Expert Systems With Applications 211 (2023) 118185

Contents lists available at ScienceDirect

Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa

A sustainable and efficient home health care network design model

under uncertainty
Mahdyeh Shiri a, *, Fardin Ahmadizar a, Dhananjay Thiruvady b, Hamid Farvaresh a
a
Department of Industrial Engineering, Faculty of Engineering, University of Kurdistan, Sanandaj, Iran
b
School of Information Technology, Faculty of Science, Engineering and Built Environment, Deakin University, Geelong, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: To cater to the increasing demands, particularly during diseases such as Covid-19, the design and planning of
Home health care home health care systems is of significant importance. The current study proposes a multi-objective mixed-
Corporate social responsibility integer linear model for a home health care network in two stages; the first is the opening of efficient health
Efficiency
centres, and the second is the routing and scheduling considering corporate social responsibility and efficiency.
Robust-fuzzy approach
Sustainability
There are multiple objectives that we consider, including minimization of total costs and inefficiency consid­
erations, and maximization of social aspects. A novel aspect of this study is the consideration of social re­
sponsibility, which includes employment opportunities and regional economic development, and efficiency in
terms of time, energy, and mismanagement of budgets. To measure efficiency, an augmented version of the data
envelopment analysis approach is incorporated into the proposed optimization model. Additionally, the TH
approach is developed as an interactive fuzzy method to deal with the proposed multi-objective model. Within
the HHC problem, costs, social factors, and service time are inherently uncertain, and hence, to solve this
problem, a robust-fuzzy approach is proposed. The ensuing model is applied to a real case study of Kermanshah
in Iran. Moreover, several problem instances motivated by real cases are generated with different characteristics
to measure the performance of the proposed model and approach. The results show that decision-makers’
preferences play a key role in human resource planning and regional development. Furthermore, the results
confirm the efficiency of the proposed approach in different instances within reasonable time frames.

1. Introduction In order to provide HHC services effectively, a number of significant

decisions need to be made, including opening health centres, assigning
Nowadays, Home Health Care (HHC) networks are an essential part nurses to patients, and routing and scheduling of nurses (Nikzad et al.,
of the health sector. These networks serve patients in their homes so that 2021). The ensuing problem of scheduling is very complex and requires
they may stay close to their family and friends while being treated in a a great deal of effort and time to obtain effective plans, which can lead to
stress-free environment compared to hospitals (Fikar & Hirsch, 2017). significant benefits to patients and hospitals. On the one hand, it allows
HHC provides a wide range of medical services in the homes of patients decreasing costs and increasing patient satisfaction (Grenouilleau,
and the elderly. These services include treatments (e.g., injections) for Legrain, Lahrichi, & Rousseau, 2019), and on the other hand, hospitals
patients with serious illnesses, care for acute injuries, visiting patients benefit from the beds being freed. Moreover, due to aging populations,
with unstable health statuses, psychiatric care, physical therapy, etc. In the demand for HHC services has increased significantly; for example,
this network, typically, a nurse travels from a health centre to patients’ the number of nurses required in the HHC network in the United States
homes to visit them within the pre-specified time windows. At the end of has doubled over the past ten years (Span, 2016). In addition, a
each visiting tour, the treatments or samples taken from patients should pandemic such as the Covid-19 outbreak has significantly raised the
be reported and submitted to a laboratory to investigate the health re­ demand for HHC services as people are required to stay at their place of
sults (Fathollahi-Fard, Ahmadi, & Karimi, 2020). In several studies, each residence while still requiring health services. For this reason, Home
tour starts from and finishes at a nurse’s home (Nikzad, Bashiri, & Health Care News (HHCN-https://homehealthcarenews.com) states that
Abbasi, 2021). 31 percent of HHC companies have been negatively impacted by the

* Corresponding author.
E-mail addresses: [email protected], [email protected] (M. Shiri).

https://doi.org/10.1016/j.eswa.2022.118185
Received 23 November 2021; Received in revised form 7 May 2022; Accepted 14 July 2022
Available online 20 July 2022
0957-4174/© 2022 Published by Elsevier Ltd.
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Covid-19 outbreak in 2020. Thus, an efficient HHC network can be of satisfaction level for decision-makers. In other words, the minimum
critical importance in dealing with this situation. degree of confidence in the possibilistic chance constraint should be
With raising awareness of sustainability among communities, in­ provided to satisfy the decision-makers. Therefore, the proposed model
dustry, and government over the last two decades, policy-makers incorporates mixed uncertainty in fuzzy data and scenarios. For these
worldwide have tried to find a way to include sustainability issues in reasons, this study presents a robust-fuzzy approach to cope with the
industrial and urban development, and the health care industry has not complexity induced by the uncertainty surrounding the parameters.
been an exception (Fiksel, 2006). Sustainability is mostly defined as an Motivated by a case study in Kermanshah, Iran, where an HHC
equilibrium between economic and social issues involved in human network helps the health system to visit the patients at their homes, we
development. The social aspect, which has not been considered in the aim to generalize HHC for the real case study and answer the following
literature a lot, relates to forcing non-governmental organizations to pertinent and important questions:
take employment opportunities and economic development for local
communities for the social impacts of their activities (Sharifi, Hosseini- • How can the HHC be designed and planned to consider several
Motlagh, Samani, & Kalhor, 2020; Zhalechian, Tavakkoli-Moghaddam, important aspects such as total network costs, efficiency, and social
Zahiri, & Mohammadi, 2016). In addition, the World Health Organiza­ responsibility?
tion emphasizes the important relationship between economic devel­ • How can a number of the important factors with different units be
opment and health (Zahiri, Zhuang, & Mohammadi, 2017), where used simultaneously to affect the selection process of efficient health
improved health directly contributes to improving the economy. For centres?
example, a worker’s good state of health increases productivity, reduces • What procedure should be followed to promote CSR issues within a
production losses due to worker sickness, and decreases absenteeism sustainable HHC network?
rates. Additionally, more employment opportunities for nurses in cen­ • How can the patients get multiple services within a specific time
tres that have been opened improve the speed of regional employment, window?
thereby affecting regional economic and employment development, • How can the uncertain scenarios and fuzzy parameters be dealt with?
particularly in less-developed districts. Hence, the motivation for our • How can an approach be devised to efficiently manage the different
study is to employ social impact explicitly as an objective, namely, types of uncertain data?
Corporate Social Responsibility (CSR).
Inefficiency, if not carefully considered, can lead to a waste of re­ Underpinned by these questions, this paper presents a new mathe­
sources, time, energy and mismanagement of budgets in the network matical model for integrated locating, assignment, routing, and sched­
(Haeri, Hosseini-Motlagh, Ghatreh Samani, & Rezaei, 2020). The uling decisions for the HHC problem. Locating is the basic phase of
opening of the centres is an initial component of an HHC network, and designing an HHC network and significantly improves the delivery of
selecting them appropriately, significantly based on several main fac­ services and development of regions. The proposed model arises from
tors, improves the efficiency of the network. To measure the inefficiency the classic Location Routing Problem (Prodhon and Prins, 2014). In the
of the network, the Data Envelopment Analysis (DEA) method has pre­ proposed model, an important strategic decision affects opening the
viously been shown to be effective (Cook, Roll, & Kazakov, 1990). This appropriate health centres with a high population, low traffic, and low
method has two main advantages compared to other methods; 1) the pollution. To do so, the DEA tool is used to select appropriate centres by
factors remain in their natural physical units, 2) multiple factors are input and output factors with various units and without any weight. In
used simultaneously to assess which candidates work most efficiently. the proposed model, patients can get multi-services, request required
Additionally, most studies focus on the total cost of the HHC network, drugs, and take tests within their pre-defined time window. The
while the efficiency of centres and their social impacts have not been employment opportunity for nurses and regional economic development
investigated simultaneously. These reasons motivate us to develop a are considered to increase social responsibility. In addition to social
Multi-Objective Linear Programming (MOLP) model for the design and responsibility, total network costs and CSR are considered as objective
planning of an HHC network. To solve the MOLP model, we utilize an functions and we use a multi-objective decision-making approach. In
interactive fuzzy method proposed by Torabi and Hassini (2008) called particular, the TH approach as an interactive strategy is developed to
the TH method. A key aspect of this method is to consider the satisfac­ solve the proposed multi-objective model. Therefore, the aim of this
tion levels of decision-makers, which are the normalization values of model is to determine the set of efficient and sustainable health centres,
objective functions or membership values. This method allows decision- while taking into account fixed and variable costs. In this HHC network,
makers to effectively trade off their preferences concerning multiple the nurses start their visiting tour from an appropriate health centre.
objectives (Torabi & Hassini, 2008). In other words, it is a promising and Based on our review of real case situations, the health centre often tries
interactive fuzzy approach that can produce high-quality solutions to allocate its nurses to patients who live as close as possible to the
based on the preferences of decision-makers, and at the same time af­ health centre. At the end of each visiting tour, nurses finish their work at
fords flexibility. a laboratory; therefore, this model is an open vehicle routing problem.
Due to the changes in the status of patients, their service times are The time windows of patients are pre-defined based on patients’ pref­
uncertain and can vary greatly, thereby affecting the design of the erences as to when they want to be visited. Additionally, the capacity of
network. Additionally, the knowledge of the nurses and the medical the nurse’s vehicle and the number of required services for each patient
history of patients provide an estimation of the patients’ service times as are known to be important points in the tactical decisions assignment,
a fuzzy and scenario-based parameter. In the literature, fuzzy pro­ routing, and scheduling. In the real world, the service time for visiting
gramming and robust approaches have been employed to deal with such each patient is undetermined and can be characterized in different
problems with uncertainty (Samani, Hosseini-Motlagh, & Homaei, scenarios based on the patient’s medical history and nurses’ knowledge.
2020). Hence, we develop a p-robust approach to tackle this problem, Moreover, the costs and social parameters in the HHC network are un­
which in particular minimizes the deviation between the objectives and certain, and there is no information and historical data for them. Thus,
optimal values under each scenario. Furthermore, data such as cost and the fuzzy logic theory is used, and these parameters are considered as
social parameters are uncertain, with no prior or historical knowledge. trapezoidal fuzzy numbers. For solving the resulting model, this study
These parameters are considered epistemic uncertainties, effectively proposes a mixed robust-fuzzy approach to cope with uncertain service
estimated using fuzzy logic and trapezoidal fuzzy numbers. To deal with times, costs, and social parameters. In the model concerned, identifying
the uncertainty, Possibilistic Chance Constrained Programming (PCCP) the location of health centres is the first decision, followed by assign­
is one of the most widely used fuzzy programming methods. This ment, scheduling, and routing. Overall, this study has three phases. In
method can be applied to possibilistic data and provide a minimum the first phase, we apply the p-robust approach to deal with the

