Industrial Engineering and Applications 815

L.-C. Tang (Ed.)

© 2023 The authors and IOS Press.
This article is published online with Open Access by IOS Press and distributed under the terms
of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).
doi:10.3233/ATDE230110

Nurse Rostering via Mixed-Integer

Programming
Eman OUDA a,1 , Andrei SLEPTCHENKO a,b and Mecit Can Emre SIMSEKLER a,b
a Department of Industrial and Systems Engineering, Khalifa University,

Abu Dhabi, United Arab Emirates

b Research Center on Digital Supply Chain and Operations Management,

Khalifa University, Abu Dhabi, UAE

ORCiD ID: Eman Ouda https://orcid.org/0000-0002-8631-5577, Andrei Sleptchenko
https://orcid.org/0000-0001-7188-5692, Mecit Can Emre Simsekler
https://orcid.org/0000-0002-1555-5012

Abstract. Nurse scheduling is a topic widely studied due to its overall effect on pa-
tient care and hospital performance. This research focuses on formulating a Mixed
Integer Programming (MIP) workforce scheduling model as a nurse rostering prob-
lem. The model incorporates multiple objectives of individual nurse preference and
qualification. This is approached by categorizing nurses into different hierarchical
levels based on their qualifications and positions; Head Nurse, Senior Nurse, Nurse,
and Assistant Nurse. Moreover, each nurse’s holiday preferences are accounted for
in the model. The proposed MIP model is solved to generate a schedule to meet
hospital demand and individual nurse preferences. A MIP solver using Python 3
was used to find the optimal solution by cutting planes. The sensitivity analysis and
computational results reflect different scenarios and scheduling to fit all hospital
environments and demands.

Keywords. MIP, Scheduling, Nurse Roster, Healthcare

1. Introduction

The challenges of providing high-quality healthcare arise due to the increasing health-
care costs of material and staff salaries. Therefore, advanced decision-making techniques
are studied to minimize expenses. One of the crucial factors that needs special atten-
tion is the scheduling of nurses. Nurses in hospitals are essential medical professionals
that significantly affect patient care and overall hospital performance. Therefore, opti-
mizing nurse-related issues is widely studied in the literature. They consist of schedul-
ing individuals over a given time horizon to fit hospital nurse demand. Throughout the
week, nurses can experience physical and psychological fatigue. Several issues must
be addressed when setting up a weekly schedule, such as individual nurse preference,
minimum working hours, and maximum working hours. In the next section, a literature
review on nurse scheduling is done to study these different factors and form a Mixed
Integer Programming (MIP) model.
1 Corresponding Author: Eman Ouda, [email protected]
816 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming

The main objectives of this research are as follows. First, we focus on scheduling
nurses and assigning nurses according to their qualifications and roles. To our knowl-
edge, previous research generally focuses on a homogeneous group of nurses rather than
creating a detailed schedule for each nurse based on their qualifications, planned vaca-
tions, and working hours preferences. The proposed MIP model explores the trade-off
between nurse demand, preferences, qualifications, and working hours. Using the MIP
proposed in Section 4 on the generated data, nurse efficiency can be improved and bet-
ter service levels obtained. Based on the constraints, the model should positively impact
nurses’ schedules to meet their requests and the hospital demand. We also examine the
performance of the model for different parameters settings. This research concludes that
minimizing overtime and meeting nurse preferences positively impact hospital perfor-
mance.
The remainder of the paper is organized as follows. Section 2 gives an overview of
how nursing problems have been approached in the literature and defines the contribu-
tion. Then Section 3 introduces the problem statement, the scope of the paper, and the
model assumptions. Section 4 describes the MIP formulation. Section 5 presents the data
used in this investigation and the solution approach. The computational results and out-
comes are also discussed. Finally, Section 6 summarizes the findings, limitations, and
future work.

