Reducing Surgical Ward Congestion through

Improved Surgical Scheduling and Uncapacitated Simulation

Vincent S. Chow1, Martin L. Puterman2, Neda Salehirad3, Wenhai Huang1, Derek Atkins3

1. British Columbia Cancer Agency.600 West 10th Avenue, Vancouver BC, Canada V5Z 4E6

2. Centre for Health Care Management, University of British Columbia. 209 - 2053 Main

Mall, Vancouver BC, Canada V6T 1Z2

3. Centre for Operations Excellence, University of British Columbia. 2053 Main Mall,

Vancouver BC, Canada V6T 1Z2

Contact Information

Vincent S. Chow [email protected] 604-877-6000

Martin L. Puterman [email protected] 604-822-8388

Neda Salehirad [email protected] 604-822-1800

Wenhai Huang [email protected] 604-877-6000

Derek Atkins [email protected] 604-822-9665

0 Abstract

High surgical bed occupancy levels often result in heightened staff stress, frequent

surgical cancellations and long surgical wait times. This congestion is in part attributable to

surgical scheduling practices which often focus on the efficient use of operating rooms but

ignore resulting downstream bed utilization. This paper describes a transparent and portable

approach to improve scheduling practices which combines a Monte Carlo simulation model and

a mixed integer optimization model. For any surgical schedule, the simulation samples from

historical case records and predicts bed requirements assuming no resource constraints. The

mixed integer optimization model compliments the simulation model by scheduling both surgeon

blocks and patient types to reduce peak bed occupancies. Scheduling guidelines were developed

from the optimized schedules to provide surgical planners with a simple and implementable

alternative to the optimization model. This approach has been tested and delivered to planners at

a health authority in British Columbia, Canada. The models have been used to propose new

surgical schedules and to evaluate the impact of proposed system changes on ward congestion.

Insights from the models are guiding future surgical schedule revisions.

Key Words: Surgical Scheduling, Mixed Integer Optimization, Monte Carlo Simulation,

Scheduling Guidelines, Hospital Bed Management

1
1 Introduction

In recent years, national governments have placed a high priority on reducing surgical

wait times; Postl (2006) provides a Canadian perspective on this issue. This pressure, in

combination with the increasing demands of a growing and aging population has led to a

significant rise in elective surgical volumes. This increase in surgical volumes has led to high

utilization rates of hospital and specifically surgical beds. High surgical bed utilization poses a

challenge to management because:

 Peaks in demand can result in periods of insufficient capacity.

 Lack of capacity in speciality specific surgical wards may result in sub-optimal

treatment and extended lengths of stay for surgical patients.

 Available capacity on surgical wards may be occupied by medical (non-surgical)

patients limiting future access to surgical patients.

The result of this is often cancellations and long wait times for surgeries.

The day on which a surgeon operates is determined by the surgical block schedule (SBS).

The SBS assigns full or half-day blocks of operating room (OR) time to surgeons over a multi-

week cycle. Usually, surgical planners consult with surgeons when constructing the SBS, while

surgeons schedule patients within the blocks. Since each surgery type has a specific post-

operative bed need, the SBS provides a significant lever for managing bed utilization. Creating

an SBS is challenging, especially in the absence of planning software. Surgical planners must

account for staff, room and equipment availability, and surgeon preferences when developing an

SBS and often require many revisions just to take these considerations into account. However,

being able to assess the impact of this schedule on downstream bed utilization is beyond the

scope of most scheduling activities.

2
 The main contribution of this paper is the development of a transparent, portable, and

rigorous approach to improve surgical scheduling. It consists of two components, a trace-driven

Monte Carlo simulation model and a mixed integer optimization model. The simulation enables

surgical planners to predict the impact of a SBS on surgical ward occupancies. The optimization

model schedules both surgeon blocks and patient mix within each block to help planners create

surgical schedules with minimal bed requirements. Since managers and planners will not be able

to run the optimization model, a set of scheduling guidelines was developed to support ongoing

scheduling revisions.

Our paper is organized as follows. Section 2 provides a review of recent literature.

Section 3 provides a description of the models. Section 4 provides a case study that illustrates the

application of our approach. Section 5 presents study results and illustrations of how the models

can be used and Section 6 concludes with general insights and future research.

2 Related Literature

Operating room planning and scheduling has been widely studied; a review paper by

Cardoen et al. (2008) describes recent advances in this area. We describe other relevant research

below.

Researchers frequently use Discrete Event Simulation (DES) to model complex patient

flow and to estimate resource utilization in hospitals that cannot be accurately characterized

using queuing models. Both Jun et al. (1999) and Jacobson et al. (2006) provide a review of DES

applications in health care. In Surgical Services, Everett (2002) developed a decision support

tool to evaluate various policies on wait lists and bed occupancies. VanBerkel and Blake (2007)

developed a DES model to evaluate surgical wait times and support capacity planning decisions.

Blake et al. (1995) proposed a surgical process management tool that utilizes summarized

3
historical patient records directly in the model. For general patient flow within hospitals, Isken

and Rajagopalan (2002) modelled obstetrical and gynaecological patient flow. Both Harper

(2002) and Pitt (1997) developed generic frameworks for portable DES models of patient flows.

Accurate DES patient flow models often require a significant amount of time and

resources to develop as they are often highly specialized for each system. Harper (2002) and Pitt

(1997) have tried to address these issues but other challenges, such as the creation of clinically

relevant groupings to define parameters within the model can be difficult (Isken and

Rajagopalan, 2002). A retrospective paper by Carter and Blake (2005) also noted the challenges

of creating a flexible DES model as in Blake et al. (1995).

Other methods of simulation have also been used. Henderson and Mason (2005) utilized

a trace-driven simulation model to support ambulance-planning. The paper argued that the ability

to maintain correlations between various parameters within the simulation outweighs the

limitations of using trace data.

A large body of research concerns development of methods to improve OR scheduling.

The scheduling process can be divided into three main scheduling levels: the discipline level, the

surgeon level, and the patient level (Cardoen et al. 2008). Discipline level scheduling determines

the amount of operating room time to allocate to each surgical specialty. Blake and Donald

(2002) developed an integer programming model to produce equitable master surgical schedules.

Beliën and Demeulemeester (2007) devised a mixed integer programming (MIP) model to

minimize the expected bed shortages by relocating specialty blocks. Santibanez et al. (2007)

optimally allocated surgical specialties across a system of hospitals to explore wait lists and

resource utilization under different objective functions.

