Brown
et al. Emerg Themes Epidemiol (2016) 13:11
DOI 10.1186/s12982-016-0053-z
RESEARCH ARTICLE
Comparison ofdementia recorded
inroutinely collected hospital admission data
inEngland withdementia recorded inprimary
care
Anna Brown
1
, Oksana Kirichek
1
, Angela Balkwill
1
, Gillian Reeves
1
, Valerie Beral
1
, Cathie Sudlow
2
, John Gallacher
3
and Jane Green
1*
Abstract
Background: Electronic linkage of UK cohorts to routinely collected National Health Service (NHS) records provides
virtually complete follow-up for cause-specific hospital admissions and deaths. The reliability of dementia diagnoses
recorded in NHS hospital data is not well documented.
Methods: For a sample of Million Women Study participants in England we compared dementia recorded in rou-
tinely collected NHS hospital data (Hospital Episode Statistics: HES) with dementia recorded in two separate sources
of primary care information: a primary care database [Clinical Practice Research Datalink (CPRD), n = 340] and a survey
of study participants’ General Practitioners (GPs, n = 244).
Results: Dementia recorded in HES fully agreed both with CPRD and with GP survey data for 85% of women; it
did not agree for 1 and 4%, respectively. Agreement was uncertain for the remaining 14 and 11%, respectively; and
among those classified as having uncertain agreement in CPRD, non-specific terms compatible with dementia, such
as ‘memory loss’, were recorded in the CPRD database for 79% of the women. Agreement was significantly better
(p < 0.05 for all comparisons) for women with HES diagnoses for Alzheimer’s disease (95 and 94% agreement with any
dementia for CPRD and GP survey, respectively) and for vascular dementia (88 and 88%, respectively) than for women
with a record only of dementia not otherwise specified (70 and 72%, respectively). Dementia in the same woman was
first mentioned an average 1.6 (SD 2.6) years earlier in primary care (CPRD) than in hospital (HES) data. Age-specific
rates for dementia based on the hospital admission data were lower than the rates based on the primary care data,
but were similar if the delay in recording in HES was taken into account.
Conclusions: Dementia recorded in routinely collected NHS hospital admission data for women in England agrees
well with primary care records of dementia assessed separately from two different sources, and is sufficiently reliable
for epidemiological research.
Keywords: Dementia, Hospital Episode Statistics, Clinical Practice Research Datalink, Electronic health record, Cohort
studies
© The Author(s) 2016. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium,
provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license,
and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/
publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Open Access
Emerging Themes in
Epidemiology
*Correspondence: jane.green@ceu.ox.ac.uk
1
Cancer Epidemiology Unit, Nuffield Department of Population Health,
University of Oxford, Richard Doll Building, Roosevelt Drive, Oxford OX3
7LF, UK
Full list of author information is available at the end of the article
Page 2 of 9
Brown
et al. Emerg Themes Epidemiol (2016) 13:11
Background
Dementia is known to have a long pre-clinical phase [1,
2]. Large prospective cohort studies with long-term fol
-
low-up through linkage to routinely-collected hospital
admissions records provide important opportunities for
epidemiological investigations of dementia. e reliabil
-
ity of diagnoses of dementia in hospital data is, however,
not well documented.
In the Million Women Study cohort, virtually complete
follow-up for hospital admissions has been established by
record linkage to routinely collected National Health Ser
-
vice (NHS) databases in England (Hospital Episode Sta-
tistics, HES) and Scotland (Scottish Morbidity Records).
e linked hospital records contain coded diagnostic
information for all inpatient and day-case admissions,
and have been shown in this cohort to be reliable for
ascertainment of vascular disease [3]. Primary care data,
which is the most comprehensive single source of NHS
information on consultations, prescriptions, diagnoses,
treatments and referrals is held by each individual’s Gen
-
eral Practitioner. Over 99% of the UK population is reg-
istered with a GP in the NHS [4]. e Clinical Practice
Research Datalink (CPRD) has for many years collected
coded information from GPs on diagnoses, prescriptions
and other factors in primary care, with active coverage of
around 7% of the UK population [5, 6].
For a sample of Million Women Study participants in
England, we aimed to compare information on dementia
recorded in hospital admission data (HES) with informa
-
tion on dementia obtained from two different sources of
primary care data: (1) through linkage to coded CPRD
records; and (2) postal survey information from a sample
of study participants’ GPs.