2
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

scenarios of service time. We utilize PCCP to cope with fuzzy parameters In another study, Fathollahi-Fard, Hajiaghaei-Keshteli, and Tavakkoli-
(costs, and social data) in the second phase. The first and second phases Moghaddam (2018) investigate a bi-objective green HHC routing
together represent a robust-fuzzy approach to deal with mixed uncertain problem and propose hybridizations of Simulated Annealing (SA) and
parameters (fuzzy and scenario-based parameters). In the third phase, Salp Swarm algorithms. The objective functions consist of environ­
we develop the TH approach to solve the multi-objective HHC model. To mental pollution and total costs. The epsilon constraint method is used
achieve these aims, we present a novel sustainable and efficient HHC to deal with the multi-objective model. Zhang, Yang, Chen, Bai, and
network design, namely Design and Planning of Home Health Care Chen (2018) study an HHCRSP for a real case in China and formulate a
(DPHHC) under uncertainty. Finally, to validate our approach, we novel MIP model with uncertain service times, match qualities, and time
consider a real case study from Kermanshah in Iran, following which we windows. They aim to minimize travel costs, waiting time, and service
design a data set with the help of expert knowledge to carry out a time, and use the weighted method to solve the multi-objective model.
comprehensive computational study. They also develop an Ant Colony Optimization method to solve large
The paper is organized as follows: Section 2 reviews similar recent instances of the problem. Carello, Lanzarone, and Mattia (2018) present
research on the HHC network. Section 3 provides the problem statement a set of Integer Linear Programming (ILP) models for a home care sys­
and mathematical formulation. Section 4 introduces a hybrid method­ tem. They model the nurse-to-patient assignment problem under con­
ology to tackle the uncertainty and an interactive fuzzy method to deal tinuity of care and consider minimizing cost, maximizing utilization,
with the multi-objective model. A real case study, results, and sensitivity and minimizing the total number of reassignments as the objective. They
analyses are defined in Section 5. Finally, findings, discussions, and use the threshold method to find the optimal solutions for the multi-
some important directions for future research are provided in the last objective model. Habibnejad-Ledari, Rabbani, and Ghorbani-Kutenaie
section. (2019) propose a multi-objective Non-Linear Programming (NLP)
model to address staff assignment problems in a home care system. They
2. Literature review aim to minimize the costs and employees for each service and maximize
the worker satisfaction level. They apply a new version of NSGA-II with
The DPHHC problem in an uncertain environment is a relatively new a heuristic initialization for solving this model.
but rapidly evolving field of research. There is a vast literature on health In another study in this category, Regis-Hernández, Carello, and
care systems, though our focus is on recent research closely related to Lanzarone (2020) formulate a multi-objective Linear Programming (LP)
the problem being investigated. The readers are referred to the latest model for HHC services and a matheuristic approach to solve a real case.
review publication by Fikar and Hirsch (2017), which presents a The model can determine the numbers of physicians, nurses, technicians
comprehensive overview of papers in the areas of the Home Health Care and devices to acquire that are needed to meet the demand. Gong, Geng,
Routing and Scheduling Problem (HHCRSP). They classify problems into Zhu, Matta, and Lanzarone (2020) propose an ILP model for the home
two classes based on whether they consist of single or multi-time care scheduling problem and develop a matheuristic approach to solving
periods. this problem. They minimize costs, the penalty of the continuity of care
In recent studies, Rodriguez-Verjan, Augusto, and Xie (2018) pro­ violation, and the preference mismatch, and then use the weighted
posed two Mixed Integer Linear Programming (MILP) models to design method to solve it. Entezari and Mahootchi (2021) propose an MILP
an HHC network with two separate aims to locate health facilities and model for staff routing and scheduling in HHC and develop a Genetic
manage the activities of the health facilities. They implement a real case Algorithm (GA) to find near-optimal solutions. They aim to minimize
study of the Loire department in France. Grenouilleau et al. (2019) their objective functions: travel times, tardiness in providing services,
propose an MILP model for HHCRSP to minimize the total costs of staff overtime, violation of care continuity, and violation of the staff’s
nurses’ overtime, routes, unscheduled visits, and idle time. They apply time windows by the weighted method. Lin, Ma, and Ying (2021) design
their model and a Large Neighborhood Search (LNS) method for a an MILP model for HHC and propose a branch and price method for
dataset from Alayacare in Montreal. Liu, Yuan, and Jiang (2020) model matching demand to supply. They minimize total costs, maximize
a periodic HHC server assignment problem to minimize the maximum customer satisfaction, and use a lexicographic method to solve the
workload between servers in various periods. They propose an efficient model. Goodarzian, Abraham, and Fathollahi-Fard (2021) propose a bi-
region-partition-based algorithm, which effectively solves large-scale objective model for HHC logistics to minimize total costs and time by
problems. In another study, Cinar, Salman, and Bozkaya (2021) considering route balancing. To solve the problem, they use the epsilon
formulate a multi-period model for the HHCRSP to visit patients based constraint method and a metaheuristic approach for large instances.
on their priorities. They develop an adaptive LNS approach to tackle the Malagodi, Lanzarone, and Matta (2021) formulate an MILP model for a
problem. home care vehicle routing problem. They aim to minimize the mis­
We review several relevant studies to find the gaps in existing matches of strict and soft preferences over the works, overtime, and total
research in Table 1. The papers related to this study can be broadly travel time for all caregivers. The weighted method is used to cope with
classified into three main categories: 1. multi-objective HHC problems, the multi-objective model.
2. HHC problems under uncertainty, and 3. multi-depot and multi-care
HHC problems. 2.2. HHC problems under uncertainty

2.1. Multi-objective HHC problems Rodriguez, Garaix, Xie, and Augusto (2015) present a two-stage
MILP model for a health care company in France. The authors aim to
Rest and Hirsch (2016) introduce a Mixed Integer Programming find the minimum number of nurses that can cover all possible routes.
(MIP) formulation for HHC daily scheduling services, where nurses use They consider uncertainty in demand and use a stochastic approach to
public transport. They utilize a weighted objective function to minimize deal with this uncertainty. In another study, Shi, Boudouh, and Grunder
the shift lengths, overtime, the number of second shifts, and over- (2017) design a vehicle scheduling problem for an HHC system and
qualification. They develop three Tabu Search (TS) strategies to apply consider patients’ demand for drugs as fuzzy data, which they tackle
to the Austrian Red Cross data. Braekers, Hartl, Parragh, and Tricoire with fuzzy chance constraint programming. Cappanera, Scutellà, Nervi,
(2016) propose a meta-heuristic based on a multi-directional local and Galli (2018) propose a cardinality-constrained robust approach
search to find a set of non-dominated schedules for a multi-objective considering routing, assignment, and scheduling decisions under un­
HHCRSP. They apply the epsilon constraint method to solve the model certain demand. They propose a decomposition approach as a feasible
with two objective functions: minimizing nurses’ routes and overtime option to increase computational efficiency and apply it to real-world
cost. They use a dataset from the public employment service of Austria. data. Shi, Boudouh, Grunder, and Wang (2018) propose a stochastic

3
 M. Shiri et al.
Table 1
A classification of recent publications.
Reference Modelling Constraint Objective Uncertainty Approach Solution Method Multi-Objective Performance Measure Case
Approach Function Method Study
Multi Multi Time Stochastic Fuzzy Robust Robust- Exact Heuristic
Depots Cares Window fuzzy

Nickel, Schröder, and Steeg (2012) NLP ⨯ ⨯ Multi ⨯ Weighted method Unscheduled tasks, Loyalty, ⨯
Time, Distance
Allaoua, Borne, Létocart, and Calvo (2013) ILP ⨯ Single ⨯ Number of nurses
Milburn and Spicer (2013) MIP Multi ⨯ Epsilon constraint Number of nurses, Workload, ⨯
method Cost
Mankowska, Meisel, and Bierwirth (2014) MILP ⨯ ⨯ Multi ⨯ Weighted method Distance, Time
Lanzarone and Matta (2014) MILP Single ⨯ ⨯ Time ⨯
Mutingi and Mbohwa (2014) LP ⨯ Multi ⨯ Fuzzy simulated Workload, Time, Clustering
evolution efficiency
Carello and Lanzarone (2014) LP Single ⨯ ⨯ Cost ⨯
Rodriguez et al. (2015) MILP Single ⨯ ⨯ Cost of staff ⨯
Fikar and Hirsch (2015) LP ⨯ Single ⨯ Time ⨯
Rest and Hirsch (2016) MIP ⨯ Multi ⨯ Weighted method Time, Shift lengths, #Shifts, ⨯
qualification
Braekers et al. (2016) MIP ⨯ Multi ⨯ Multi-directional Cost, Client inconvenience ⨯
local search
Yalçındağ, Matta, Şahin, and Shanthikumar MIP Multi ⨯ Weighted method Time ⨯
(2016)
Errarhout, Kharraja, and Corbier (2016) LP Single ⨯ Time, Workload
Decerle, Grunder, El Hassani, and Barakat MILP ⨯ Single ⨯ Cost
(2016)
4

Shi et al. (2017) MIP ⨯ Single ⨯ ⨯ Distance

Cappanera et al. (2018) MILP Single ⨯ ⨯ Workload
Shi et al. (2018) LP ⨯ Single ⨯ ⨯ Cost
Fathollahi Fard et al. (2018) MILP ⨯ Multi ⨯ Simulated Environmental pollution, Cost
Annealing
Veenstra, Roodbergen, Coelho, and Zhu (2018) MILP Single ⨯ Cost ⨯
Rodriguez et al. (2018) MILP Single ⨯ Cost ⨯
Decerle, Grunder, El Hassani, and Barakat MIP Single ⨯ ⨯ Cost ⨯
(2018)
Fathollahi-Fard et al. (2019) MILP ⨯ ⨯ Multi ⨯ Cost, Environmental pollution
Erdem and Koç (2019) MILP ⨯ ⨯ Single ⨯ Time
Shiri, Ahmadizar, Mahmoudzadeh, and Bashiri MILP ⨯ Single ⨯ ⨯ Cost
(2019)