4
 Surgeon level scheduling determines the allocation of surgeon blocks to the available

operating room time within each specialty. Beliën et al. (2009) expanded their previous model

(2007) to level bed occupancy and variance by reassigning surgeons to different ORs on different

days and integrated a deterministic support tool developed by Beliën et al. (2006) to visualize

planned surgical occupancies under various surgical schedules. The support tool reports the

average occupancy, standard deviation, and expected total bed shortage on each day.

Patient level scheduling involves the allocation of patients within the surgical blocks.

Van Oostrum et al. (2008) used a two-phase decomposition approach with column generation

and a MIP to determine the patient mix and the OR scheduled each day. Vissers et al. (2005)

determined the OR time and the number of each patients in each patient category scheduled on

each day of the week in a cardiothoracic surgery department. Other work, based on heuristics,

includes maximizing operating room use within the day (Dexter and Traub, 2002) and

determining various surgical sequencing procedures to reduce idling and overtime (Denton et al.

2007).

Various papers have integrated simulation and optimization techniques. Testi et al.

(2007) used a three-phase approach to scheduling ORs. The first two phases involved the

allocation of OR time and OR blocks to specialties. The third phase involved in the execution of

the schedule using a DES model to evaluate the sequencing of surgical activities. Persson and

Persson (2009) developed a DES model to test policy changes so as to ensure elective surgeries

wait no more than 90 days. An optimization component was integrated into the model to decide

the scheduling of patients each week.

Our approach uses two models: the Bed Utilization Simulator (BUS), a trace-driven

unconstrained Monte Carlo simulation model that predicts surgical bed occupancy and the

5
Surgical Schedule Optimizer (SSO), a MIP model to level surgical bed occupancy. Advantages

of BUS over other patient flow models include a short development time and increased

portability. The base formulation of the SSO model is similar to Beliën et al. (2009) but differs

from all other surgical scheduling models by scheduling both surgeon blocks and the patient mix

within each block. Both of these models are used in an integrated approach to improve surgical

scheduling. Finally, our paper seeks to use the models to develop general scheduling guidelines

for smoother ward occupancies.

3 Models

Figure 1 gives an overview of the proposed framework. The SSO model helps develop

improved surgical schedules. Surgical planners can either obtain a SBS directly from the SSO, or

use surgical scheduling guidelines derived from the SSO to help construct the block schedules.

Surgical planners would then revise these schedules as necessary and test the schedules in BUS

to capture both unplanned patients and variability in the surgical system. Results from BUS are

analyzed and the schedules can be readjusted and re-simulated until a final SBS is obtained. Both

BUS and SSO models are described below. Some surgical scheduling guidelines are described in

Section 5.

6
Figure 1. The proposed surgical scheduling framework. Models are shaded in grey.

3.1 The Bed Utilization Simulator (BUS)

BUS was designed to provide surgical planners with an easy-to-use tool to investigate the

impact of an SBS on downstream ward volumes. Unlike other models which include “hard”

ward capacities and decision rules to handle situations when wards are full, BUS predicts the

daily demand for ward beds in an uncapacitated system. Hence, the primary output of BUS is the

predicted daily bed occupancy in each downstream surgical ward, ignoring all other competitive

demands for beds. This enables planners to determine the true needs of their system and set

capacities and schedules to reflect these needs. This approach also allows for a simple model

logic that can be broadly applied.

7
 Create planned arrivals Generate patient path
SURGICAL SCHEDULE WEEK 1
01/01/2007 - 05/30/2007

OR 1
AM
PM
MONDAY
SURGEON 1

SURGEON 1
TUESDAY
SURGEON 2

SURGEON 2
WEDNESDAY
SURGEON 3

SURGEON 3
THURSDAY
SURGEON 4

SURGEON 4
FRIDAY
SURGEON 1

SURGEON 1
and length of stay
OR 2
AM
PM
SURGEON 5

SURGEON 5

SURGEON 11
SURGEON 7

SURGEON 7

SURGEON 12
SURGEON 8

SURGEON 8

SURGEON 13
SURGEON 9

SURGEON 9

SURGEON 14
SURGEON 10

SURGEON 10

SURGEON 15
Generate Output
OR 3
AM
PM SURGEON 11 SURGEON 12 SURGEON 13 SURGEON 14 SURGEON 15
Surgical Ward X
Patient…unit…length of stay…
Enter surgical schedule
Patient…unit…length of stay…
with surgeons and case Patient…unit…length of stay… # Beds
types Patient…unit…length of stay… occupied

Day
Random selection of
Create unplanned arrivals Output estimate of true
historical records containing
demand bed occupancy
patient paths and length of
and other statistics
stays specific to surgeon
Generate using historical and patient type
distributions (by day of the
week and specialty)

Figure 2. An overview of the BUS model logic.

Figure 2 illustrates the model logic. There are two sources of arrivals, planned patient

arrivals are generated according to a user provided SBS and unplanned patient arrivals are

generated to follow historical arrival rates. Planned patients are further classified into patient

types based on their downstream bed requirements. The number of patient types can be adjusted

to reflect variation in operation between various hospitals. For each surgical block, the number

of patients of each patient type may be either specified or sampled from historical surgeon slates.

A slate is a combination of patient types performed during a block. One slate may consist of one

“Patient Type 1” case and three “Patient Type 2” cases while another slate may consist of one

“Patient Type 1”, one “Patient Type 2” case and one “Patient Type 3” case. These slates would

be obtained from historical slates, specific to each surgeon block. Our model assumes that all

planned surgeries take place. The arrival of unplanned patients can be generated from a Poisson

or empirical distribution with day of week specific arrival rates.

The model then generates ward occupancy patterns for each patient by randomly

sampling an appropriate patient record from historical data. Planned patients are selected from

8
records with the same surgeon and patient type, while unplanned patients are selected from

records with the same specialty and patient type. For surgeons with limited data, records are

selected from a similar surgeon or from a pool of records of the same specialty and patient type.

Each record contains information on the sequence of surgical wards and length of stay in each

ward the patient visited relative to the day of surgery. Finally, lengths of stays are aggregated for

each ward to generate the daily demand for ward beds.

Figure 3 illustrates the structure of a historical patient record, and how the information is

used to update ward occupancies. The attributes are read sequentially and written into tables

corresponding to surgical wards. In this example, a patient of Dr. Smith spends one night in the

ICU and the next three nights in Ward 1. The “Rel.Start” fields indicate the first night a patient

occupies the unit relative to the operation date. Day t represents the day in which the operation is

performed. It is possible to occupy beds prior to the operation. In such cases, some of the

“Rel.Start” field would be negative. This process is repeated for each day of the simulated

period.

Definition: Surgeon Patient Type Ward1 LOS1 Rel.Start1 Ward2 LOS2 Rel.Start2 ...

