J Med Syst (2012) 36:541–547

DOI 10.1007/s10916-010-9499-7

ORIGINAL PAPER

Queueing for Healthcare

R. Kannapiran Palvannan & Kiok Liang Teow

Received: 9 February 2010 / Accepted: 12 April 2010 / Published online: 8 May 2010
# Springer Science+Business Media, LLC 2010

Abstract Patient queues are prevalent in healthcare and service system and short customer wait time. In healthcare,
wait time is one measure of access to care. We illustrate patient wait time is one measure of access to care and
Queueing Theory—an analytical tool that has provided excessive waiting may lead to poorer health outcomes.
many insights to service providers when designing new Hence, the imperative to reduce it to a clinically acceptable
service systems and managing existing ones. This estab- level before treatment should begin.
lished theory helps us to quantify the appropriate service Queues exist everywhere but some are not very visible.
capacity to meet the patient demand, balancing system A few examples in the hospital are described. As patients
utilization and the patient’s wait time. It considers four key are prioritized and stabilized in the Emergency Department
factors that affect the patient’s wait time: average patient (ED), patients of lower acuity wait longer until their first
demand, average service rate and the variation in both. We consult. When new patients of higher acuity arrive later,
illustrate four basic insights that will be useful for managers they may push them behind and the average wait time may
and doctors who manage healthcare delivery systems, at be affected. Small surges in ED patient arrivals can lead to
hospital or department level. Two examples from local several patients breaching the maximum wait time limit for
hospitals are shown where we have used queueing models admission, due to the fixed capacity of inpatient beds. In
to estimate the service capacity and analyze the impact of the outpatient setting, the queue or appointment wait list of
capacity configurations, while considering the inherent subsidized and private patients seeking their first consult is
variation in healthcare. both an indicator of the demand and has implications on the
allocation of resources for service differentiation. In an
Keywords Queueing theory . Simulation . Healthcare imaging department, diagnostic tests may not be read one at
applications . Endoscopy planning . Infection control a time as they arrive, but may be batched together into a
queue to reduce the ‘set up’ cost of specialists’ time, but this
has to be balanced with the wait time for the test result. Also,
Introduction every batching exercise may have a downstream impact on
those service providers who receive the batched results in
Queues are commonplace in an urban setting. We partici- intervals. The batching effect is also seen in ward rounds that
pate in two roles: either as a quality conscious and cost are done once a day fixing the decision cycle to 24 h.
aware service provider or as the person waiting in line with The purpose of this paper is to highlight the benefits of
some frustration, be it in traffic, counters or surgery. But the queueing analysis to healthcare managers and doctors. It is the
manager has to find a balance between high utilisation of a authors’ observation that the formidable mathematics in
queueing models does not make them accessible to the
resource allocators who need them most. So they have been
avoided and the intuition behind the quantitative relationships
R. K. Palvannan (*) : K. L. Teow
is developed. Interested readers will find the derivations in
National Healthcare Group,
Singapore, Singapore books on queueing theory like [1]. This paper focuses on the
e-mail: [email protected] factors that affect the patient’s wait time and provide basic
542 J Med Syst (2012) 36:541–547

insights that will improve our understanding of the care

Patients Wait Service Leave
delivery process i.e. the time that one waits for a service to
start for him, depends on the average patient demand or Demand Queue Capacity

arrival rate, average service rate and the variability in both.

