See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224366103

DGHPSim: Supporting smart thinking to improve hospital performance

Conference Paper · January 2009

DOI: 10.1109/WSC.2008.4736228 · Source: IEEE Xplore

CITATIONS READS
10 136

2 authors, including:

Murat M. Gunal
Naval Science and Engineering Institute
42 PUBLICATIONS   718 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Simulation for Industry 4.0 View project

Hospital management View project

All content following this page was uploaded by Murat M. Gunal on 19 May 2014.

The user has requested enhancement of the downloaded file.

Proceedings of the 2008 Winter Simulation Conference
S. J. Mason, R. R. Hill, L. Mönch, O. Rose, T. Jefferson, J. W. Fowler eds.

DGHPSIM: SUPPORTING SMART THINKING TO IMPROVE HOSPITAL PERFORMANCE

M.M. Günal
M. Pidd

Dept. of Management Science

Lancaster University
Lancaster LA1 4YX, U.K.

ABSTRACT as Primary Care Trusts (PCTs), are separate from those

that provide it. PCTs cover defined geographical areas
DGHPSim is a suite of discrete event simulation models and have functions that partially mirror HMOs in the
that enable managers and clinicians to investigate im- USA. PCTs receive NHS funds, mainly on a per capita
provement scenarios for UK general hospitals. The mod- basis and are responsible for buying care from provider
els were developed in Micro Saint Sharp and are config- units, most of which are NHS hospitals. The NHS hospi-
ured using hospital data and nationally available health tals themselves belong to NHS Trusts, which may have
episode statistics. The models can be separately but func- only a single hospital or several and may operate with a
tion best as a single, overall system model that allow us- wide case mix or only highly specific cases (e.g. for can-
ers to develop understanding of the interaction effects of cer or orthopaedics). GPs are the gateway to elective hos-
possible changes. An example of the use of DGHPSim is pital care and usually advise patients which hospital is
given, using UK NHS data, demonstrating how it can be most appropriate for their needs. GP income from the
used to investigate improvement options whilst keeping NHS depends on the number of patients they serve and
an eye on side effects. whether they meet certain performance targets (e.g. % of
infants immunised against common diseases).
1 PERFORMANCE MODELLING OF As part of the commissioning process, PCTs must
HOSPITALS work with GPs to establish priorities for healthcare, re-
sulting in contracts with provider Trusts and the private
1.1 The NHS context sector. Within the NHS, prices are fixed according to a
national tariff, though private sector care can be priced
Patients usually wait for elective hospital care in publi- somewhat differently. PCTs are also responsible for help-
cally funded systems and this has been a long-standing ing their providers to meet NHS performance targets.
problem in the UK National Health Service (NHS). Wish- Since 2006 these are included in the ‘Annual Health
ing to reduce these waiting times and to see other im- Check’ (for details see Healthcare Commission (2008)),
provements, the UK Government launched a two-pronged which refers not to patients but to the performance of pro-
attack some years ago. The first element of the improve- vider Trusts and PCTs. The annual health check has two
ment programme was a substantial increase in the funding main components: quality of services and use of re-
available to the NHS, which, over the last 5 years, has in- sources. The latter refers to the financial status of the
creased annually in real terms by an average of 7.4%, Trust. Quality of services is assessed on three main di-
from about £66 Billion to about £96 Billion (Wanless et al mensions:
(2007). It seems reasonable to hope that at least some of
this extra cash led to improved performance by reducing 1. Core standards: covering patient safety, clinical
waiting times. The second element of the improvement effectiveness and cost effectiveness, Trust gov-
programme has been a performance management frame- ernance, patient focus, accessible and responsive
work that includes waiting time targets that must be met care, the care environment and amenities and, fi-
by NHS organisations, including hospitals. nally, public health
Most medical care in the NHS is provided by general 2. Existing national targets: left over from the pre-
practitioners (GPs), hospitals and community services vious ‘Star Ratings’ (see below).which include
(Office of Health Economics, 2008). Under the current waiting times and clinical standards that specify
arrangements, agencies that buy care for patients, known

978-1-4244-2708-6/08/$25.00 ©2008 IEEE 1484

 Günal and Pidd

desirable outcomes, e.g. on acquired infection • A maximum two-week wait for rapid access
rates chest pain clinics.
3. New national targets: added to the previous tar- • A maximum of 18 weeks from GP referral to
gets hospital treatment.

