Journal of Hospital Infection 79 (2011) 231e235

Available online at www.sciencedirect.com

Journal of Hospital Infection

journal homepage: www.elsevierhealth.com/journals/jhin

Data linkage between existing healthcare databases to support hospital

epidemiology
L. García Álvarez a, *, P. Aylin a, J. Tian a, C. King a, M. Catchpole a, S. Hassall b, K. Whittaker-Axon b,
A. Holmes a, b
a
National Centre for Infection Prevention and Management, Imperial College, London, UK
b
Imperial College Healthcare NHS Trust, London, UK

a r t i c l e i n f o s u m m a r y

Article history: Enhancing the use of existing datasets within acute hospitals will greatly facilitate hospital
Received 27 January 2011
epidemiology, surveillance, the monitoring of a variety of processes, outcomes and risk factors,
Accepted 2 June 2011
by J.A. Child and the provision of alert systems. Multiple overlapping data systems exist within National
Available online 24 August 2011 Health Service (NHS) hospitals in the UK, and many duplicate data recordings take place
because of the lack of linkage and interfaces. This results in hospital-collected data not being
Keywords: used efficiently. The objective was to create an inventory of all existing systems, including
Alert systems administrative, management, human resources, microbiology, patient care and other plat-
Healthcare-associated infections
forms, to describe the data architecture that could contribute valuable information for
Influenza
Record linkage a hospital epidemiology unit. These datasets were investigated as to how they could be used to
Risk predictive models generate surveillance data, key performance indicators and risk information that could be
Syndromic surveillance shared at board, clinical programme group, specialty and ward level. An example of an output
of this integrated data platform and its application in influenza resilience planning and
responsiveness is described. The development of metrics for staff absence and staffing levels
may also be used as key indicators for risk-monitoring for infection prevention. This work
demonstrates the value of such a data inventory and linkage and the importance of more
sophisticated uses of existing NHS data, and innovative collaborative approaches to support
clinical care, quality improvement, surveillance, emergency planning and research.
Ó 2011 The Healthcare Infection Society. Published by Elsevier Ltd. All rights reserved.

Introduction Promoting transparency and accountability of hospitals are top

Healthcare-associated infections (HAI) represent a huge health
and economic burden to the public healthcare system (National
Health Service, NHS) and the Department of Health in the UK. The
latest national prevalence survey of HAI carried out in the UK
estimated the prevalence of HAI in England to be 8.19%, costing the
NHS more than £1 billion per year.1,2 The control of HAI has been
a national priority over the last decade, with the Department of
Health launching a variety of initiatives focused on reducing the
number of meticillin-resistant Staphylococcus aureus (MRSA)
bloodstream infections (BSI) and Clostridium difficile infections.3
* Corresponding author. Address: Imperial College, Charing Cross Campus,
3rd Floor, Reynolds Building, St Dunstan’s Road, London W6 8RP, UK. Tel.: þ44 (0)
207 5949988; fax: þ44 (0)207 5940854.
E-mail address: l.garcia-alvarez@imperial.ac.uk (L. García Álvarez).
items on the political agenda, with further recent changes intro-
duced in infection reporting (e.g. weekly rather than monthly
reporting of MRSA BSI and C. difficile cases per individual hospital).4
The structure of the NHS is rapidly changing, as is our under-
standing of infection prevention in healthcare. This requires a rapid
adaptation of information management and technology to support
comprehensive healthcare delivery and to enhance hospital
surveillance. Data duplication due to overlapping recording
systems within the NHS represents a waste of resources, and
systems are not exploited to their maximum potential.5 Linkage of
different sources of hospital data provide a platform to tackle
infection within acute care settings at local and potentially national
level. This approach is likely not only to enhance infection
prevention and control intelligence, but to allow surveillance of HAI
beyond the use of data sources collected by infection control
services. For example, the staff:patient ratio in wards, an important

