Accepted Manuscript

A Two-Stage Stochastic Programming Approach for Nurse Scheduling Problem

in a Real Word Hospital

Mohsen Bagheri, Ali Gholinejad Devin, Azra Izanloo

PII: S0360-8352(16)30050-X
DOI: http://dx.doi.org/10.1016/j.cie.2016.02.023
Reference: CAIE 4270

To appear in: Computers & Industrial Engineering

Received Date: 30 January 2015

Revised Date: 1 February 2016
Accepted Date: 24 February 2016

Please cite this article as: Bagheri, M., Devin, A.G., Izanloo, A., A Two-Stage Stochastic Programming Approach
for Nurse Scheduling Problem in a Real Word Hospital, Computers & Industrial Engineering (2016), doi: http://
dx.doi.org/10.1016/j.cie.2016.02.023

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Mohsen Bagheri1, Ali Gholinejad Devin*2, Azra Izanloo*3
1
Department of Industrial Engineering
Sadjad University of technology, Mashhad, Iran
M_Bagheri@sadjad.ac.ir
2
Department of Industrial Engineering
Sadjad University of technology, Mashhad, Iran
Al.Gh130@sadjad.ac.ir
3
MSc in Medical Education
Research and Education Department, Razavi Hospital, Mashhad, Iran
A.izanloo@yahoo.com

[*]: corresponding Author

1
A Two-Stage Stochastic Programming Approach
for Nurse Scheduling Problem in a Real Word
Hospital
ABSTRACT
Given its complexity and relevance in healthcare, the well-known Nurse Scheduling Problem (NSP) has
been the subject of several researches and different approaches have been used for its solution. The
importance of this problem comes from its critical role in healthcare processes as NSP assigns nurses to
daily shifts while respecting both the preferences of the nurses and the objectives of hospital. Most
models in NSP literature have dealt with this problem in a deterministic environment, while in the real-
world applications of NSP, the vagueness of information about management objectives and nurse
preferences are sources of uncertainties that need to be managed so as to provide a qualified schedule. In
this study, we propose a stochastic optimization model for the Department of Heart Surgery in Razavi
Hospital, which accounts for uncertainties in the demand and stay period of patients over time. Sample
Average Approximation (SAA) method is used to obtain an optimal schedule for minimizing the regular
and overtime assignment costs, with the numerical experiments demonstrating the convergence of
statistical bounds and moderate sample size for a given numerical experiment. The results confirm the
validity of the model.

Keywords: Nurse Scheduling, Uncertainty, Stochastic programming, Sample average approximation,

Recourse action

1 Introduction:
The NSP is a well-known combinatorial optimization problem that has encouraged many researchers
to develop exact and (meta-) heuristic approaches to obtain a qualified solution. The NSP involves
constructing a schedule for nursing staff and assigning nurses to different shifts based on individual and
system preferences within the framework of government regulations.
Given its various constraints, objectives and many possible combinations, NSP is a complex
problem. In the NSP, the nurse manager creates a schedule based on nurse preferences and/or scheduling
requirements. Here, the problem is finding a schedule that both supports the preferences of nurses and
satisfies the individual and system preferences as well as government regulations.
Since there are a plethora of constraints and many possible solutions for NSP, different approaches
such as optimization, artificial intelligence and heuristic and meta-heuristic approaches have been used to
solve it. In the following, an overview of literature is presented.
In the early 1970s, (Warner, 1967) presented a multiple-choice programming in which each nurse
described a group of variables each of which serving as a possible schedule for that nurse in the planning
horizon. (Millar & Kiragu, 1998) provided a mathematical model for minimizing nurse assignment costs
in both cyclic and non-cyclic types for NSP.