For the decade to 2005, NHS Trusts were assessed Since hospitals and healthcare providers have limited
against national targets and their performance was sum- resources, decisions must be made on the allocation of
marised in Star Ratings. Rather like hotels, the better the those resources, their better use and improve configura-
performance of a Trust against the targets the more stars it tion. Essentially, managers and clinicians must take a sys-
was awarded. Chief Executives of low and no star Trusts tems view of a patient journey rather than treating it as a
often lost their jobs, though it less clear whether this af- sequence of disjointed events. This is far from straight-
fected the view that patients took of their hospital care. As forward when applied to the whole of hospital care and
an example of a target that was in the Star Ratings and requires tools that support dynamic systems thinking and
continues in the Annual Health Check, no patient should analysis.
spend more than 4 hours in an accident and emergency
department (A&E – ER in the USA) from their time of 2 THE DGHPSIM PROJECT
arrival to their discharge or admission as an inpatient.
New national targets are added to the Annual Health The DGHPSim project has the full title: Modelling for
Check to reflect changing clinical and political priorities. performance measurement and improvement to meet sto-
Non waiting-time targets include commitments to sub- chastic demand for public services: a study of acute hos-
stantially reduce mortality rates by 2010 (from a 1995/97 pitals. DGHPSim is a suite of discrete event models, writ-
baseline) from heart disease and stroke and related dis- ten in Micro Saint Sharp, that simulate individual patients
eases by at least 40% in people under 75, with a 40% re- as they flow through a hospital. The models are based on
duction in the inequalities gap between the fifth of areas a typical district general hospital and enable clinicians,
with the worst health and deprivation indicators and the managers and planners to see the effect of different ac-
population as a whole. tions on waiting times for A&E, inpatients and outpa-
tients.
1.2 The need for smart thinking The DGHPSim models should be parameterised to fit
a particular hospital using two data sources. The first is
For good reasons (to provide better care) and less good local data from the hospital’s Patient Administration Sys-
reasons (job preservation) senior managers and clinicians tem (PAS), which was developed for local administration
in NHS Trusts need to maintain good performance as as- and for billing. Not all hospitals have the same PAS, but
sessed in the Annual Health Check. This requires smart, most are very similar in what they record. The second da-
holistic thinking. It is not enough to concentrate on ta source is a national data set, collected by the UK De-
achieving a single target because most hospitals are high- partment of Health as Health Episode Statistics (HES).
ly congested systems that run close to capacity for much All hospitals are required to submit data to the HES data-
of the year. In this sense, they resemble inflated balloons sets, which contain anonymised data on each inpatient
– squeeze them hard in one place and the pressure moves and outpatient episode. We have developed a software
elsewhere, or the balloon bursts. However, there may be tool, the Health Activity Data Analyser (HADA) to en-
more to it than this. For example, some commentators able the integration of PAS and HES data and to convert
have alleged that apparent improved performance in A&E it into the form used by the DGHPSim suite.
may be caused by delaying entry to patients brought to the Previous WSC papers (Gunal and Pidd 2005, 2006,
hospital in ambulances (Guardian, Feb 17th, 2008). Oth- 2007) have described the DGHPSim project as it has de-
ers have alleged that patients are being shifted from A&E veloped. In its final form, the DGHPSim suite consists of
into assessment wards as the 4-hour target approaches. If four sub-models (see figure 1), models 1 to 3 can be run
true, the latter has a further advantage to the admitting independently if desired, but running all 4 in concert pro-
NHS Trust, because it may receive a higher payment for vides a holistic, dynamic view of a hospital’s perform-
such an admission. ance.
Examples of core standards and targets include:
• A four hour maximum wait in A&E from arrival 1. Accident and emergency department: most gen-
to admission, transfer or discharge. eral hospitals have these units, which accept pa-
• A two-week maximum wait from urgent GP re- tients who arrive themselves or are brought in by
ferral to first outpatient appointment for all ur- ambulance. Most A&E patients do not require
gent suspected cancer referrals. admission as inpatients and, in many hospitals,
those that do are initially admitted to assessment