0195-6701/$ e see front matter Ó 2011 The Healthcare Infection Society. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.jhin.2011.06.016
232 L. García Álvarez et al. / Journal of Hospital Infection 79 (2011) 231e235
risk factor for HAI, particularly in intensive care units (ICUs) and
high dependency units (HDUs), can be closely monitored and alert
systems and key performance indicators developed.6e8 Further-
more, the infrastructure and methodology can be adjusted to
support any future administrative and structural changes likely to
happen within the NHS.
This paper presents an innovative approach in which disparate
hospital datasets held within one of the largest NHS Trusts in the UK
are being integrated into a unique linked system to provide
a powerful tool for infection prevention and control. The uses of these
linked sets of data are described in hospital epidemiology, syndromic
surveillance, workforce monitoring, reporting and planning, and
applied research. The scheme’s immediate testing and application for
H1N1 influenza is illustrated, and further applications are outlined for
monitoring different presentation patterns, emerging threats, and
unintended consequences resulting from changes in policy, practice
and activity across healthcare settings.
Methods
Data sources
Imperial College Healthcare NHS Trust is comprised of five
hospitals and three microbiology laboratories, which have recently
become a single central laboratory. Large amounts of data are
routinely collected and stored in local hospital databases main-
tained by different clinical and non-clinical departments within the
Trust. There are up to 750 official databases, of which more than
250 are stand-alone. In addition, there is an unknown number of
other unofficial in-house departmental data systems. Examples of
the wide range of information contained within these Trust
systems include: inpatient and outpatient records, infection control
data, antimicrobial usage data and antibiotic susceptibility test
results, laboratory and imaging records, accident and emergency
(A&E) data, ward records on hand hygiene, human resources data,
and environmental and cleaning information.
Data auditing and record linkage
An inventory of available information sources was constructed
and evaluated to choose systems of potential value in hospital
infection control, as well as to develop appropriate indicators and
metrics for infection and outbreaks. Staff-identifiable data were
removed. Permission was granted to use these local data from the
hospital administration and the information and communication
technologies teams.
Structure
Fields and tables were developed based on existing data, and
managed within a Microsoft Structured Query Language (SQL)
Server 2008 (Microsoft Corporation).
Information was linked on microbiological testing and drug
sensitivity tests provided by the laboratory information management
systems (Telepath and Mysis), with data retrieved from the Patient
Information Management System (PIMS) and the General Informa-
tion and Management System (GIMS), which include patient
admission, discharge, diagnosis and procedure data. Microbiological
data and PIMS data were linked to the Infection Control Nurse data-
base (ICNet) (Figure 1). ICNet receives automated alerts of positive
results from the microbiological databases thrice daily. Each positive
result is transferred to ICNet on verification, with transfer occurring
as soon as 1 h (to a maximum of 12 h) following verification,
depending on the time of the next automated alert scheduled on that
day. This allows the Trust to respond to nearly real-time diagnosis of
patients, and provides robust nursing documentation for each alerted
organism. In addition, there are non-automated real-time systems in
place where microbiologists directly inform infection control staff of
any interim results of concern.
Similarly, infection prevention audit data were incorporated,
such as staff hand hygiene and inpatient pressure ulcers reports,
from Synbiotix Ltd (http://www.synbiotixhealth.net/) into the
system and this information was linked to infection cases
(Figure 1). Synbiotix enables the organization to collect and collate
audit of practice, including environmental and care bundle
compliance data, over a total of 148 wards and units.
The system was programmed to load staff absence and staff
personnel data from human resources into the database. These data
were subsequently linked to the microbiological and patient data to
provide a variety of workforce absence metrics and infection rates
in wards. The daily suspected H1N1 patient data were stored within
a table, regularly updated and linked to inpatient information.
Stored procedures were created to report H1N1 cases among staff
and patients at clinical programme group (CPG), specialty and ward
level. Closure of bays and ward cleaning information were also
integrated and reported in conjunction with infection rates. Drug
daily doses at CPG level were calculated based on data collected by
pharmacy, and this information was subsequently linked via
a database to the infection data stored within the system (Figure 1).
It is intended in the near future to link Urgent Care (Adastra,
Advanced Health & Care, http://www.advancedcomputersoftware.
com/) and A&E (Symphony, Ascribe, http://www.ascribe.com/)
data to the operational database. Urgent Care data will be used to
complement the A&E data to assess real-time admissions on
emergencies and urgent care, as well as to monitor re-admissions,
which are significant quality indicators.
Outputs
Routine reports were developed based on the following defini-
tions for infections and clinical practices: (i) C. difficile cases:
patients who develop diarrhoea after 72 h in hospital (Trust cases)
or in the first 72 h of admission (non-Trust cases), and who have not
been diagnosed as positive for C. difficile toxin gene in the previous
28 days; (ii) MRSA BSI cases: patients who develop MRSA BSI 48 h
after hospital admission (Trust cases) or in the first 48 h of
admission (non-Trust cases); (iii) meticillin-susceptible S. aureus
(MSSA) BSI episodes: patients who develop MSSA BSI 48 h after
hospital admission; A&E specimens are not included; (iv) hand
hygiene and ‘bare below the elbows’ compliance data are received
for each specific area, and a weekly average score is calculated if
a ward submits more than 10 observations in a week; (v) MRSA
elective screening: percentage of admissions in a month where
a screening swab has been taken from the same patient at most 92
days prior to admission or on the day of admission.
Data on each MRSA BSI case and each case of C. difficile infection
in patients aged 2 years, which are reported within the Trust, are
subsequently entered into the Health Protection Agency website as
part of the national mandatory surveillance schemes in the UK.9
Development of a database for applied research
The availability of a fully integrated operational database within
the Trust allows the development of a relational research database.
This research platform was built from an anonymised extract of the
Trust database, with records from 1 April 2007 to 17 May 2010.
Five main steps were involved in the building of the research
database: (i) hardware and software purchase and installation;
(ii) database configuration and construction within a secure
network server to allow access to researchers via a secure login;
 L. García Álvarez et al. / Journal of Hospital Infection 79 (2011) 231e235 233