2
 (Venkataraman & Brusco, 1996) presented a mixed-integer liner programming for evaluating nurse
preference and management regulations. (Ozkarahan, 1989) proposed a flexible decision support system
that sought to satisfy the preferences of both hospitals and nurses. (Jaumard, et al, 1998) proposed a
generic binary linear programming model for NSP with the aim of minimizing salary costs and
maximizing both nurse preferences and team balance so as to satisfy the demand coverage constraints.
(Klinz, et al, 2007) proposed two mathematical models to minimize the total number of work shifts. (Bard
& Purnomo, 2005) offered an integer programming model for nurse assignment for both regular and pool
nurses under different conditions to satisfy the demands stipulated in the planning horizon.
(Hattori, et al, 2005) presented a nurse scheduling system based of Constraint Satisfaction Problem
(CSP) with different levels of importance and subject to dynamic change. (Parr & Thompson, 2007) used
Sawing and Noising with simulated annealing in NSP to ensure the sufficiency of nurse demands in each
shift. (Punnakitikashem, et al, 2008) proposed a stochastic integer programming model for NSP to
minimize the workload penalty on nurses and satisfy the expected demands in the planning horizon. (Fan,
et al, 2013) used binary integer linear programming to maximize nurse preferences and hospital
regulations. (Li & Aickelin, 2004) used the Bayesian optimization and classifier systems for NSP to
minimize the total preference cost of nurses in the planning horizon.
(Topaloglu & Selim, 2007) used a fuzzy goal programming model for NSP to measure uncertainty in
an objective evaluation of hospital regulations and nurse preferences. (Topaloglu & Selim, 2010)
proposed a multi-objective integer programming for NSP to both produce an equitable schedule for nurses
and satisfy hospital management objectives. (Maenhout & Vanhoucke, 2007) presented a novel
electromagnetism meta-heuristic technique for the NSP to minimize the total pattern penalty costs in the
planning horizon. (Landa-silva & Le, 2008) used a multi-objective approach to cope with real-world
uncertainties in NSP. To do so, they used an evolutionary algorithm to achieve high-quality non-
dominated schedules.
(Tsai & Li, 2009) presented a two-stage mathematical modeling approach for the NSP with respect to
hospital management requirements, government regulations, and nurse preferences. (Ohki, et al, 2010)
used a new approach using cooperative genetic algorithm (CGA) to solve NSP. (Zhang, et al, 2011)
proposed a hybrid Swarm-based optimization algorithm in hospital environments that incorporated
genetic algorithm and variable neighborhood search to address highly-constrained NSP with respect to
hospital management requirements.
In Table 1. A brief classification of models in literature is presented.

{Please insert Table.1 about here}

In the real-world applications of NSP, the vagueness surrounding the target values of management
and nurse preferences are a source of uncertainties that need to be addressed to provide a high-quality
schedule. For this purpose, the basic parameters such as the demand and patient-stay period are stochastic
in nature and the distribution of these parameters is determined from historical data.
The rest of this article is organized as follows: in Section 2, the proposed optimization model for NSP
is presented and the structure of the model is investigated. In Section 3, the solution approach is
introduced with a detailed description of SAA method. Numerical experiments are presented in Section 4.
Finally, concluding remarks are made in Section 5.

3
2 The Model for Nurse Scheduling Problem
In the following, the indices, parameters, variables and our mathematical model describes Stochastic
Nurse Scheduling Problem (SNSP).
{Please insert Table.2 about here}

4
 The main objective of the model is to minimize the regular and overtime assignment costs. Let
be the corresponding probability of scenario and . (1), ensures that nurse
1 and 2 are assigned to shift one in allowed dates. According to hospital regulations, nurses 1 and 2 (head
nurses) should be assigned to shift one (morning shift) in working days. In this hospital, no one is allowed
to work on two consecutive afternoon and night shifts. (2) Applies this constraint. If a nurse is assigned to
a night shift, he/she is not allowed to work in the following days. (3) Considers this limitation. (4) Shows
that every nurse should work at least 26 shifts, knowing that shift 3 (night shift) has a double work load
compared to shifts 1 and 2 (morning and afternoon shifts). (5) Shows how empty beds in shift j at date t
can be calculated from total beds available and the number of patients that are present in shift j in date t.
(6-9) calculate the number of accepted patients in shift j in date t considering the remaining capacity. (10)
Shows the total number of patients present in shift j in date t. (11-14) computes additional nurses in shift j
in date t. For this purpose, if the number of available nurses is less than required, the number of overtime
nurses will be a positive value, as calculated by (14).
In this research and are assumed to be uniformly (discretely) distributed: and
for
An exact solution can be achieved by enumeration for a small-size problem. However, if the size of
problem is increased, the model may become unmanageable. In this article, Sample Average
Approximation (SAA) algorithm has been used as a solution strategy for Stochastic Nurse Scheduling
Problem. To apply SAA algorithm, the SNSP is reformulated using recourse action model and the basic
properties of the new formulation are investigated.