1485
 Günal and Pidd

units for observation from which they may be and specialties. General hospitals serve many different
discharged or fully admitted as inpatients. types of patient and typical in-patient specialties include
2. Inpatients: this model simulates the operation of general surgery, cardio-vascular medicine, trauma, ortho-
the main wards of a hospital into which patients paedics, ophthalmology, obstetrics, gynaecology and
are admitted for care as emergencies or electives. urology. There are several classification systems in use,
Patients may move between wards whilst in the for example ICD-10 (World Health Organisation, 1992),
hospital. which allows causes of mortality and morbidity to be
3. Outpatients: which receives referrals from GPs, classified and coded for statistical analysis. Others were
sees patients through outpatient appointments created to support payment systems, for example Health
and diagnostics and, if necessary, places them on Resource Groups (HRGs) are used in the UK to determine
waiting lists for inpatient care. charges to be levied by a provider to a PCT for healthcare
4. Waiting list: this model cannot be run alone, but episodes (NHS Information Centre, 2008). Both ICD-10
serves as a bridge between outpatients and inpa- and the latest version of HRGs offer too fine a level of
tients. Patients are added to a waiting list and granularity for the performance modelling in DGHPSim,
then admitted for inpatient care according to in which the user is allowed to specify their own level of
rules defined by the user. granularity with a recommendation not to go below spe-
cialty level – otherwise, the data requirements become too
Thus, the models simulate individual patients, who may onerous. Thus, HADA determines the current case mix of
be emergencies (admitted via A&E or as direct referrals) a hospital using HES and PAS data as inputs. As patients
or electives (admitted via outpatients). Taken together, the enter the simulated system, sampling routines allocate
models provide a dynamic systems view of hospital per- them to a specialty, which determines their resources
formance. needed and also, by further sampling process the charac-
teristics of their treatment such a length of stay.

3 USING DGHPSIM

3.1 General uses for DGHPSim

The DGHPSim suite can be used to help managers, clini-

cians and planners to investigate questions such as:
• For a given investment what level of waiting can
be delivered?
• What advice should be given to hospitals seeking
to achieve performance?
The answer to these questions is not obvious, as the
Figure 1: DGHPSim suite: 4 linked models relationship of waiting to resourcing is non-linear (twice
the investment does not equal half the waiting time).
System dynamics models of health care systems are Moreover, hospitals are complex systems dealing with
common: for example, Taylor and Dangerfield (2005), stochastic demand flows in which actions which appear
Brailsford et al (2004) and Lane et al (2000). Hence, it sensible when viewed in the context of one specialty (e.g.
might reasonably be asked why DGHPSim employs dis- carving out bed space for patients with a particular condi-
crete event simulation rather than system dynamics, given tion) may often appear less sensible when viewed from
that a holistic, dynamic system model is needed. The an- the point of view of the interests of the organisation as a
swer is that it is well-known that, in congested systems, whole. DGHPSim enables them to add quantitative detail
stochastic effects become very important. For example, to the qualitative insights on which policy is too often
queuing theory results suggest that even in a single server based, and provides useful information both to those re-
queue, variances of the arrival and service times have a sponsible for setting targets, and for those whose role is to
major effect. The DGHPSim models simulate individual help hospitals achieve higher levels of performance.
patients and their stochastic arrivals, referrals, treatments DGHPSim can be used to explore a number of ques-
and other factors. This allows stochastic performance ef- tions, such as the following:
fects to be modelled properly. It is possible that simple, • Given this hospital’s level of resourcing, what
deterministic spreadsheet models would suffice if hospi- sort of performance characteristics (length of
tals were not congested systems. stay, use of day-case surgery) would be required
A further question to be addressed when modelling a for it to meet the 18 week wait target?
multi-specialty hospital is the representation of case mix

1486
 Günal and Pidd

• Given this hospital’s performance characteristics, By 2008, the total delay across all 3 stages suffered
what sort of level of resourcing would be re- by patients should be no more than 18 weeks. To place
quired for it to meet the 18 week wait target? this in context, a previous (2005) target required that stage
• Given hospitals are required to hold buffers 1 be complete by 17 weeks. This would leave just one
against emergency demand, how does perform- week for stages 2 and 3 were stage 1 delays not reduced
ance against elective targets trade-off against further. It is not uncommon to hear health service manag-
targets for emergency admissions? And within ers speak of reducing stage 1 delays to 4 weeks, stage 2 to
the elective wait, how does performance against 4 weeks and stage 3 to 8 weeks. On reflection, this is very
waits for those whose journey stops as outpatient strange, since only stage 2 of the 3-stage process is active
trade-off against those whose journey goes right and sensible time compression would focus on stages 1
up to inpatient admission? and 3, especially stage 3.
• The development agency for the NHS, the Insti- So as not embarrass any hospital, the example of
tute for Innovation and Improvement, makes a DGHPSim use described here is based on an analysis of
number of recommendations (e.g. concerning stage 3 elective waits, using 2004 data. It is important to
combining queues and outlying patients) which realise that the hospital in question is now performing
are thought to impact waiting time performance. much better than it was in 2004, though is still some way
Are these impacts substantial or are they dwarfed from reaching the 18-week target across all 3 stages. Fig-
by uncontrollable factors (e.g. seasonal fluctua- ure 3 shows the actual stage 3 (waiting for admission)
tions in demand)? performance of the hospital in 2004/5. The case mix is di-
vided into 6 super-specialties: trauma and orthopaedics;
3.2 An example of DGHPSim use general surgery; ear, nose and throat; urology; paediatrics;
ophthalmology; general medicine. During the period, the
One success of the target regime in that patients wait for hospital treated nearly 3,000 elective patients (excluding
much shorter times in A&E than was the case some years day-cases) and almost 16,000 emergency patients.
ago; this is because NHS hospitals have worked hard to
ensure that patients spend no longer than 4 hours in A&E. 500