Expert Systems With Applications 211 (2023) 118185

Entezari and Mahootchi (2021) MILP ⨯ ⨯ Multi ⨯ Weighted method Cost
Fathollahi Fard et al. (2020a) MILP ⨯ ⨯ Multi ⨯ ⨯ Weighted method Cost, Environmental pollution
Fathollahi Fard et al. (2020b) MILP ⨯ ⨯ Multi ⨯ ⨯ Weighted method Gas emissions, Costs
Shiri, Ahmadizar, and Mahmoudzadeh (2021) MILP ⨯ Multi ⨯ ⨯ Nimbus method Cost, Qualification, ⨯
Qualitative factors
Shahnejat-Bushehri, Tavakkoli-Moghaddam, NLP ⨯ Single ⨯ ⨯ Cost
Boronoos, and Ghasemkhani (2021)
Liu, Dridi, Fei, and El Hassani (2021) MILP ⨯ ⨯ Single ⨯ Cost
Fathollahi-Fard et al. (2021) NLP ⨯ ⨯ ⨯ Multi ⨯ ⨯ Red Deer Algorithm Cost, Unemployment time,
Continuity of care
Yalçındağ and Lanzarone (2021) MILP Multi ⨯ Weighted method Utilization rate, Workload
Our research MILP ⨯ ⨯ ⨯ Multi ⨯ ⨯ TH Method Cost, Efficiency, Social impact ⨯
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

programming model for vehicle scheduling within an HHC routing caregiver inconsistency, and minimization of workload imbalance as
problem. They consider travel and service times as stochastic parameters objective functions.
and devise an efficient SA method to solve the problem. In another
study, Khodaparasti, Bruni, Beraldi, Maleki, and Jahedi (2018) present a 2.3. Multi-depot and multi-care HHC problems
multi-period location-allocation model for nursing home network
planning. They consider uncertain demand and use a robust approach to Bahadori-Chinibelagh, Fathollahi-Fard, and Hajiaghaei-Keshteli
cope with this uncertainty. They aim to minimize the number of un­ (2019) formulate a novel multi-depot HHC routing model to minimize
visited times for patients and consider a real case study of Shiraz city. total costs. They assume an equal number of laboratories and pharma­
Nikzad et al. (2021) consider a two-stage model for planning of re­ cies in the model and present two simple constructive algorithms as
sources in an HHC problem and propose a variant of the progressive solution methods. Erdem and Koç (2019) propose an analysis of electric
Frank, Wolfe and Hedging algorithms. This study aims to minimize the vehicles in the HHC problem by considering multiple depots. They as­
travel costs considering uncertain travel and service times. Addis et al. sume a team of nurses performs a number of patients’ demands via
(2015) present a cardinality-constrained model to handle health care electric vehicles. In addition, Fathollahi-Fard, Govindan, Hajiaghaei-
management problems under uncertainty. This model is a robust opti­ Keshteli, and Ahmadi (2019), Fathollahi-Fard, Ahmadi, & Karimi
mization approach, and surgery duration is uncertain. Lanzarone, Matta, (2020), and Fathollahi-Fard, Ahmadi, Goodarzian et al. (2020) investi­
and Sahin (2012) develop several mathematical programming models gate the multi-depot ability for the HHCRSP. They consider pharmacies
for home care services and assume that the demands of patients are and laboratories as depots. Among the multi-care HHC problems,
either uncertain or deterministic. The authors use Stochastic Program­ Manavizadeh, Farrokhi-Asl, and Beiraghdar (2020) develop a model for
ming to deal with uncertainty. Lanzarone and Matta (2012) balance the HHCRSP and solve it with SA. Multiple (interdependent) services are
workload between the operators to minimize the expected cost value in considered for the benefit of patients. Nasir and Kuo (2020) develop an
the home care network. They consider the patients’ demands as sto­ MILP model for HHC logistics planning to serve elderly people. They aim
chastic and deterministic. Carello and Lanzarone (2021) introduce the to minimize the costs of route assignment for vehicles and nurses and
nurse-to-patient assignment problem in the HHC system under uncertain travel costs. They propose a hybrid GA to tackle the problem. Entezari
service times. They formulate the uncertainty set and use the and Mahootchi (2021) study staff routing and scheduling in HHC in­
implementor-adversary method to solve the robust model. Shi, Bou­ dustries, where a patient can require more than one service (indepen­
douh, and Grunder (2019) develop a model for an HHCRSP under un­ dent or interdependent).
certain travel and service times. The authors use a robust optimization In the following, we summarize several studies in routing and
based on budget uncertainty. They present multiple solution approaches scheduling to help identify gaps. Ghannadpour and Zarrabi (2019)
to solve the problem, including a commercial mixed-integer program­ develop an MIP model and evolutionary approaches for the multi-
ming solver, namely Gurobi, TS, Variable Neighborhood Search, and an objective heterogeneous VRP and vehicle scheduling problem. They
SA algorithm. have two scenarios formulated as objectives: minimizing the total
In the papers related to both multi-objective problems and uncer­ number of vehicles for serving customers and the vehicle fuel, maxi­
tainty, Fathollahi-Fard, Ahmadi, and Karimi (2020) propose an MILP mizing the customers’ satisfaction rate, and minimizing the number of
model for a multi-period, multi-depot, and multi-objective HHC rental vehicles, travel distance, and fuel of personal vehicles while
network. They consider travel and service times as fuzzy parameters and maximizing customer satisfaction. Rahbari, Nasiri, Werner, Musavi, and
use Jimenez’s method (Jiménez, 1996) to deal with the uncertainty. Jolai (2019) design a model for a VRP and cross-dock scheduling
They consider total costs and patient satisfaction objectives and employ problem under uncertain travel times. To deal with uncertainty, they use
the epsilon constraint method to solve the bi-objective model for small a budget of uncertainty approach to bound the variations between the
instances. Moreover, they show that the Non-dominated Sorting Genetic uncertain and nominal values of parameters. Kisialiou, Gribkovskaia,
Algorithm (NSGA-II) can be adapted to solve the problem efficiently. and Laporte (2019) provide reliable supply vessel planning and sched­
Fathollahi-Fard, Ahmadi, Goodarzian, & Cheikhrouhou (2020) develop uling by minimizing costs. Furthermore, the authors consider the effect
a bi-objective robust optimization model for HHCRSP. The authors of uncertain demand on schedule performance and impose requirements
consider travel and service times as uncertain parameters and use a on the reliability of voyages through the construction of vessel schedules
robust approach underpinned by the Keshtel algorithm to cope with the using an ALNS metaheuristic. Barma, Dutta, and Mukherjee (2019)
uncertainty. This study minimizes two objectives, namely greenhouse study a multi-depot VRP with homogeneous vehicles to minimize the
gas emissions and total costs. They consider weights for the objective total routing distance in the network. They tackle this problem using the
functions based on their priorities and use the weighted method to solve discrete Ant Lion optimization algorithm and investigate a 2-opt local
the bi-objective model. Fathollahi-Fard, Ahmadi, and Karimi (2021) search algorithm. Weiszer, Burke, and Chen (2020) develop a multi-
propose a mixed-integer non-linear program for a multi-objective HHC objective shortest path algorithm for routing and scheduling an air­
problem. They consider minimizing the total network cost, minimizing port’s ground movements. This problem aims to obtain the optimal (or
the unemployment time for the worker, and minimizing the continuity near-optimal) routes for a fixed aircraft sequence. The study considers
of care by reducing the number of patients visited by the worker as fuel consumption and taxi time as objective functions. Wang, Liao, Li,
objective functions. They utilize the Red Deer Algorithm for solving the Yan, and Chen (2021) investigate a multi-objective model for a dynamic
multi-objective functions. Additionally, they use the Mulvey method to VRP with time windows. They develop a novel dynamic evolutionary
deal with uncertain travel and service time parameters based on sce­ algorithm using ensemble learning to solve the multi-objective model.
narios. Yang, Ni, and Yang (2021) present a multi-objective HHCRSP. The authors compare their approach with four algorithms in the litera­
They minimize costs and improve workload balance and service con­ ture and consider minimizing route distances and customer waiting
sistency. The authors consider travel and service times as uncertain. times as objective functions.
They develop a multi-objective artificial bee colony metaheuristic to Table 1 summarizes the studies in HHC, including the proposed
solve the model and use the Pareto dominance strategy to compare the approach, type of constraint, type of uncertainty, and whether or not the
solutions of multi-objective optimization. Zheng, Wang, Li, and Wu research tackles a real case study. In addition to the usual considerations
(2021) propose two-stage stochastic programming considering maxi­ in the problem, aspects such as mixed uncertainty, employment op­
mizing the expected income for HHCRSP under demand uncertainty. portunities, and economic development have not been considered by
The first stage is capacity planning and service authorization, and the previous studies. We aim to address these gaps by providing a compre­
second stage is resource allocation. They consider the minimization of hensive formulation of the problem. Moreover, the proposed robust-
the total operating cost for all customers, minimization of the value of fuzzy method effectively deals with fuzzy and scenario-based data,