2.1 Stochastic Programming with Recourse Action

The most important group of stochastic programming models, known as recourse models, is calculated by
allowing recourse actions after realizations of the random variables ( ). Considering a first-stage
decision for all possible realizations, (q, T, h) of (q, T, h). Are compensated at minimum costs
by selecting the second-stage decision as an optimal solution of the second-stage problem.

In the second-stage problem, is the recourse action unit cost vector and the recourse matrix
determines the available technology. We will use to characterize all randomness in the
problem. The objective function of this second-stage problem, which determines the minimum recourse
action costs as a function of the first-stage decision x and a realization of , is defined by ; the
expectation = provides the recourse action expected costs associated with a first-stage
decision . Thus, the two-stage recourse action model is:

5
 Where, indicates the total expected costs of a decision , (Stougie, et al, 2003). The SNSP
can be formulated using the following recourse model.

Where is the recourse action function, and

6
 Given the large random data vector , it is highly difficult to evaluate the value of as it
involves solving a large number of similar Integer Linear Programming (ILP). (Birge, J. Louveaux, F,
1997) Since, it is difficult to provide an exact solution of the proposed model, the following structure of
the SNSP2 is proposed to give an approximation.
3 Solution approach: Sample Average Approximation
Suppose we have a sample of N replications of the random vector so that each ,
has the same probability distribution. We can approximate the value of expectation
function by = averaging . Thus, the true (expectation)
problem is equal to (Shapiro, A. and Philpot, A., 2007). The following
mathematical model explains the SAA problem of the SNSP with a sample size of N.

St:

7
 (Ahmed, S., et al, 2002) proposed a general SAA algorithm for this type of Stochastic Discrete
Optimization Problem (SDOP). The procedure is as follows:
Suppose is the number of replications in the sample, is the number of scenarios in the sample
problem, and is the number of sample used to estimate for a given feasible
solution .
1. For , perform step 1.1 through 1.3
1.1 Generate sample .
1.2 Solve the proposed problem and let and for optimal objective function value and
optimal solution.
1.3 Generate independent random samples and evaluate the objective function value
and variance for feasible solution . There are different methods to obtain a feasible
solution , including a deterministic problem with expected value parameters known the
Expected Value Problem (EVP). Another method is solving a problem without stochastic
constraints, which is used for a two-stage stochastic problem.

2. Calculate and .

3. For each solution , calculate the optimal gap by and variance

of . Finally, choose one of the M candidate solutions.

In the algorithm, we can obtain a lower bound (LB) and an upper bound (UB) for the true optimal
value by estimating and respectively, (Mak, et al, 1999). Is an unbiased estimator of
optimal objective function that Here is an unbiased estimator of
the true objective value, but

8
4 Numerical experiments
4.1 Example data
This section describes a case study conducted in the Department of Heart Surgery in Razavi Hospital.
In this department, there are sixteen nurses ( ), a head nurse and an assistant head nurse
( ) with 31 working days each divided into 3 shifts (Morning, Afternoon and Night). The
system capacity (total available beds) is 25 beds.
To evaluate the discrete distribution of the number of new patients arriving in shift j in date t and
the number of shifts that a patient k remains in system, the actual data of the hospital were analyzed.
The disruption of these stochastic parameters is estimated as: and
for the estimated regular and overtime assignment costs
for each shift are 15$ and 18$ respectively. Coverage factor of a nurse is considered , meaning
that each nurse can support five patients.

