GPRD 2010

Using electronic patient level data in Primary Care
Research: stories from our General Practice
Research Database experience
Evangelos Kontopantelis1
Tim Doran1
Stephen Campbell1
Jose Valderas2
Martin Roland3
Mark Harrison1
David Reeves1
1
National Primary Care Research and Development Centre
University of Manchester
2
Department of Primary Health Care, University of Oxford
3
General Practice and Primary Care Research Unit, University of Cambridge
NPCRDC, 15th June 2010
Kontopantelis (NPCRDC) GPRD & patient level data 15 June 2010 1 / 39

Outline
1 Background
Electronic patient records (EPR)
Quality and Outcomes Framework
Primary Care databases
2 GPRD
Data details
Extracting the information
3 Our research
Synopsis
Disease prevalences
Clinical quality indicator performance
Exception reporting
Co-morbidity and workload
4 Summary

Background Electronic patient records (EPR)
Times are changing!
Tarzan no want computer

Advantages of using EPRs
...and disadvantages
They have the potential to bring huge beneﬁts to patients.
can speed up clinical communication.
reduce the number of errors.
assist doctors in diagnosis and treatment
Quality of research can be augmented with the added level of detail.
patient level factors can be taken into account.
subgroup analyses are made easy.
statistically, analyses can be more powerful.
But...
(even more) conﬁdentiality issues arise.
the structure of the data might require much work and advanced computer
skills.

What is happening in the UK
...it may take a while though
Implementing EPR systems is one of the main aims of the 10-year
National Programme for Information Technology (NPfIT), launched in
2002.
Connecting for Health, the organisation responsible for delivering NPfIT.
The main idea is to create a shared patient record divided into two levels:
the Detailed Care Record (DCR) - held locally.
the Summary Care Record (SCR) - held nationally. Initially, it will hold only
basic info (allergies, adverse reactions & prescriptions).
Project core is the development of a national system which will involve
replacing obsolete local IT systems across the NHS and linking
up-to-date systems together.

Background Quality and Outcomes Framework
Computers and GPs DO mix!
if lucky you can still get scribbled illegible notes on your prescription
A voluntary pay-for-performance program (QOF) kicked off in 2004 with
the introduction of a new GP contract, which required practices to
become computerised.
A high percentage of practices was already computerised by then
(helped by PCTs, own initiative etc).
Initial investment £1.8 bn for 3 years (increasing GP income by up to
25%) motivated laggard practices.
Now over 99.9% of English practices are computerised and participating
in QOF (but using various systems: Emis, Seetec, ViSion etc).
READ codes - a coded thesaurus of clinical terms - enable GPs to make
effective use of the systems and leave freetext behind.

QOF details
according to the rumour mill the scheme that pays GPs to do their job...twice
General practices rewarded for achieving a set of quality targets for
patients with several chronic conditions.
Aim was to increase overall quality of care and to reduce variation in
quality between practices.
146 quality indicators.
Clinical care for 10 chronic diseases (76 indicators).
Organisation of care (56 indicators).
Additional services (10 indicators).
Patient experience (4 indicators).
Into the 7th year now (01Mar10/31Apr11); cost for the ﬁrst 6 years was
well above the estimate at £5.8 bn approx.
QOF is reviewed at least every two years.

Some of the indicators for diabetic patients.
Percentage of diabetics...
with a record of HbA1c in previous 15 months (3p).
in whom last HbA1c is ≤7.4 in previous 15m (16p).
who have a record of BP in the past 15m (3p).
in whom the last BP is ≤145/85 (17p).
with a rec of serum creatinine testing in previous 15m (3p).
who have a record of total cholesterol in previous 15m (3p).
whose last measured total cholesterol in previous 15m is ≤5mmol/l (6p).
who have had inﬂuenza immunisation in the preceding 1Sep-31Mar (3p).

Background Primary Care databases
The General Practice Research Database
GPRD - not for proﬁt
Established in 1987, with only a handful of practices.
Since 1994 owned by the Secretary of State Health.
In April 2010:
545 active practices (ViSion system only).
11.20M patients.
Sample of 100k patients can be obtained for ’free’ (MRC funds up to 50
approved academic proposals per year).
Costs vary for larger samples; our 600k patients sample cost £32,000
and is tied to a speciﬁc - albeit vague - research proposal (QOF related).
Access to the whole database is offered (ability to extract data for
approved projects) and costs £127,000 per annum.