Fig. 1 Basic queue structure
We will explore the structure of a queueing system, share
fundamental insights and illustrate two examples where
queueing models had been useful in capacity sizing and
impact analysis in hospital settings in Singapore. We then two ways. Firstly it helps to quantify the appropriate service
discuss the limitations of such analytical models and describe capacity so that the patient’s wait time is acceptable, e.g.
how complementary discrete event simulation follows when planning the number of inpatient beds needed in a
queueing analysis to handle more complexity. specialty to meet an acceptable admission wait time for the
emergency patient. Secondly it guides in analyzing the
impact of change in demand or service factors, e.g. when
Origin and structure of queueing analysis estimating the impact on the wait time of the emergency
patient indicated for surgery if there is a 3% increase in
The word queue comes from the Latin word cauda or tail patient load. In these questions, there is an emphasis on
and the study of queueing is a branch of applied probability the patient wait time in terms of the patient demand and
called “Queueing Theory”. Queueing Theory has its origins service rate. In short, ‘queueing theory’, is the equation
in early 1900s when studies were undertaken to econom- that defines the relationship between demand, capacity
ically design the capacity of a telephone exchange [2]. A. and queues/wait time when there is significant variability.
K. Erlang, a Danish engineer, observed that the demand for In most cases, it is straightforward to identify the
the telephone exchange is characterized by phone calls that demand and capacity, though some care is required to
come in randomly and the service by the random duration define the demand exactly. Demand is computed from the
of each call. Since the demand could not be scheduled and point at which the service is first requested and not started.
had to be responded to by the telephone exchange, some For example, when studying the dynamics of ED care from
call requests could not be served. Erlang realized that it was patient presentation to disposition, the demand for consul-
uneconomical to service all incoming calls at all times but tation is calculated as the time at which the patients
could only accept a proportion of them. He contributed by complete triage and not when the doctor’s consultation
developing exact mathematical relationships between cus- starts. In the process of admitting the patient, the demand is
tomer demand, service rate and customer queue or wait calculated as the time at which the ED physician decides to
time, while considering the significant variation in demand admit the patient and not the actual admission time. Table 1
and service. This provided transparent equations for the illustrates a few more examples. If time of demand request
manager to plan the service capacity to meet an acceptable is not recorded it has to be prospectively collected for
service level. Cost considerations were added to judge the analysis.
balance between system utilization and wait time. There is an emphasis on collection of time-related data,
Since then, queueing theory has been well developed [3] given that we are studying the patient’s wait time. With
and applied in many areas: traffic analysis of air, road and sufficient data of the patients’ arrival times and service
Internet usage, service design in call centers, supermarket, durations, they are summarized using appropriate probability
banking, restaurant and manufacturing processes. It is also distributions that capture the essential information of the mean
being studied for healthcare applications [4, 5]. They and variance of the patient demand and service rate. Two
include the study of wait lists [6], deployment of ambu- complementary distributions are used as approximations of
lances [7], planning critical care resources [8] and bed the random process: the Poisson distribution for describing
planning [9]. Generally speaking, the emergent presentation the rate of demand and the negative exponential distribution
of a patient to a doctor constitutes the demand and his for describing the service duration. For example, if on average
unique condition which requires a subtly different treatment 20 patients visit a walk-in clinic between 11 am to 12 pm on
and duration constitutes the service. With multiple tests and Mondays, it does not mean that the inter-arrival time is exactly
procedures involving other health care providers in the 3 min. The distribution of the inter-arrival time data will be
diagnosis and therapy, the care delivery process has right skewed suggesting the negative exponential distribution
significant variability with patients waiting for busy health as a suitable approximation. Similarly the number of patients
care providers. who walk in per 10 min window is approximated to be a
Figure 1 shows the typical structure with the demand, Poisson random variable. Both Poisson and the negative
capacity and the queues where many patients have to wait if exponential distributions refer to the same random dynamic
the service provider is busy. Queueing analysis can guide in process and it depends on whether we are studying the rate
J Med Syst (2012) 36:541–547 543

Table 1 Common queues in healthcare

Description Demand Capacity Queues

Seeing an ED patient ED registration rate Triage, consult throughput Waiting for triage, consult etc.
Admitting an ED patient Bed request rate Discharge rate Waiting for admission
Servicing an outpatient appointment Call up rate Patients seen per day Appointment wait list
Emergency Operation Surgery indication rate Specialty surgery rate Waiting for surgery excluding preparation time