Selected Record: Smith SDSA ICU 1 0 Ward 1 3 1 ...

ICU Ward 1
Day Bed Occupancy Day Bed Occupancy
t-1 t-1

t +1 t

t+1 t+1 +1

t+2 t+2 +1

t+3 t+3 +1

t+4 t+4

t+5 t+5

... ...

Figure 3. Structure of a historical record how a patient is used to update ward occupancies

9
 The metric bed-days over capacity summarizes a surgical schedule’s performance. This

metric captures and quantifies the frequency at which patient off-servicing and/or surgery

cancellation would be necessary if the system were capacitated. The metric is computed by

summing the total number of beds in excess of a user-specified bed capacity for each day of the

scheduling period within each surgical ward. The user-defined capacities are an input to the

model and can vary by day of the week if required. We emphasize that the only purpose for

including these capacities is to derive the metric; they don’t impact patient flow in any way since

the model is uncapacitated. The model also produces distributions and descriptive statistics of

ward bed occupancies.

Since this model relies on patient specific input data, the approach to selecting historical

cases can highly influence results. Therefore, it is crucial that only cases with paths and length of

stays that reflect clinical necessity be included in the historical record database. Historical data

should be screened to remove planned cases with patient paths that are inconsistent with known

patient requirements.

BUS was developed in MS Excel using Visual Basic for Applications (VBA) to facilitate

portability, reduce software cost and reduce the need for simulation software expertise. The

model interface was designed in collaboration with surgical planners and executed through a

series of VBA forms and spreadsheets. BUS includes an enhanced spreadsheet to define surgical

schedules (Figure 4A) and an output spreadsheet to display the simulation results (Figure 4B).

The numbers of user-definable simulation parameters are kept to a minimum to facilitate use by

non-experts. The two main user-definable parameters are the number of warm up days and the

number of replications. Recommended values are set as defaults.

10
 A

Ward 1

Ward 1 Bed Occupancy

Figure 4. A) A screenshot of the block schedule input interface. Here, the user can define
surgeons, patient volumes and ward capacities. B) A screenshot of the output report.
Calculated metrics are presented on the left hand table. The distribution of daily bed
occupancies appears in the box plot on the right.

When one uses historical records directly in a simulation, the simulation is often said to

be "trace-driven". Using historical records preserves some correlation patterns within the data.

This can be either a disadvantage or an advantage. Law (2007) notes that when using trace-

driven simulation, the model can only reproduce what has happened historically and may in fact

be the result of system constraints or policy decisions that the simulation is being used to

explore. On the other hand, one might wish to see how new system settings might have

11
performed in the past so the trace driven approach will provide a reasonable base case. The BUS

model has two types of correlation to consider: that within patients and that between patients.

Within patient correlations are between patient type, length of stay and patient path. Ideally,

these correlations should be preserved; especially if the data is cleaned so that non-desirable

paths are removed. On the other hand, the correlation between patient arrivals should be

avoided. Our approach preserves the within patient correlations and through random sampling of

cases, eliminates between patient correlations. This allows us to capture the best of both worlds.

Other potential shortcomings of this modelling approach are that representative datasets

are required for each surgeon and patient-type configuration, and that the model cannot be used

to determine cancellation rates or numbers of patients off-serviced, i.e., those who are not

assigned to the most appropriate ward. We addressed (see Section 4) the first shortcoming by

careful design of data inputs and development of approaches for addressing cases with little data.

We regard our choice of the metric “bed-days over capacity” as a surrogate for off-service and

cancellation rates even though it cannot distinguish between the two. While discrete event

simulations could determine these rates, strong assumptions would be required about

cancellation and redirection policies which may not be explicit or available. Nonetheless varying

the SBS to reduce bed-days over capacity should reduce cancellations and off-servicing which

are two key objectives of these models.

3.2 The Surgical Slate Optimizer (SSO)

While BUS enables planners to investigate bed requirements of new surgical schedules, it

remains extremely challenging for surgical planners to create a SBS that reduces day-to-day

variability in bed occupancy across all surgical wards. Therefore, two integer optimization

models were developed to generate improved surgical schedules. The base model assigns

12
surgical blocks to days to achieve smoother bed occupancies. The slate model generalizes the

base model by also determining the patient type mix (see above) within each block.

Mathematical formulations of these models appear in Appendix 1 and 2. Each model assumes a

multi-week block schedule.

In the base model, binary decision variables indicate whether a specific surgical block is

scheduled on a specific weekday of a multi-week schedule. Each surgical block represents a

unique a surgeon and duration combination that can be scheduled multiple times throughout the

schedule. The duration of each block is expressed in terms of OR-Days. A surgeon using one OR

for one day would be assigned 1 OR-Day while a surgeon using one OR for half a day would be

assigned 0.5 OR-Days. It is important to note that for the same surgeon, a half day block is not

equivalent to half of a full day block. Both the patient mix and length of stay can vary

significantly between these two surgical blocks and they are modelled separately to capture this

difference. Note, if each surgeon has only one block length, then the number of surgeons and

surgeon blocks would be the same. We simplified the models by not including the decision of

which operating room (OR) to assign to a surgical block; that is left for the surgical planners. If

there are additional constraints where only specific types of specialties can be performed in

specific ORs, then the decision variables must be expanded to assign surgical blocks to both days

and ORs.

The objective of the model is to minimize the total maximum bed occupancy across all

wards. By doing this we expect that day to day variation in ward occupancies will be minimized

without resorting to nonlinear objectives. Base model constraints include:

 Total OR capacity cannot be exceeded each day.

13
  A surgeon cannot have more OR-Days than he/she can possibly do on one day.

Usually this will be 0.5 or 1 but if a surgeon can operate in two ORs in sequence then

this could be also 2.

 An upper bound on the number of each surgical block scheduled per week.

 An equality constraint which specifies the total number of blocks to be assigned to a

surgeon over the entire scheduling cycle. Usually this will be chosen to be consistent

with the “as is” frequencies at the time of the study.

 A bookkeeping constraint which defines the maximum number of beds needed in

each ward on each day. Due to the cyclic nature of the schedule, any surgical blocks

that are scheduled near the end of the schedule would result in additional bed demand

in the beginning of the surgical schedule.

The slate model builds upon the base model by taking into account the patient mix within

each surgical block. To reduce model size, we selected a few pre-defined slates from historical

data to restrict the feasible patient mixes for each surgical block. Hence the binary decision

variables determine whether a specific surgical block with a specific slate is assigned on a

specific day. It was believed that with this added flexibility, downstream bed requirements could

be reduced.