4.2 Experimental results

In this section, the experimental results achieved from the implementation of our approach in
Department of Heart Surgery in Razavi Hospital are discussed. In our approach, we used N = [1-2-3-
4-5-6-7-8-9-10-20-30-40-50-100], M = 20 and = 30,000 for SAA algorithm presented in Section 3.
{Please insert Figure.1 about here}

Figure 1 demonstrates the SAA algorithm optimal solution changes relative to different sample
sizes of N. Although the original sample space in SNSP model is extremely large, a high-quality
solution can be obtained by a relatively small sample size. For the solution of numerical experiment, a
sample size of 100 can provide an acceptable solution.
The sample size (i.e. the size of scenario set ) in numerical experiments is calculated by the
sample size of and which are the possible realization of the number of new patients arriving in
shift j as well as the number of shifts that patient k remains in the system respectively.
All experiments are not needed to evaluate the effects of variable sizes in our SNSP model. It is
proven by (Shapiro & Mello, 2001), however, the sample size required to obtain optimal solutions
and optimal values functions is logarithmical to the variable size. Therefore, the required sample size
increases linearly proportional to a rise in the number of new patients arriving in shift j and number of
shifts that patient k remains in system. The results concerning the achievement of optimal solutions
and optimal values show the convergence of the SAA solutions with exponential rates.
The EVP solution is obtained by substituting random parameters and by their mean values
and then solving the deterministic problem. The results of solving SNSP model shows that the
deterministic problem (EVP) yields unsatisfactory solutions.
We can obtain a lower bound (LB) and an upper bound (UB) for true optimal solutions and
optimal values as functions by estimating and respectively. The former is an unbiased
estimator of the optimal objective function in which In estimating the
lower bound, we can generate independent samples of the uncertain parameters, each with the size

9
 and by solving the corresponding SAA problems, the optimal objective values can be
obtained Then, it can be estimated that:

It is an unbiased estimator of , and therefore a statistical lower bound to . is an

unbiased estimator of the true objective value where . In calculating each
solution of random sample , the objective functions are evaluated and a statistical
upper bound to the optimal value ( ) of the stochastic programming with integer recourse is
provided. For a fixed sample size, the proposed statistical and deterministic lower and upper bound
techniques can be useful to validate the quality of a candidate optimal solution.
In implementation of numerical experiments, the value of M (number of replications in the
simulation procedure) is set to 20. Table 3 shows statistical lower and upper bounds, indicating that
20 replications suffice to obtain a reasonable confidence interval for statistical lower and upper
bounds.
If the variances of statistical lower and upper bounds are too large, the value of M should be
increased.
{Please insert Table.3 about here}

The assignment of nurses to optimal sequence in the planning horizon is demonstrated in Table 3.
When assigning nurses to regular and overtime shifts, the proposed deterministic constraints should
be considered.
{Please insert Table.4 about here}

10
 A summary of the simulation procedure and the gap calculation between stochastic nurse
scheduling problem (SNSP) and expected value problem (EVP) is given in Figure 2. Arbitrarily, the
run length is determined as 100 iterations. The results of solving SNSP model show that the solution
presented by the deterministic problem (EVP) is less desirable than the one provided by the SAA
problem, and there is an evident difference between SNSP and EVP solutions.
{Please insert Figure.2 about here}

4.3 Evaluation of stochastic solutions

Here, our proposed SNSP procedure and EVP methods are compared via simulation. The summary
of the simulation procedure is given in Figure 3. According to the simulation results, it can be argued
that the solution provided by SNSP is better than EVP solution in uncertain environments.
{Please insert Figure.3 about here}