(Poisson is intuitive for demand) or duration (negative tracing to the right of the x-axis) if the average patient
exponential is intuitive for service). demand has increased or the system capacity has been
reduced. With this a few observations follow.
Firstly, as the system gets busier, the patient waits
Insights from queueing theory longer, and the waiting gets inexorable as the system is
close to congestion. An increase from 80% to 90% has a
This section highlights fundamental insights of the rela- larger impact on patient wait time than an increase from
tionship between patient demand, service rate and waiting 70% to 80%. The relationship is subtle and non-linear and
time when there is significant variability. Though these can can be counter intuitive as proportional assumptions do not
be written as equations, they are shown graphically to hold at high utilization. If we operate at a high utilisation of
illustrate counter intuitive relationships. 85%, we are sensitive to small changes in demand and
capacity i.e. a small increase in utilisation will cause a large
1. Queues are formed even when capacity is greater than increase in waiting. This is typically faced in a hospital that
demand, when there is variation operates with a high bed occupancy rate (BOR), as a small
surge in bed requests means ‘suddenly’ several patients
Few would argue with the converse—that when demand may have to wait for admission for a very long time [10],
is higher than capacity, there is interminable wait. But even triggering much operational response. On the other hand, a
if we service faster than the rate at which patients are small increase in bed capacity at high BOR—possibly by
coming in, there will still be queues. This happens as using temporary beds reserved for other use—tends to
arrivals and services have variation: they cannot be exactly restore the admission wait time very quickly.
predicted in time and therefore cannot be synchronized. At Secondly, to achieve a low wait time, the system must
times, the service rate is higher than the demand, and operate at a very low utilisation by investing in slack
momentarily the service provider will have some free time. capacity and therefore have a high cost structure. Different
This is less likely in a busy service. As the unused capacity response systems operate at different utilisation levels to
cannot be stored and recovered, the average utilization is cater to their patients’ urgency. An emergency response
less than 100%. On the other hand, most of the time, the system (e.g. ambulance) may have to operate at about 20%
capacity is less than the demand and the service providers and a relatively urgent admitting unit (e.g. ICU) at 75%.
are busy seeing one patient after another while the new The graph also shows that when we operate at a low
ones are waiting. The result is a situation where though the
20
average capacity was higher than the average demand
giving a high but not 100% utilisation, the patients have to 18

wait. This is an elementary but a practical observation on 16

the nature of a service process where the time of arrival and 14

duration of services have variation.
Average wait time

12

10
2. Short wait time demands low system utilization and
8
high cost structure.
6

Figure 2 shows the classical relationship between system 4

utilisation and patient’s average wait time. It is a schematic 2

plot of how the patient’s average wait time for a service 0

varies with the system’s utilisation. Recalling that system 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Utilisation
utilisation is the ratio of the average patient demand to
average service rate or capacity, it can only increase (by Fig. 2 Average wait time versus system utilisation
544 J Med Syst (2012) 36:541–547

48
utilization, it takes a large increase in capacity to reduce the
44
wait time targets, as we have diminishing returns to capacity 40
20-bed
40-bed
increase. 36
100-bed

32

Waiting time (hrs)

3. Variation in demand and service rate affects patient 28
24
wait time significantly
20
16
Two service units with the same capacity and demand may 12
experience different patient wait times. One reason could be 8

the marked difference in the variation of their patient arrivals 4

0
or their service durations. Figure 3 illustrates schematically 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
the impact of variation on the average wait time. Curve A is Utilisation Rate (BOR)

typical of a system with higher variation compared with Fig. 4 Impact of size and partitioning of service capacity on average
Curve B. For the same utilisation level we see a longer wait time
patient wait time. Line C shows an ideal setting with no
variation, which is typical of a synchronized automated bed unit, as the smaller bed unit will face bed shortages
arrival and service mechanism where congestion happens much more frequently.
only when demand exceeds capacity. Similarly this principle of scale applies when a 100 bed unit
This figure suggests a guide to decide the appropriate is partitioned into smaller units for some reason. If the patients
operating utilisation level for a service. We start with the are homogeneous—having the same arrival and service
clinical decision that defines the acceptable average patient patterns—and the service is split into two units there may be
wait time, marking a point on the y-axis. Next we look at a significant impact on patient wait time. So a unit of 100 beds
sample historical data to estimate the variation in arrival will have a shorter wait time than when split into five 20 bed
and service rates—this will help to map out the suitable units, although the total capacity is the same. Splitting a
operating curve. Reading off the x-axis, we can identify the service is a common intervention that we consider and it
region of operating utilisation and not far to its right will be would be good to analyze the impact on patient wait time.
its congestion point. Putting it altogether, the expression below shows how
the four factors affect the patient’s average wait time. This
4. Size and partitioning impacts patient wait time. is an approximation for a system with high utilization as
exact equations are not known [3]. The expression shows
It is known that a larger service unit has economies of that the average wait time is proportional to the square of
scale as the fixed overhead is spread thinner, supporting an the variability in patient arrival or demand (σx) and the
argument for expansion, until some threshold is reached service duration (σc) but lessened by the number of service
[11]. Additionally the larger the pool of resources, the better providers (c). And the higher the arrival rate (λ) or lower
is the absorption of variation in arrival and service duration. the service rate (μ), the longer is the wait time.
Figure 4 illustrates that a 100 bed unit is able to operate at a  
s2
higher utilisation for the same patient wait time than a 20 l s 2x þ c2c
Patient0 s average wait time   
l
20
2 1  cm
18