The slate model requires two additional constraints. The first constraint ensures that only

one slate is chosen for each block. The second constraint ensures that the number of surgical

cases performed by each surgeon is as least as great as the average number of surgical cases

performed in the past. This ensures that no surgeon is penalized with fewer surgeries under the

optimized schedules. The objective function of the slate model remains the same as the base

model.

14
4 Case Study

The approach was applied at the Royal Jubilee Hospital (RJH) in Victoria, B.C. RJH is

the largest tertiary care hospital for Vancouver Island’s 750,000 residents and as well serves as a

regional hospital for approximately 350,000 residents in Victoria. In 2007, RJH ran 16 operating

rooms (ORs) and 428 inpatient beds, of which about 100 are “reserved” for surgical patients. At

the time of our study, high variability in demand for surgical beds had resulted in an increased

risk of surgical cancellations during peak occupancy periods. Also, shortages of medical beds

during times of increased emergency department congestion placed increasing pressure to use

surgical beds to accommodate medical (non-surgical) patients who require a bed for monitoring

and/or treatment. The consequence of this was that beds planned for use by surgical patients

become blocked. The intent of the study was to reduce surgical cancellations through smoothing

bed requirements for planned surgical cases. Other benefits of doing this include lower costs or

increase access.

4.1 Background

Figure 5 provides an overview of surgical patient flow in the system studied. It shows

two primary inputs: planned elective surgeries scheduled up to a week in advance and unplanned

surgeries that result primarily from emergencies. After surgery, most patients stay temporarily in

the Post Anaesthesia Recovery Room for monitoring prior to being transported to one of six

surgical wards. Patients in unstable condition or who have received major cardiac surgery stay

instead in the Intensive Care Unit or the Cardiac Care Unit prior to being redirected to surgical

wards. The six surgical wards at RJH are categorized by the acuity of the surgery. Day Care or

Short Stay Wards are for patients with minor surgeries or surgeries requiring overnight stay.

15
Both wards operated only on weekdays, and patients who did not meet discharge criteria are

transferred to Wards 1-4 on weekends. Wards 1-4 are for patients with major surgeries and are

assigned to specific surgical specialties.

Figure 5. A high level process map of surgical patient flow at RJH.

RJH uses a 4-week cyclical SBS that assigns full or half-day blocks to surgeons. Planned

patients are categorized into three patient types: Day Care (DC), Short Stay (SS) and Same Day

Surgical Admit (SDSA). Post-surgery, DC and SS patients stay in corresponding wards. SDSA

are inpatients that typically sojourn to Wards 1-4 depending on the type of surgery.

4.2 The Problem

Figure 6 provides a nine month time series of surgical bed occupancy, staffed surgical

bed capacity, and surgical cancellations in Wards 1-4 and clearly illustrates the challenges health

system managers faced in operating this system. Observe that:

16
  surgical volumes were highly variable and that the system was always operating

close to or above capacity,

 a significant portion of the surgical capacity was occupied by non-surgical (medical)

patients who were assigned to these units when space was available, and

 towards the end of 2006, there were frequent surgical cancellations when the system

was over capacity and beds were not available for surgical patients in these wards.

(Cancellations may also occur for patients who cannot be off-serviced to an alternate

surgical ward when their target ward is at capacity)

120
14

100
12
Beds in Surgical Wards

80 10

Cancellations
8
60
Occupancy by medical patients
6
Occupancy by surgical patients
40
Cancellations
4
Capacity
20
2

0 0
04/2006 05/2006 06/2006 07/2006 08/2006 09/2006 10/2006 11/2006 12/2006

Date

Figure 6. Time series of bed occupancy, staffed bed capacity, and surgical cancellations in
Wards 1-4. The bed capacity is adjusted for bed closures due to staffing issues,
maintenance, and outbreaks.

Further analysis of occupancy patterns within surgical wards reveals that unplanned

patients occupy a larger proportion of surgical beds (54%) than planned patients (46%).

However, when analyzing this occupancy across the days of the week, volumes of unplanned

17
patients varied around a common mean, while planned patients exhibited systematic day to day

variation. Figure 7 provides box plots of daily bed occupancy of planned patients in a typical

ward which shows that median occupancy varied between 6 on Mondays to 10 on Fridays and

also that there was considerable within day variability. Given the stochastic nature of unplanned

arrivals and the imbalance in planned patient occupancy, we focused our efforts on improving

the scheduling of planned patients; the only lever directly available to managers in this system.

16

14

12
Beds Occupied

10

0
Sun Mon Tue Wed Thu Fri Sat

Day of the week

Figure 7. A box-plot of daily bed occupancy by planned patients in the surgical Ward 1.

4.3 Applying the Models

A large portion of the study concerned obtaining and analyzing data. Our primary data

sources were; the Admissions, Discharge, Transfer System (ADT) and the Operating Room

Scheduling Office System (ORSOS). ADT is a patient tracking software that stores admission,

transfer or discharge time stamps. ORSOS is a scheduling and management system for surgical

services within the hospital. Both databases were linked via patient identification and surgery

18
times to create the dataset for modelling. This data was then crosschecked with a patient

discharge database and a manual nightly bed census count to insure data integrity. The final

dataset contained the following information for each case:

 responsible surgeon and surgical specialty

 patient type (DC, SS, SDSA)

 method of entry to the hospital (elective or emergency surgery)

 surgery date

 sequence of post-surgical wards occupied, and the length of stay within each

We observed that planned cases with patients off-servicing to non-designated surgical

wards were largely caused by operational as opposed to clinical reasons. Upon reviewing data

with ward managers, 7.7% of the data had paths that were inconsistent. Removing these cases

resulted in minor change to the average total length of stay of inpatients (1.5% decrease) with the

difference being statistically insignificant at the 0.05 level. A comparison between each

surgeon’s mix of cases in the final database and the historical mix ensured that a similar

proportion of procedures were represented.

The data was then entered into the BUS model and validated against historical records. A

direct validation of the model-reported bed occupancies was not possible since BUS was an

unconstrained model and intended to determine the impact of a SBS in an ideal world. Since the

logic was simple and historical data was used, we had confidence in our results. On the other

hand, we validated results graphically for a specific ward which had few patient relocations and

cancellations (Figure 8). We observed that when using historical surgical schedules to generate

planned arrivals and randomly generating unplanned arrivals using historical distributions, the

simulated bed occupancy patterns are similar to historical patterns. Results for other wards were

19
reviewed by hospital managers and planners to determine whether the model reported over/under

capacity patterns reflected the level off-servicing and cancellations observed in reality.