5
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

which is the first attempt at such a problem. Moreover, the simultaneous • A nurse starts from a center to visit patients, and after serving all
consideration of multi-care and multi-depot factors is the first attempt in patients, the nurse finishes the route at the laboratory;
the literature. • The number of patients and locations of patients’ homes are pre­
defined (i.e., patients give all information to the HHC network before
3. Problem description getting the service);
• The required services for a patient should be provided by nurses with
We formally define the problem for the design and planning of HHC, consideration of the desired time windows for respective patients;
called Design and Planning of Home Health Care (DPHHC), and then • All patients need services in a certain time window;
present a multi-objective MILP model for the problem. In this problem, • The potential locations of center candidates and the location of the
nurses start their routes at centers, and visit patients in their homes. laboratories are predefined;
After the visits, the nurses finish their routes at a laboratory where they • There is one nurse at least, and some centers need more than one
report the status of patients to the laboratory. The patients must be nurse;
visited during a pre-specified time window based on their preferences. • No nurse arrives at patients’ homes before the opening time and after
More than one center (multi-depot) is typically opened in this problem, the closing time specified by the time windows of the patients;
and a multi-care option is also considered, where patients can request • The vehicles of nurses are heterogeneous and have different capac­
more than one service. The location of the laboratories and the potential ities, so different types of vehicles are taken into consideration. These
locations of centers are predefined. Based on this description, we pro­ vehicles may be the nurses’ own vehicles or a third-party logistic
pose a model with the following details: organization may be used to provide vehicles;
The proposed model is a two-stage model that includes strategic and • All patients must be visited at least once, and some patients need
tactical decisions. In the first stage, the strategic decision is determined multiple services during their time window;
that involves opening p locations among candidates n ∈ N as centers. • The service time, costs, and social parameters are considered as
The fixed cost for opening center n is equal to fn . In the second stage, the uncertain parameters;
tactical decisions are defined, such as transportation decisions, assigning • The patient’s demand and the capacity for drugs in each nurse’s
nurses to patients, and routing and scheduling of nurses. At first, the vehicle are known;
nurse v ∈ V moves from center n ∈ N to serve patient m ∈ M within the • There is no direct movement of nurses between the centers and
specified time window [am , bm ]. After visiting all patients assigned to laboratories;
nurse v ∈ V based on their scheduling, the nurse travels to laboratory h ∈ • The social impacts and efficiency of candidate locations are
H to transfer medical tests or treatments taken from patients. The travel considered.
time and travel cost from node i ∈ I to node j ∈ I are represented by trij
and coij , respectively. The number of services required by patient m ∈ M
is shown by βm , and each patient can receive multiple services (inter­ 3.1. Deterministic mathematical model
dependent services).1 Besides, patients demand a number of drugs in
each service (dm ), and the vehicles transporting nurses have limited We now provide a comprehensive mathematical programming
capacity (capv ). model of the DPHHC problem in the deterministic environment. We first
In the problem, the factors of the input set r ∈ R and output set g ∈ G define the sets, parameters, and decision variables used in the mathe­
are defined by experts as important criteria for selecting efficient cen­ matical formulation.
ters. These factors are used to measure the inefficiency of candidates
based on the DEA method. This method can handle such factors (traffic, 3.1.1. Notations
pollution, population density, etc.) on different scales. For candidate n, The notations related to the deterministic proposed model are
the quantities of input factor r ∈ R and output factor g ∈ G are given by described as follows:
knr and ong , respectively. The problem aims to optimize three objective Sets
functions simultaneously, which is typical in real settings. The first v∈V The set of all nurses
objective function concerns the total cost, including costs of opening m∈M The set of patients
centers and transportation. The second objective minimizes the in­ n∈N The set of center candidates
efficiency of centers, and the third objective considers social impacts to h∈H The set of laboratories
i, j ∈ I The set of all nodes (patients, center candidates, and
maximize the employment opportunities of nurses as human capital and laboratories,M ∪ N ∪ H)
economic development. g∈G The sets of input factors
A schematic example problem is presented in Fig. 1. It shows 10 r∈R The sets of output factors
center candidates, 11 patients, three nurses, and one laboratory. The Parameters
fn The fixed cost for opening center candidate n
first decision is to determine the selected centers; in this example, can­
dm The demand for drugs for patient m
didates 3 and 8 are chosen to be opened (strategic decision). The second capv The capacity of nurse’s vehicle v
set of decisions shows the path of each nurse to carry out their activities. tsiv The service time for node i visited by nurse v (where tsnv = tshv = 0 for
For example, in the dashed route, a nurse starts her/his activity at Center all h ∈ H and n ∈ N)
8 and visits Patients 4, 1, 5, 7, and 2. Patient 2 needs to receive two coij The travel costs for moving from node i to node j
trij The traveling time from node i to node j
services, and hence, nurses of Centers 3 and 8 visit Patient 2 during the
ong The quantity of input factor g for center candidate n
patient’s time window. Moreover, medical tests and treatment are knr The quantity of output factor r for center candidate n
conducted for patients and transported to the laboratory, which is the am The earliest time to visit patient m, i.e., the starting time of the time
endpoint of a route. For the purposes of this study, we make the window for patient m
following assumptions to clarify the scope of the study undertaken: bm The latest time to visit patient m, i.e., the ending time of the time
window for patient m
knn The number of employment opportunities at candidate location n
cen The employment rate, which is the number of nurses employed in
candidate location n in a year
jen The regional economic value at candidate location n
den The factor of regional development, i.e., a value of development for a
1 candidate location n that is between 0 and 1
For example, giving medication to a patient before/after a meal with a
(continued on next page)
predetermined time required between services.

6
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Fig. 1. An example shows the opening of centers, assigning nurses to patients, and nurses’ routes. Three centers are opened among 10 candidates, and a nurse travels
from each center to visit patients (highlighted routes). Some patients require two services (red circles).

(continued ) Constraint (1) is the sub-tour elimination constraint. Constraint (2)

Sets guarantees that patients requiring medical services are visited by nurses
βm The number of services required for patient m
within the patients’ time windows. In Constraint (3), nurses must leave
p The number of center candidates that are allowed to be opened the patient’s home after completing a service. Constraint (4) states that
M A large number each nurse starts a route from an opened center to travel to a patient’s
ω A small number home. Constraint (5) determines that all nurses finish their routes at the
Variables
laboratory after meeting all patients. Constraint (6) denotes that the
yn A binary variable, which is 1 if center candidate n is selected, and
0 otherwise. service for the next patient can commence after serving the previous
xijv A binary variable, which is 1 if nurse v travels from node i to node j, patient. Constraint (7) prohibits the violation of each patient’s time
and 0 otherwise. window and limits the start time of services for each patient, i.e., the
stiv A positive continuous variable, which is the starting time for visiting patients must be visited during their time window. Constraint (8) pre­
node i by nurse v
φn A positive continuous variable, which is the summation of the negative
vents exceeding the capacity of the vehicle-carrying nurses. Constraint
and positive weight deviation for candidate location n (9) chooses a certain number of potential location candidates that must
ari A continuous variable, which is used to eliminate sub-tours be opened. Constraint (10) requires that if a center is opened, a route can
wong A positive continuous variable, effectively the weight or importance of be started by a nurse from that center. Constraint (11) defines binary and
output factor g for candidate location n
positive decision variables.
winr A positive continuous variable, effectively the weight or importance of
input factor r for candidate location n
∑ ∑
(12)
′
on′ g wong − = n′
kn′ r winr + φn = 0 ∀n, n ∈ N, n ∕
g∈G r∈R

∑
We now discuss the constraints related to the DPHHC problem and detail ong wong + φn = yn ∀n ∈ N (13)
each of these one by one. g∈G

ari − arj + capv xijv ≤ capv − dj ∀v ∈ V, i, j ∈ I (1) ∑

knr winr = yn ∀n ∈ N (14)
∑∑ r∈R
xmjv = βm ∀m ∈ M (2)
j∈I v∈V
ωyn ≤ winr ∀n ∈ N, r ∈ R (15)
∑ ∑
ximv − xmjv = 0 ∀v ∈ V, m ∈ M (3) ωyn ≤ wong ∀n ∈ N, g ∈ G (16)
i∈I j∈I
Constraint (12) guarantees that the maximum efficiency score ob­
∑∑
xnmv = 1 ∀v ∈ V (4) tained for each center is achieved by adding the total deviations (in­
n∈N m∈M efficiency or φn ). Constraint (13) computes the total quantity of output
factors plus the inefficiency of each center if that center is opened.
∑ Constraint (14) indicates the total quantity of input factors for each
m+n+1
xihv = 1 ∀v ∈ V, h ∈ H (5)
i=n+1 center if that center is opened. Constraints (15) and (16) guarantee that
the weights assigned to the output and input factors of each center have
( )
stiv + tsiv + tij − M 1 − xijv ≤ stjv ∀i, j ∈ I, v ∈ V (6) at least a few values if that center is opened. There are three objective
functions considered in this study, which were briefly discussed earlier.
am ≤ stiv ≤ bm ∀i ∈ I, m ∈ M, v ∈ V, i = m (7) We provide the details of these functions in the following.
∑ ∑
dm xmjv ≤ capv ∀v ∈ V (8) 3.2. Objective function 1: Total Cost
m∈M j∈I

∑ The total cost is composed of two different costs. The first is the fixed
yn = p (9) costs for opening the centers, and the second is the traveling costs, which
n∈N
are the cumulative cost of the movements of nurses from one location to
∑∑ another. The total cost is:
xniv ≤ M yn ∀n ∈ N (10) ∑ ∑∑∑
i∈I v∈V
MinZ1 = fn yn + coij xijv (17)
n∈N i∈I j∈I v∈V
xijv , yj ∈ {0, 1}, stiv ≥ 0 ∀i, j ∈ N, v ∈ V (11)

7
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

where coij are calculated by c × dij in which c is the travel cost for
s.t : wog , wir ≥ ω ∀g ∈ G, r ∈ R (19)
traveling one kilometer and dij is the distance between two nodes i and j.
where the maximum score of efficiency for each DMU is equal to 1.
3.3. Objective function 2: Inefficiency Based on this optimization method (GP), Equation (18) can be rewritten
as follows to identify a set of common weights.
To provide accessible, high-quality, and equitable care, presenting
∑
N
( )
an appropriate method to evaluate the inefficiency of centers is critical Min φ−n + φ+
n (20)
to identify ideal design interventions that promote efficiency. For this n=1