16

14
Case examples
Average wait time

12
We will move on to see how the application of simple
10
A B C analytical queueing models, has helped to develop insights
8
into capacity sizing and impact analysis in hospital settings.
More variation
6 They were analyzed using QTSPlus® from [1], an Excel
4 add-in tool for queueing analysis.
2

0
Example 1—Capacity sizing to meet target patient wait time
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Utilisation
The endoscopy centre of Tan Tock Seng Hospital, Singapore
Fig. 3 Impact of variability in demand and service on average wait time considered expanding the number of endoscopy theatres from
J Med Syst (2012) 36:541–547 545

Endoscopic operation suites Table 3 Patient arrival and length of stay data by cluster

Cluster Admission rate (patients/day) Length of Stay (days)

Complete Complete Enter Ave Variance Ave Std dev

Other
discipline Post recovery area Preparation area
DSOT Medical 32.4 53.1 5.9 8.0
Surgical 11.1 16.2 5.0 8.5
Orthopaedics 7.2 8.7 5.0 6.1
Bed
recycle Cardiology 9.8 14.1 7.1 9.3
Oncology 3.0 2.7 8.3 11.4
Registration counter

not considered the variation in the patient exiting the

Patient arrival Patient departure Patient arrival
procedure and the variation in the recovery process. If at
Fig. 5 Patient flow in endoscopy centre most 1% of the patients can wait, we move up the column
and read off that 20 recovery beds are needed.
4 to 6 to cater for increasing patient load [12]. Figure 5 shows Several observations can be made. Firstly the wait time
the patient flow as they move from pre-operative sedation, target can be stated in explicit units of time or as percentage of
procedure and then recovery in the post-recovery beds for patients who have to wait, as in this case. Secondly planning
close monitoring before discharge. The administrator needed the capacity based on average without considering the
to estimate the number of post-recovery beds that would be variation in demand and service will result in significant wait.
required such that no patient would be left waiting after Thirdly the relationship between capacity and patient wait
procedure for recovery. time is not linear. The number of additional recovery beds
To analyze this, the historical data was used to estimate required increases much more when one needs a lower
the mean and variance of the arrival of patients who need probability of loss; we notice this as we move up the column,
recovery beds as well as the time they need to recover. This so we have much diminishing return to increase in capacity.
estimates the patient demand and service rate for the Finally we also observe some economies of scale, as the
queueing system. Ideally no patient should be waiting for demand increases (more sedation cases) the relative increase
recovery beds after procedure in the theatre. But rather than in capacity for recovery is lower. For instance, having 10 beds
stating an extremely low acceptable patient wait time, the to meet an average demand of 7 patients sees a 7.9% loss of
objective is specified as the percentage of patients who patients, while doubling both the capacity to 20 beds to meet
have to wait as a small proportion e.g. 1%. This is similar to 14 patients has a much lower loss (proportionately) at 3%—
the classic case that Erlang used to estimate the capacity this illustrates the size and pooling effect.
of the telephone exchange and the queueing model is
called the Erlang’s loss model. It was analyzed using the Example 2—Impact of ring fencing on patient wait time
M/G/c/c module in QTSPlus®.
Table 2 shows the percentage of patients who will have Partitioning or ring fencing a fixed number of inpatient
to wait for a recovery bed. The row heading is the average beds into 2 groups increases the admission wait time, and
number of beds required. If 12 beds are required as more beds are required just to restore the wait time to the
highlighted in the marked column, reading down the original level. The National University Hospital, Singapore
column, and we operate with 12 beds, about 19.9% of the wanted to study the impact of patient cohorting for
patients will end up waiting for a recovery bed, which is infection control [13] on ED admission wait time. Infected
unacceptable. This is the flaw of using averages as it has or colonized patients are said to be cohorted when they are