This dataset was also used to compute the bed demand parameters and surgical block

compositions for the SSO. However, there are several unique characteristics at RJH that required

minor modifications to the model. ORs are not a limiting constraint at RJH and Urology

surgeons are assigned two ORs when they operate. These surgeons alternate between both ORs

to minimize their idle time caused by the setup and cleanup. Thus a urologist is assigned two

blocks on days when he/she operates. In addition, two ORs were equipped to perform all

specialties; while the remaining 14 ORs can be used by any specialty except urology. Therefore,

a constraint is placed to limit the number of ORs allocated to Urology each day. See Appendix 3

for details.

35

30
Surgical Occuapncy: Ward 1

25

20
Simulated
Actual
15

10

0
05/2005 07/2005 09/2005 11/2005 01/2006 03/2006
Date

Figure 8. Actual vs. Simulated occupancies for Ward 1. Simulated data was generated from
20 runs using historical arrival times of planned patients and random arrival times of
unplanned patients. Each simulated case was randomly selected by the model.

20
5 Case Study Results and Analysis

Both SSO and BUS were used to explore opportunities to reduce peak surgical bed

utilization and smooth out day-to-day variability in bed occupancy. The impact of using the two

versions of SSO will be discussed first. Next, the development of surgical scheduling guidelines

and the resulting rules will be presented. Finally, the use of BUS as a standalone tool to evaluate

various ‘what-if’ scenarios is illustrated with examples.

5.1 Optimized Surgical Schedules

In our application, SSO included 47 surgeons, 74 unique surgical blocks, and 8 surgical

wards over a 4 week scheduling cycle. The base model consists of 1488 decision variables and

1574 constraints while the slate model with 2 slate choices in each block consists of 2968

decision variables and 3195 constraints. The models were solved with CPLEX 11 using GAMS

on an Intel Q6600 with 2 GB of RAM. Both the base model and the slate model provided good

feasible solutions (optimality gap ~5%) within 10 minutes. Optimality could not be reached after

6 hours of computation and only minor decreases were observed in the optimality gap (<1%).

Methods to reduce computational time through heuristics and alternate model formulations are

areas for future work.

The SSO base model output, prior to evaluation with BUS, suggests that significant

reduction in average peak occupancy can be achieved in each ward. Figure 9 shows that prior to

optimization there was considerable variability in average bed occupancies between days in all

wards. For example, in Ward 3, the maximum average occupancy was 11.5 beds on Day 23 and

the minimum average occupancy was 4 beds on Day 16. Post optimization, the maximum bed

occupancy dropped to 9 and the minimum bed occupancy increased to 7. This effectively

21
reduced peak occupancies by 2.5 beds and reduced the range of average bed occupancies from

7.5 to 2. Reductions in peak occupancies were observed across all wards and the total of these

decreases was predicted to be 8.5 beds. Consequently there is strong evidence that variation

caused by planned surgical patients can be reduced using the optimized schedule.

12
Before
10
Ward Bed Occupancy

2
Day Care
Short Stay
0
Ward 1
0 5 10 15 20 25
Ward 2
Day
Ward 3
12 Ward 4
After
Cardiac Care
10 Intensive Care
Ward Bed Occupancy

0
0 5 10 15 20 25
Day

Figure 9. Modelled average occupancy by planned patients in each surgical ward before
and after optimization using the base case optimization method.

The optimized schedule was then entered into BUS to assess the impact of variability and

unplanned patients on ward occupancies. Wards were initialized to zero at the beginning of every

22
run/replication. A warm up was used to allow the wards to reach a pseudo “steady state” prior to

collecting performance statistics. To determine an adequate duration for the warm up period, the

ward occupancies of several runs were analyzed graphically to determine the number of

simulated days necessary to achieve stable bed occupancies. It was found that 6 scheduling

cycles (24 weeks) was sufficient under all scenarios. Statistics were then collected from the

seventh surgical schedule cycle. This entire process is repeated for subsequent replications.

In our study, the total number of bed-days over capacity prior to optimization was 80

over a total of 277 inpatient cases in a 4 week period. The optimization resulted in a reduction of

16% or 13 bed-days over capacity. The practical significance of this is that up to 13 instances of

patient relocation to inappropriate wards and/or surgery cancellations could be avoided over a 4

week period.

The same analysis was repeated using the SSO slate model. The total number of cases

performed with the optimized schedule was higher than the initial schedule due to the lower

bound constraint on the number of cases performed by each surgeon. The optimized model

resulted in an increase of 15 surgical cases (4 SDSA and 11 SS) per 4 week period. These

additional cases increased surgical volume by 5%. The total number of bed-days over capacity

using this schedule was reduced by 9% to 7. This result is significant, demonstrating that the

hospital could both serve additional surgical cases and reduce the number of bed-days over

capacity simultaneously.

Our decision to limit choice to two slates per surgeon was driven primarily by

computational issues. However because of the limited number of slates, the resulting model

might be infeasible because the minimum demand constraint could not be satisfied. Slates were

23
pre-screened to ensure a feasible solution could be found. An alternative method to address this

issue would have been to make the demand constraint a soft constraint.

5.2 Surgical Scheduling Guidelines

One of the limitations of SSO is its inaccessibility to hospital planners. Past experience

suggests that optimization models are difficult to operate without expert knowledge; especially

when dealing with issues regarding infeasible solutions. Also, modification of constraints and

continual upkeep of the model to changing operational parameters would be difficult for planners

to perform. Therefore, it was concluded that the development of scheduling guidelines would be

beneficial to planners.

Optimized schedules from both model formulations exhibited characteristics that were

significantly different than surgical schedules in use at the time of the study. This was especially

evident in the slate optimized schedules where the selection of surgical blocks and slates

exhibited recurring patterns. These patterns have been analyzed and formulated into the

following scheduling guidelines:

1. Group surgical blocks with similar ward requirements together. Usually a group will

consist of one surgical specialty, but a group can include several specialties if these

specialties share the same ward.

2. Within each group, schedule surgeon blocks with high patient volumes and long

length of stay requirements (i.e. SDSA patients) at the beginning and the end of the

week. This would increase the occupancy of under-utilized surgical wards at the

beginning of the week and over the weekend. Surgeons with lower inpatient bed

requirements can be scheduled in the middle of the week to maintain utilization rates

across the week.

24
 3. For wards that close on weekends, schedule surgeons with high demand for short

length of stay cases (2 days) on Mondays and Wednesdays. Scheduling primarily on

Mondays and Wednesdays maximizes ward utilization and minimizes patient off-

servicing to inpatient wards on the weekend.

