Computers & Operations Research 32 (2005) 491 – 507

www.elsevier.com/locate/dsw

A 0-1 goal programming model for nurse scheduling

M.N. Azaieza;∗ , S.S. Al Sharif b
a
Industrial Engineering Department, College of Engineering, King Saud University, P.O. Box 800,
Riyadh 11421, Saudi Arabia
b
Planning and Development Department, Riyadh Armed Forces Hospital, P.O. Box 10880 Riyadh 11443,
Saudi Arabia

Abstract

In this study, a computerized nurse-scheduling model is developed. The model is approached through a 0-1
linear goal program. It is adapted to Riyadh Al-Kharj hospital Program (in Saudi Arabia) to improve the current
manual-made schedules. The developed model accounts both for hospital objectives and nurses’ preferences,
in addition to considering some recommended policies that are displayed in the literature. Hospital objectives
include ensuring a continuous service with appropriate nursing skills and sta6ng size, while avoiding additional
costs for unnecessary overtime. Nurses preferences, which are deduced from a survey conducted on-purpose
for the sake of this study, include mainly fairness considerations, in terms of ratio of night shifts and weekends
o7, in addition to avoiding isolated days on and o7. The model is implemented in an experimental phase of
six-month period using LINGO and is considered to perform reasonably well, based both on some quality
criteria displayed in the literature and on the feedback obtained from a second survey, that has been developed
to assess the scheduling system performance.
? 2003 Published by Elsevier Ltd.

1. Introduction

The objectives in nurse scheduling are multiple. These include developing a systematic procedure
for allocating nurses to work shifts and workdays in a way to ensure a continuous and appropri-
ate service of patient care, and satisfying organizational scheduling policies, such as speci=c work
requirements while using minimum sta6ng to avoid wasted manpower.
The problem is further complicated by such factors as: variation in patient demand, nurse qual-
i=cation and specialization, acuity of patient illnesses, organizational characteristics (e.g., minimum

∗
Corresponding author. Tel.: +966-1-467-6826; fax: +966-1-467-8657.
E-mail address: [email protected] (M.N. Azaiez).

0305-0548/$ - see front matter ? 2003 Published by Elsevier Ltd.

doi:10.1016/S0305-0548(03)00249-1
492 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

required coverage and days o7 policies), unpredictable absenteeism, and personal requests for
vacations, work stretch, and work pattern. Moreover, some of these considerations may conEict
with others, such as employee requests versus the need to balance workload.
Nurse scheduling is a di6cult and time-consuming task. The schedule should determine the
day-to-day shift assignments of each nurse for a speci=ed horizon of time in a way that satis-
=es the given requirements. The schedule should also be fair enough to everyone and not disruptive
to nurses’ health, families, or social lives.
Nursing talents exist at a variety of levels. Some individuals are trained to handle special needs,
such as intensive care and rehabilitation therapy. Depending on their training, individuals can function
at di7erent positions such as registered nurses (RNs), licensed practical nurses (LPNs) or aids (AIDs)
[1]. In Riyadh-Al-Kharj Hospital (RKH), the categories of nurses in a decreasing hierarchy are nurses
in charge, sta7 nurse 1, sta7 nurse 2, and nurse aid.
In RKH, nurse scheduling is performed manually. It takes approximately one workday for a head
nurse to build the schedule each month. Fairness is not addressed while making the schedule. In
fact, some sample schedules show very important discrepancies in the fraction of day shifts worked
in some 4-week schedules. Also, the number of weekends o7 or consecutive days o7 a nurse obtains
per year is very unbalanced. Moreover, nurses’ preferences are never considered. This often causes
nurses’ frustration leading to either working under high stresses or quitting their jobs. In either case,
the quality of the nursing service may highly be a7ected. In addition, a manual schedule is not
expected to minimize overtime and utilize the nursing sta7 e6ciently.
The current study attempts to develop a computerized nurse scheduling system for RKH that
utilizes e7ectively the nursing personnel. The system will also rely on fairness bases among nurses
and will consider nurses’ preferences to maximize their satisfaction. This will help them provide a
proper quality of service.
A =rst survey is conducted in order to gain understanding on nurses’ preferences. Fairness bases
are considered both from the survey results and the suggested policies in the literature. Given that
satisfying all preferences while making an e7ective utilization of nurses seems infeasible, a number of
priority levels are considered in developing the scheduling system. Required policies are formulated
as model constraints. The remaining policies are modeled as soft constraints with di7erent importance
weights. After implementing the model in a 6-month trial period, a second survey was conducted to
assess the model adequacy.
RKH consists of two hospitals and several clinics. The main one is in Riyadh City and the other
one is in Al-Kharj City. The clinics are scattered in di7erent places in Riyadh and belong to the
main hospital. Thus, RKH has a very large nursing sta7 size that amounts to 1798 nurses in Riyadh
only. RKH started a nurse scheduling system with three shifts per day of 8 h each. At the end of
the 1980s, RKH has opted for the two-shift day system of 12 h per shift. Scheduling is performed
manually through trial and error. Each schedule is made for a 4-week period. No written policies
exist for this system. However, the main hospital obligates all nurses to work their contracted time
(of 176 h per a 4-week period) and any additional hours-worked are considered as overtime. Policies
and guidelines in the scheduling patterns are left to the nursing administration and head nurses to
arrange them. Head nurses account on their experience, knowledge and department agreements for
producing schedules. However, they have no o6cial ergonomics standards to follow. RKH has more
than 55 head nurses that make schedules for their units. Each unit consists of a number of specialties.
Each specialty has two types of shifts.
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 493

1. One 8-h shift for clinics for 5 days (Saturday–Wednesday) from 7:30 AM to 4:00 PM and half
a day (Thursday) from 7:30 AM to 12:00 PM, with half an hour break.
2. Two 12-hour shifts for hospital wards that consist of a day shift from 7:00 AM to 7:00 PM and
a night shift from 7:00 PM to 7:00 AM.