purpose, the DEA is an appropriate approach that is widely used as a ∑G

non-parametric method for measuring the inefficiency of a set of Deci­ g=1 wog ogn
s.t : ∑R + φ−n − φ+
n = An ∀n ∈ N (21)
sion Making Units (DMUs) by a combination of qualitative factors (Cook r=1 wir krn
et al., 1990). There is no set standard for factors; therefore, the DEA
presents a relative inefficiency measure. Meanwhile, this approach has where φ−n and φ+ n are positive continuous variables and reflect the
two advantages over other techniques as it simultaneously uses multiple negative and positive deviations from the nth goal, respectively. In the
input and output factors to find which units (districts/center candidates) optimization model, An is the maximum level for the nth objective
are inefficient. Furthermore, the weights of the factors do not need to be function as a goal for the nth DMU. WhenAn = 1, it implies that the nth
determined, and being on different scales does not pose a problem. The objective function approaches its goal and variable φ+ n cannot have a
DEA model compares weighted input factors to weighted output factors, positive value. Therefore, Equation (21) can be rewritten as:
common to all DMUs. To do this, we consider the DMUs as the center ( )
candidates in the DPHHC, and the input and output factors for each ∑G ∑R ∑R
wog ogn + φ−n wir krn = wir krn × An ∀n ∈ N (22)
center are identified via an expert team’s knowledge. In the previous
g=1 r=1 r=1
studies, researchers applied efficiency to evaluate the design of a
network like a supply chain (Chen, Liang, & Yang, 2006; Liang, Yang, Due to the nonlinear Equations (20)–(22), the GP approach is utilized
Cook, & Zhu, 2006; Parmigiani, Klassen, & Russo, 2011). Omrani, again to linearize the model. In the revised model, An =1 is considered in
Adabi, and Adabi (2017) and Klimberg and Ratick (2008) incorporated Equation (21), and the efficiency score of the nth DMU is calculated as
∑G
the efficiency into location-allocation models by using the DEA model. wog ogn +φ+ n
∑g=1 . Therefore, for decreasing inefficiency, the numerator must
This study considers two input factors, traffic and pollution, and two R
r=1
wir krd − φn
−

output factors, population density and appropriate workplace. Based on increase and the denominator decrease. For this purpose, φ+
n is added to
the employees’ judgment in the municipality, these factors are consid­ the numerator, and φ−n is subtracted from the denominator. Conse­
ered important in selecting the best candidates for centers and are quently, Equation (18) can be reformulated as:
therefore considered here:
∑
N
( )
Min φ−n + φ+ (23)
1) Traffic: This factor indicates the ease of nurses’ access to each center n=1
n

before starting their routes. The lower the congestion in a district, the
∑G
lower the energy and time required to arrive at each center, g=1 wog ogn + φ+
(24)
n
reflecting the advantage and superiority of the center. Therefore, s.t : ∑R
− φ−n
=1 ∀n ∈ N
r=1 wir krd
traffic is considered as an input factor measured based on the average
time a nurse takes to arrive at a center. The objective function can be further simplified as follows, where φn
2) Pollution: Districts with a lower level of air or noise pollution are is replaced with (φ−n + φ+
n ) and, as a result, Equations (23) and (24) can
more desirable for the nurses in terms of health and safety. Thus, be rewritten as follows:
pollution is also considered as an input factor for centers.
∑
N
3) Population density: The number of people that live in each district Min Z2 = φn (25)
is the population density of that district. The greater the population n=1

density, the greater the demand for services; hence, the greater the
significance of centers in those districts. Therefore, population den­ ∑ ∑
G R
s.t : wog ogn − wir krn + φn = 0 ∀n ∈ N (26)
sity is used as an output factor for centers. g=1 r=1
4) Appropriate workplace: Acceptable temperature, humidity levels,
natural light, etc., are needed for nurses to work effectively. Hence, where φn is a positive continuous variable. Let (wo*g , wi*r , φ*n ) be the
an appropriate workplace is used as an output factor for centers.
optimal solution, the efficiency score of the nth DMU is obtained from:
∑G
Herein, to calculate the inefficiency of the center candidates, a *
g=1 wog ogn φ*
common weight of DEA is chosen that leads to a Goal Programming (GP) θ*n = ∑R *
= 1 − ∑R n * ∀n ∈ N (27)
wi k
r rn r=1 wir krn
model. GP is a branch of multi-objective optimization, which in turn is a
r=1

branch of multi-criteria decision analysis (Jones & Tamiz, 2016). Finally, Equation (25) is the second function of the proposed model
Essentially, the idea is to minimize the deviation of common weights that is the deviation of common weights from the value calculated by the
from the obtained values of the DEA model. The DEA model is specified basic DEA model for centers.
as follows:
Given a set of N DMUs (n = 1,⋯,N), R input factors (r = 1,...,R), and 3.4. Objective function 3: Social impacts
G output factors (g = 1, ..., G), Equations (18)-(19) denote a Multi-
Objective Fractional Programming (MOFP) model, with the aim of The third objective is a “social” objective, which measures the social
maximizing the efficiency scores for DMUs at the same time: impacts/CSR of centers, and can be formulated as:
{∑G ∑G ∑G } ∑ ∑
Maxw = ∑R
g=1 wog og1 g=1 wog og2
, ∑R
g=1 wog ogn
, ⋯, ∑R (18) MaxZ3 = w1 ( knn cen yn ) + w2 ( jen den yn ) (28)
r=1 wir kr1 r=1 wir kr2 r=1 wir krn
n∈N n∈N

where w1 and w2 are the weights of the employment opportunity and

8
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

economic development, respectively. These weights are adjusted ac­ ( ) ( )

(α − λ) ( s1 ) (1 − α) s2
cording to their importance, where w1 +w2 = 1. In Equation (28), the stsiv + tij + tsiv + (tsiv ) − M 1 − xsijv ≤ stsjv ∀i, j
first part specifies the employment opportunities gained from opening (1 − λ) (1 − λ)
centers in line with the employment rate. In other words, knn and cen are ∈ I, v ∈ V, s ∈ S
defined as the number of employment opportunities and the employ­ (29)
ment rate (which is the number of nurses employed in the candidate
location in a year) at candidate location n. By multiplying knn in cen , the where α and λ reflect a minimum confidence level and an optimi­
employment opportunities gained from opening centers are in line with stic–pessimistic parameter, respectively, that are related to the PCCP
the employment rate (Sharifi et al., 2020). The second part of the approach (with respect to Phase 2 in Section 4.2). Constraint (29) states
equation calculates economic development, which aims to improve that service for the next patient can commence after serving the previous
districts’ economic, fiscal, and social conditions. It is computed as the patient, considering uncertain scenarios related to the patient’s service
product of the two regional development and economic values of each time. Other constraints under the uncertain environment are presented
district. Regional development is a broad term but can be seen as a way in Appendix A.
to improve living standards and enhance well-being in the district, with
the regional economic value being a measure of the benefits of a service 3.5.1. Objective function 1: Total Cost in uncertain environment
to a district. This objective function minimizes the total cost, including set-up and
transportation costs, under fuzzy and scenario-based parameters. The
3.5. Mixed Robust-Fuzzy model first item of the cost objective function is the fixed costs for the opening
centers, and the second item is the travel costs to move the nurses from
The model presented in Section 3.1 is formulated in the deterministic one node to another under scenario s. Thus, the total cost is equal to:
environment. Hence, in this section, we modify the model to develop a ∑((1 − λ) ( ) λ( )
) ∑∑∑((1 − λ) (
hybrid robust-fuzzy formulation that incorporates uncertainty by Z1s =
2
fn1 + fn2 + fn3 + fn4 yn +
2 2
co1ij
considering the scenarios of service time and several fuzzy parameters. n∈N i∈I j∈I v∈V
) λ( ))
In particular, the costs, number of employment opportunities, employ­ + co2ij + co3ij + co4ij xsijv
ment rate, and regional economic and regional development values are 2
imprecise and uncertain, estimated by municipal employees using (30)
trapezoidal fuzzy numbers. In the following, we provide details of the
sets, parameters, and decision variables, then update the constraints and where f̃n and co
̃ ij are the trapezoidal fuzzy numbers. As can be seen, the
revised objective functions. Note that Objective 2 (associated with in­ cost objective function is based on scenarios. Therefore, we use a
efficiency) does not depend on uncertainty and is exactly the same as p-robust approach to calculate the expected value of Z1s . According to
that presented in the previous section. In the following, the set of sce­ Phase 1 in Section 4.1, Equation (31) and Constraint (32) are proposed.
narios is denoted by s ∈ S. In real settings, the service time for a patient is ∑
not deterministic because a patient’s status is changeable; therefore, it is MinE[Z s ] = prs Z1s (31)
s s∈S
uncertain data and shown by t̃smv under scenario s. The parameters and
decision variables used in the robust-fuzzy model are as follows: s.t : Z1s ≤ Z1s* (1 + ρ) ∀s ∈ S (32)
Parameters

f̃n The fixed cost for opening center n obtained as a trapezoidal fuzzy where ρ is defined as the desired level of robustness and Zs* 1 is the
number (̃f = f 1 , f 2 , f 3 , f 4 )
n n n n n
optimal value of the first objective function under scenario s.
s
t̃smv The service time for patient m visited by nurse v with a trapezoidal
s
fuzzy number under scenario s (t̃smv = tss1
mv , tsmv , tsmv , tsmv )
s2 s3 s4
3.5.2. Objective function 3: Social impacts in uncertain environment
co
̃ ij The traveling cost from node i to node j with a trapezoidal fuzzy The third objective associated with the social impacts of centers in­
̃ ij = co1ij , co2ij , co3ij , co4ij )
number (co
cludes the employment opportunities and economic development and is
kn The number of employment opportunities at center n with a
measured within the fuzzy setting as follows:
̃n
trapezoidal fuzzy number (kn ̃n = kn1 , kn2 , kn3 , kn4 )
n n n n
ce
̃n The employment rate at center n with a trapezoidal fuzzy number ∑((1 − λ) ( ) λ( )
)(
(1 − λ) ( 1
̃ n = ce1n , ce2n , ce3n , ce4n )
(ce Z3 = w1 kn1n + kn2n + kn3n + kn4n cen
2 2 2
The regional economic value at center n with a trapezoidal fuzzy
n∈N
je
̃
n ) (
̃ = je1 , je2 , je3 , je4 ) ) λ( ) ∑ (1 − λ) ( ) λ(
number (je n n n n n + ce2n + ce3n + ce4n yn + w2 je1n + je2n + je3n
de
̃n The regional development factor at center n with a trapezoidal fuzzy 2 n∈N
2 2
number (dẽ n = de1 , de2 , de3 , de4 ) )( )
n n n n ) (1 − λ) ( 1 ) λ
prs The probability of scenario s + je4n den + de2n + (de3n + de4n ) yn (33)
ρ The level of desired robustness (ρ ≥ 0) 2 2
Variables
arsi A continuous variable used to eliminate sub-tours where w1 and w2 are assigned to parts 1 and 2 and reflect the importance
stsmv A positive continuous variable, the starting time of a visit for patient m of each part. Equation (33) incorporates uncertain parameters, where
by nurse v under scenario s
xsijv A binary variable, 1 if the nurse v moves from node i to node j under
kn
̃n , ce
̃n , je
̃ and de
̃n are the fuzzy parameters. This objective function
n
scenario s, and otherwise 0. consists of two parts, employment opportunities and economic
development.