The main principle behind these guidelines is that surgical blocks should be scheduled

based on both surgical ward requirements and patient mix. Blocks of the same speciality tend to

have similar ward requirements but their patient mix can vary significantly. Some surgeons may

operate exclusively on outpatient cases while others on inpatient cases, both of which have

highly different downstream requirements. Only by understanding the differences in these

requirements can surgical planners make more informed scheduling choices to smooth

occupancies.

The proposed scheduling guidelines are general enough to apply to other hospitals with

similar ward configurations. While true optima are not achieved using these guidelines, surgical

planners can still create improved and implementable surgical schedules. In this way the

optimization results guide practice, but does not dictate it.

5.3 Scenario Analysis Using the Bed Utilization Simulator

The Bed Utilization Simulator is also a powerful standalone tool that can be used on a

“what-if?” basis to investigate schedule changes. We now describe how it was used to evaluate

new surgical schedules, assess the impact of increasing knee replacement cases, assess the

impact of changing ward capacities, and determine the number of surgical beds to protect in a

shared surgical and medical ward. For most scenarios, two hundred replications were required to

insure that the 95% confidence intervals for daily mean bed occupancies and bed-days over

25
capacity were within acceptable levels (<0.2 beds, <0.5 bed-days over capacity). Additional

replications were performed if these levels were not achieved.

5.3.1 Introducing a New Surgical Schedule

Hospital management was interested in determining the impact of a new surgical

schedule that was planned to be implemented shortly after our study was completed. This

schedule was evaluated using BUS; some results appear in Figure 10.

Ward 1 Ward 2
35 25

30
Ward Bed Occupancy

20
Ward Bed Occupancy

25

20 15

15
10
10
5
5

0 0
1 2 3 4 1 2 3 4
Week Week

Figure 10. Simulated bed occupancies for Ward 1 and Ward 2. The dotted line represents
the approved surgical capacity for the ward.

BUS identified several unique characteristics of the new surgical schedule. Results from

Ward 1 show a consistent weekly pattern of low bed utilization at the beginning of the week and

high bed utilization at end of the week. This result was expected as the new surgical schedule did

not differ significantly from previous schedules under which similar patterns had been observed.

The results also suggest that bed occupancies were slightly higher and more variable on the

second and fourth week of the schedule. Closer inspection of the new surgical schedule showed

that two orthopaedic surgeons whose cases required long lengths of stay were scheduled to

26
operate on the same day during these two weeks. Simulation results for Ward 2 reveals relatively

smooth surgical bed occupancies and lower utilization in comparison to Ward 1. Results also

indicate that the first week has lower bed occupancies and upon reviewing the surgical schedule,

this was attributed to a lack of surgeons with high inpatient demands scheduled on week 1.

5.3.2 Increasing the weekly volume of Knee Replacement Cases

Hospital management wished to investigate the impact of scheduling two additional knee

replacement procedures every Monday to reduce current knee replacement wait times. To

represent this scenario in BUS, a new surgeon who only performs knee replacements was created

and added to the schedule. Results showed that scheduling two additional knee replacement

cases would increase bed utilization across the week and as well, increase the likelihood that

Ward 1, an orthopaedics ward, would be over capacity. A comparison of average median bed

occupancies also indicates a significant change, increasing from 21.1 beds to 22.4 beds, while

bed-days over capacity increased by 8. To reduce these effects, management could either add one

additional bed to this ward to keep bed-days over capacity near previous levels or reschedule

other surgeons to decrease the anticipated peak bed occupancies in this ward (increase bed-days

over capacity by 3).

5.3.3 Changing Ward Capacities

BUS can also be used to evaluate whether the current allocation of beds across wards can

be improved. Using bed-days over capacity as the main metric, sensitivity analysis can be

performed on the current bed allocations to determine the impact of redistributing bed capacity.

Results from three wards appear in Table 1. Change in bed-days over capacity given an

increase or decrease change in current bed capacity. They show that the largest reduction in bed-

27
days over capacity can be achieved by adding one additional bed to Ward 1. If a bed were to be

eliminated, removing it from Ward 2 would have the least impact on bed-days over capacity.

Similar sensitivity results (less than 0.5 bed-days over capacity difference) are observed when

changing capacities while using optimized schedules.

It is important to note that the actual change in capacity would also depend on the cost of

operating one bed in each unit and whether there was staff and physical capacity to do so. This

information can be used as one measure by managers to help redistribute current beds or plan for

future bed capacities across wards.

Table 1. Change in bed-days over capacity given an increase or decrease change in current
bed capacity

Bed-Days Over Capacity

Surgical Ward
Current Increase One Bed Decrease One Bed

Ward 1 32.4 -8.4 +10.5

Ward 2 11.7 -4.1 +5.7
Ward 3 27.8 -7.3 +9.2

5.3.4 Setting Surgical Bed Protection Levels

Many hospitals use utilization rates to determine target ward capacities. However, this

provides no indication on the accessibility of this ward to patients. Instead, it has been suggested

that operating targets (Green, 2002) or specific access levels (Proudlove et al., 2006) be

established for wards. We use BUS to explore these approaches. We define the access level as

the probability that the demand for beds on a given day is less than or equal to the staffed ward

capacity. Thus, the higher the access level, the less the chance of the ward being over capacity.

28
 In the context of our case study we explore improving access for surgical patients in

Ward 2. At the time of the study, Ward 2 was a shared ward with 31 beds, 20 of which were

protected for surgical patients. Assuming that other patients can be properly managed, the

number of protected beds for surgical patients can be determined using BUS. If a 75% access

level for surgical patients is desired, the model suggests that 18 beds should be protected. If a

95% access level is desired, then 25 beds should be protected.

6 Conclusions

This paper describes our development and use of the Bed Utilization Simulator and the

Surgical Schedule Optimizer, to support surgical scheduling and bed management. These tools

are transparent to users and are easily adaptable to other hospital systems. Compared to other

simulation models described in the literature, the benefits of BUS include simple logic, short

development time, an intuitive user interface and that it MS Excel based. Despite its simplicity,

BUS is capable of analyzing many important "what-if?" questions that continually challenge

surgical managers. In addition, bed management concepts such as the impact of variability on

ward accessibility can be captured and conveyed to managers and planners. Thus, BUS doubles

as a valuable teaching tool.

The SSO model demonstrates that significant reduction in off-servicing and surgical

cancellations can be achieved through enhanced SBSs. Results from using the slate model shows

that surgical throughput can be increased while decreasing peak bed occupancies. However, SSO

is challenging for non-technical users for many reasons so scheduling guidelines were developed

to allow for incorporating the optimization results into practice.

The results of our study have influenced surgical operations at Royal Jubilee Hospital.

The planners are using results from the project to support future surgical schedule revisions.