Local Hospital Databases

Microbiology Infection Control

Infection Control Nurse Death Certificate
Drug Sensitivity Audit Returns
(ICNet) Information
(Telepath and Mysis) (Synbiotix)

Human Resources Patient Information

(EROSTER, EHR, POST) (PIMS/GIMS) H1N1 Flu

Nurses’ absence Demographic Laboratory Result

Other absence Diagnosis / Procedures Attendance (A&E)
Personnel information Admission / Discharge Admission (inpatient)
Other information

Occupational Health Pharmacy Ward Management

Suspected H1N1 staff cases Drug Daily Doses Bay closure
Staff flu query cases Ward cleaning

WORK IN PROGRESS: Accident and Emergency (Symphony)

Urgent Care (Adastra)

Integrated Database
DATA
Development of Test Live WAREHOUSE
database database database

REPORTING
Database
Hospital real-time surveillance reports
administrator
Data transferred via email

Figure 1. Example of local data architecture and application of record linkage to develop an integrated database within Imperial College Healthcare NHS Trust. Telepath and Mysis
are laboratory information management systems; ICNet, Infection Control Network software; PIMS, Patient Information Management System; GIMS, General Information and
Management System; EROSTER, EHR and POST are human resources data systems.

(iii) application for ethical and National Information Governance hospital data into the integrated database. The systems are
Board research approvals to use patient data at the individual level validated on a weekly basis.
without prior consent; (iv) data anonymization to preserve patient
and staff identity; and (v) development of database technical Results
specifications.
Description and uses of the hospital operational database

Quality control and system maintenance
Existing data governance schemes within the Trust are
responsible for ensuring data quality and maintenance. Data
profiling, cleaning and outdating is undertaken prior to loading the
Linkage of information from disparate resources into a unique
data warehouse enables innovative uses of healthcare data within
a wide range of areas and disciplines besides infection reporting
alone. The size of this integrated database is about 75 GB and
contains records from 2004.
234 L. García Álvarez et al. / Journal of Hospital Infection 79 (2011) 231e235