The current waiting time challenge is to reduce the time 450 T&O
taken from GP referral to the start of elective treatment.
When the drive to reduce waiting times began, many pa- 400 GENERAL SURGERY

tients suffered extremely long waits for elective treatment. 350

ENT
Waits of two years for hip replacement were far too
No .of Patients

300
UROLOGY
common, as were long waits for cataract surgery. Since
250
1997, a series of targets have been applied to these waits PAEDIATRICS
for elective care and the current target requires this care to 200
OPHTHALMOLOGY
be provided within 18 weeks by the end of 2008. The 150

waiting period can be divided into the three stages shown 100
GENERAL MED ICINE
in figure 2. Stage 1 is the delay between a GP deciding
that a patient needs to be seen by a specialist and the first 50

out-patient appointment with that specialist. Stage 2 is the 0

period of out-patient consultation, which may involve

1

7
10

13

16

19

22

25
28

31

34

37

40
43

46

49

52
Time on Waiting List (weeks)
several visits to a clinic and diagnostic procedures. Stage
3 is the delay between a specialist deciding that a patient
will require inpatient care and the admission of that pa- Figure 3: Actual waiting time performance in stage 3
tient for care. (2004/05)

To demonstrate the use of DGHPSim, 3 different

scenarios are imposed on the 2004/5 admissions data and
hospital resources:

1. Suppose it was possible to reduce length of stay

in all specialties, what effect would this have on
stage 3 waiting times?
2. Suppose the hospital allocated 30% more of its
in-patient beds to elective patients rather than
Figure 2: Stages in 18-week RTT journey.