29
Surgical planners recognize that BUS can assist them in testing new schedules on a "what-if?"

basis. The scheduling guidelines also challenged previous assumptions. Prior to the project,

planners believed scheduling specialties evenly across the days of the week would smooth bed

occupancy; our models clearly showed this was not the case. These guidelines are now being

used on a daily basis to support ongoing scheduling decisions. A newly established Operations

Research department has since been placed in charge of the surgical scheduling project and is

evaluating the use of these tools at other facilities within the health authority.

Ongoing challenges exist in ensuring that models such as these are used in the future.

While BUS requires relatively basic data for its operation, hospitals also need to have an IM/IT

infrastructure that collects the appropriate information to support it. A substantial amount of time

was required to clean and validate data prior to generating the BUS input database. Therefore to

allow for the dissemination of BUS, ongoing work to integrate operations based data into current

information systems is required to lay the foundation for current and future operations research

based studies.

There are several fruitful areas of research arising from this work. Investigations into

methods for improving the slate version of the SSO to increase the number of possible slates and

reduce computation time are warranted. In addition, there is an opportunity to improve

management of unplanned surgical patients. While little variation in mean bed occupancies of

unplanned patients was observed across the week, large variations in bed occupancies exist

within each day. Further analysis into possible options of managing unplanned surgeries by the

division of patients into urgent and emergent should be considered.

30
Acknowledgements

We wish to thank Eileen Goudy, David McCoy, Andrea Boardman, Sheri Yager and all

Vancouver Island Health Authority staff involved in this study for their help and continued

support. Also, we want to acknowledge Yixin Chen and Mahesh Nagarajan for their earlier work

on surgical scheduling. The results reported here were developed in the Centre for Operations

Excellence in the University of British Columbia’s Sauder School of Business.

References

Blake J., M. Carter, L. O'Brien-Pallas, L. McGillis-Hall. 1995. A surgical process management

tool. Medinfo, 8(2), 1527-1531.

Blake J., J. Donald. 2002. Mount Sinai Hospital Uses Integer Programming to Allocate

Operating Room Time. Interfaces, 32(2), 63-73.

Beliën J., E. Demeulemeester. 2007. Building cyclic master surgery schedules with levelled

resulting bed occupancy. European Journal of Operational Research, 176(2), 1185-1204.

Beliën J., E. Demeulemeester, B. Cardoen. 2009. A decision support system for cyclic master

surgery scheduling with multiple objectives. Journal of Scheduling, 12(2), 147-161.

Beliën J., E. Demeulemeester, B. Cardoen. 2006. Visualizing the demand for various resources

as a function of the master surgery schedule: A case study. Journal of Medical Systems,

30(5), 343-350.

Carter M., J. Blake. 2005. Using Simulation in an Acute-care Hospital: Easier Said Than Done.

Operations Research and Health Care: A Handbook of Methods and Applications, Ed. M.

Brandeau, F. Sainfort, W. Pierskalla. Kluwer, New York, 191-215.

31
Cardoen B., E. Demeulemeester, J. Beliën. 2008. Operating room planning and scheduling: A

literature review. Technical Report KBI 0807, Katholieke Universiteit Leuven, Belgium.

Denton B., J. Viapiano, A. Vogl. 2007. Optimization of surgery sequencing and scheduling

decisions under uncertainty. Health Care Management Science, 10(1), 13-24.

Dexter F., R.D. Traub. 2002. How to Schedule Elective Surgical Cases into Specific Operating

Rooms to Maximize the Efficiency of Use of Operating Room Time. Anesthesia and

Analgesia, 94(4), 933-942.

Everett J.E. 2002. A Decision Support Simulation Model for the Management of an Elective

Surgery Waiting System. Health Care Management Science, 5(2), 89-95.

Green L.V. 2002. How many hospital beds? Inquiry, 39(4), 400-412.

Harper P.R. 2002. A Framework for Operational Modelling of Hospital Resources. Health Care

Management Science, 5(3), 165-173.

Harper P.R., A.K. Shahani. 2002. Modelling for the planning and management of bed capacities

in hospitals. The Journal of the Operational Research Society, 53(1), 11-18.

Henderson S.G., Mason A.J. 2005. Ambulance Service Planning: Simulation and Data

Visualization. Operations Research and Health Care: A Handbook of Methods and

Applications, Ed. M. Brandeau, F. Sainfort, W. Pierskalla. Kluwer, New York, 77-102.

Isken M.W., B. Rajagopalan. 2002. Data Mining to Support Simulation Modeling of Patient

Flow in Hospitals. Journal of Medical Systems, 26(2), 179-197.

Jacobson S., S. Hall, R. Swisher. 2006. Discrete-Event Simulation of Health Care Systems.

Patient Flow: Reducing Delay in Healthcare Delivery. Ed. R.W. Hall. Springer, Los

Angeles, 211-252.

32
Jun J.B., S.H. Jacobson, J.R. Swisher. 1999. Application of discrete-event simulation in health

care clinics: A survey. The Journal of the Operational Research Society, 50(2), 109-123.

Persson M., J.A. Persson. 2009. Health economic modeling to support surgery management at a

Swedish hospital. Omega, 37(4), 853-863.

Pitt M. 1997. A Generalised Simulation System to Support Strategic Resource Planning In

Healthcare. Proceedings of the 1997 Winter Simulation Conference, 1155-1162.

Postl B.D. 2006. Final Report of the Federal Advisor on Wait Times. http://www.hc-

sc.gc.ca/hcs-sss/pubs/system-regime/2006-wait-attente/index-eng.php. Accessed Nov 08.

Proudlove N.C., S. Black, A. Fletcher. 2006. OR and the challenge to improve the NHS:

modelling for insight and improvement in in-patient flows. The Journal of the Operational

Research Society, 58, 145-158.

Santibanez P., M. Begen, D. Atkins. 2007. Surgical block scheduling in a system of hospitals: an

application to resource and wait list management in a British Columbia health authority.

Health Care Management Science, 10(3), 269-282.

Testi A., E. Tanfani, and G. Torre. 2007. A three-phase approach for operating theatre schedules.

Health Care Management Science, 10(2), 163-172.

Van Oostrum J.M., M. Van Houdenhoven, J.L. Hurink, E.W. Hans, G. Wullink, G. Kazemier.

2008. A master surgery scheduling approach for cyclic scheduling in operating room

departments. OR Spectrum, 30(2), 355-374.

VanBerkel P.T., J.T Blake. 2007. A comprehensive simulation for wait time reduction and

capacity planning applied to general surgery. Health Care Management Science, 10(4),

373-385.