2. Literature Review

The review examines MIP models that address nurse scheduling problems, which are
typically formulated with an objective function aimed at either maximizing or minimiz-
ing certain factors. Maximization is typically used to prioritize nurse performance pref-
erences, while minimization typically targets various system costs.
Effective management of the healthcare system is highly dependent on proper
scheduling of nurses. Once the scheduling model is formulated, various methods are
available to obtain the optimal schedule. Guessoum et al. [1] propose a two-phase model
with predefined constraints for nurse rostering. In the first phase, a generic variable fixing
heuristic is utilized to form a second sub-problem, which, in its turn, can be solved by a
general-purpose MIP solver. This leads to a swift and efficient optimization approach for
nurse rostering. However, it is crucial to consider the individual preferences of nurses as
a significant factor [2]. This can only be accomplished by planning a sufficient number
of nurses fairly for each duty. Additionally, the qualifications of the employees must be
considered when creating MIP models and generating schedules. A separate study ex-
plores the integration of various nurse qualifications and competencies in a MIP-based
nurse scheduling problem [3]. The model was validated and solved using actual data
from a hospital, where it was successful in meeting the hospital’s demand.
Overall, effectively managing nurses and their schedules has a significant impact
on nurse productivity. Another MIP model was developed and solved to enhance nurse
productivity, using extensive datasets from a Chinese hospital [4]. The study demon-
strated that there is a trade-off between performance criteria. Moreover, different con-
straints must be defined as either hard or soft for a well-defined nurse roster. The models
presented in this study are discussed in research that addresses the International Nurse
Rostering Competition [5], where some of the essential scheduling constraints for nurses
 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming 817

and healthcare providers are utilized in our research. Additionally, another recent work
[6] has focused on modifying typical constraints found in the literature to promote the
concept of balanced work routines for all nurses. The authors present a sophisticated and
intricate schedule arrangement that aligns with the actual requirements of hospital units.
On the application side, Belin et al. [7] integrate nurse scheduling and operating
room scheduling using an MIP model. The proposed branch-and-price approach uses col-
umn generation to find the exact solution. Therefore, they were able to achieve promis-
ing cost savings results with a large number of computational experiments. Similarly,
Maenhout2010 et al. [8] used the branch-and-price approach for a different MIP model.
Another interesting set of studies integrates nurse scheduling with patient availability.
Schoenfelder et al. [9] consider scheduling changes and unit assignments of cross-trained
float nurses, where patient availability is an uncertain parameter. Adyas et al. [10](2021),
in their turn, use historical data to estimate non-stationary patient arrivals. The resulting
dynamic nurse staffing policy had 3% cost savings compared to fixed staffing policies.
Nurse scheduling problems have also been upgraded to be solved by metaheuris-
tic optimization algorithms. A genetic algorithm method was used to obtain the optimal
nurse schedule [11]. Two-point crossover and random mutation were used. Eventually,
the schedule was compared to an existing hospital schedule, resulting in savings of 12%
and 13% in staffing expenses per month and the number of nurses, respectively. A recent
review, [12], provides a comprehensive exploration of potential research prospects on
nurse scheduling. Mathematical models are solved using math-heuristics and other ad-
vanced methods. The authors also emphasize the importance of operational research for
healthcare communities. Therefore, this also highlights the importance of nurse schedul-
ing problems and their evolution for future research to aid in managing nurses efficiently.
The presented research focuses on formulating an MIP model for a nurse scheduling
problem. The initial objective is to minimize the costs associated with nurse working
hours. This objective includes quantitative measures and results in lower costs. In ad-
dition, qualitative measures are included in this research to increase healthcare quality.
According to Maenhout et al. [8], during nursing personnel scheduling, the head nurses
often face problems when there are multiple goals and priorities. Therefore, this research
focuses on scheduling nurses considering their qualifications and their vacation prefer-
ences. The problem formulated is solved to make it easy and efficient to find the optimal
schedule.

3. Problem Definition

In this article, we consider a hospital with a set of nurses. Each nurse has a different
position depending on her or his qualifications. According to Bester et al. [13] research,
there are six different nurse positions with decreasing qualifications. These are: Head
Nurse, Senior Nurse, Nurse, Staff Nurse, Senior Nurse Assistant, and Nurse Assistant.
In the presented research, these categories are abbreviated into four: Head Nurse, Se-
nior Nurse, Nurse, and Nurse Assistant. The hospital has a specific demand for different
nurses on every shift. However, to maintain the best care provided by the hospital, the
preference for the working hours of the nurses should also be considered. Therefore, a
schedule must be generated that considers the hospital’s demand, nurses’ qualifications,
and nurses’ preferences.
818 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming

This is done by formulating and solving an MIP model. The main objective of the
model is to minimize the total allocations of nurse slots. Several assumptions are consid-
ered to form the objective function and the set of constraints illustrated in Section 4. The
following is a summary of some of the assumptions.