5 Conclusions

In the real-world applications of the NSP, multiple sources of uncertainty derived from
management objectives and nurse preferences should be considered to provide qualified schedules. In
this paper, the uncertainties related to the demand and stay period of patients have been addressed in
this study. The distribution of stochastic variables is determined from historical data. This study
proposes a two-stage stochastic optimization model for nurse scheduling in accordance with Razavi
Hospital regulations in an uncertain environment. Sample Average Approximation (SAA) algorithm
was used to solve the proposed model and numerical experiments were conducted for a model with
real data in Razavi Hospital. Then, the convergence of statistical bounds with the moderate sample
size was studied. Finally, a simulation study was conducted to show the advantages of SNSP over
EVP.
There are several venues of research in the field of nurse scheduling problem. First, since patients
may have multiple diagnoses during their stay in a hospital unit, different types of priority, demand
and stay period can be considered. Second, given the stochastic demand for resources in a hospital
unit, a model can be designed that regard the service level in determining the number of nurses.
Finally, the education, experience and skill of nurses should be included as important factors in their
assignments, so that we can evaluate the restriction of nurse assignments in the first stage based on
these factors.
REFERENCES
Warner, D. M. (1976). Scheduling Nursing Personnel According to Nursing Preference: A Mathematical
Programming Approach. Operations Research, 24(5), 842-856. doi: 10.1287/opre.24.5.842
Millar, H. H., & Kiragu, M. (1998). Cyclic and non-cyclic scheduling of 12 h shift nurses by network programming.
European Journal of Operational Research, 104(3), 582-592. doi
Venkataraman, R., & Brusco, M. J. (1996). An integrated analysis of nurse staffing and scheduling policies. Omega,
24(1), 57-71.
Ozkarahan, I. (1989). A flexible nurse scheduling support system. Comput Methods Programs Biomed, 30(2-3),
145-153.

11
Jaumard, B., Semet, F., & Vovor, T. (1998). A generalized linear programming model for nurse scheduling.
European Journal of Operational Research, 107(1), 1-18. Doi
Klinz, B., Pferschy, U., & Schauer, J. (2007). ILP Models for a Nurse Scheduling Problem. In K.-H. Waldmann &
U. Stocker (Eds.), Operations Research Proceedings 2006 (Vol. 2006, pp. 319-324): Springer Berlin
Heidelberg.
Bard, J. F., & Purnomo, H. W. (2005). Short-term nurse scheduling in response to daily fluctuations in supply and
demand. Health Care Manag Sci, 8(4), 315-324.
Hattori, H., Ito, T., Ozono, T., & Shintani, T. (2005). A Nurse Scheduling System Based on Dynamic Constraint
Satisfaction Problem. In M. Ali & F. Esposito (Eds.), Innovations in Applied Artificial Intelligence (Vol.
3533, pp. 799-808): Springer Berlin Heidelberg.
Parr, D., & Thompson, J. M. (2007). Solving the multi-objective nurse scheduling problem with a weighted cost
function. Annals of Operations Research, 155(1), 279-288. doi: 10.1007/s10479-007-0202-4