33
Vissers J.M.H., I. Adan, J.A. Bekkers. 2005. Patient mix optimization in tactical cardiothoracic

surgery planning: A case study. IMA Journal of Management Mathematics, 16(3), 281-304.

Appendix 1: Surgical Scheduling Optimizer Base Model

The binary decision variables Xbi,w are used to indicate whether a surgical block b is

scheduled on a day i of week w in the schedule. Each surgical block represents a unique

combination of a surgeon d and duration (NumORb). This duration is expressed in terms of OR-

Days. A surgeon who requires one OR for half a day would require 0.5 OR-Days while a

surgeon who requires two ORs for an entire day would require 2 OR-Days. Blocks

corresponding to surgeon d are part of the set B(d). The model assumes that the assignment of

blocks to ORs can be managed by surgical planners.

The Bedbp,u,j,i,w parameter captures the expected bed demand (in nights) for a patient (of

block b and type p) in each ward u and on each day j, given the block is scheduled on day i of

week w in the cyclic surgical schedule. For example suppose that a surgical block b and patient

type p combination always resulted in 50% of the patients staying for 3 nights in Ward 1, and the

other 50% of the patients staying for 1 night in Ward 1 and 1 night in Ward 2 before discharge.

Then Bedb p,u,j,i,w would equal 1 in Ward 1 on day 1, 0.5 in Ward 1 and Ward 2 on day 2, and 0.5

in Ward 1 on day 3. In outpatient wards (i.e. Day Care Ward), Bedb p,u,j,i,w is expressed as bed-

days. A 6 hours stay post operation would be represented as 0.25 bed-days on day 1. To capture

the expected occupancies of a block, this parameter is multiplied by the expected number of

patients of each patient type in the block (NumCasesbp).

The model includes several constraints. The first constraint (1) is an OR capacity

constraint where the total OR-day requirements on any given day cannot exceed the total OR-

days available that day (ORperDay i,w). The second constraint (2) limits total number of OR-days

34
allowed for each surgeon on each day (ORperDaySurgeond i,w). The third constraint (3) limits the

number of instances scheduled for each surgical block in each week (WeekBlockbw) to distribute

blocks across the scheduling horizon. The fourth constraint (4) ensures that the total number of

surgical blocks scheduled equals a predefined volume (TotalBlockb). This is usually chosen to be

consistent with the “as is” frequencies at the time of the study. An additional inequality (5)

defines the maximum bed utilization in each ward MDu. The objective of the model (6) is to

minimize the sum of maximum bed occupancy over surgical wards.

Sets

b: Surgical blocks u: Wards

i: Weekdays (1…5) d: Surgeons

w: Weeks of the surgical schedule p: Patient types

j: Days of the surgical schedule (1…7·w) B(d): Blocks associated with surgeon d

Parameters

Bedb p,u,j,i,w: Expected number of bed-nights/days used by one patient of type p in ward u

on day j due to surgical block b scheduled on day i of week w.

NumORb : OR-days required for each surgical block b

ORperDay i,w : OR-days available on day i of week w

ORperDaySurgeond i,w : OR-days available on day i of week w for surgeon d

WeekBlockbw : Number of Blocks b available in week w

TotalBlockb : Total number of blocks b in the surgical schedule

NumCasesbp: Number of cases for each patient type p in surgical block b

Decision Variables

Xbi,w : 1 if block b is scheduled on day i of week w

35
 0 otherwise

MDu : Maximum number of beds in use in ward u over the scheduling period

Constraints

Daily OR-day capacity: X

b
b
i,w
 NumORb  ORperDay i ,w i, w (1)

Daily OR-day capacity for each surgeon:

X  NumOR b  ORperDaySu rgeon d i, w, d

i ,w i,w
b (2)
bB ( d )

X  WeekBlockb b, w
i,w w
Weekly surgical block capacity: b (3)
i

 X  TotalBlock b b
i,w
Surgical blocks balance: b (4)
w i

Definition of the maximum bed utilization across the scheduling period in each ward:

 X  NumCasesb  Bed b  MDu j , u

i ,w p p ,u , j ,i , w
b (5)
w i b p

Objective

Minimize the summation of the maximum bed occupancy in each surgical ward:

Min MDu (6)

u

Appendix 2: Surgical Scheduling Optimizer Slate Selection Model

In this model, the decision variables are modified to include slate choice s. The parameter

NumCasesbp,s replaces NumCasesbp which now stores the number of each patient type for each

slate. Two additional constraints are also added. The first constraint (7) ensures only one slate is

selected for each instance of a surgical block and the second constraint (8) ensure the number of

36
cases performed for each patient type is greater than historical demand for each surgeon

(Totalcasesdp).

Sets

s : Model addition - Slate choices

Parameters

TotalCasesd p : Model addition - Number of historical cases that needs to be met for each

surgeon d and patient type p

NumCasesb p,s : Replaces NumCasesbp - Number of cases for each patient type p in

surgical block b with slate s

Decision Variables

Xbi,w,s : Replaces Xbi,w 1 if block b is scheduled on day i of week w with slate s

0 otherwise

Constraints

X 1 i, w, b
i , w, s
Model addition - Choose at most one slate: b (7)
s

Model addition - Number of cases must be at least equal to historical volumes:

   X  NumCases b  TotalCases d p, d

i , w, s p,s p
b (8)
w i bB ( d ) s

Replaces (1):  X
b s
b
i , w, s
 NumORb  ORperDay i , w i, w (9)

 X  NumORb  ORperDaySu rgeon d i, w, d

i , w, s i ,w
Replaces (2): b (10)
bB ( d ) s

 X  WeekBlockb b, w
i , w, s w
Replaces (3): b (11)
i s

 X  TotalBlock b b
i , w, s
Replaces (4): b (12)
w i s

37
  X  NumCasesb  Bed b  MDu j , u (13)
i ,w,s p ,s p ,u , j ,i , w
Replaces (5): b
w i b p s

Objective

The objective function remains the same as (6).

Appendix 3: Surgical Scheduling Optimizer Additional RJH Constraints

A constraint on the number of OR-days available on each day of the schedule for

Urology (ORperDayUroli,w) is applied in the RJH setting. The following additions are made to

both the base model and the slate model.

Sets

y: Model addition - Urology specialty

B(y): Model addition - Blocks associated with Urology

Parameter

ORperDayUrol i,w : Model addition - OR-day available on day i of week w for Urology

Constraints

Base model addition - OR-day capacity for Urology specialty:

X  NumORb  ORperDayUrol i , w i, w

i,w
b (14)
bB ( y )

Slate model addition - OR-day capacity for Urology specialty:

 X
bB ( y ) s
b
i , w, s
 NumORb  ORperDayUrol i ,w i, w (15)

38