Methods
e Million Women Study (www.millionwomenstudy.
org) has been described elsewhere [7, 8]. Between 1996
and 2001, over 1.3 million UK women aged 50–64years
were recruited through NHS breast screening pro
-
grammes in England and Scotland. Women in the study
gave written consent to follow-up through their NHS
records. Linkages to routinely collected NHS records
are done by matching women using their unique NHS
number, together with other identifying details includ
-
ing date of birth and postcode. Follow-up for deaths is up
to 31/12/2014 and, at that time, only 1% had been lost to
follow up.
Electronically linked hospital admissions data from
HES for the period 1 April 1997–31 March 2011 for
the 1.25 million women recruited in England were pro
-
vided by the Health and Social Care Information Cen-
tre (HSCIC) [9]. e HES records include admission
and discharge dates and coded diagnostic data for any
number of clinical conditions. Diagnostic data are rou
-
tinely extracted from hospital medical records and coded
by trained NHS clinical coders using the 10th Revision of
the International Classification of Diseases (ICD-10 [10]).
For this study, dementia in HES records and in death
certificates was defined as any of the following ICD-10
codes: E512, F00, F01, F02, F03, F10.6, F10.7, G30, or
G31.0. Some analyses were restricted to codes for Alz
-
heimer’s dementia (ICD-10: F00, G30), vascular demen-
tia (ICD10: F01) and dementia, not otherwise specified
(NOS; ICD 10: F03).
e CPRD is a computerised UK research data
-
base containing linked anonymised patient records for
patients registered with an NHS GP. Active coverage is
around 7% of the UK population, with research-usea
-
ble data available for some 11m people [5, 6]. Records
are coded by the individual’s GP using the Read code
system. e database consists of longitudinal medical
records with varying periods of observation, depending
on when each individual joins or leaves a GP who con
-
tributes data to CPRD. Linked coded CPRD records for
Million Women Study participants for the period 1 Janu
-
ary 1990–31 December 2012 were provided by the CPRD
division of the Medicines and Healthcare products Regu
-
latory Agency (MHRA), with data linkage performed by
HSCIC. Dementia in CPRD was defined here as any of
97 specific Read clinical codes and/or as a code for a drug
specifically prescribed for dementia, i.e. donepezil, galan
-
tamine, memantine and rivastigmine (Additional file1:
Code list 1).
A further 92 Read codes (Additional file1: Code list 2)
that we considered compatible with, but not sufficient to
define, dementia in CPRD records (e.g. codes for mem
-
ory loss, or for assessment of cognitive function) were
used to investigate further cases where there was uncer
-
tain agreement (neither definite agreement nor definite
disagreement; see later) between HES and CPRD records
of dementia as defined above.
In the postal survey of GPs we wrote asking for infor
-
mation about 333 study participants with a HES record
of dementia before March 2008, and about 1004 study
participants without a HES record of dementia by March
2008. GPs were selected to ensure a broad geographi
-
cal coverage across England and, in these areas, random
samples of women were selected for study. GPs were
asked to complete a short questionnaire and to provide
copies of relevant documents, such as letters from hos
-
pital clinics. e questionnaire asked GPs to confirm the
hospital admission diagnosis of dementia (Alzheimer’s,
vascular, or other); to report that they had no record of
such a diagnosis; or to state if they were unable to com
-
ment, for example because of incomplete or unavailable
records.
Page 3 of 9
Brown
et al. Emerg Themes Epidemiol (2016) 13:11
Analyses
For comparisons with information from CPRD and from
GPs, diagnoses of dementia in HES were classified as
fully agreeing (evidence in primary care records to con
-
firm a diagnosis of dementia, of any type); not agreeing
(clear evidence in primary care records against a diagno
-
sis of dementia), or of uncertain agreement (neither clear
agreement nor disagreement) with primary care records.
Agreement was assessed independently by at least two
researchers (J.Gr. and V.B.) and discrepancies resolved by
discussion. Where agreement with primary care data was
uncertain, all available sources of additional information
were used; and dementia mentioned on a death certifi
-
cate was taken to confirm a HES record of dementia (and
classified as fully agreed).
As the periods of observation in CPRD and HES differ,
comparison of dementia recorded in the two databases
was restricted to women with overlapping observation
periods: the observation period in CPRD was required to
cover at least 12months before and 12months after the
first HES record of dementia. For these women, all availa
-
ble CPRD records between 1.1.1990 and 31.12.2012 were
examined for dementia diagnoses.
e first mention of dementia is likely to be in primary
care rather than in hospital admissions records. To esti
-
mate the time lag we calculated the difference between
the date of first mention of dementia in CPRD and first
mention in HES for women who had a record in both.