4. Solution methodology

To deal with the complexity of the above model with uncertain pa­
rameters, a novel solution technique is proposed. As shown previously,
In the uncertain version of the model, Equations (1)-(8) and (10)-(11) the proposed model aims to optimize three different objectives: mini­
from the deterministic model should be updated using uncertain pa­ mizing the total cost of the network, minimizing inefficiency, and
rameters and variables. In particular, Equation (6) is rewritten incor­ maximizing the employment opportunities and economic development
porating uncertainty as follows: arising from opening centers. The inherent uncertainty of the model

9
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Fig. 2. Flowchart of the solution methodology including three phases: in Phase 1) to deal with the scenarios of service time, we apply the p-robust approach, in Phase
2) PCCP is utilized to cope with fuzzy parameters that are the cost and social data. Phases 1 and 2 together represent a hybrid robust-fuzzy approach to deal with
mixed uncertainty, and in Phase 3) the TH approach is presented to solve the multi-objective model.

derives from the service times, costs (including the fixed cost of the considered in our model simultaneously, incorporating the location-
opening centers and transportation cost), the number of employment allocation and transportation decisions. The compact form of the
opportunities for each district, the employment rate for nurses, the p-robust programming model is as follows:
regional economic value, and regional development value. All these ∑
aspects further complicate the problem, and thus the resulting model. Min E[Z s ] = Ay + ps Bxs (34)
s∈S
The solution approach for the DPHHC splits the solution process into
three phases, as depicted in Fig. 2. Phase 1 presents a p-robust model to Z s (y, xs ) − Z s*
deal with imprecise parameters by using a scenario-based approach. In s.t: ≤ρ ∀s ∈ S (35)
Z s*
Phase 2, the PCCP is applied to tackle the possibilistic data and possi­
bilistic chance constraint. In Phase 3, the Torabi and Hassini (TH) where A and B reflect the fixed and transportation costs, respectively.
approach is used, specifically designed for solving a multi-objective And y is a binary variable corresponding to strategic decisions to be
model. made, and xs is a vector of assignment variable (positive continuous
variable) under scenario s that is related to tactical decisions. Index s
4.1. Phase 1: The P-robust model illustrates the scenario associated with the service times of patients, and
ps is the probability of scenario s. The function E[Zs ] is the expected value
Due to the uncertainty of the service time parameter, the nurses of the objective function under scenario s, and Zs* is the optimal value
report data as scenarios based on their knowledge and the patient status. resulting from solving the deterministic model under each scenario s.
s s
Zs*
In this regard, the p-robust approach is defined, which is utilized to Note that Z (y,xZs*)− is the maximum regret under each scenario.
model the DPHHC problem under uncertain scenarios. The robust
approach tries to minimize the maximum regret or the expected value of
cost, in which the obtained solution is optimal for realizing uncertainties 4.2. Phase 2: Possibilistic chance constrained programming
in predefined sets. Therefore, we aim to build a reliable design and
planning for the HHC under different scenarios. Let us consider a set of As mentioned previously, costs and social parameters are fuzzy, and
scenarios s ∈ S in which no disruption is seen (Snyder & Daskin, 2006). hence the PCCP approach is used to deal with fuzzy data. This approach
In this regard, the p-robust model keeps the value of the objective is considered as one of the well-known fuzzy programming methods to
function within 100 % and 100 +ρ% of the optimal objective function tackle uncertain models (Pishvaee, Torabi, & Razmi, 2012), particularly
under each scenario, whereρ ≥ 0 (the desired level of robustness). As possibilistic data and possibilistic chance constraints. In this context, a
mentioned previously, two strategic and tactical decisions are minimum confidence level can be obtained for decision-makers to meet

10
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

the chance constraints. Here, two measures and standards can be

x ≤ My
adopted, namely possibility Pos and necessity Nec, and we now provide
the details of its implementation.
where Z1 , Z2 , and Z3 are cost, inefficiency, and social impact functions,
Let ψ
̃ be a trapezoidal fuzzy number that is ψ
̃ = (ψ 1 , ψ 2 , ψ 3 , ψ 4 ), where
respectively. And A,
̃ B,
̃ C̃ and K
̃ are inherently uncertain parameters that
ψ 1 (most pessimistic) < ψ 2 (pessimistic) < ψ 3 (optimistic) < ψ 4 (most
reflect the fixed costs of centers, the variable costs of the network, social
optimistic) and the membership function is as follows:
parameters, and service times for the patients, respectively. Vector y
⎧ x − ψ1
⎪ ψ1 ≤ x ≤ ψ2 represents the binary decision variable, and x is the positive continuous
⎪
⎪ ψ2 − ψ1
⎪
⎪
⎪ variable. Additionally, D, G, F, H, and I are the coefficient matrices, and
⎪
⎨1 ψ2 ≤ x ≤ ψ3 M is a large constant (“Big M”), typically used in integer programming
μ(x) = (36) formulations. The uncertain parameters in the objective functions were
⎪
⎪ ψ 4 − x
⎪
⎪ ψ3 ≤ x ≤ ψ4 replaced by their corresponding expected values E[Z], while the measure
⎪
⎪ ψ − ψ3
⎪
⎩ 4 Me was employed in the possibilistic chance constraints. Also, certain
0 otherwise.
scenarios for service time are generated using a nurse’s preference.
The possibilistic values of the possibility Pos, necessity Nec, and the Therefore, according to Equation (38), model (40) can be reformulated
expected value of the fuzzy number ψ ̃ are calculated based on Inuiguchi with model (41) as follows:
and Ramı́k (2000) and Liu and Iwamura (1998). In the following, to ( )
(1 − λ) λ ∑ ((1 − λ)
fluctuate between these two optimistic and pessimistic extremes, a fuzzy Min E[Z1 ] = (A1 + A2 ) + (A3 + A4 ) y + ps (B1
compensatory measure, namely the Me measure, was proposed by Xu 2 2 s∈S
2
)
and Zhou (2013). The Me measure enables decision-makers to impose a λ
+ B2 ) + (B3 + B4 ) xs
level of optimism and pessimism regarding their preference by choosing 2
any convex combination between the Pos and Nec measures: (41)
Me{e} = λ Pos{e} + (1 − λ) Nec{e} (37)
MinZ2 = y
where e denotes an event, and λ defines an optimistic–pessimistic
(1 − λ) λ
parameter that can be changed to fall within the interval [0, 1] based on MaxE[Z3 ] = ( (C1 + C2 ) + (C3 + C4 ))y
2 2
the decision-makers’ preference. In this respect, when the decision-
makers’ preferences are extremely pessimistic (λ = 0), the measure Nec s.t : Dy = G
is yield, which shows the possible level of minimum occurrence for the
possibilistic event e. Alternatively, when the decision-makers’ prefer­ Fy ≤ 1
ence is optimistic (λ = 1), the measure Pos is yield, which shows the
maximum occurrence possibility level for the possibilistic event e. (α − λ)K1s + (1 − α)K2s
xs ≤
Consequently, according to Xu and Zhou (2013), the measure Me as well 1− λ
as the expected value of ψ
̃ can be obtained consideringα ≥ 0.5 andψ ≥ 0.
As a result, whenλ < 0.5, it needs to be close to Nec to back a decision- Hxs ≤ I
maker’s pessimistic attitude. The following constraints show the re-
xs ≤ My
written formulation of the Me measure:
x − ψ3
ψ ≤ x} ≥ α ⇔ λ + (1 − λ) ×
Me{̃ ≥α⇔x 4.3. Phase 3: Torabi and Hassini approach
ψ4 − ψ3
(α − λ)ψ 4 + (1 − α)ψ 3
≥ (38) A major component of this research is a development of an optimi­
1− λ
zation model that integrates different decisions and objectives (cost,
ψ2 − x inefficiency, and social impact) in HHC. To solve the ensuing MOLP,
ψ ≥ x} ≥ α ⇔ λ + (1 − λ) ×
Me{̃ ≥ α⇔ x
ψ2 − ψ1 there are many approaches available in the literature, with several fuzzy
(α − λ)ψ 1 + (1 − α)ψ 2
≤ (39)
1− λ

where α is a minimum confidence level for decision-makers to meet the

chance constraints. In the following, the compact form of the PCCP
model can be represented as:
[ ] [ ]
̃ y+E B
MinE[Z1 ] = E A ̃ x (40)

MinZ2 = y
[ ]
̃ y
MaxE[Z3 ] = E C

s.t : Dy = G

Fy ≤ 1
{ }
Me K̃ ≥x ≥α

Hx ≤ I Fig. 3. A virtual map of Kermanshah in Iran. The map shows the locations of
the laboratory, patients’ homes (rectangles), and potential locations of cen­
ters (circles).