Weekly, monthly and yearly reports are produced and sent to H1N1 hospital surveillance subsystem
wards, CPGs and the Trust (Table I). Comparisons between wards
and CPGs can be made, which allows designing targeted control An example of a real-time application of this database is the
measures and intervention policies to promote best practices at all monitoring of H1N1 within the Trust to detect patterns of
levels and reduce infection rates. presentation among patients and staff. The H1N1 patient report
Reports on staff sickness, staff absence ratio, and C. difficile cases and the staff absence report from EROSTER (Allocate Software,
together with staff absence ratio alerts provide powerful tools for http://www.allocatesoftware.com/) are sent each day via e-mail
workforce monitoring, reporting and planning, with particular and automatically saved as an attachment. These reports are loaded
attention to staff deployed in ICU and HDU. This information is used into the database as soon as they are saved. A staff sickness report
by the Trust to support monitoring of staffing and decision-making and a staff absence report are generated daily from EHR (Electronic
in relation to activation of contingency plans when required. Human Resources, Equiniti-ICS, http://www.equiniti-ics.com/) at
Information on patient movements combined with MRSA and 10:30, and data checked and loaded into the database at 20:00 each
C. difficile cases offers a way to track disease spread within hospital, day. A weekly staff personnel report is sent by email and auto-
and allows prompt implementation of control measures and matically saved as an e-mail attachment, with the data loaded into
management strategies to prevent infection. the database as soon as the report is saved. The outputs generated
This integrated electronic database can be used for real-time from this subsystem include an H1N1 hospital patient report;
syndromic surveillance to enable the early detection of outbreaks a staff absence numbers report by CPG, division and ward; a staff
and emerging threats, prior to clinical diagnosis, as well as monitoring sickness numbers report by CPG, division and ward; and a hospital
disease trends within the Trust. Syndromic surveillance relies on the patient polymerase chain reaction influenza report.
classification of symptoms and diagnosis into syndromes.10 The This subsystem enables the Trust to understand the impact that
availability of data on ICD-10 (International Classification of influenza has on staff sickness and provides information about the
Diseases) permits the categorization of syndromes within the ability to maintain adequate quality of care on the wards.
system.11
This system can be linked to pathogen-typing data to investi- Description and uses of the research database
gate clusters of infection, caused by endemic or emerging strains,
and to monitor disease spread within and between hospitals. The The research database comprises patient hospital information,
surveillance data can be employed further to map disease within such as ethnicity, gender, age on admission, admission method,
and across wards, clinical areas and patient pathways, discharge method and destination, details of diagnosis (ICD-10) and
enabling early identification of spatiotemporal patterns of infec- procedures, specialty and ward. This system also contains micro-
tion, generation of warnings and risk monitoring within the biological and drug sensitivity test results and hospital staff sick-
Trust. These processes can be particularly important in ness data. All the inpatient data are anonymized with identifiers
relatively large organizations such as Imperial College Healthcare removed and a random identification number generated. This new
NHS Trust in which there are about 170,000 patient admissions identifier enables record linkage between five inpatient-related
per year. tables. The research database contains 370,599 inpatient records
to date, which correspond to 157,482 unique inpatient identities.
This database can be used: (i) to enhance surveillance and
management of HAI in patients; (ii) to investigate previously
Table I
Examples of outputs currently being generated by the operational integrated
recognized risk factors of HAI and explore the data to identify, test
database developed within Imperial College Healthcare NHS Trust and validate new ones; (iii) to develop measurable predictors of re-
admission for HAI; (iv) to build surveillance subsystems in high risk
Report Frequency of reporting
populations; (v) to measure the impact of new processes, practices
1. ICNet download of Clostridium difficile Daily
and pathways; (vi) to link these data to primary care and
cases and MRSA bloodstream
infection cases community data to allow cross-boundary research; (vii) to estab-
2. Trust and non-Trust C. difficile cases Daily; weekly; monthly lish systems to monitor antibiotic resistance and prescribing
3. MRSA bacteraemia episodes Weekly; monthly practice and facilitate the investigation of antibiotic strategies
4. MSSA bacteraemia episodes Monthly within the acute care setting; (viii) to develop predictive models for