Punnakitikashem, P., Rosenberger, J., & Buckley Behan, D. (2008). Stochastic programming for nurse assignment.
Computational Optimization and Applications, 40(3), 321-349. doi: 10.1007/s10589-007-9084-2
Fan, N., Mujahid, S., Zhang, J., Georgiev, P., Papajorgji, P., Steponavice, I., . . . Pardalos, P. (2013). Nurse
Scheduling Problem: An Integer Programming Model with a Practical Application. In P. M. Pardalos, P. G.
Georgiev, P. Papajorgji & B. Neugaard (Eds.), Systems Analysis Tools for Better Health Care Delivery
(Vol. 74, pp. 65-98): Springer New York.
Li, J., & Aickelin, U. (2004). The Application of Bayesian Optimization and Classifier Systems in Nurse
Scheduling. In X. Yao, E. Burke, J. Lozano, J. Smith, J. Merelo-Guervs, J. Bullinaria, J. Rowe, P. Tio,
A. Kabn & H.-P. Schwefel (Eds.), Parallel Problem Solving from Nature - PPSN VIII (Vol. 3242, pp.
581-590): Springer Berlin Heidelberg.
Topaloglu, S., & Selim, H. (2007). Nurse Scheduling Using Fuzzy Multiple Objective Programming. In H. Okuno &
M. Ali (Eds.), New Trends in Applied Artificial Intelligence (Vol. 4570, pp. 54-63): Springer Berlin
Heidelberg.
Topaloglu, S., & Selim, H. (2010). Nurse scheduling using fuzzy modeling approach. Fuzzy Sets and Systems,
161(11), 1543-1563.
Maenhout, B., & Vanhoucke, M. (2007). An electromagnetic meta-heuristic for the nurse scheduling problem.
Journal of Heuristics, 13(4), 359-385. doi: 10.1007/s10732-007-9013-7
Landa-silva, D., & Le, K. (2008). A Simple Evolutionary Algorithm with Self-adaptation for Multi-objective Nurse
Scheduling. In C. Cotta, M. Sevaux & K. Srensen (Eds.), Adaptive and Multilevel Metaheuristics (Vol.
136, pp. 133-155): Springer Berlin Heidelberg.
Tsai, C.-C., & Li, S. H. A. (2009). A two-stage modeling with genetic algorithms for the nurse scheduling problem.
Expert Systems with Applications, 36(5), 9506-9512.
Ohki, M., Uneme, S.-y., & Kawano, H. (2010). Effective Mutation Operator and Parallel Processing for Nurse
Scheduling. In V. Sgurev, M. Hadjiski & J. Kacprzyk (Eds.), Intelligent Systems: From Theory to Practice
(Vol. 299, pp. 229-242): Springer Berlin Heidelberg.
Zhang, Z., Hao, Z., & Huang, H. (2011). Hybrid Swarm-Based Optimization Algorithm of GA & VNS for Nurse
Scheduling Problem. In B. Liu & C. Chai (Eds.), Information Computing and Applications (Vol. 7030, pp.
375-382): Springer Berlin Heidelberg.
Stougie, L. Maartin, H. Vlerk, V. (2003). Approximation in Stochastic Integer Programming. Mathematics Subject
Classification: 90C15, 90C11.pp. 457-468
Birge, J. Louveaux, F. (1997). Introduction to Stochastic Programming. Springer, New York.
Shapiro, A and Philpott, A. (2007). A Tutorial on Stochastic Programming. Springer, New York.
Kleywegt, A. Shapiro, A. Homem-de-Mello, T. (2001). The sample average approximation method for stochastic
discrete optimization. SIAM Journal on Optimization 12 (2), 479502.
Ahmed, S., Shapiro, A., & Shapiro, E. (2002). The sample average approximation method for stochastic programs
with integer recourse. Submitted for publication.
Mak, W.K., Mortonand, D.P., Wood, R.K., (1999). Monte Carlo bounding techniques for determining solution
quality in stochastic programs. Operations Research Letters 24, 4756.