11
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185
Table 2
The results of the case study using the parameter settingsρ = 0.5, α= 0.5, λ= 0.2, andγ = 0.5.
Combination Cost Inefficiency Social Opened Objective Function Routes
Impact Centers
Cost (E Inefficiency Social Scenario 1
+ 7) Impact

1 ✓ 459 2.19 3 17.28 N1 : 9,14,25,13,15,26,29,31N2 : 9,18,15,20,31N3 : 5,20,12,16,30,27,24,11,31N4 :

4,22,17,23,19,12,21,31N5 : 4,28,11,24,19,31
2 ✓ 1710 2.92 0 18.91 N1 : 1,15,25,31N2 : 1,30,27,31N3 : 1,13,18,31N4 : 7,12,17,23,21,31N5 :
10,21,19,27,16,20,24,18,14,12,25,28,29,22,11,26,13,31
3 ✓ 135 2.88 3 20.72 N1 : 3,20,15,28,27,13,22,31N2 : 3,16,11,17,25,31N3 : 1,22,17,26,18,31N4 :
5,29,19,30,23,12,15,24,31N5 : 5,25,14,12,21,26,31
4 ✓ ✓ ✓ 159 2.41 1.41 19.45 N1 : 5,11,24,31N2 : 5,20,12,14,25,17,22,26,29,31N3 : 1,19,30,15,13,18,31N4 :
9,23,16,12,27,24,21,31N5 : 5,28,11,16,31
5 ✓ ✓ 145 2.43 1.28 19.15 N1 : 5,28,11,23,31N2 : 4,20,12,27,24,21,31N3 : 5,19,20,21,31N4 :
1,19,30,15,13,18,31N5 : 1,23,16,12,14,25,17,22,26,29,31
6 ✓ ✓ 159 2.41 1.41 19.45 N1 : 5,17,31N2 : 5,20,12,19,30,15,13,18,31N3 : 1,23,16,12,27,24,21,31N4 :
9,14,25,17,22,26,29,30,31N5 : 5,28,11,18,25,29,31
7 ✓ ✓ 1510 2.63 0.46 20.06 N1 : 10,14,25,17,22,26,29,31N2 : 5,20,12,19,30,15,13,18,31N3 : 5,28,11,14,31N4 :
1,23,16,12,27,24,21,31N5 : 5,23,11,31

methods being widely adopted. These methods enable decision-makers ⎧

to effectively trade off their preferences concerning multiple objec­ ⎪
⎪
⎪ 1 Z2 ≤ Z PIS
2
⎪
⎪
tives (Torabi & Hassini, 2008). Herein, an efficient fuzzy technique is ⎨ Z NIS − Z
(47)
2
used to solve the proposed MOLP model. This interactive fuzzy tech­ μ2 (x) = 2
Z PIS NIS
2 ≤ Z2 ≤ Z 2
⎪
⎪
⎪ Z NIS
2 − Z2
PIS

nique introduced by Torabi and Hassini (2008), the so-called Torabi and ⎪
⎪
⎩ 0 Z2 ≥ Z NIS
Hassini (TH) approach, is capable of obtaining an efficient solution and 2

works as follows: ⎧
Let’s consider the following compact MOLP form for the proposed ⎪
⎪
⎪ 1 Z3 ≥ Z PIS
3
⎪
⎪
mathematical model: ⎨ Z − Z NIS
(48)
3
μ3 (x) = 3
Z NIS
3 ≤ Z3 ≤ Z PIS
3
MinZ1 (Cost) (42) ⎪ Z PIS
⎪ 3 − Z3
NIS
⎪
⎪
⎪
⎩ 0 Z3 ≤ Z NIS
3
MinZ2 (Inefficiency)
Step 3: According to Equations (49) and (50), a model with a single
Max Z3 (CSR) objective can be written with respect to the TH aggregation function. By
changing the value of the compensation coefficient (γ), Equations (49)
s.t : A x ≤ B and (50) result in a trade-off between the minimum satisfaction level of
each objective and their relative importance within the feasible district.
x∈X Additionally, γ controls the minimum satisfaction level of each objective
as well as the trade-off between the objectives. The following formula­
where A and B are the coefficient matrices.
tion determines the compensation coefficient for each objective function
Step 1: For each objective function h (h={1,2,3}), Positive Ideal
by γ, which helps balance2 the optimized solution. The first term in
Solutions (PIS) and Negative Ideal Solutions (NIS) need to be calculated.
Equation (49) is (γ × Z0 ), which calculates the minimum satisfaction
If the PIS of the proposed MOLP is available (ZPIS PIS
h , xh ), the NIS can be level of objective function h (Z0 = min{μh (x)}). In the second term, the
obtained from Equations (43)-(45). aggregation of μh (x) is weighted by θh , based on the decision-maker’s
{ ( ) (
Z1NIS = Max Z1 x2PIS , Z1 x3PIS
)}
(43) preference.
∑
{ ( ) ( )} Max Z(x) = γ × Z0 + (1 − γ) θh × μh (x) (49)
Z2NIS = Max Z2 x1PIS , Z2 x3PIS (44) h

{ ( ) ( )}
Z3NIS = Min Z3 xPIS
1 , Z3 xPIS
2 (45) s.t: Z0 ≤ μh (x) ∀h = 1, 2, 3 (50)

Step 2: For each objective function h (h={1,2,3}), the satisfaction where θh and γ represent the relative weights (priorities) of the satis­
level (normalization) should be calculated under solution vector x, faction level for each objective function and compensation coefficient,
represented by μh (x). Following this, the linear membership functions respectively. To have the same positive scale, these weights are deter­
are: mined between 0 and 1 according to the decision-makers’ preferences,
∑
⎧ where h θh = 1. Moreover, the satisfaction level of objective function h
⎪
⎪ 1 Z1 ≤ Z PIS
⎪
⎪
⎪
1
for vector x is denoted by μh (x). Moreover, the lower bound of μh (x) is
⎨ Z NIS − Z
defined by Constraint (50).
(46)
1
μ1 (x) = 1
Z 1 ≤ Z1 ≤ Z NIS
PIS
Step 4: Finally, to generate an efficient solution for the proposed
NIS PIS 1
⎪
⎪ Z − Z
⎪
⎪ 1 1
⎪
⎩ 0 Z ≥ Z NIS MOLP model, the values of parameters should be set based on the de­
1
cision-makers’ preference.
1

2
Balance is defined achieving similar satisfaction levels among the mem­
bership functions μ based on changes in γ (Torabi & Hassini, 2008).

12
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Routes Time
(s)
Scenario 2 Scenario 3

N1 : 4,28,20,12,26,22,17,25,13,15,16,29,31N2 : 9,14,12,21,31N3 : N1 : 5,20,12,27,24,21,31N2 : 4,28,18,20,31N3 : 5,19,12,14,25,17, 13.7

5,19,23,30,27,24,18,31N4 : 5,11,24,19,31N5 : 5,11,20,15,31 23,13,15,30,16,29,31N4 : 5,11,15,19,31N5 : 9,22,26,24,31
N1 : 7,12,30,31N2 : 1,27,21,18,31N3 : 7,29,20,19,24,28,22,11,12,27,31N4 : N1 : 10,29,26,13,19,31N2 : 1,12,17,20,13,28,21,31N3 : 7,16,24,30,18,27,31N4 : 13.8
10,18,15,17,25,13,16,13,14,26,31N5 : 1,23,21,25,31 7,12,11,22,25,14, 15,18,25,31N5 : 7,23,27,21,31
N1 : 5,27,23,16,13,15,12,30,25,24,15,26,28,29,21,31N2 : 3,11,17,12,14,31N3 : N1 : 1,13,27,16,22,23,30,12,28,24,15,26,19,20,29,14,11,31N2 : 3,15,31N3 : 13.9
1,18,26,31N4 : 1,22,20,25,31N5 : 1,19,17,22,31 5,18,25,26,31N4 : 3,12,17,31N5 : 1,21,17,22,25,31
N1 : 5,11,16,31N2 : 9,23,16,12,14,25,17,22,26,29,31N3 : 1,19,30,15,13,18,31N4 : N1 : 5,28,11,24,31N2 : 5,11,16,31N3 : 5,20,12,27,24,21,31N4 : 82.3
5,28,11,24,31N5 : 5,20,12,27,24,21,31 9,23,16,12,14,25,17,22,26,29,31N5 : 1,19,30,15,13,18,31
N1 : 5,28,20,19,31N2 : 1,19,30,15,13,18,31N3 : 4,23,16,12,14,25,17,22,26,29,31N4 : N1 : 5,11,31N2 : 5,28,19,31N3 : 1,23,16,12,27,24,21,20,31N4 : 1,19,30,15,13,18,31N5 : 39.5
5,11,21,23,31N5 : 5,20,12,27,24,21,31 4,20,12,14,25,17,22,26,29,21,23,31
N1 : 5,11,18,30,31N2 : 5,20,12,27,24,21,31N3 : N1 : 5,28,11,31N2 : 1,23,16,12,19,30,15,13,18,31N3 : 5,20,12,27,24,21,31N4 : 47.7
1,23,16,12,19,30,15,13,18,29,25,31N4 : 9,14,25,17,22,26,29,31N5 : 5,28,17,31 5,17,18,25,29,31N5 : 9,14,25,17,22,26,29,30,31
N1 : 5,11,23,31N2 : 1,23,16,12,27,24,21,31N3 : 5,28,11,31N4 : N1 : 5,11,14,31N2 : 5,20,12,27,24,21,31N3 : 1,23,16,12,19,30,15,13,18,31N4 : 55.7
10,14,25,17,22,26,29,31N5 : 5,20,12,19,30,15,13,18,14,31 10,14,25,17,22,26,29,31N5 : 5,28,11,23,31