Table 2 Percentage of patients who have to wait for a recovery bed Table 4 Impact of cohorting at cluster and hospital level on
Prob of not Demand = # of theatres * % req beds * Recovery time / Ops duration admission wait time
finding a bed 5 6 7 8 9 10 11 12 13 14
22 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.1% 0.3% 0.7% 1.2%
Input parameters Additional bed capacity required (%)
21 0.0% 0.0% 0.0% 0.0% 0.0% 0.1% 0.2% 0.6% 1.1% 2.0%
20 0.0% 0.0% 0.0% 0.0% 0.1% 0.2% 0.5% 1.0% 1.8% 3.0%
# of Recovery beds

19 0.0% 0.0% 0.0% 0.0% 0.1% 0.4% 0.8% 1.6% 2.8% 4.4% Cohorted Cohorted Scenario 1 Cluster Scenario 2
18 0.0% 0.0% 0.0% 0.1% 0.3% 0.7% 1.5% 2.7% 4.3% 6.3%
17 0.0% 0.0% 0.1% 0.2% 0.6% 1.3% 2.5% 4.1% 6.2% 8.6% group (%) patients level cohorting Hospital level
16 0.0% 0.0% 0.1% 0.5% 1.1% 2.2% 3.9% 6.0% 8.6% 11.5% ALOS (days) cohorting
15 0.0% 0.1% 0.3% 0.9% 2.0% 3.6% 5.9% 8.6% 11.6% 14.8%
14 0.0% 0.2% 0.7% 1.7% 3.4% 5.7% 8.5% 11.7% 15.1% 18.6%
13 0.1% 0.5% 1.4% 3.1% 5.4% 8.4% 11.9% 15.5% 19.2% 22.8% 5% 8 5.4% 1.6%
12 0.3% 1.1% 2.7% 5.1% 8.3% 12.0% 15.9% 19.9% 23.7% 27.5%
11 0.8% 2.3% 4.8% 8.1% 12.1% 16.3% 20.6% 24.8% 28.7% 32.4% 10% 12 7.7% 2.1%
10 1.8% 4.3% 7.9% 12.2% 16.8% 21.5% 26.0% 30.2% 34.1% 37.7%
546 J Med Syst (2012) 36:541–547