Background Primary Care databases
The Health Improvement Network database
THIN - commercial
A collaboration between In Practice Systems Ltd (INPS) and EPIC.
In April 2010:
428 active practices (ViSion system only, 50-60% overlap with GPRD).
8.70M patients.
Usually offered under a 4-year licence which costs £119,000.
Similar to GPRD (judging by a very small sample) and they promise to
offer more patient characteristics (using the patient’s home postcode).

GPRD Data details
The single source of truth...
...not
Sorry, you are not in the database

GPRD Data details
Why we needed GPRD
Interested in longitudinal data from 1999 to 2007: 270 active practices
with GPRD in the whole period.
We set out to measure the effect of QOF on incentivised and
non-incentivised aspects of clinical quality of primary care.
We needed GPRD since there are considerable advantages over the
NHS published QOF data (QMAS):
data availability prior to the introduction of QOF
ability to construct non-incentivised indicators of quality
ability to focus on a gender or speciﬁc age groups.
Sampled 1,000 patients from each of 100 ’representative’ UK practices.

GPRD Data details
The 100,000 patients ’free’ sample
Database broken down to numerous tables, because of the volume of the
data (4GB).
Text ﬁles need to be imported into powerful analysis/database
management software.
Some of the information available:
Patient birthyear, sex, marital status, smoking/drinking status, height, weight
and BMI.
Clinical, referral, therapy, test, immunisation and consultation events.
All events are entered in codes (lookup tables available).
Everything (likely to be recorded by a GP) can be identiﬁed; provided one
knows which codes to look for and in which tables!
BUT a manual search on all the codes is not possible (the READ codes
alone are 98,031) and automated processes are required.

GPRD Data details
The main GPRD tables
and relationships between them
Event ﬁles.
Clinical: all medical history data (symptoms, signs and diagnoses).
Referral: information on patient referrals to external care centres.
Immunisation: data on immunisation records.
Therapy: data relating to all prescriptions issued by a GP.
Test: data on test records.
Lookup ﬁles.
Medical codes: READ codes, 98,031 available.
Product codes: 77,198 available.
Test codes: 304 available.

GPRD Extracting the information
Our approach
Size of the tables prohibits looking at codes one by one.
Instead we use search terms to identify potentially relevant codes in the
lookup tables and create draft lists.
Example (Search terms for diabetes)
String search in Medical codes: ’diab’ ’mell’ ’iddm’ ’niddm’.
READ code search in Medical codes ﬁle: ’C10’ ’XaFsp’.
String search in Product codes ﬁle: ’insulin’ ’sulphonylurea’
’chlorpropamide’ ’glibenclamide’.

Our approach
...continued
Clinicians go through the draft lists and select the relevant codes.
Three sets of codes are created, corresponding to:
QOF criteria.
Conservative criteria.
Speculative criteria.
Using the ﬁnalised code lists we search for events in the Clinical,
Referral, Immunisation, Therapy and Test ﬁles.
The whole process involves much work in code writing, therefore usage
of an appropriate statistical package like STATA or SAS is essential.
Diabetes code example

Moving on to the bigger sample
Once all the processes were in
place and we were ’conﬁdent’
about our estimates...
...code lists were
communicated to GPRD and
they extracted a snapsot of the
database using the speciﬁc
codes.
Final GPRD sample holds data
on 660,565 patients, from 150
’representative’ UK practices.
Power analyses on the original
sample informed our decision
on the numbers of practices
and patients required.

Our research Synopsis
It all starts with a grant...
They're harmless when they're alone but get a
bunch of them together with a research grant
and watch out

Our research Synopsis
What GPRD has done for us
after some convincing...
Estimated year on year prevalence scores for various conditions.
Generated trends for practice performance on clinical quality indicators.
Examined the effect of QOF on incentivised and non-incentivised aspects
of quality of primary care.
Investigated the reasons of exception reporting in QOF (and their timing).
Created co-morbidity mappings of the available patients.
Measured the workload associated with each condition (including
co-morbidities).

Our research Disease prevalences
Prevalence
Based on conservative criteria
0.00%
2.00%
4.00%
6.00%
8.00%
10.00%
12.00%
14.00%
16.00%
18.00%
1999/00 2000/01 2001/02 2002/03 2003/04 2004/05 2005/06 2006/07
Percentage
Prevalence
Asthma
CHD
COPD
Depression
Diabetes
Hypertension
Hypothyroidism
Osteoarthritis
Osteoporosis
Stroke

Our research Clinical quality indicator performance
QOF measurement/recording indicator example
DM11: % who have a record of BP in the previous 15 months (3p).