For clinics, the regular load per nurse is 44 h per week. For the hospital wards, it is of 176 h per
4-week period. In case of nurses’ shortage in any ward, the corresponding head nurse may borrow
nurses from inside the unit or from the same specialty in the clinics. In the latter case, the borrowed
nurse will work her regular time (till 4:00 PM).
The organization of the paper will be as follows. Section 2 will present some literature review.
Section 3 will discuss the =rst survey’s description and results. Section 4 will present the linear
goal programming (GLP) model along with results and discussions. Section 5 will discuss the
implementation of the model. Finally, Section 6 will present some concluding remarks and directions
for future research.

2. Literature review

Investigations in health care systems may be classi=ed into four interrelated nursing human-
resources decisions [2].

• Sta6ng decisions, which specify the number of full time equivalent nursing personnel of each
class of skills to be =lled for each nursing unit.
• Scheduling decisions, which specify when each nurse will be on and o7 duty in the scheduling
period and minimum number of nurses of each class of skills required for each shift on each
day.
• Allocation decisions, which allocate a pool of available Eoating nurses to accommodate the Euc-
tuating demand for nursing care and for absenteeism.
• Assignment decisions, which assign nurses to individual shifts.

Modeling nurse scheduling is not a new idea. Until the 1960s, scheduling tools consisted only of
graphical devices such as Gantt Chart. Howell [3] outlined the steps necessary to develop a cyclical
schedule. Howell’s method is a step-by-step procedure for accommodating the work patterns and in-
dividual preferences of nurses. In the early 1970s, scheduling systems began to be based on heuristic
models [21,22]. These models were deemed more appropriate because they could theoretically take
into account of all scheduling constraints in solving the problem. Maier-Rothe and Wolfe [4] de-
veloped a cyclical scheduling procedure that assigns di7erent types of nurses to each unit based on
average patient-care requirements, hospital personnel policies, and nursing sta7 preferences.
Several nurse-scheduling models were based on linear programming [1,5], penalty-point algo-
rithms and mixed-integer programming [23]. Other optimization techniques have been used in nurse
scheduling particularly for the non-cyclical type. These include the assignment problem [6], integer
programming [24], stochastic programming [7], non-linear programming [8], and goal programming
[1,9,10], etc.
494 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

Recently, goal programming (GP) has received the most attention among optimization techniques,
as it attempts to optimize a number of objectives simultaneously. These objectives include: maximiz-
ing utilization of full-time sta7, minimizing understa6ng and oversta6ng costs, minimizing payroll
costs, as well as minimizing deviations from desired sta6ng requirements, nurse preferences, and
nurse special requests.
Trivedi [11] has developed a MIGP model for expense budgeting in a hospital nursing department.
Objectives are based on cost and quality nursing care considerations. Musa and Saxena [12] have
used a 0-1 goal programming formulation for nurse scheduling in one unit of a hospital. Goals with
di7erent priority levels represent hospital policies and nurses preferences. Berrada et al. [13], in their
0-1 GP model for nurse scheduling, have used for hard constraints administrative and union contract
speci=cations, while work patterns and nurses preferences have been formulated as soft constraints.
Moores et al. [14] have formulated the student nurse allocation problem using also a 0-1 GP. The
problem was to produce a 3-year schedule for student nurses to comply with the minimum practical
and theoretical standards while being used as part of the hospital work force.
Some authors use arti=cial intelligence techniques, such as knowledge-based systems [15] and
declarative programming using Prolog [25].
Chen and Yeung [15] have set =ve rules for evaluating a good schedule, namely physiological
adjustment, well being (sleep, fatigue and appetite), personal and social problems, health (gastroin-
testinal and nervous disorders), and performance and accidents. The authors have also recommended
a number of ergonomic constraints including limiting the succession of night shifts (maximum of
three night shifts is preferable), avoiding isolated working days, alternating weekends o7, considering
preferences on days o7, preferences on shifts, requests for emergency days o7, and assigning 40 h
per week for full time nurses.
Oldenkamp and Simons [16] has suggested =ve factors for assessing a schedule quality. These
factors are given below.

• Optimality factor: represents the degree to which nursing expertise is distributed over the di7erent
shifts.
• Completeness factor: represents the degree to which quantitative demands for occupation per shift
are met.
• Proportionality factor: represents the degree to which each nurse has been given about the same
amount of night shift, evening shifts and weekends o7.
• Healthiness factor: represents the degree to which it has been taken care of the welfare and health
of the nursing sta7.
• Continuity factor: represents the degree to which there is some continuity in nursing crew during
the di7erent shifts.

3. The rst survey

The nursing sta7 in RKH constitutes a mixture of nurses from di7erent countries. This includes
nurses from the West, Philippines, South Africa, and Arab countries, in addition to nurses from inside
the Kingdom of Saudi Arabia. It is therefore expected that the nurses’ preferences would widely
di7er. In order to gain understanding about their preferences and to incorporate these preferences
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 495

in the nurse-scheduling model, a survey was conducted during summer 2000, for the purpose of
the current study. This survey involved 400 questionnaires addressed to nurses working in hospital
wards (having 12-h shift patterns). The questionnaires were distributed to all nurse levels. Some
procedures have been used to motivate nurses in order to obtain a large response rate. The procedure
was successful and 354 out of 400 forms (i.e. 88.5 %) are =lled out and returned. The respondent
categories are as follows:

• 33 person in charge
• 165 sta7 nurse 1
• 95 sta7 nurse 2
• 58 nurse aide
• 3 unspeci=ed.

