%) 152 (13.1%) 3659 (13.6%)
10-15% 25 (8.3%) 170 (14.6%) 3874 (14.4%)
15-20% 55 (18.3%) 194 (16.7%) 3974 (14.7%)
20+ % 215 (71.4%) 585 (50.3%) 13361 (49.5%)
Positive factors influencing the GP decision on participation in eLung
1. Consistency with local prescribing guidance (N=18)2. Relevance of topic (N=14)3. Feasibility to recruit (N=13)
- Number of patients (N=9)- Minimal impact of rescue packs (N=4)
4. Computer-based pop up alerts (N=10)5. Hot recruitment method (N=6)6. Demands of study design on GP time (N=4)7. Benefit to patients (N=3)
Negative factors influencing the GP decision on participation in eLung
1. Feasibility to recruit (N=10)– High impact of rescue packs (N=4)– Patient views on ‘no antibiotics’ (N=4)– Number of patients (N=2)– Lack of referrals from other GPs (N=1)
2. Hot recruitment method (N=9)3. Relevance of topic (N=4)4. Time to conduct study at point of invitation (N=4)5. Demands of study design on GP time (N=4)6. Computer-based pop up alerts (N=3)7. Consistency with local prescribing guidance (N=2)
Most important factors influencing the GP decision to accept or decline
Outer circle: Generic factors for any research project Inner circle: Project-specific factors
Qualitative feedback of eLung participants
• The main reason for agreeing to take part in eLung was the hope that it might improve their own health (6/10) or other people’s health in the future (4/10). Seven of the ten patients cited their excellent doctor-patient relationship as a key influencing factor in their decision-making process and their trust that it will be in their best interest to participate in the study if their GP has asked them
• All 10 patients considered use of EHRs to collect trial outcome data to be acceptable. This included one patient who thought routine medical records would provide more useful information than patient reported information and one patient who preferred not to complete too many questionnaires due to suffering from arthritis. All patients preferred the data to be anonymised for use by a research team and one patient expressed their explicit support for longitudinal data analysis if the data would not be used for any other purpose
Medical records: please release them, let them go• With the voluntary releasing of medical records, people
can play as active a role in healthcare as they do in politics• Absurdly, it's easier to choose arbitrary treatments for
everyone, than to suggest one treatment for half the people who need it and another treatment for the other half
• The only thing to change would be that an arbitrary decision about how to treat you would be based on a random allocation instead. It's cheap, easy, and a great way to see uncertainties over treatments resolved.
Cluster Randomised Trials and Electronic Patient Records
Data Quality of routinely collected data – different views
• Only prospectively collected data are valid: site visits / study-specific CRFs / central adjudication of outcomes
• Gold-standard approach + validation• ‘Pragmatic’ approach: mixture of statistical techniques, clinical review of
potential cases and confirmation of known associations• Big Data: lots of data with application of statistical models
Þ Simple outcomes!Þ Duration of follow-up / statistical power / costs also affect data quality Þ High quality of data recording may be reached when the conclusion is no
different than if all of these elements had been recorded without error [innovative approaches to clinical trials IOM]
Discussion
• EHR point-of-care trials are feasible but complexity of trial approvals
• Randomisation to address uncertainty should be a matter of routine
• Farr could be ideal place to further expand this method