5. Vancomycin-resistant enterococci Monthly
bacteraemia episodes
early identification of at-risk patients; and (ix) to provide measures
6. Staff sickness cases Weekly; monthly; yearly of the burden of HAI to the Trust.
7. Staff absence ratioa Weekly
8. Staff cases of diarrhoea and vomiting Weekly; monthly; yearly Discussion
9. Staff flu cases Weekly; monthly; yearly
10. H1N1 patient monitoring Weekly; monthly
(attendance and admission) This paper investigates innovative uses of existing NHS data-
11. Bay closures Weekly bases to enhance local and national infection surveillance capa-
12. Death certificate information e HAI-related Monthly bility, and facilitate early detection of infection, targeted research
13. MRSA screening coverage (%) Weekly; monthly and rapid development of predictive models to identify at-risk
14. Synbiotix: hand hygiene compliance Weekly; monthly
with WHO ‘Five Moments’
conditions, individuals and groups. The methodology allows the
15. Synbiotix: pressure ulcer incidences Weekly; monthly development of new indicators and measures for infection and
and returns outbreaks, including proxy indicators, syndromic monitoring, and
16. Monitoring of bed moves vs minimum Monthly organizational factors.
allowance per procedure/process
The value of the use of electronic surveillance in hospital
17. Ward-cleaning processing and reporting Monthly
prevention and control has been reported since the 1990s.12e19
ICNet, Infection Control Network software; MRSA/MSSA, meticillin-resistant/ However, and in spite of the availability of a broad range of local
susceptible Staphylococcus aureus; HAI, healthcare-associated infection; WHO,
World Health Organization.
databases and computer resources within healthcare settings, this
a
Number of absence hours divided by total number of working hours within the type of surveillance is not exploited to maximum advantage and its
department. use is not widespread.20 Electronic surveillance presents
 L. García Álvarez et al. / Journal of Hospital Infection 79 (2011) 231e235
advantages over more traditional monitoring approaches, and has
been found to possess a relatively high utility in detection of HAI.20
Besides obvious benefits relating to time and cost of data collection
and analysis, automated systems provide a powerful tool to target
a more extensive spectrum of hospital infections, and enable
monitoring of processes which can influence infection control
directly or indirectly. Moreover, these systems are resilient to future
changes within the NHS, and represent an economic and pragmatic
approach to hospital surveillance and infection control.
Previous studies have developed computerized databases via
record linkage of healthcare and social care datasets for a variety of
purposes, including: (i) to build surveillance systems and enable
early detection of nosocomial infections;13,14,16 (ii) to determine the
financial burden of hospital infections;17,19 (iii) to monitor antimi-
crobial resistance patterns and prescribing practice;13,19 and (iv) to
identify mortality trends associated with chronic diseases.18 These
systems have proved to be valuable although uses of data can be
extended further.
This work presents more comprehensive applications of these
datasets, such as the development of syndromic surveillance and
proxy indicators, and improves previously described frameworks
by integrating staff absence and staff sickness data.21 The real-time
operating system monitors influenza patterns, influences patient
care management decision-making, and addresses emerging
threats, for which actions can be taken as the infrastructure is
already in place. Creative use of data overcomes the disadvantages
of simpler electronic surveillance systems regarding the uncer-
tainty attached to the diagnosis of some infections, for which
microbiological data alone were not sufficient.20 The development
of a research database to investigate HAI risk factors and build more
sophisticated risk prediction and surveillance models for HAI
provides insights and intelligence for intervention strategies at
local and potentially national level.
There are some limitations in relation to the data itself, such as
missing values due to data recording errors. Future applications of
urgent care and A&E data will encounter difficulties regarding
discrepancies in patient identification. Both Adastra and Symphony
use different patient identifiers than those assigned to the inpatient
databases, and potential missing values on NHS number or patient
name may impair appropriate linking of these databases to the
integrated system.
Further applications and improvements of this methodology will
include the use of such a system to monitor other infectious diseases
with particular relevance to hospital staff (e.g. norovirus), as well as the
incorporation of pathogen typing data to detect clusters of infection
and monitor disease spread within and between hospitals in real-time,
and in particular in high risk areas such as neonatal ICU. Variation in
infection occurrence due to changes in policy, practice and activity
including prescribing practice may be detected via alert systems as