The form has 15 questions with multiple choice answers related mainly to work patterns, evaluation
of current schedules, preferences on day and night shifts, and preferences on days and weekends o7.
About half the respondents (46.6%) were sta7-nurses 1. The remaining were sta7-nurses 2 (26.55%),
nurse aids (16.81%) and persons in charge (9.1%). Married nurses were 56% while single nurses
were 44% of respondents.
The main results related to nurses preferences show that about one third of the nurses negotiate
their schedules. The 12-h shift is preferred to the 8-h shift to most of the nurses (77%). Also, 52%
of respondents prefer day shift against 19% prefer night shift and 29% of them are indi7erent. Also,
37% of the respondents prefer to decrease night shifts, while 18% prefer to have more night shifts.
Concerning preferences over days o7, 47% of the nurses prefer having their days o7 on weekends,
58% of them prefer full weekend o7, while 36% are indi7erent to when to have their days o7.
For a 4-week schedule, 78% of the nurses prefer to have at least 2 weekends o7, while 10% are
undecided. Also, 47% prefer working no more than 4 consecutive days against 20% and 16% that
prefer working no more than 5 and 6 consecutive days, respectively. About the same results were
obtained for night shifts. Isolated days o7, isolated days on, and day–night (of the following day)-o7
patterns are not preferred to most of the respondents.
The results suggest in particular having some balanced schedules, where the fraction of night shifts
should be about the same for all nurses. Also, isolated days on and o7 should be avoided. In addition,
fairness bases should be applied to all nurses to have them bene=t from more weekends o7 and if
possible of full weekends o7. Finally, no more than 4 consecutive working days should be assigned
to nurses. This last policy is consistent with the current practices of nurse scheduling at RKH.

4. The model

4.1. Nursing policies

First, nursing policies will be developed. These policies will be based mainly on current hos-
pital practices that the head nurses consider as implicit requirements, the results of the survey,
and published policies. The importance in incorporating some of the published policies relies on
accounting for ergonomic considerations. In fact, the human being has various physical limitations
496 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

and the lack of ergonomic consideration causes frustration and reduces productivity quantitatively
and qualitatively.
Because of the large number of constraints that the schedule attempts to satisfy, it is possible that
no feasible solutions to such a nurse-scheduling problem would exist. For this reason, the constraints
are divided into two classes: hard constraints that must be satis=ed and soft constraints that may be
violated. However, the model will minimize these violations by reducing the deviations in the soft
constraints from their respective targets. The schedule will extend over a 4-week period. The set of
all constraints combined is given as follows.

4.1.1. Present hospital constraints

• The unit is covered by two 12-h shifts (day and night shifts for 7 days a week, and 24 h a day).
• Each nurse has to work at least 176 h per schedule (4 weeks), which is equivalent to 14.67 days
per schedule.
• The regular working days are between 14 and 15 days per schedule. Any additional worked hours
above 176 h per schedule are considered as overtime.
• No nurse can work for more than 4 consecutive working days.
• A nurse should not work in any case two consecutive shifts.
• Minimum sta7 level requirements must be satis=ed.
• Night shifts must constitute at least 25% of total workload for each nurse.

4.1.2. Published policies (based mainly on ergonomic considerations)

• Avoid any isolated days on (o7–on–o7).

• Avoid any isolated days o7 (on–o7–on).
• Avoid working in a day shift and the night shift of the following day.
• Preference on days o7 should be considered.
• All nurses should have the same amount of night shifts and weekends o7.

4.1.3. Survey results (after feedback and adjustments)

• Day shifts should exceed night shifts for each nurse and schedule.
• At least 2 weekends o7 in the schedule is preferred.
• Full weekend is preferred.
• Isolated days o7 (on–o7–on) are to be avoided.
• Isolated days on (o7–on–o7) are to be avoided.
• (day shift–night shift–day o7) are to be avoided.

4.2. Development of the 0-1 LGP model

4.2.1. Model hard and soft constraints

The problem consists on scheduling a number of nurses for a 4-week period in a cyclical way
for a particular unit, while satisfying minimum sta6ng requirements, along with other constraints.
The nurse-scheduling model will also attempt to satisfy several goals including reducing oversta6ng
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 497

(and hence overtime cost) as well as incorporating nurses’ preferences and establishing fairness bases
among nurses. These suggest using a zero-one linear goal programming (LGP) approach.
The scheduling problem contains a total of 11 scheduling sets of constraints. It is not expected
however that a feasible solution may be obtained while satisfying all sets of constraints. Therefore,
these sets are divided into two groups: one group consists of sets of hard constraints that must
be satis=ed. The other group consists of the remaining sets of constraints that are considered as
soft constraints. The model will attempt to satisfy these soft constraints. If not possible, the model
will reduce to a least the violations of these soft constraints based on the importance of each set.
Classifying constraints into hard and soft constraints as well as assigning importance weights have
been made through consulting the head nurses that are in charge of nurse scheduling. The sets of
hard and soft constraints are given below.

4.2.1.1. Hard constraints

• The =rst set of constraints ensures that the daily minimum sta7 level is met. Because the daily
requirements may di7er from one day to another, the model will allow the user to insert the daily
minimum requirement for each day and night shift.
• The second set of constraints assigns for each nurse and each day a day shift, a night shift or a
day o7. Also, this set of constraints disallows assigning two consecutive shifts (from 7:00 AM to
7:00 PM and from 7:00 PM to 7:00 AM or from 7:00 PM to 7:00 AM and from 7:00 AM to
7:00 PM). In other words, no nurse can be assigned 24 h of continuous work.
• The third set of constraints ensures that no nurse is assigned more than four consecutive days on.
• The fourth set of constraints ensures that each nurse is assigned at least 4 days o7 during weekends
in each 4-week schedule.
• The =fth set of constraints ensures that each nurse is assigned 14 days on and at most 16 days on
per 4-week schedule. Note that the 176 h considered by the hospital, as regular load is equivalent
to 14.67 days. This set of constraints will reduce overtime and establish some fairness among
nurses regarding monthly loads. Note that the upper bound is not needed, as the =rst set of soft
constraints below call for 15 days per schedule for each nurse. It is however placed for convenience
to reduce the search space. Our computational experience show a tremendous increase in model
implementation time when deleting the upper bounds of 16 days.
• The sixth set of constraints ensures that each nurse is assigned a minimum number of night shifts,
Nmin (Nmin will be taken as 25% of total working shifts). This will limit day shifts to some Dmax
(in the application, Dmax will be taken as 75% of total working shifts). This set of constraints will
establish some balance in the fraction of night shifts for all nurses, leading also to more fairness
among nurses.