12
 Table 1 A brief review of literature on the NSP

Author Objective Property Approach

Maximizing objective functions Considering skill and
Multiple choice
1 Warner, D. M. (1976). with respect to nurse preference characteristics of nurses in
programming
and hospital regulation different schedules
Millar, H. H., & Kiragu, Minimizing nurse assignment A cyclic network which leads to
2 Mathematical model
M. (1998). costs efficient solution
Minimizing nurse assignment Determining aggregate labor
Venkataraman, R., & Mixed integer linear
3 costs in terms of agency and requirement for 6-month
Brusco, M. J. (1996). programming
overtime cost Planning horizon
Minimizing the weekly labor Improving flexible work pattern
4 Ozkarahan, I. (1989). Goal programming
costs of the total workers in scheduling problems
Minimizing salary costs and Considering nurse preference
Jaumard, B., Semet, F., &
5 maximizing both employee and staff balance and satisfying Column generation
Vovor, T. (1998).
preferences and team balance the demands
Considering nurse preference,
Klinz, B., Pferschy, U., & Minimizing the total number of Integer linear
6 hospital regulations and
Schauer, J. (2007). work shifts programming (ILP)
satisfying the demand
Proposing a schedule for
Minimizing the sum of costs of a set of rosters for regular and
Bard, J. F., & Purnomo, Integer
7 each alternative available to pool nurses with regard to the
H. W. (2005). programming (IP)
satisfy shortages. expected demand of planning
horizon
Proposing an effective re-
Dynamic weighted
Hattori, H., Ito, T., Minimizing scheduling
max constraint
8 Ozono, T., & Shintani, T. the total weight of the to achieve solution stability in
satisfaction problem
(2005). unsatisfied constraints the re-scheduling process with
DW-Max CSP
dynamic constraints
Proposing an approach to Sawing and noising
Parr, D., & Thompson, J. Minimizing nurse assignment
9 ensure the sufficiency of with simulated
M. (2007). costs
nurses on each shift annealing
Punnakitikashem, P., Stochastic
Proposing a stochastic nurse
Rosenberger, J., & Minimizing workload penalty programming-
10 assignment with a recourse
Buckley Behan, D. on nurses benders
penalty function
(2008). decomposition
Fan, N., Mujahid, S.,
Zhang, J., Georgiev, P., Maximize the nurse Considering nurse preferences
Binary integer linear
11 Papajorgji, P., satisfactions and hospital and hospital regulations to
programming
Steponavice, I., Pardalos, regulations satisfy demands
P. (2013).
Considering hospital Bayesian
Li, J., & Aickelin, U. Minimizing the total cost of
12 regulations with satisfying the optimization and
(2004). nurses preference
demands classifier systems
Considering uncertainties in the
Fuzzy mathematical
Topaloglu, S., & Selim, Minimizing nurse assignment hospital target-values
13 programming, fuzzy
H. (2007). costs management, nurse preferences
goal programming
and constraints
Considering uncertainty in the
Minimizing deviations of nurse Fuzzy mathematical
Topaloglu, S., & Selim, hospital target-values
14 preferences and hospital programming, fuzzy
H. (2010). management, nurse preferences
regulations goal programming
and constraints
Proposing a procedure for
Electromagnetic
Maenhout, B., & Minimizing the total costs of maximizing the overall quality
15 meta-heuristic, local
Vanhoucke, M. (2007). pattern penalty of the schedule while taking
search
various constraints into account
Satisfaction of nurse Generating a qualified schedule
Landa- silva, D., & Le, Evolutionary
16 preferences and work with respect to the nurse
K. (2008). algorithms
regulations preferences
Tsai, C.-C., & Li, S. H. Minimizing the violation of Proposing a two- stage Genetic algorithms,
17
A. (2009). government regulations, programming with respect to interactive

13
 Author Objective
hospital management and the
scheduling fairness.
Ohki, M., Uneme, S. y., Minimizing penalty function for
18
& Kawano, H. (2010). evaluating shift schedules
Maximizing the quality of
Zhang, Z., Hao, Z., &
19 objectives with respect to the
Huang, H. (2011).
importance of constraints
Property Approach
nurse preferences and hospital information system
regulations
Creating an effective mutation
operator and an effective
parallel processing algorithm Cooperative genetic
for cooperative genetic algorithm (CGA)
algorithm (CGA) to solve the
NSP
Proposing a combination of Genetic algorithm,
different heuristic and variable
meta-heuristic methods to solve neighborhood
NSP search

Table 2 Notation summery

Indices and Sets

Index of available nurses
Index of shift
Index of patient
Index of date in planning horizon
Index of scenario
Index of allowed dates for assigning nurse 1 and 2

Parameters
Normal cost of nurse assignment in shift j
Overtime cost of nurse assignment in shift j
Total number of beds in each shift
Empty capacity of shift j in date
Number of new patient arriving in shift j in date
Maximum number of allowed patients in shift j in date
Number of shifts that patient k in date (entered in shift j) remains in the system
Coverage factor of a nurse ( is considered )
A binary variable that indicates whether patient k (presented shift w) is still at shift j or
not.
Total number of patients in shift j in date
Variables
1 if nurse is assigned to shift j in date
0 otherwise
Additional nurses in shift j in date
Demand for nurses in shift j in date

Table 3 Experimental results

Sample size N 95% CI of LB 95% CI of UB Optimally gap

1 (19800.78279 , 19809.42921) (20115.17679 , 20123.82321) 29.3111
2 (19803.41741 , 19806.21404) (20010.05541 , 20018.41259) 27.9879