QOF treatment indicator example
DM18: % who have had inﬂuenza immunisation in the preceding 1 Sep-31 Mar (3p).

QOF outcome indicator example
DM6: % in whom last HbA1C is ≤7.4 in the previous 15 months (16p).

Inc vs non-inc clinical aspects of primary care
Two aspects to clinical indicators:
a disease condition (e.g. diabetes, CHD).
a care activity (e.g. inﬂuenza vaccination, BP control).
Three indicator classes, in terms of incentivisation:
(A) Condition & process incentivised within QOF (28 ind)
(B) Condition or process incentivised (13 ind)
(C) Neither condition nor process incentivised (7 ind)
Three different types of activities:
clinical processes related to measurement (PM/R).
e.g. blood pressure measurement
clinical processes related to treatment (PT).
e.g. inﬂuenza immunisation
intermediate outcome measures (I).
e.g. control of HbA1c to 7.4 or below
We end up with 48 indicators in six indicator groups.

A-PM/R: fully inc recording
process.
Hypertension & BP
A-PT: fully inc treatment process.
COPD & inﬂuenza immunisation
A-I: fully incentivised
indermediate outcome.
DM & BP of 145/85 or less
B-PM/R: partially inc recording
process.
PAD & total cholesterol
B-PT: partially inc treatment
process.
CRD & inﬂuenza immunisation
C-PT: non-inc treatment process.
back pain & strong analgesics
20
30
40
50
60
70
80
90
100
%
2000/01 2001/02 2002/03 2003/04 2004/05 2005/06 2006/07
Year
A−PM/R (17) A−PT (6) A−I (5)
B−PM/R (9) B−PT (4) C−PT (7)
using group means of indicator means (by practice)
in brackets, the number of indicators in each group
Percentage scores
Indicator group performance

Short term (2004/05):
Overall, all three groups of fully incentivised indicators exhibited
performance above the pre-QOF expectation.
(from 1.1% to 38.2% with 4 smoking indicators having uplifts of over 30%).
Partially incentivised Measurement/Recording indicators demonstrated
signiﬁcantly lower than expected gains, on average.
Long term (2006/07):
Overall, the three fully incentivised groups continued to perform above the
expectation, although none exceeded 4%.
The three partially incentivised and non-incentivised groups displayed
signiﬁcantly negative uplifts, on average.

Our research Exception reporting
Exceptions
timing
Exception reporting
is considered a
safeguard against
patient
discrimination.
There is interest in...
the patients who
are excepted from
QOF.
whether practices
use ER as a
’gaming’ tool
(timing/reason).
’met exceptions’.
0200400600800
Frequency
April
04M
ay
04
June
04July
04
August04
Septem
ber04
O
ctober04
N
ovem
ber04
D
ecem
ber04
January
05
February
05
M
arch
05April
05M
ay
05
June
05July
05
August05
Septem
ber05
O
ctober05
N
ovem
ber05
D
ecem
ber05
January
06
February
06
M
arch
06April
06M
ay
06
June
06July
06
August06
Septem
ber06
O
ctober06
N
ovem
ber06
D
ecem
ber06
January
07
February
07
M
arch
07
first record of exception reporting in year, for each patient
using GPRD
Exceptions over time, ages 60+

Exceptions
timing
0
10
20
30
40
50
60
70
80
90
100
%
01ja
n2006
01apr2006
01ju
l2
006
01oct2006
01ja
n2007
01apr2007
Date
Reported achievement, excluding all exceptions IQR
Total exception reporting rate IQR
% of unmet exceptions in exception total
Excluding all exceptions
RA & ER for STROKE8
STROKE8: % of stroke patients with last chol meas ≤5mmol/l