4.2.1.2. Soft constraints

• The =rst set of soft constraints attempts to assign to each nurse a total of 15 days as per schedule.
This will create more balance in the workload of the di7erent nurses.
• The second set of soft constraints attempts, again for fairness reasons, to assign more day shifts
than night shifts to each nurse in each 4-week schedule.
498 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

• The third set of soft constraints attempts to avoid assigning a day shift followed by a night shift on
the next day. This would help nurses adjust their time to sleep. This policy is also recommended
in the literature [26].
• The fourth set of soft constraints attempts to avoid isolated days on (o7–on–o7 patterns).
• The =fth set of soft constraints attempts to avoid isolated days o7 (on–o7–on patterns).

4.2.2. Notations and assumptions

The schedule is assumed to start at the =rst day of a week. A working day starts from 7:00 AM
to 7:00 AM of the next day (two shifts per 24 h). Thus, the origin of time is shifted from midnight
to 7:00 AM for convenience. A weekend consists of two days. The length of a schedule is 28 days
(4 weeks). The following notation is introduced.
n: number of days in a schedule (n = 28)
m: number of nurses available for the unit of interest
i: index for days, i = 1; 2; : : : ; n
k: index for nurses, k = 1; 2; : : : ; m
Di : sta7 requirement for day shift of day i; i = 1; 2; : : : ; n
Ni : sta7 requirement, for night shift of day i; i = 1; 2; : : : ; n
Additional notation will be introduced when appropriate.

4.2.3. Decision variables

XDi; k = 1 if nurse k is assigned a day shift for day i; i = 1; 2; : : : ; n; k = 1; 2; : : : ; m
= 0 otherwise

XNi; k = 1 if nurse k is assigned a night shift for day i; i = 1; 2; : : : ; n; k = 1; 2; : : : ; m

= 0 otherwise

XRi; k = 1 if nurse k is assigned a day o7 for day i; i = 1; 2; : : : ; n; k = 1; 2; : : : ; m

= 0 otherwise

4.2.4. Formulating model constraints

Hard constraints:
Satisfy daily sta7 requirements

m
XDik ¿ Di ; for all i = 1; n; (4.1)
k=1

m
XNik ¿ Ni ; for all i = 1; n: (4.2)
k=1
Avoid any two consecutive shifts
XDik + XNik + XRik = 1; for all i = 1; n and k = 1; m; (4.3)

XDi+1; k + XNik 6 1; for all i = 1; n − 1 and k = 1; m: (4.4)

 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 499

No more than 4 consecutive days on

XRik + XR(i+1)k + XR(i+2)k + XR(i+3)k + XR(i+4)k ¿ 1; for all i = 1; n − 4 and k = 1; m: (4.5)
At least 4 days o7 in a 4-week-schedule during weekends
XR6; k + XR7; k + XR13; k + XR14; k + XR20; k + XR21; k + XR27; k + XR28; k ¿ 4; for k = 1; m: (4.6)
Minimum and maximum working days per four-week schedule

n
(XDik + XNik ) ¿ 14; for all k = 1; m; (4.7)
i=1

n
(XDik + XNik ) 6 16; for all k = 1; m: (4.8)
i=1

Minimum night shifts in the scheduling period (25% of the total)

n
XNik ¿ 4 for all k = 1; m: (4.9)
i=1

4.2.5. Formulating goals

In order to incorporate the soft constraints in the scheduling model, we will include the following
goals, which are consistent respectively with the above soft constraints. The problem has therefore
=ve goals as follows.
Goal 1: It minimizes the deviations between the sum of actual days on and the minimum required
days on. This goal ensures that all nurses are scheduled to have 15 days as possible in the 4-week
schedule.
 n 


−
(XDik + XNik − (d1+ k − d1k ) = 15; for all k = 1; m: (4.10)
i=1
−
Here, d1k (respectively d1+k ) is the amount of negative (respectively positive) deviation from goal 1
for nurse k. Only positive deviations are penalized.
Goal 2: It attempts to have in the schedule more day shifts than night shifts for all nurses.

n
n
−
XDik − XNik − (d2+ k − d2k ) = 1; for all k = 1; m: (4.11)
i=1 i=1
−
Here, d2k (respectively d2+k ) is the amount of negative (respectively positive) deviation from goal 2
for nurse k. Only negative deviations from goal 2 are penalized.
Goal 3: It avoids assigning a nurse to work a day shift and the night shift of the following day.
−
XDik + XN(i+1)k − (d3+
ik − d3ik ) = 1; for all i = 1; n − 1 and k = 1; m: (4.12)
−
Here, d3ik (respectively d3+
ik )
is the amount of negative (respectively positive) deviation from goal 3
for day i and nurse k. Only positive deviations are penalized.
Goal 4: It avoids o7–on–o7 patterns. This goal attempts to have minimum isolated day or night
shifts, for all nurses.
−
XRik + XD(i+1)k + XN(i+1)k + XR(i+2)k − (d4+
ik − d4ik ) = 2; i = 1; n − 2; k = 1; m: (4.13)
500 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

Here, d4− +
ik (respectively d4ik ) is the amount of negative (respectively positive) deviation from goal 4
related to isolated day/night shifts on, for day i and nurse k. Only positive deviations are penalized.
Goal 5: This goal consists on minimizing isolated days o7.
−
XDik + XNik + XR(i+1)k + XD(i+2)k + XN(i+2)k − (d5+
ik − d5ik ) = 2; i = 1; n − 2; k = 1; m:
(4.14)
Here, d5− +
ik (respectively d5ik ) is the amount of negative (respectively positive) deviation from goal
5 related to isolated days o7 for the di7erent combinations of days on preceding and following the
day o7, for day i and nurse k. Only positive deviations are penalized.