separated from those who are not. This is expected to tative approximation. When evaluating or planning a new
improve the adherence to hand hygiene protocol of health- service system we start by estimating the wait time and
care providers. But it is also felt that the more hard queue, using the queueing models. It is then followed by a
partitions are imposed within a homogeneous ward, the discrete event simulation which handles greater complexity.
greater will be the chances of smaller ring fenced wards The queueing analysis guides a subsequent simulation
being full and therefore increases patient’s admission wait analysis by assessing the feasibility of resource require-
time. We will see how we can quantify the impact using the ments and screening the alternatives of a service design,
partitioning concepts we had looked at earlier. thereby reducing the exploration space when simulating.
Table 3 shows the statistics of the ED patients who were Since they require aggregate data they do not impose on
admitted. It shows the weekday bed request rate and length high resolution data requirements at an early stage of
of stay of patients in the hospital by clusters (groupings of engaging the managers. But the queueing models, at the
specialties e.g. oncology, orthopedics). There is significant benefit of being tractable, have limitations. Classical
variation in demand and service which is expected as these models do not handle the time variation in the arrival rate
come from unscheduled emergency patients. The hospital’s over the course of the day or day of the week and plan for
average length of stay (ALOS) was 5 days and the average peak load only. Much information is lost when the
bed wait of the patient in the ED was 2 h. The classical waiting probability distributions are not Poisson and complex
model with Poisson arrival and service rate with multiple routing logic in a queueing network is not addressed.
servers was used to estimate the effect of partitioning. It was While the academic community has developed the
analyzed using the M/M/c module in QTSPlus®. theory of queueing models to sophistication, the authors
Moving forward, we want to see the impact of partitioning believe that they have not been exploited by the resource
every cluster into 2 groups: one for the clean patients and one managers in healthcare. Much queueing development has
for those who are colonized. We looked at 2 scenarios of been done for performance modeling of computer networks
cohorting: one at a cluster level and one at a hospital level. It is and some of these may be applicable to the vast queueing
assumed that the ALOS of the cohorted patients is in the range networks in healthcare at hospital and health system level.
of 8–12 days and that the percentage of cohorted patients is in On the other hand, some simulation analyses may not
the range of 5%–10% as this may vary across the clusters. The precede their computational effort by exploiting queueing
clusters are partitioned and the new wait time is computed models to leverage their analysis.
which is expected to be longer. After that we computed how In summary, inherent variation in patient presentation
much of additional bed capacity one needs to add to restore and treatment course means that patient queues have to be
the bed wait time back to 2 h. The additional bed capacity is managed through appropriate capacity sizing and impact
another way of expressing increase in bed wait time due to analysis. We have seen how queueing theory and the
partitioning. fundamental principles of queueing systems offer practical
The results are shown in Table 4. The first 2 columns are the insights on the factors that impact the patient’s wait time.
2 unknowns: cohorted patient load and cohorted patients’ Patient flow is nevertheless complex and we will have to
ALOS. The next 2 columns show the results of the 2 scenarios: resort to computational simulation studies that require
cohorting at a cluster level and hospital level. So when we extensive data to get a more accurate picture of the
have 5% of cohorted patients and ALOS of 8 days and we workflow and resource bottlenecks. But a preliminary
partition each cluster into a clean and cohorted ward, the queueing analysis can help us to design new and manage
impact on the wait time will be such that, an additional 5.4% existing queueing systems in hospitals to a large extent.
bed capacity is required to restore the original 2 h wait time.
Hospital level cohorting means an impact of only 1.6% of Acknowledgements We thank Ms Grace Chiang (Director Opera-
additional bed capacity, naturally due to the large pooling tions, National University Hospital), Sister Y P Chia and Sister H H Tan
effect. In short, cluster level cohorting has a significant impact (Tan Tock Seng Endoscopy Centre) for permission to use their data.
ranging from 5%–7% while hospital level cohorting is only
about 1%–2%. This was a counter intuitive observation as the
increase in the bed capacity is to absorb the increase in the wait References
time due to partitioning for the same overall patient demand.
1. Gross, D., Shortle, J. F., Thompson, J. M., and Harris, C. M.,
Fundamentals of queueing theory, Wiley series, 2009.
Discussion 2. Halstrøm, H. L., and Jensen, A., The life and works of A.K.
Erlang. By E. Brockmeyer, Transactions of the Danish Academy of
Technical Sciences, 1948 No. 2.
This paper has highlighted the usefulness of analytical 3. Kleinrock, L., Queueing systems: Volume I & II—Theory. Wiley
queueing models in giving qualitative insights and quanti- Interscience, New York, 1975.
J Med Syst (2012) 36:541–547 547

4. Green, L., Queueing analysis in healthcare, in Patient flow: 9. Green, L., How many hospital beds? Inquiry 39:400–412, 2002.
reducing delay in healthcare delivery, Hall, R.W., ed., Springer, Winter 2002/2003.
New York, 281–308, 2006. 10. Bagus, A., Place, M., and Posnett, J. W., Dynamics of bed use in
5. Fomundam, S., and Herrmann, J. W., A survey of queuing accommodating emergency admissions: stochastic simulation
theory applications in healthcare. Technical report, University model. BMJ 319:155–158, 1999.
of Maryland, 2007. 11. Teow, K. L., Practical operations research applications for health-
6. Worthington, D. J., Queueing models or hospital waiting lists. J. care managers. Ann. Acad. Med. Singapore 38:564–566, 2009.
Oper. Res. Soc. 38(5):413–422, 1987. 12. Teow, K. L., Chia, Y.P., Tan, H. H., Zhu, Z., Heng, B. H., A pure
7. Restrepo, M., Henderson, S. G., and Topalogu, H., Erlang loss loss queueing model for endoscopy recovering bed planning.
models for the static deployment of ambulances. Health Care NHG Annual Scientific Congress 2009. 16–17 Oct 2009.
Manage. Sci. 12:67–79, 2009. 13. Palvannan, R. K., Teow, K. L., Zhu, Z., Heng, B. H., Estimating
8. McManus, M. L., Long, M. C., Cooper, A., and Litvak, E., impact of service partitioning with demand variability using
Queueing theory accurately models the need for critical care queueing theory. NHG Annual Scientific Congress 2009. 16–17
resources. Anesthesiology 100(5):1271–1276, 2004. Oct 2009.