Exceptions
QOF conditions multimorbidity
Multimorbidity count No Yes Total % ER No Yes Total % ER No Yes Total % ER
0 0 934 934 ‐ 0 624 624 ‐ 0 1,085 1,085 ‐
1 70,177 1,843 72,020 2.6% 72,411 2,216 74,627 3.0% 71,278 2,889 74,167 3.9%
2 22,910 1,330 24,240 5.5% 24,239 1,558 25,797 6.0% 25,896 1,815 27,711 6.5%
3 10,003 765 10,768 7.1% 10,514 892 11,406 7.8% 11,987 1,049 13,036 8.0%
4 4,084 448 4,532 9.9% 4,345 487 4,832 10.1% 5,491 627 6,118 10.2%
5 1,421 232 1,653 14.0% 1,581 223 1,804 12.4% 2,204 296 2,500 11.8%
6 450 82 532 15.4% 498 100 598 16.7% 845 144 989 14.6%
7 103 27 130 20.8% 155 28 183 15.3% 298 69 367 18.8%
8 28 6 34 17.6% 31 9 40 22.5% 75 13 88 14.8%
9 3 1 4 25.0% 4 2 6 33.3% 20 5 25 20.0%
10 ‐ 2 0 2 0.0% 5 1 6 16.7%
Total 109,179 5,668 114,847 4.9% 113,780 6,139 119,919 5.1% 118,099 7,993 126,092 6.3%
2006/072004/05 2005/06
Exception reported

Exceptions
reasons and ’met exceptions’
40
50
60
70
80
90
100
Percentages of 'met exceptions' in total / by reason of exception
0
10
20
30
40
50
60
70
80
90
100
Percentages of 'met exceptions' in total / by reason of exception
no consent/refusal unsuitable/contraindicated

Our research Co-morbidity and workload
Co-morbidity
co-morbidites Thursday October 1 12:14:12 2009 Page 1
___ ____ ____ ____ ____tm
/__ / ____/ / ____/
___/ / /___/ / /___/
Statistics/Data Analysis
User: Evan
Project: GPRD
conservative estimates (where applicable) and QOF year 2006/07
CHD DM Strk 45+ HRT Lith Dem Depr Asth COPD PAD
HT 17.6 17.2 8.4 94.2 3.1 0.2 1.2 22.8 9.0 4.4 2.4
56.2 56.4 59.0 29.3 19.9 20.6 40.0 19.8 17.7 40.1 58.2
CHD 20.3 12.4 98.1 1.9 0.1 1.7 24.9 10.5 7.5 4.6
20.8 27.4 9.5 3.7 5.3 17.5 6.8 6.5 21.3 35.1
DM 8.5 88.6 1.7 0.3 1.2 24.0 9.4 4.4 3.1
18.2 8.4 3.4 9.7 12.3 6.3 5.6 12.1 23.4
Stroke 97.5 1.3 0.1 4.9 27.6 9.0 7.1 4.7
4.3 1.2 2.0 22.6 3.4 2.5 9.1 16.3
45orover 4.4 0.2 0.9 21.5 7.5 3.4 1.2
91.3 78.5 99.7 59.8 47.3 97.9 96.7
HRT 0.3 0.1 38.7 11.2 2.0 0.5
4.9 0.5 5.2 3.4 2.8 1.8
Lithium_therapy 2.4 76.9 9.7 2.2 0.8
0.6 0.5 0.2 0.2 0.2
Dementia 32.0 5.2 4.7 2.9
0.8 0.3 1.3 2.2
Depression 9.9 2.9 1.0
22.6 30.7 28.3
Asthma 9.3 0.8
42.8 10.2
COPD 4.4
12.0
first row for each condition/activity: denominator indicated by row
second row for each condition/activity: denominator indicated by column
i.e. first observation indicates the % of patients with HyperTension, who also have CHD

Workload - consultation type
number of appointments
Freq. Percent Freq. Percent Freq. Percent
Acute visit 1,955 0.35 1,889 0.31 1,771 0.27
Administration 88,119 15.97 102,014 16.69 119,900 18.24
Casualty Attendance 1,021 0.19 985 0.16 897 0.14
Clinic 22,694 4.11 23,901 3.91 24,506 3.73
Discharge details 1,614 0.29 1,476 0.24 1,580 0.24
Emergency Consultation 940 0.17 759 0.12 731 0.11
Follow‐up/routine visit 1,226 0.22 864 0.14 746 0.11
Letter from Outpatients 31,075 5.63 37,500 6.14 39,482 6.01
Mail from patient 482 0.09 954 0.16 953 0.14
Mail to patient 477 0.09 757 0.12 711 0.11
Night visit, Deputising service 18 0 15 0 12 0
Night visit, Local rota 17 0 4 0 2 0
Other 89,128 16.15 103,406 16.92 112,331 17.09
Out of hours, Non Practice 854 0.15 994 0.16 1,289 0.2
Out of hours, Practice 58 0.01 58 0.01 115 0.02
Repeat Issue 109,711 19.88 110,196 18.03 107,029 16.28
Results recording 66,433 12.04 78,722 12.88 88,764 13.51
Surgery consultation 122,964 22.29 131,625 21.54 136,425 20.76
Telephone call from a patient 6,664 1.21 7,552 1.24 8,026 1.22
Telephone call to a patient 3,443 0.62 3,977 0.65 4,130 0.63
Third Party Consultation 2,841 0.51 3,456 0.57 7,861 1.2
Total 551,734 100 611,104 100 657,261 100
Freq. Percent Freq. Percent Freq. Percent
Acupuncturist 0 0 359 0.06 467 0.08
2006/07
2006/07
2005/06
2005/06
2004/05
2004/05