4.2.6. Assigning importance weights

Importance weights are assigned to each goal reEecting the relative importance of that goal com-
pared to the others. Penalty levels for violating the corresponding goals express these importance
weights. These levels are denoted respectively by P1 ; P2 ; P3 ; P4 , and P5 . For the sake of the appli-
cation of RKH Hospital, goal 1 has been considered by far the most important. Goals 2–5 were
considered relatively comparable with highest importance to goal 2, then, goal 3, and =nally goals 4
and 5 simultaneously. After a number of comparisons and assessments in the same spirit as analytical
hierarchy process (AHP) the importance weights have been assigned so that the highest importance
P1 =20, then P2 =5, P3 =3 and =nally P4 =P5 =1. Sensitivity analysis shows that these weights need
not be very accurately estimated, as the solution remains unchanged for reasonably large ranges of
these weights for this particular problem. The reader is referred to Romero [17] for an interesting
and extended discussion on weighting issues.

4.2.7. Objective function

The objective function consists on minimizing the sum of the weighted deviations from the cor-
responding goals. Note that goal 1 minimizes in particular the unnecessary overtime and therefore
the corresponding cost. The expression of the objective function is given by

m
m
Z = P1 +
d1k + P2 d2− k
k=1 k=1
 n− 1 m   n− 2 m 





+ P3 d3+
ik + P4 d4+
ik (OBJ)
i=1 k=1 i=1 k=1
 n− 2 m 



+ P5 d5+
ik :
i=1 k=1

4.2.8. Underlying LGP model

The LGP model consists on minimizing the above objective function (OBJ) under the hard and soft
constraints (4.1)–(4.9) and (4.10)–(4.14), respectively. The model has a total of 9mn–10m decision
variables (3mn binary variables and 6mn–10m non-negative deviation variables). For the case of a
28-day schedule, the number of binary decision variables will be 84m and the number of deviation
variables will be 158m. The total number of model constraints (excluding the non-negativity and
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 501

the binary constraints) is 6mn + 3n − 5m (3mn + 2n − m) hard constraints and 3mn + n − 4m soft
constraints. This corresponds to 83m + 56 hard constraints and 80m + 28 soft constraints for a 28-day
schedule.

4.3. Issues on model implementation

4.3.1. Introduction
The psychiatry unit is =rst selected for illustration. In this unit, there are a total of 15 nurses (6
from sta7 nurse 1, 8 from sta7 nurse 2, and 1 nurse aid). Table 1 shows a manual schedule made
by the head nurse. In the corresponding schedule, two nurses were on vacation (one from sta7 nurse
1 and one from sta7 nurse 2). Thus, the schedule accounts only for the 13 remaining nurses. The
minimum requirement of nurses per shift is 3 nurses for all day and night shifts and for all days
including weekends. Among these 3 nurses at least one nurse must be from sta7 nurse 1 and 1 nurse
must be from sta7 nurse 2. The problem consists of 1135 hard constraints and 1068 soft constraints.
It also consists of 1092 binary decision variables and 2054 non-negative deviation variables. The size
of the problem in the current application is computationally prohibitive. Therefore, some heuristic
must be used for obtaining a satisfactory solution.

4.3.2. Subgrouping
Huarng [18,19] proposes for NP-hard scheduling problems the approach of subgrouping by splitting
nurses and workloads into several subgroups, so that each subgroup will be of manageable size. By
aggregating these subgroups, all the hard constraints must be satis=ed. According to the computational
experience in Huarng [18,19], the solutions obtained by such an approach are very satisfactory.
However, there is no systematic way for subgrouping and the approach is model-dependent. In fact,

Table 1

ManualSchedule for Psychiatry Unit

% of Nights

Nurse
% of Days

Week1 Week2 Week3 Week4

i
Name
Staff Nurse 1 Level

Night

Totall
Day

S S M T WT F S S M T WT F S S M T WT F S S M T WT F
1 G.M N N N N N N N D D D D D D D D 8 7 15 53% 47%
2 E.H D D D D D D D D N N N N N N N 8 7 15 53% 47%
3 M.B D D D D D D N N N N N N N N N 6 9 15 40% 60%
4 M.R N N N N D D N D D D D D D D D 10 5 15 67% 33%
6 D.M N N N N N N N D D D D D N N N 5 10 15 33% 67%
1 V.A N N N D D D D D N N N N N N N 5 10 15 33% 67%
2 E.B D D D N N N N N N N N N N N N 3 12 15 20% 80%
NA Staff Nurse 2

4 M.D D D D D D D D D N N N N N N N 8 7 15 53% 47%

5 N.T N N N N N N D D D D D D D D D 9 6 15 60% 40%
6 R.S N N N N N N N N N N D D D D D 5 10 15 33% 67%
7 C.C N N N N D D D D D D N N N N N 6 9 15 40% 60%
8 R.C N N N N N N N D D D D D D D D 8 7 15 53% 47%
1 M.N D D D D D D D D D D D D D D D 15 0 15 100% 0%

TDS 2 4 4 2 2 3 3 3 3 5 2 3 4 4 3 4 4 4 5 4 4 3 4 4 2 3 4 4
TNS 4 4 4 4 4 4 4 4 4 3 3 4 4 4 3 3 3 3 3 3 3 3 3 4 4 4 3 3

TDS:Total Day Staff, TNS:Total Night Staff, MDS:Min. Day Staff ,MNS:Min. Night Staff
D - Day Shift, N -Night Shift, SN1-Staff Nurse 1, SN2-Staff Nurse 2, NA- Nurse Aid
502 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