1. Nurses are classified into different positions according to their qualifications.

2. There are four types of nurse positions in a hospital (Head Nurse, Senior Nurse,
Nurse, and Nurse Assistant).
3. There is a demand for nurses who have different qualifications.
4. Nurses enter their working hour availability and holiday preference.
5. A nurse cannot exceed the maximum daily working hours.
6. A nurse must meet the minimum daily working hours requirement.
7. Nurses have maximum overtime hours they can not exceed.
8. The created schedule is based on the results of the MIP model.

4. Mathematical Formulation

The defined problem is formulated as an MIP mathematical model. The model consists
of indices, parameters, decision variables, an objective function, and constraints. The
optimal solution that meets hospital demand is obtained, and a nurse schedule is devel-
oped depending on the model’s time span. The model is then solved to form an optimum
schedule for nurses to meet the hospital’s demand.
First, the model indices are the set of nurses, the shift times, and the days. Moreover,
a set of pre-defined parameters is listed below.

Indices:

n – nurse index (n = 1, . . . , N),

q – qualification index (q = 1, . . . , 4),
t – time index (t = 0, . . . , T ),
d – day index (d = 0, . . . , D).

Parameters:

Det,d – demand for nurses at time t on day d,

αn – minimum working hours for nurse n ∈ N per week,
βn – maximum working hours for nurse n ∈ N per week,
δ – minimum working hours for a nurse per day,
π – maximum working hours for nurse per day,
τ – maximum overtime hours,
σ – maximum shortage hours,
γn,d – preferred number of hours for nurse n ∈ N to work on day d ∈ D,
λq – demand for nurses with qualification q ∈ Q,
Pq,n – binary indicator that nurse n ∈ N has the qualification q ∈ Q.
 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming 819

For this model, we expect an outcome telling us the shifts each nurse is assigned
to. Hence, the two main binary decision variables are defined. The first, Xn,t,d , indicates
whether nurse n works at time t on day d. The second, Yn,t,d , indicates that defines the
time each nurse starts. The decision variables, model objective function, and constraints
are illustrated below.

Decision variables:

1 Nurse n works at time t on day d
Xn,t,d =
0 otherwise

1 Nurse n begins work at time t on day d
Yn,t,d =
0 otherwise

Mixed Integer Programming Model formulation:

min ∑ ∑ ∑ Xn,t,d (1)

n∈N t∈T d∈D

s.t. ∑ Xn,t,d ≥ Det,d ∀t ∈ T, d ∈ D (2)

n∈N

∑ Yn,t,d ≤ 1 ∀n ∈ N, d ∈ D (3)
t∈T

Yn,t,d ≥ Xn,t,d − Xn,t−1,d ∀n ∈ N,t ∈ T, d ∈ D (4)

∑ Pq,n Xn,t,d ≥ λq ∀q ∈ Q,t ∈ T, d ∈ D (5)

n∈N

∑ Xn,t,d ≤ 2π ∀n ∈ N, d ∈ D (6)
t∈T

∑ Xn,t,d ≥ 2δ ∑ Yn,t,d ∀n ∈ N, d ∈ D (7)

t∈T t∈T

∑ ∑ Xn,t,d ≤ 2(βn + τ) ∀n ∈ N (8)

t∈T d∈D

∑ ∑ Xn,t,d ≥ 2(αn − σ ) ∀n ∈ N (9)

t∈T d∈D

∑ Xn,t,d ≤ 2γn,d ∀n ∈ N, d ∈ D (10)

t∈T

The objective function (1) is defined to minimize the total number of slots to which
nurses are assigned. A set of MIP constraints composed to enforce the demand and fair-
ness requirements. Constraints (2) guarantee that general hospital demand is met and
that there are enough nurses available. Constraints (3) ensure that nurses’ shifts start at
most once a day. Constraints (4) connect the beginning of nurses’ shifts to the conse-
quent time slots. Constraints (5) require that there be enough nurses for each qualifica-
tion q. Constraints (6) ensure that nurses work a maximum of π hours per day. Since the
slots in the model are 30 minutes each, the right-hand side of some of the constraints is
820 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming

doubled. Constraints (7) take care of the minimum working hours per day, assuming that
nurse n starts working that day. Constraints (8) and (9) consider the entire week and the
minimum and maximum overtime and shortage hours that each nurse is allowed to take.
Finally, constraints (10) take into account the preferences of each nurse in working hours
for each day.