1487
 Günal and Pidd

keeping them for emergencies, what effect would lay treatment seems unlikely, since that would require a
this have? level of resourcing beyond that which is economically and
3. Suppose that the hospital were able to treat a fur- politically feasible. In the UK, the government has pro-
ther 12% of its surgical patients as day-cases, vided incentives and extra money for healthcare providers
what effect would this have? to reduce waiting times from levels that were unaccept-
able. The incentives are based on targets to be achieved
Table 1 summarises the results of 20 replicated simu- and there has been concern that the target regime has en-
lations of these scenarios and compares them with the couraged managers to focus their efforts on meeting these
base case. The results suggest that the biggest bang for the targets lowering performance elsewhere.
buck comes from creating greater elective capacity by de- The DGHPSim suite can be used by managers and
voting 30% more of the beds to elective admissions. This clinicians to investigate options for improvement and ser-
policy allows the hospital to admit more elective patients vice reconfiguration. Not only can the models show the
(3490 as against 2880) and reduces those waiting more effect of the policies on meeting the targets but can also
than 8 weeks to 34%, as against 58%. Table 2 also shows show the effects on other, very important aspects of hos-
the number of elective admissions cancelled and the num- pital performance. For example, the simulations of 3 sce-
ber of emergency patients who could not be admitted, narios for reducing stage 3 waiting also show the effects
which the simulations assume will be diverted to other on emergency patients and seem to show that, as at
hospitals. This is not enough to reduce typical stage 3 2004/5, a considerable reduction in stage 3 waits would
waits to 8 weeks, but is a large step in the right direction. have been possible with very little effect on bed availabil-
It would, of course, be possible to experiment with com- ity for emergency admissions.
bined scenarios such as more beds allocated to electives
and a reduced length of stay – whether this is feasible in ACKNOWLEDGEMENTS
reality is another question altogether.
The DGHPSim project is funded by the Engineering and
Table 1: Comparative outputs of scenarios [std.dev.] Physical Sciences Research Council under grant
Base LoS 30% 12% in- EP/C010752/1. We are also grateful to staff of the Royal
down more crease day- Lancaster Infirmary for their co-operation in the project
20% electives cases and for the contributions of Professor Gwyn Bevan and
% Waiting Dr Alec Morton(LSE), Professor Peter C. Smith (Univer-
sity of York) and Iván Castilla Rodríguez (University la
> 5 wks 72.6 58.7 48.1 62.4 Laguna).
[1.2] [3.5] [2.5] [1.8]
> 8 wks 58.2 43.8 34.7 48.2 REFERENCES
[1.6] [3.2] [2.4] [1.5]
> 10 wks 46.8 33. 25.5 37.4 Brailsford S.C., Lattimer V.A., Tarnaras P. and Turnbull
[1.7] [2.9] [1.9] [1.2] J.C. 2004. Emergency and on-demand health care:
wait > 18 28.5 17.6 13.4 22.2 modelling a large complex system. Jnl Opl Res Soc,
wks [1.71] [1.7] [1.3] [1.1] 55, 34-42.
Side effects Gunal M.M. and Pidd M. 2005. Simulation modeling for
performance measurement in healthcare. Proceedings
Elect pa- 2880 3396 3490 2654 of the Winter Simulation Conference, M. E. Kuhl, N.
tients [21] [72] [93] [72] M. Steiger, F. B. Armstrong, and J. A. Joines,
Elect 90 73 114 51 eds.,Orlando, FLA.
cancelled [22] [15] [23.3] [17] Gunal M.M. and Pidd M. 2006. Understanding Accident
(3.1%) (2.1%) (3.9%) (2.5%) and Emergency Department Performance Using Si-
Emerg 469 405 505 469 mulation. Proceedings of the 2006 Winter Simulation
outliers [86] [63] [68] [86] Conference. L. F. Perrone, F. P. Wieland, J. Liu, B.
(3%) (2.7%) (4%) (3.2%) G. Lawson, D. M. Nicol, and R. M. Fujimoto, eds.
Monterey, CA.
4 BRINGING IT ALL TOGETHER Gunal M.M. and Pidd M. 2007. Interconnected DES
models of emergency, outpatient, and inpatient de-
One important element in providing acute health care is partments of a hospital. Proceedings of the 2007
ensuring that appropriate treatment is provided when it is Winter Simulation Conference, S. G. Henderson, B.
needed and that patients do not have to wait for unneces- Biller, M-H. Hsieh, J. Shortle, J. D. Tew, and R. R.
sarily long periods for their treatment. Universal zero de- Barton, eds. Washington DC.

1488
 Günal and Pidd

Healthcare Commission. 2008.

<http://www.healthcarecommission.org
.uk/healthcareproviders/serviceprov
iderinforma-
tion/annualhealthcheck/annualhealth
check2007/2008.cfm>
Lane DC, Monefeldt C and Rosenhead JV 2000. Looking
in the wrong place for healthcare improvements: A
system dynamics study of an accident and emergency
department. J Opl Res Soc 51: 518-531.
NHS Information Centre. 2008. Health resource groups.
<http://www.ic.nhs.uk/our-
services/standards-and-
classifications/casemix/healthcare-
resource-groups>
Office of Health Economics. 2008. The NHS - organisa-
tion and structure.
<http://oheschools.org/ohech4pg2.htm
l>
Taylor K and Dangerfield B. 2005. Modelling the feed-
back effects of reconfiguring health services. Jnl Opl
Res Soc, 56, 659–675.
World Health Organization. 1992. The ICD-10 classifica-
tion of mental and behavioural disorders: clinical de-
scriptions and diagnostic guidelines. Geneva.

AUTHOR BIOGRAPHIES

MIKE PIDD is Professor of Management Science in the

Department of Management Science at Lancaster Univer-
sity Management School. He is known for his work in 2
areas: computer simulation and the complementary use of
soft and hard OR. His text ‘Computer simulation in man-
agement science’ is now in its 5th edition and his books
on complementarity include ‘Tools for thinking’ (3rd edi-
tion in progress) and ‘Systems modelling: theory and
practice’. All are published by John Wiley. He has been
an ESRC-funded Research Fellow in the UK’s Advanced
Institute of Management Research, examining perform-
ance measurement in the public sector. Email to
<[email protected]>, website at
<http://www.lancs.ac.uk/staff/smamp/>.

MURAT GUNAL is the Research Associate working on

the DGHPSim project and is also a PhD student at Lan-
caster University. He received his MSc degree in Opera-
tional Research from the same university in 2000. Before
returning to Lancaster he worked for the Turkish Navy as
a simulation analyst. His research interests are currently
discrete event simulation and its use in modelling com-
plex systems such as hospitals. Email to
<[email protected]>.

1489

View publication stats