Table 2

Nurse Main Model - Subgroup 1

Night
Total
Day
Name
Level

Days
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
G.M N N N N N N D D D D D D D D N 8 7 15
E.H D D N N N N D D N N N D D D D 8 7 15
SN1

M.B D D D D D D D D N N N N N N N 8 7 15
M.R D N N N N N N N D D D D D D D 8 7 15
D.M N N D D N N D D D N N N D D D 8 7 15

Day 2 1 1 1 1 1 1 1 1 1 1 1 1 2 3 2 2 1 1 1 1 2 2 2 2 2 1 2 40
Night 1 1 1 2 3 3 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 35

Table 3

Nurse Main Model - Subgroup 2

Night
Total
Day
Name
Level

Days
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
V.A D N N N N N N D D D D D D D N 8 7 15
SN2

E.B D D D D N N N N N N N D D D D 8 7 15
M.D D D D N N D D D N N D D N N N 8 7 15
N.T N N N D D D D D D D D N N N N 8 7 15

Day 2 1 2 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 32
Night 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28

by having a large number of subgroups, the model may become unbalanced, as the larger the number
of subgroups, the less Eexible the allocation of workloads. In addition, as the number of subgroups
increases, the deviation from optimality tends to increase. On the other hand, a small number of
subgroups may result in ine6cient computational time per subgroup. This approach will be used in
the current study.

4.3.3. Illustration
We opted for subgrouping to solve the scheduling problem at the psychiatry unit using LINGO.
The computational time required to determine the optimal solution for a schedule with more than
six nurses is found to be too long using a PC/Pentium 700 MHz. The model takes more than 2 h
to solve for a subgroup of size six. It takes 10 min to solve for a subgroup of size 5, and only few
seconds to solve for subgroups of size of 4 or less. Therefore, we select 3 subgroups of respectively
5, 4, and 4 nurses for the 13-nurse psychiatry unit. Subgroup 1 consists of 5 nurses from sta7 nurse
1. Subgroup 2 consists of 4 nurses all from sta7 nurse 2, and subgroup 3 consists of 4 nurses, where
3 nurses are from sta7 nurse 2 and 1 nurse aid. At every shift, at least one nurse from each subgroup
will be assigned. This guarantees that the aggregate minimum requirement is met, including the types
of nurses per shift. Tables 2–4 show the nurse scheduling for subgroups 1–3 respectively. Table 5
shows the combined nurse scheduling for the whole unit.
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 503

Table 4

Nurse Main Model - Subgroup 3

Night
Total
Day
Name
Level

Days
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
R.S D N N N N N N D D D D D D D N 8 7 15
SN2

C.C D D D D N N N N N N N D D D D 8 7 15
R.C D D D N N D D D N N D D N N N 8 7 15
NA

M.N N N N D D D D D D D D N N N N 8 7 15

Day 2 1 2 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 32
Night 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28
D - Day Shift, N -Night Shift, SN1-Staff Nurse 1, SN2-Staff Nurse 2, NA- Nurse Aid

Table 5

Nurse Main Model - Entire Unit

Night

Total
Day
Name

Days
Level

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
G.M N N N N N N D D D D D D D D N 8 7 15
Subgroup 3 Subgroup 2 Subgroup 1

E.H D D
N N N N D D N N N D D D D 8 7 15
SN 1

M.B D D D D D D D D N N N N N N N 8 7 15
M.R D N N N N N N N D D D D D D D 8 7 15
D.M N N D D N N D D D N N N D D D 8 7 15
V.A D
N N N N N N D D D D D D D N 8 7 15
SN2

E.B D D D D N N N N N N N D D D D 8 7 15
M.D D D D N N D D D N N D D N N N 8 7 15
N.T N N N D D D D D D D D N N N N 8 7 15
R.S D
N N N N N N D D D D D D D N 8 7 15
SN2

C.C D D D D N N N N N N N D D D D 8 7 15
R.C D D D N N D D D N N D D N N N 8 7 15
NA

M.N N N N D D D D D D D D N N N N 8 7 15

Day 6 3 5 3 3 3 3 3 3 5 3 3 3 4 5 4 4 3 3 3 3 4 4 4 6 4 3 4 104
Night 3 3 3 4 5 5 3 3 4 4 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 91

D - Day Shift, N -Night Shift, SN1-Staff Nurse 1,SN2-Staff Nurse 2, NA- Nurse Aid

4.3.4. Comments
The solution obtained in this illustration for the entire unit is optimal since all goals are met
(objective function is zero). As mentioned above, the total computational time is about 10 min. In
this computerized schedule, all nurses work for 15 days, as recommended by the =rst soft constraint.
This is also the case for the manual-made schedule. However, The manual-made schedule is very
unbalanced with respect to day and night shifts. In fact, the proportion of day shifts varies among
nurses between 20% and 100%. Note however that in the computerized schedule all nurses have
the same proportion of day shifts. In addition, the sta7 requirement (considered in the model as a
hard constraint) is violated in the manual schedule (Tuesday of the second week). Moreover, the
soft constraint of avoiding a day shift followed by a night shift in the next day is violated 3 times
in the manual schedule. Isolated days on and o7 are also encountered in the manual schedule.
This is also the case for the computerized schedule, except that these violations occur in the
504 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

computerized schedule only at the end of the schedule. In this case, these violations may be avoided
simply by assigning days on/o7 properly at the beginning of the next schedule (see the application
below).
To assess the adequacy of selecting subgroups of sizes 5 – 4 – 4, we implemented the model with
sizes of 5 –5 –3. In this latter case, in addition to almost doubling the computational time, the
subgroup of size 3 had a much larger fraction of day shift than for the other subgroups leading to
an unbalanced aggregate schedule, as discussed above. This suggests avoiding other combinations,
such as 4 –3–3–3. Consequently, the choice made in the original model is appropriate.
A sensitivity analysis is carried out for penalty weights. A variety of ranges of penalty weights
o7er also optimal solutions with slightly di7erent schedules. Therefore, these penalty weights are
not considered as sensitive. However, the original values have the advantage of broadly reEect-
ing the relative importance as seen by the decision-makers (several head-nurses) for the di7erent
goals.