5. Discussion of results

5.1. Input Data

The data used in this research are based on available in literature and on logical assump-
tions. Initially, it is assumed that there are 18 nurses available in a specific department.
Two are qualified as nurse assistants, and ten are nurses. Only 4 and 2 are qualified to
be senior and head nurses, respectively. In addition, Excel generates random numbers
that reflect the preferences of each nurse for the working days and the number of hours.
However, hospital demand may differ from one department to another [13]. The four cat-
egories of nurses have a demand that fluctuates between 1-4 nurses per shift. Hence, us-
ing this data, a demanding schedule may be generated. Initially, Six days of the week are
managed; however, this value may change based on the scheduling requirements. More-
over, each nurse’s maximum and minimum working hours per week are set up like a typ-
ical 48 hours per week working schedule, with a maximum of 60 hours and a minimum
of 48.

5.2. Solving Approach and Results

As discussed earlier, MIP models can be solved in multiple ways, such as branch-and-
price, column generation, different (meta)heuristics, etc. In this study, Gurobi Optimizer
9.1.2 was used to solve the proposed MIP model. Although it is an automatic solver that
uses libraries to solve the model, it is fast and efficient. The solver uses cutting planes
within minutes to find an optimal solution for the generated data. The first set of data
resulted in a minimum solution of 1728. This optimal solution implies the costs incurred
by nurses. The slots covered are the nurse timings, and their sum is the slots covered by
the nurses all week. After the optimal solution is found, the decision variables, X and Y ,
are used to generate a plan for all nurses. Figure 2 illustrates a schedule generated after
solving the MIP model. The schedule includes all four categories and the timing of each
nurse and shift.
Different scenarios are applied to change the roster schedule for further analysis and
the optimal solution and schedule are compared. The first case study examines how the
number of nurses hired affects the optimal solution and the increase in costs. There is a
continuous increase in total cost as the number of nurses is hired. This is illustrated in
the plot in Figure 1.
The following case study studies the relationship between demand per shift and total
cost. Two types of shift demands are considered. During the day, the nurse demand is
assumed to be higher. Therefore, 7 AM to 7 PM has a different nurse demand than 7 PM
to 7 AM. The initial demand per shift is 6 for the first 12 hours of the day and 5 for the
next. However, this has changed and increased by around 15%. When the demand per
 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming 821

1,900

Total Cost
1,800

1,700

1,600
15 16 17 18 19 20
Nurse Availability
Figure 1. Total cost versus the total number of nurses available in the department

shift increases, the total cost does not increase drastically. The difference in the optimal
solution is presented in Table 1. Therefore, adding more demand per shift is more man-
agerial than hiring new nurses in this case. A similar analysis can be applied to other case
studies, hospital demands, and departments to make the right decision.

Table 1. Total demand per shift.

Total Demand Total Cost
7AM to 7 PM: 5 nurses 1728
7PM to 7 AM: 4 nurses
7AM to 7 PM: 6 nurses 1729
7PM to 7 AM: 5 nurses

However, not all nurses prefer to work the same number of hours. Some nurses do
not mind working 12 hours, while others may lose function after 10 hours. This has
been applied after generating random preferences in Excel. The number of hours per
shift is based on employee preference. Table 2 summarizes the relationship between the
minimum solution and the difference in nurse preferences. It can be concluded that there
is no difference in the optimal solution when nurses have different choices in the number
of working hours per shift.

Table 2. Variability in the number of hours per shift per nurse.