Age-specific rates for dementia were estimated using
HES and CPRD data for the 8% of the cohort linked to
CPRD. e CPRD rates used the specified periods of
observation in CPRD from 1 January 1990 up to the first
mention of dementia or to 31 December 2012, which
-
ever came first. e HES rates were calculated from the
date of entry into the cohort, up to whichever came first
out of the first mention of dementia, death or 31 March
2011 (the last date of complete HES data). In a sensitiv
-
ity analysis, age-specific rates using HES data were esti-
mated assuming that dementia had been diagnosed
1.6 years before the first mention of dementia in the
hospital records (the time difference between first men
-
tion of dementia in CPRD and first mention in HES, as
described above).
Results
Figure1 summarises the study design and the number of
women in each group.
HES hospital admission data were available for
1,248,973 Million Women Study participants recruited in
England. Linked CPRD primary care data were available
for 102,076 (8%) of the study participants who also had
HES data, among whom 340 women had a HES dementia
1,248,973 Million Women Study parcipants recruited in
England and linked to rounely collected NHS hospital
admission data (Hospital Episode Stascs, HES)
102,076 women linked to primary
care database (Clinical
Pracce Research Datalink,
CPRD)
819 haddemena first recorded
in HES at ages 55-79years before
31/3/2011 (Table 3)
340* women with HESrecord of
demena and overlapping
period of observaon in CPRD
hospital admission-coded
primary care (CPRD)
comparison (Table 1)
244* women with HES record of
demena and useable
informaon provided by GP
hospital admission-GP survey
comparison (Table 2)
333 women with a HES record of
demena selected at random
survey quesonnaire posted to
each woman’sgeneral praconer
(GP) asking about demena
1004 women withouta HES
record of demenaselected
at random
survey quesonnaire posted to
each woman’sgeneral praconer
(GP) asking about demena
866women with no HES record of
demenaand useable
informaon provided by GP
hospital admission-GP survey
comparison of women with
no HES demena (see text)
Fig. 1 Million Women Study participants included in data comparisons. * indicates by chance, three women were included both in the coded
primary care (CPRD) and in the GP survey comparison
Page 4 of 9
Brown
et al. Emerg Themes Epidemiol (2016) 13:11
code and an overlapping period of observation in CPRD
(Fig.1).
Results of comparisons with CPRD records for these
340 women with dementia coded in HES are shown
in Table1. HES diagnoses of dementia fully agreed for
288 (85%, 95% CI 80–88%) women (278 agreed with
the CPRD codes listed in Additional file1: Code list 1
and another ten had dementia coded as cause of death).
Agreement was greatest in women with more than one
HES admission mentioning dementia (92%, 89–97%).
For only four women (1%) did CPRD codes definitely
disagree with the HES code at the time of hospital admis
-
sion, e.g. the CPRD code showed an acute confusional
state associated with sepsis. For the remaining 48 (14%)
women agreement was uncertain; although 79% of them
(38/48) had one or more of the dementia-compatible
codes in CPRD listed in Additional file 1: Code list 2,
such as memory loss or confusion.
Table 1 also shows the results of the comparison
between type of dementia coded in hospital admissions
data and mention of dementia (of any type) in CPRD.
Agreement was significantly greater (p < 0.05 for all
comparisons) for specific HES diagnoses of Alzheimer’s
disease (95%, 90–97%) or of vascular dementia (88%,
79–94%) than for a HES diagnosis of dementia, not oth
-
erwise specified (70%, 60–78%). Agreement with a CPRD
record of any dementia was lowest, at 60%, for women
with just one HES record of dementia, not otherwise
specified.
In the comparison of HES records with information
provided directly by GPs, informative responses were
received for 73% (244/333) of the sample of women with
a HES record of dementia (Fig.1). No reply was received
from the GP for 35 women (11%) and for 54 women
(16%) the GP returned the survey form but without use
-
able information on dementia. In most such cases the GP
commented that the patient had died or moved, and the
practice no longer had access to full records.
Comparisons with HES data are shown in Table2, using
the same format as in Table1. HES diagnosis of demen
-
tia fully agreed for 208 (85%, 95% CI 80–89%) women
(204 confirmed by GPs, and a further four by death cer
-
tificates). As found in comparisons with CPRD records,
agreement with GP reports was greatest in women with
more than one HES record of dementia (94%, 90–98%).