5. Application: implementation for a 6-month trial period

Before engaging in a costly project of developing a friendly user software package based on the
current model, the hospital opted for a 6-month trial period of the model in some 12 units of the
hospital. The largest unit, the surgical unit, has 22 nurses and the smallest, the pediatric unit, has
12 nurses. One of the authors (the 2nd) has taken in charge running the model in all selected
units for the 6-month period and making the necessary changes when requested for. For this pilot
experiment, head nurses were very reluctant to approve late requests of days o7 and other requested
changes in the schedules in order not to disturb much the user of the model. Thus, this test-period
was characterized by less Eexibility of head-nurses than usual. This has been pointed out by several
respondents to the second survey.
Before running the model, a computer code has been developed by the authors and has interfaced
with LINGO to avoid violations when linking two schedules. The serious types of violations that
may occur are to assign to a given nurse a night-shift at the last day of one schedule and a day-shift
on the =rst day of the next schedule leading to a 24-h shift. A second type of violation would be
to assign for a nurse say the last 3 days of a schedule and the =rst 3 days of the following one
resulting into 6 consecutive days on, which violates the no-more than 4 consecutive days on rule.
The computer code essentially keeps record of the last days of the schedule for each nurse and adds
additional constraints (hard and soft when appropriate) for the following run to avoid (or reduce)
violations.
Out-of the 72 (6 times 12) runs, optimality (i.e., zero objective value) has been obtained in
64 occasions. In the remaining cases, soft constraints violations have occurred only for the lowest
importance weights (Pi = 1). The largest running time was of about 20 min obtained for the surgical
unit (having 22 nurses), for which subgrouping was of the sizes 5 –5 – 4 – 4 – 4.
Surprisingly, no data was found regarding the nursing overtime cost per unit or per category of
nurses. However, for the last few years, total nursing overtime cost at the entire hospital level was
around 720,000 US $ per year, leading to an average over time cost of around 400 $ per nurse
(independently of the nurse category) per year, given that the hospital accounts for 1798 nurses.
The results of the pilot experiment has shown slightly more than 14% overtime reduction in the
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 505

average based on a conservative estimation. If this result will hold true when applying the model
at the hospital level, then a potential annual saving of about 100,000 US $ would occur, indicating
that the developed model is promising in this regard.
To evaluate the e7ect of this computerized nursing system on nurses preferences, a second survey
was produced and distributed to 100 nurses among those that had the pilot experiment. A response
rate of 74% has been obtained. Among the respondents, about 70% (52 nurses) were fully satis=ed
with the new system. Also, 65 nurses (i.e., 88%) believe that the new system is signi=cantly better
than the manual one with respect to fairness bases and 61 nurses (i.e., 82%) believe that it is signif-
icantly better with respect to nurses preferences. Two drawbacks were recorded. In fact, 39 nurses
(52%) pointed out that the current system is less Eexible than the old one. Also, 22 nurses com-
plained from the fact that the new system o7er them less overtime than usual and therefore deprive
them from a good source of income. It should be noted however that inEexibility is only temporary
and as soon as the head nurses would manipulate by themselves the system (in its user-friendly
version, see [20]), then they would be able to make modi=cations at will. The overtime problem
conEicts with the hospital objectives and it is even considered (for the hospital administration) as
an advantage of this computerized system.

6. Conclusions and directions for future research

In this study, a 0-1 linear goal-programming model is developed for nurse scheduling in RKH
hospital. Currently, the head nurse at each unit/department makes a manual schedule through a
trial-and-error approach. This pencil-and-paper approach is not only costly, but also ine6cient in
producing satisfactory schedules. These manual schedules do not satisfy a number of important
criteria for e6cient scheduling. These include balanced schedules, fairness considerations, and nurses’
preferences, in addition to ergonomic considerations, and sta6ng requirements both in quality and
size. The developed model provides important improvements in this regard besides the fact that it
o7ers a practical computerized tool.
The developed model considers nurses’ preferences by relying on a survey’s results reEecting
these preferences. This survey is produced for the sake of this study. Other sources used in building
scheduling policies are the current applied policies in the hospital, as well as recommended policies
displayed in the literature and that account for ergonomic factors. Satisfying simultaneously all the
suggested policies need not be feasible. Consequently, some of these policies are taken as hard
constraints that must be satis=ed. Hard constraints are selected based on feedback from head nurses
and other specialized sta7 in the hospital. The rest of the constraints are taken as soft constraints with
di7erent importance weights. Also, these importance weights are assessed based on the judgement
of several head nurses.
For measuring a schedule quality, Oldenkamp and Simons [16] develop =ve factors as speci=ed
below.

1. Optimality: represents the degree in which nursing expertise is distributed over the di?erent
shifts. The way subgrouping is made in the developed 0-1 goal program accounts for this factor.
2. Completeness: represents the degree in which the quantitative demands for occupation per shift
are met. In the 0-1 model, this is formulated as hard constraint and is always satis=ed.
506 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507

3. Proportionality: represents the degree in which each nurse has been given about the same amount
of day shifts and night shifts. This is amply satis=ed in the developed 0-1 model as illustrated
by the example above and during the trial-period.
4. Healthiness: represents the degree in which it has been taken care of the welfare and health of
the nurses. This is also largely considered in the model by incorporating several related factors
as hard and soft constraints.
5. Continuity: represents the degree in which there is continuity in the nursing crew during the
di?erent shifts. This is also satis=ed in the developed model.