soon as drug daily doses are available at ward and patient level, with
measures taken accordingly. This multi-disciplinary approach may be
transferred to other specialties beyond infectious disease control as it
represents a suitable framework for optimizing the use of routinely
collected data in healthcare management.
In conclusion, this work has a direct impact on patient care and
hospital management. This system supports hospital infection
prevention, patient safety, emergency preparedness and responses,
as well as organizational learning and resilience. It is currently
designed to be used at local level but the model could be applied
throughout the entire public healthcare system and exploited to
develop a national reporting framework. Some would argue that, in
an ideal world, there would be a single complete information system
based around the electronic patient record, but until this becomes
235
a reality, this work has shown that there is much to be gained from
linking and using current disparate information systems.
Acknowledgements
We thank the infection control and prevention and the infor-
mation technology teams within Imperial College Healthcare NHS
Trust. In particular, we thank M. Richards, J. Sampson, S. Roden-
hurst, A. Shearin and J. Khan for their contribution to this work.
Conflict of interest statement
None declared.
Funding sources
The UKCRC funds the National Centre for Infection Prevention
and Management at Imperial College London; also supported by
the UK National Institute for Health Research Biomedical
Research Centre funding scheme.
References
1. Smyth ET, McIlvenny G, Enstone JE, et al. Four country healthcare associated
infection prevalence survey 2006: overview of the results. J Hosp Infect
2008;69:230e248.
2. Plowman R, Graves N, Griffin MA, et al. The rate and cost of hospital-acquired
infections occurring in patients admitted to selected specialties of a district
general hospital in England and the national burden imposed. J Hosp Infect
2001;47:198e209.
3. National Audit Office. Reducing healthcare associated infections in hospitals in
England (HC 560), Report by the Comptroller and Auditor General, Session
2008e2009. London: NAO; 2009.
4. Department of Health. Weekly infections statistics gives power to patients.
London: DoH; 2010.
5. Rosen R. Can the NHS cut costs without substantially damaging the quality of
health care? Yes. BMJ 2010;340. c1967.
6. Alonso-Echanove J, Edwards JR, Richards MJ, et al. Effect of nurse staffing and
antimicrobial-impregnated central venous catheters on the risk for blood-
stream infections in intensive care units. Infect Control Hosp Epidemiol
2003;24:916e925.
7. Cimiotti JP, Haas J, Saiman L, Larson EL. Impact of staffing on bloodstream
infections in the neonatal intensive care unit. Arch Pediatr Adolesc Med
2006;160:832e836.
8. Hugonnet S, Chevrolet JC, Pittet D. The effect of workload on infection risk in
critically ill patients. Crit Care Med 2007;35:76e81.
9. Health Protection Agency. Surveillance of Healthcare Associated Infections
Report: 2008. London: HPA; 2008.
10. Henning KJ. What is syndromic surveillance? Morb Mortal Wkly Rep
2004;53(Suppl.):5e11.
11. Barta A. ICD-10-CM official coding guidelines. J AHIMA 2009;80:70e71.
12. Broderick A, Mori M, Nettleman MD, Streed SA, Wenzel RP. Nosocomial
infections: validation of surveillance and computer modeling to identify
patients at risk. Am J Epidemiol 1990;131:734e742.
13. Evans RS, Abouzelof RH, Taylor CW, et al. Computer surveillance of hospital-
acquired infections: a 25 year update. AMIA Annu Symp Proc 2009;2009:178e182.
14. Evans RS, Wallace CJ, Lloyd JF, et al. Rapid identification of hospitalized patients
at high risk for MRSA carriage. J Am Med Inform Assoc 2008;15:506e512.
15. Lyons RA, Jones KH, John G, et al. The SAIL databank: linking multiple health
and social care datasets. BMC Med Inform Decis Mak 2009;9:3.
16. Pittet D, Safran E, Harbarth S, et al. Automatic alerts for methicillin-resistant
Staphylococcus aureus surveillance and control: role of a hospital information
system. Infect Control Hosp Epidemiol 1996;17:496e502.
17. Vrijens F, Hulstaert F, Van de Sande S, Devriese S, Morales I, Parmentier Y.
Hospital-acquired, laboratory-confirmed bloodstream infections: linking
national surveillance data to clinical and financial hospital data to estimate
increased length of stay and healthcare costs. J Hosp Infect 2010;75:158e162.
18. Weng C, Coppini DV, Sonksen PH. Linking a hospital diabetes database and the
National Health Service Central Register: a way to establish accurate mortality
and movement data. Diabet Med 1997;14:877e883.
19. Wisniewski MF, Kieszkowski P, Zagorski BM, Trick WE, Sommers M,
Weinstein RA. Development of a clinical data warehouse for hospital infection
control. J Am Med Inform Assoc 2003;10:454e462.
20. Leal J, Laupland KB. Validity of electronic surveillance systems: a systematic
review. J Hosp Infect 2008;69:220e229.
21. Brown C, Richards M, Galletly T, et al. Use of anti-infective serial prevalence
studies to identify and monitor hospital-acquired infection. J Hosp Infect
2009;73:34e40.