Only 9 (4%) disagreed, where the GP provided evidence
that the woman did not have dementia. ese included,
for example, a diagnosis of dementia suspected at time of
HES admission, but not confirmed on subsequent inves
-
tigation; other women had diagnoses such as encepha-
litis, pneumonia or urinary tract infection with acute
confusional state. For 27 (11%) the diagnostic compari
-
son remained uncertain after review of all available data.
e uncertain group is largely comprised of those whose
GP did not confirm the HES diagnosis, but where it was
not clear if the GP still had access to relevant records (if
a woman dies or leaves the GP practice, including some
moves to institutional care, the primary care records gen
-
erally move with her).
Table2 also shows the results of comparisons by type
of dementia recorded in HES. As in Table1, agreement
between HES records and GP reports (of any dementia)
was significantly greater (p< 0.05 for all comparisons)
for a specific HES diagnosis of Alzheimer’s disease (94%,
89–97%) or of vascular dementia (88%, 76–94%) than for
dementia, not otherwise specified (72%, 62–81%). For
women with just one HES record of dementia, not other
-
wise specified, agreement was 63%.
GPs were also asked whether any of a randomly
selected sample of 1004 women without a hospital
admissions record of dementia (Fig. 1) had dementia.
Informative replies were received for 86% (866/1004)
of women and only one (0.1%) was reported by her GP
to have dementia. No reply was received for 68 women
(7%), and replies with no useable data on dementia diag
-
nosis for the remaining 70 women (7%).
Table3 shows estimated age-specific rates for demen
-
tia per 1000 women per year in 5year age groups from
55–59 to 75–79 for the 102,076 study participants linked
both to HES and to CPRD. ere were insufficient data in
the cohort to estimate rates at other ages. Dementia rates
are strongly dependent on age: based on CPRD data,
rates increased 80-fold between ages 55–59 and 75–79,
from 0.1 to 8 per 1000 per year (Table3A). e CPRD
age-specific rates for dementia in this cohort are similar
to other published rates using CPRD data [11].
Age-specific rates calculated using HES data (Table3B)
are, as expected, lower than the CPRD rates (Table3A).
However, among the women with dementia recorded
both in the HES and in the CPRD data, the first mention
of dementia was an average of 1.6 (SD 2.6)years earlier in
CPRD than in HES. In a sensitivity analysis we assumed
that, for women with a HES record of dementia, the
dementia had been diagnosed 1.6years earlier; under this
assumption age-specific rates are similar to those based
on CPRD data (Table3C; Fig.2).
Of the 340 women with both a HES and a CPRD record
of dementia, 64% (216) died before 31 December 2014
and dementia was mentioned on the death certificate
for 37% of them (71 as the underlying cause of death and
eight as a contributory cause of death).
Discussion
Our results suggest that dementia recorded in routinely-
collected coded NHS hospital admission data in Eng
-
land agrees well with dementia recorded in primary care.
Page 5 of 9
Brown
et al. Emerg Themes Epidemiol (2016) 13:11
Table 1 Comparison ofdementia recorded inHospital Episode Statistics (HES) withinformation fromthe primary care Clinical Practice Research Datalink (CPRD)
a
All comparisons are with CPRD record of any dementia diagnosis
Women withany HES record
fordementia
N (% oftotal)
Women withHES record forAlzhei‑
mer’s dementia
N (% oftotal)
Women withHES record forvascu‑
lar dementia
N (% oftotal)
Women withHES records only
fordementia, not otherwise
specied
N (% oftotal)
Number ofadmissions
mentioning dementia
Number ofadmissions
mentioning dementia
Number ofadmissions
mentioning dementia
Number ofadmis‑
sions mentioning
dementia
All 1 2+ All 1 2+ All 1 2+ All 1 2+
CPRD diagnosis fully agreed
a
288 (85%) 93 (73%) 195 (92%) 155 (95%) 41 (93%) 114 (95%) 66 (88%) 14 (78%) 51 (90%) 65 (70%) 33 (60%) 32 (84%)
Diagnostic agreement uncertain 48 (14%) 30 (24%) 18 (8%) 9 (5%) 3 (7%) 6 (5%) 9 (12%) 4 (22%) 6 (10%) 24 (26%) 18 (33%) 6 (16%)
CPRD diagnosis disagreed 4 (1%) 4 (3%) 0 0 0 0 0 0 0 4 (4%) 4 (7%) 0
Total (=100%) 340 127 213 164 44 120 75 18 57 93 55 38