Therefore, the developed model performs quite well based on the quality criteria of Oldenkamp
and Simons [16]. This conclusion was also supported by the results on the feedback obtained from
the second survey that has been designed to assess nurses satisfaction upon the 6-month trial period
of the model implementation in 12 units of the hospital. The model has been found not only to
satisfy hospital’s objectives but also, and to some large extent, nurses’ preferences (proportionality,
weekends o7, isolated days on and o7, etc.). The implementation also shows a 14% reduction in
overtime over the tested units. This proves some potential for important savings in nurses costs if
the model is to be implemented at the entire hospital level.
The solution has been obtained through subgrouping as the computational time is prohibitive for
an entire unit. Subgroups sizes are selected not to exceed 5 nurses. The implementation time is within
10 min per subgroup. The solution is found most often optimal (all goals are satis=ed). Suboptimality
has been encountered only in few occasions, where violations of soft constraints occurred only for
those with low-importance weights.
Future work may focus on building a friendly-user computer package. Work in progress by the
authors [20] extend the current model to account for other important scheduling aspects, namely
vacations and days o7 requests. For requests of few days o7, the model accepts them by adding
additional constraints on days o7, as requested. If the number of requests becomes fairly high by
several nurses, then some rules are developed (such as priority for assigning days o7, =rst requested
=rst accepted with a maximum number of requested days) to keep the schedule feasible. Long
vacations are incorporated by taking out the corresponding nurses from the schedules or from a
number of weeks in the schedules. However, other rules are under considerations to ensure that the
nurses, after taking vacations, would have a balanced remaining schedule.

References

[1] Ozkarahan I. A Eexible nurse scheduling support system. Computer Methods and Programs in Biomedicine
1989;30 (2/3):145–53.
[2] Warner DM. Nurse sta6ng, scheduling, and reallocation in the hospital. Hospital and Health Services Administration
1976a;21(3):77–90.
[3] Howell JP. Cyclical scheduling of nursing personnel. Hospitals 1966;40:77–85.
[4] Maier-Rothe C, Wolfe HB. Cyclical scheduling and allocation of nursing sta7. Socio-Economic Planning Science
1973;7:471–87.
[5] Ozkarahan I. An integrated nurse scheduling model. Journal of the Society for Health Systems 1991;3(2):79–101.
[6] Caron G, Hansen P, Jaumard B. The assignment problem with seniority and job priority constraints. Operations
Research 1999;47(3):449–53.
 M.N. Azaiez, S.S. Al Sharif / Computers & Operations Research 32 (2005) 491 – 507 507

[7] Abernathy WJ, Balo7 N, Hershey JC, Wandel S. A three-stage manpower planning and scheduling model—a
service-sector example. Operations Research 1973;22(3):693–711.
[8] Warner DM. Scheduling nursing personnel according to nursing preference: a mathematical programming approach.
Operations Research 1976b;24(5):842–56.
[9] Arthur JL, Ravindran A. A multiple objective nurse scheduling model. IIE Transactions 1981;13(1):55–60.
[10] Ozkarahan I, Bailey JE. Goal programming model subsystem of a Eexible nurse scheduling support system. IIE
Transactions 1988;20(3):306–16.
[11] Trivedi VM. A mixed-integer goal programming model for nurse service budgeting. Operations Research
1981;29(5):1019–34.
[12] Musa AA, Saxena U. Scheduling nurses using goal-programming techniques. IIE Transactions 1984;16(3):217–21.
[13] Berrada I, Ferland JA, Michelon P. A multi-objective approach to nurse scheduling with both hard and soft
constraints. Socio-Economic Planning Sciences 1996;30(3):183–93.
[14] Moores B, Garrod N, Briggs G. The student nurse allocation problem: a formulation. Omega 1978;6(1):93–6.
[15] Chen J, Yeung TW. Hybrid expert-system approach to nurse scheduling. Computers in Nursing 1993;11(4):183–90.
[16] Oldenkamp JH, Simons JL. Quality factors of nursing schedules. Planning and Evaluation-Proceeding AMICE
1995;95:69–74.
[17] Romero C. Handbook of critical issues in goal programming. Oxford: Pergamon Press; 1991 [Chapter 8].
[18] Huarng F. Integer goal programming model for nurse scheduling: a case study, Lecture Notes in Economic and
Mathematical System, Vol. 448. Hagen, Germany: Springer; 1997. p. 634 – 43.
[19] Huarng F. A primary shift rotation nurse scheduling using zero-one linear goal programming. Computers in Nursing
1999;17(3):135–44.
[20] Azaiez MN, Al-Sharif S. An extended 0-1 linear goal programming model for nurse scheduling, 2003, in preparation.
[21] Smith DL, Wiggins A, Bird D. Post-implementation experience with computer-assisted nurse scheduling in a large
hospital. INFOR 1979;17(4):309–21.
[22] Isken MW, Hancock WM. A heuristic approach to nurse scheduling in hospital units with non-stationary, urgent
demand, and =xed sta7 size. Journal of the Society for Health Systems 1991;2(2):24–40.
[23] Harmeier PE. Linear programming for optimization of nurse scheduling. Computers in Nursing 1991;9(4):149–51.
[24] Tobon Perez MV. An integrated methodology to solve sta6ng, scheduling and budgeting problems in a nursing
department. Ph.D. dissertation, University of Pittsburgh, 1984.
[25] Okada M, Okada M. Prolog-based system for nursing sta7 scheduling implemented on a personnel computer.
Computers and Biochemical Research 1988;21:53–63.
[26] Tucker P, Smith L, Folkard S. Distribution of rest days in 12 hour-shift systems: impact on health, well being, and
on shift alertness. Occupation Environment Med 1999;56:206–14.