14
 3 (19806.21404 , 19811.71366) (20020.27404 , 20028.34196) 25.4349
4 (19811.71366 , 19831.85828) (20029.06666 , 20036.84534) 24.9471
5 (19831.85828 , 19852.00291) (20011.48128 , 20018.97072) 22.4337
6 (19852.00291 , 19882.14753) (20010.63391 , 20017.83409) 22.1149
7 (19882.14753 , 19885.32416) (20006.66853 , 20013.57947) 22.0253
8 (19885.32416 , 19886.62078) (20017.09316 , 20023.71484) 21.8301
9 (19886.62078 , 19885.23541) (20011.06778 , 20017.40022) 21.8120
10 (19885.23541 , 19906.34397) (20016.00841 , 20022.05159) 21.3209
20 (19906.34397 , 19907.13565) (20013.89303 , 20019.64697) 19.8278
30 (19907.13565 , 19919.08928) (20023.02565 , 20028.49035) 18.8230
40 (19919.08928 , 19939.1599) (20019.92428 , 20025.09972) 18.2574
50 (19939.1599 , 19947.30453) (20006.6709 , 20011.5571) 17.4816
100 (19947.30453 , 19803.41741) (20000.12453 , 20004.72147) 11.5266

15
 Table 4 The assignment of nurses in optimal sequence in plannig horizon

13 13 92 92 92 92 92 92 91 99 93 93 32 81 81 81 81 81 81 81 88 81 9 1 1 1 2 1 1 1 8
M M M M M M M M M M M M M M M M M M M M M M M M M M 3
M M M M M M M M M M M M M M M M M M M M M M M M M M 9
- M/N M/A M M/A M/A - N - M/N M/A M/A M/A M/A - M/N M/A M A M M/A M - N A - M/N M/A M/A - M/N 1
- N A A - N M - N A A M - M/N A M A - M/N M - - N - M/N M 2
- N - M/N - M/N M/A - N M/A M/A - M/N - N M/A M/A - M/N M/A M/A A - N M/A M/A A M/A M/A M/A 2
M/N M/A A M A - M/N M - N A A A M -- N M - N - N M/A A A A - N A 2
M/A A - N M/A - N M/A A - N M M/A - N M/A M/A M/A A - N A - M/N - N M A 2
- N M A - N M/A A M/A A M/A A M/A - M/N A - N - M/N M/A A - N A A A A 2
A M/A - N A M/A A A A A - N A - N - N A - N A A - N A 2
N A A M/A M - M/N A - N - N A - N - N M M A A - N M A M 33
N - N - N A A A - M/N M/A M A M M A - N A A - N A 33
A A - N A - N M - N - N A - N - N - N M/A A - N A - N 39
N A N - N A - M/N A - N A A - N A - N - N A A - N A 31
A A A - N - N - N - N A - N - N - N - N - N - N A - N A 32
M/N - N A - N A A - N A A A A A M M - M/N M - N A 32
M M/A M/A A A A - N - N - N - N A A - N A A - N A A A - N 32
M - N - M/N - N A - N - N A - N A A M A - N - N A - N 32
A A - N A - N - N M A - N A - N - N M/A - N A - N - N 32

16
 4
x 10
2.015
EVP
LB
2.01
UB

2.005
Objective Value

1.995

1.99

1.985

1.98
0 10 20 30 40 50 60 70 80 90 100
SAMPLES

Figure 1 Optimal solution changes relative to the sample size N

x 10
4 Gap calculation
2.05
EVP
SNSP
2.04

2.03
Total Cost

2.02

2.01

1.99
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Iteration

Figure 2 Simulation results: total costs

17
 Figure 3 Simulation procedure

18
Acknowledgement
We would like to express our gratitude to the Research and Education Department of Razavi
Hospital for their support throughout this study. Also, the authors acknowledge the Surgery Group of
Razavi Hospital for their sincere assistance during this research.

19
Highlights
Modeling nurse scheduling problem which considers uncertainties.
Sample Average Approximation (SAA) method is used to obtain an optimal schedule.
A case study conducted in Department of Heart Surgery in Razavi Hospital.
The main objective of the model is to minimize the total costs.
The problem SNSP was formulated using the recourse model.

20