Workload - health worker
number of appointments
Freq. Percent Freq. Percent Freq. Percent Freq. Percent Freq. Percent Freq. Percent
ncturist 0 0 359 0.06 467 0.08 Hospital Nurse 22 0 29 0.01 21 0
strator 39,242 7.71 49,713 8.71 53,876 8.71 Interpreter/Link Worker 7 0 7 0 20 0
nt 6,614 1.3 7,754 1.36 8,779 1.42 Locum 10,816 2.13 10,310 1.81 10,550 1.71
ate 4,026 0.79 5,127 0.9 3,652 0.59 Maintenance staff 55 0.01 8 0 15 0
ss Manager 923 0.18 1,020 0.18 1,417 0.23 Midwife 167 0.03 282 0.05 349 0.06
odist 189 0.04 92 0.02 59 0.01 Other Health Care Professional 24,900 4.9 31,283 5.48 36,573 5.91
ercial Deputising service 22 0 18 0 35 0.01 Partner 93,973 18.47 104,501 18.32 115,918 18.74
unity Nurse 1,651 0.32 1,930 0.34 2,569 0.42 Pharmacist 2,157 0.42 2,195 0.38 2,040 0.33
unity Psychiatric Nurse 209 0.04 133 0.02 133 0.02 Physiotherapist 299 0.06 349 0.06 240 0.04
ter Manager 3,125 0.61 3,964 0.69 4,933 0.8 Practice Manager 20,396 4.01 21,616 3.79 21,337 3.45
tant 414 0.08 342 0.06 444 0.07 Practice Nurse 54,697 10.75 60,036 10.52 61,118 9.88
llor 833 0.16 779 0.14 765 0.12 Receptionist 150,926 29.67 165,525 29.02 187,014 30.23
an 263 0.05 291 0.05 277 0.04 School Nurse 166 0.03 136 0.02 150 0.02
ser 12,444 2.45 15,482 2.71 15,779 2.55 Secretary 31,762 6.24 31,376 5.5 30,654 4.96
Manager 695 0.14 189 0.03 58 0.01 Senior Partner 37,152 7.3 42,289 7.41 43,544 7.04
istrar 8,116 1.6 10,374 1.82 12,385 2 Social Worker 2 0 0 0 3 0
Education Officer 34 0.01 26 0 49 0.01 Sole Practitioner 2,208 0.43 2,716 0.48 3,131 0.51
Visitor 176 0.03 221 0.04 204 0.03 Total 508,681 100 570,472 100 618,558 100
2004/05 2005/06 2006/07 2004/05 2005/06 2006/07

Summary
Overview
So, GPRD then?
Advantages...
Patient level data; breakdown by age, sex, year of diagnosis etc
Data on many time points are available.
We are able to extract/create data not available anywhere else.
Available now (with trustworthy data for a few years back).
Disadvantages...
Too much work!
Not cheap.
Only a sample of all English practices participates.
Practices are anonymised and controlling analyses for practice
characteristics can be a struggle.
Absolute reliance on codes and GPs getting them right...
Quality of data before the introduction of QOF is questionable.
Too much work! (but you become everyone’s new best friend...)

Summary
Still a long way to go!
Something goes around something but
that's as far as I've got...
Comments: e.kontopantelis@manchester.ac.uk

GPRD 2010

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Was ist angesagt?

Was ist angesagt? (20)

Andere mochten auch

Andere mochten auch (17)

Ähnlich wie GPRD 2010

Ähnlich wie GPRD 2010 (20)

Mehr von Evangelos Kontopantelis

Mehr von Evangelos Kontopantelis (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

GPRD 2010