Maximum hours/shift Total Cost
12 1729
10/12 1728

Finally, the demand per nurse qualification is defined in the problem. Initially, there
was a demand of 2,2,1,1 for Assistant Nurse, Nurse, Senior Nurse, and Head Nurse,
respectively. However, the demand for the primary nurse type, Nurse, is analyzed. Figure
3 shows the change in the objective function plot. The optimal solution increases slightly
as the demand increases. However, the total cost increases significantly once it reaches a
specific number, 6. This is due to the limited availability of nurses hired.
822 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming

Mon Tue Wed Thu Fri Sat Sun

Assistant Nurse 1
Assistant Nurse 2
Nurse 3
Nurse 4
Nurse 5
Nurse 6
Nurse 7
Nurse 8
Nurse 9
Nurse 10
Nurse 11
Nurse 12
Senior Nurse 13
Senior Nurse 14
Senior Nurse 15
Senior Nurse 16
Head Nurse 17
Head Nurse 18

Figure 2. One week employee plan (Nurse Roster)

1,900
Total Cost

1,850

1,800

1,750

2 3 4 5 6
Number of Nurses Qualified

Figure 3. Total cost versus the demand for qualified nurses

6. Conclusion

In conclusion, this study develops a framework to reduce the number of nursing slots
scheduled for all nurses. Data were compiled on the basis of assumptions from the lit-
erature. Additionally, a schedule was made to meet the hospital’s demand for different
nurse qualifications and preferences after formulating and solving the MIP quickly and
efficiently.
This study faces two main limitations. First, the data used are not real-life hospi-
tal data due to the limitations of accessing hospital information and confidentiality. An-
other limitation was the costs of having nurses work and their wages, which were not
considered when building the MIP model.
In terms of quality of healthcare care, future work includes integrating nurse
scheduling with other facilities such as operation room scheduling. Moreover, other col-
 E. Ouda et al. / Nurse Rostering via Mixed-Integer Programming 823

umn generation methods may be used to have optimal solutions when solving the MIP
model. Finally, the study can be extended to use efficient heuristics or metaheuristics.

References

[1] Guessoum F, Haddadi S, Gattal E. Simple, Yet Fast and Effective Two-Phase Method for Nurse Roster-
ing. American Journal of Mathematical and Management Sciences. 2020;39.
[2] Dowsland KA. Nurse scheduling with tabu search and strategic oscillation. European Journal of Oper-
ational Research. 1998;106.
[3] Widyastiti M, Aman A, Bakhtiar T. Nurses Scheduling by Considering the Qualification using Inte-
ger Linear Programming. TELKOMNIKA (Telecommunication Computing Electronics and Control).
2016;14.
[4] Guo M, Wu S, Li B, Rong Y. Maximizing the efficiency of use of nurses under uncertain surgery
durations: A case study. Computers and Industrial Engineering. 2014;78.
[5] Santos HG, Toffolo TAM, Gomes RAM, Ribas S. Integer programming techniques for the nurse roster-
ing problem. Annals of Operations Research. 2016;239.
[6] Böðvarsdóttir EB, Bagger NCF, Høffner LE, Stidsen TJR. A flexible mixed integer programming-based
system for real-world nurse rostering. Journal of scheduling. 2022;25(1):59-88.
[7] Beliën J, Demeulemeester E. A branch-and-price approach for integrating nurse and surgery scheduling.
European Journal of Operational Research. 2008;189.
[8] Maenhout B, Vanhoucke M. Branching strategies in a branch-and-price approach for a multiple objective
nurse scheduling problem. Journal of Scheduling. 2010;13.
[9] Schoenfelder J, Bretthauer KM, Wright PD, Coe E. Nurse scheduling with quick-response methods: Im-
proving hospital performance, nurse workload, and patient experience. European Journal of Operational
Research. 2020;283.
[10] Aydas OT, Ross AD, Scanlon MC, Aydas B. New results on integrated nurse staffing and scheduling:
The medium-term context for intensive care units. Journal of the Operational Research Society. 2021;72.
[11] Leksakul K, Phetsawat S. Nurse scheduling using genetic algorithm. Mathematical Problems in Engi-
neering. 2014;2014.
[12] Ngoo CM, Goh SL, Sabar N, Sze S, Abdullah S, Kendall G. A Survey of the Nurse Rostering Solution
Methodologies: The State-of-the-Art and Emerging Trends. IEEE Access. 2022 01;10:1-1.
[13] Bester MJ, Nieuwoudt I, Vuuren JHV. Finding good nurse duty schedules: A case study. Journal of
Scheduling. 2007;10.