5. Experimental evaluation optimize each of the three objectives, one at a time. Fig. 4a shows that
Districts 4, 5, and 9 are selected as the least expensive Centers to open in
The experimental evaluation in this study consists of two parts. First, Kermanshah city, whereas, if we consider Inefficiency or Social Impact,
we present a case study motivated by the real data in Kermanshah city in Centers 4 and 9 are excluded (Fig. 4b and c). In particular, Centers 1, 7,
Iran. The aim here is to demonstrate the validity of our approach in a and 10 are ideal considering the Inefficiency measures (Fig. 4b), as they
real-world setting. Second, we conduct a numerical study by varying have low traffic, low pollution, a high population density, and an
different aspects of the problem, including sensitivity analysis with the appropriate workplace. According to Fig. 4 and Table 2, Centers 7 and
parameter λ within the possibilistic chance constraints and a sensitivity 10 are not inexpensive or ideal social locations; hence they are not
analysis on γ in the TH approach. This second part aims to explore chosen to be opened. Considering Social Impact, the model selects Cen­
different problem characteristics that we may encounter in reality, and ters 1, 3, and 5 to be opened (Fig. 4c), as these Centers (especially Center
to measure the performance of our approach in these different contexts. 3) have excellent employment opportunities and economic develop­
Our proposed mathematical formulation is implemented within GAMS ment. We see that Center 1 is ideal considering Inefficiency and Social
29.1.0 using the CPLEX solver (https://www.ibm.com/au-en/analyt Impact, while Center 5 has low Cost and excellent Social Impact.
ics/cplex-optimizer). All computational experiments were carried out Combination 4: Equal weighting of all objectives. To reiterate,
on a PC with Intel Core i7, CPU 2.67 GHz, and RAM 16 GB. the overall goal of the model is to find the best-fit locations for opening
Centers so as to minimize the total Cost and Inefficiency while maxi­
mizing Social Impact. For this purpose, according to the steps of the TH
5.1. Case study
method outlined in Section 4.3, the model is solved considering all three
objectives with equal weights (θ1 =θ2 = θ3 ) of their satisfaction levels.
We investigate a real case study of Kermanshah in Iran within our
The results of solving this model are shown in Fig. 5a. We see that among
proposed framework. The city of Kermanshah is divided into 10 main
the 10 candidate districts in Kermanshah, three Centers, namely 1, 5,
districts, an overview of which can be seen in Fig. 3. Ten locations –
and 9, are chosen to be opened, which simultaneously leads to a mini­
Districts 1 to 10 – have been chosen as potential locations for opening
mum total Cost, minimum Inefficiency, and maximum employment op­
centers by the municipality of Kermanshah. The data for the case study
portunities and economic development. This analysis combines the first
(in a de-identified form) was provided by Taleghani hospital and the
three Combinations (Combination 1 is Cost, Combination 2 is In­
municipality of Kermanshah. There are 20 patients, 5 nurses, and 10
efficiency, and Combination 3 is Social Impact), where we observed that
potential center candidates. Among the centers, 3 are allowed to open.
Centers 1, 5, and 9 are ideal for opening up, and Center 9 is the best of
Furthermore, there are 2 input factors (traffic and pollution), 2 output
the remaining options.
factors (population density and appropriate workplace), 3 scenarios,
Combination 5: Equal weighting of Cost and Inefficiency mea­
and a single laboratory. These three scenarios, each of which corre­
sures. In this setting, we optimize the Cost and Inefficiency measures and
sponds to a different service duration, are: pessimistic, most likely, and
ignore Social Impact. This leads to three Centers being opened up,
optimistic. These scenarios are set by a team including four nurses and
namely 1, 4, and 5 as shown in Fig. 5b. In Table 2, the value of Social
one physician in the Taleghani hospital. This team decides based on
Impact in Combination 5 is lowest compared to its values in Combina­
patients’ health conditions and medical history.
tions 4 (equal weighting of all objectives), 6 (equal weighting of Cost and
In the following, we present the results of applying our proposed
Social Impact), and 7 (equal weighting of Inefficiency and Social Impact),
model to the case study. Then, we carry out a sensitivity analysis of key
since we explicitly avoid optimizing this objective. The first (just Cost)
parameters of interest. This analysis is particularly useful in providing
and second (just Inefficiency) Combinations are combined in this section,
valuable insights for health managers so that they can make informed
which shows among all the opened Centers in those Combinations (Cost:
decisions on the best location/s to open centers.
4, 5, 9 and Inefficiency: 1, 7, 10), that Centers 1, 4, and 5 are selected with
equal weight importance (0.5) considering Cost and Inefficiency. In other
5.1.1. Effect of different objective weights
words, Centers 4 and 5 have low Cost and reasonable levels of In­
In this section, we solve our model to identify which locations are
efficiency, while Center 1 has a low level of Inefficiency and reasonable
ideal for opening up new centers in the case study. We analyze the
Cost.
objective functions, total Cost, Inefficiency, and Social Impact, in various
Combination 6: Equal weighting of Cost and Social Impacts. We
Combinations that are shown in Table 2. In addition, the route of each
now focus on total Cost and Social Impact while ignoring Inefficiency.
nurse (Nv ) in each scenario is reported.
Three Centers 1, 5, and 9 are opened, as shown in Fig. 5c. These Centers
Combinations 1–3: Optimizing single objectives. To identify
are also selected in Combination 4 (equal weighting of all objectives).
which Centers are best considering one objective, we run the model and

13
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Fig. 4. Best fit locations for the centers considering only one objective function. Based on Cost: 4, 5 and 9; Inefficiency: 1, 7 and 10; Social Impacts: 1, 3 and 5
candidates are opened.

Centers 1, 5, and 9 have relatively good levels of Cost and Social Impact Inefficiency and Social Impact and decided that their relative importance
(Combination 6) simultaneously. Moreover, if we consider Inefficiency in can be split as 0.5 for Cost, 0.4 for Inefficiency, and 0.1 for Social Impact.
addition to Cost and Social Impact, the same Centers are chosen for the Additionally, we set the value of the compensation coefficient (γ) for the
same reasons previously discussed. Centers 1, 5, and 9 are in districts TH function to 0.5, which is a mid-point leading to a solution that is
with low Cost of living, and Center 1 has a high Social Impact. neither balanced nor unbalanced. We note that the p-value (ρ) of the Cost
Combination 7: Equal weighting of Social Impacts and In­ objective is set to 0.5, which allows the Cost to be up to 50 percent of the
efficiency measures. Here, we investigate Social Impact and Inefficiency best-known Cost in each scenario, while the solution remains robust. The
measures with equal weightings. This leads to selecting three Centers 1, decision-makers assume that the optimistic–pessimistic parameter (λ) is
5, and 10, as seen in Fig. 5d. This analysis combines Combinations 2 equal to 0.2. When λ is<0.5, this shows that the attitude of the decision-
(Inefficiency) and 3 (Social Impact), where Centers 1, 7, 10 and 1, 3, 5 makers is pessimistic, and their goal is to evaluate the model under
were opened for Inefficiency and Social Responsibility, respectively. pessimistic conditions. In the following, an analysis of λ is presented in
Centers 1 and 10 are very efficient, and Center 5 has a high Social Impact; Table 3, where α is set to 0.5.
hence these are chosen to be opened. As can be seen, the first candidate First, we solve the case study with different values of λ, while the
(Moallem) is selected in Combinations (1–7), which is because Moallem other parameters are kept constant (ρ = 0.5,α = 0.5, γ= 0.5, θ1 = 0.5,θ2
is located in the northwest of Kermanshah with the lowest levels of = 0.4, andθ3 = 0.1). The results are shown in Table 3, where the first
traffic. Moreover, its population density, economic development, and column shows the values of λ, the next four columns show the overall
employment opportunities are high relative to other candidates. (TH) Cost, Inefficiency, and Social Impacts objective functions, respec­
This analysis shows that three objective functions can be considered tively. The final column shows the time required to solve the model. We
in different Combinations, which allow decision-makers to decide on the see that with increasing values of λ, Cost and Social Impact increase while
best candidates for Centers. For example, if the first and second objective Inefficiency is not affected (not surprising since uncertainty does not
functions are considered based on the decision-makers’ preferences and directly affect this objective). When a decision-maker’s preferences are
the third objective function is ignored, according to Combination 5 (Cost pessimistic (λ close to 0), we see a minimum “possibility” level con­
and Inefficiency), candidate locations 1, 4, and 5 are selected as Centers. cerning a possibilistic event, leading to the values of Cost and Social
Impact being at the minimum level. Alternatively, from an optimistic
5.1.2. A sensitivity analysis of the case study viewpoint (λ close to 1), we see the opposite, where Cost and Social
We further analyze the case study to understand the behavior of the Impact increase.
model and approach. The relative importance weighting parameter (θh ) We have seen that Cost and Social Impact are very sensitive to λ, and
for each satisfaction level of objective functions (θ1 , θ2 , θ3 ) is determined the attitude of decision-makers can be effectively modeled by varying
∑
by the decision-makers ( h θh = 1, h= 1, 2, 3). In consultation with the this parameter. Fig. 6 shows the trade-off between the objectives by
decision-makers, we identified that Cost has a higher priority than varying λ, where Social Impact improves and Cost increases with

14
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Fig. 5. Best-fit candidate locations for Centers with different Combinations of weights for three objective functions.

Table 3
An overview of performance according to the various levels of λ.
Instance λ Objective Function Time (s)

TH Cost Inefficiency Social Impact

1 0 0.583 2.338E + 7 1.282 17.69 207.6

2 0.1 0.581 2.387E + 7 1.282 18.42 190.8
3 0.2 0576 2.416E + 7 1.416 19.45 192.4
4 0.3 0.575 2.486E + 7 1.282 19.88 263.1
5 0.4 0.573 2.535E + 7 1.282 20.06 275.5
6 0.5 0.572 2.563E + 7 1.416 21.72 286.4

increasing λ. We see that there is no point that dominates another point.

As a final note, we find that decision-makers tend to be pessimistic. In
the following, we analyze the performance of the proposed model on
different values of ρ in Table 4. Note here that Inefficiency and Social
Impact are not dependent on the scenarios, and ρ does not affect them
directly. The results are shown in Table 4, including all the objective
functions and the maximum regret for the DPHHC model. These mea­ Fig. 6. Effect of λ on Cost against Inefficiency and Social Impact.
sures are calculated to examine the model performance while varying ρ
in Constraint (32). This analysis is a trade-off between the p-robustness satisfaction levels of the objective functions (θh ). Table 5 demonstrates
value and Cost. We also see that the Inefficiency and Social Impact tend to that the analysis of the TH method aims to balance the values of the
improve when the Cost increases. membership functions, namely Cost, Inefficiency, and Social Impact. In
The next parameter of interest is γ (compensation coefficient) within other words, decision-makers prefer to find solutions that are balanced
the TH method. Higher γ values lead to balanced solutions or similar considering the objectives. Finally, the results show that increasing
satisfaction levels for all objective functions, while decreasing the value values of γ lead to relatively equal satisfaction levels of just over 0.5.
of γ leads to unbalanced solutions that emphasize the importance of the Moreover, as Fig. 7 shows, when the value of γ approaches 1.0, the

15
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Table 4
An overview of performance according to the various levels of ρ.
ρ Objective Function Maximum Regret Time (s)

TH Cost Inefficiency Social Impact

0 < 0.1 Infeasible Infeasible Infeasible Infeasible Infeasible Infeasible

0.10 0.343 2.181E + 7 2.234 17.28 0.0008 142.1
0.15 0.458 2.378E + 7 1.416 18.19 0.0910 149.5
0.20 0.538 2.416E + 7 1.416 19.45 0.1086 897.7
0.25 0.565 2.417E + 7 1.416 19.45 0.1088 208.3
0.30 0.538 2.416E + 7 1.416 19.45 0.1086 632.1
0.35 0.578 2.436E + 7 1.282 19.15 0.1178 349.5
0.40 0.579 2.436E + 7 1.282 19.15 0.1178 219.2

The final parameter of interest is θh , which specifies the relative

Table 5
weightings of Cost, Inefficiency, and Social Impact and prioritizes weights.
An overview of performance according to the various levels of γ.
The results are shown in Table 6. We consider different Combinations of
γ Satisfaction Level 1 Satisfaction Level 2 Satisfaction Level 3 the objective functions and give them weights between 0.1 and 0.7 as a
(μ1 ) (μ2 ) (μ3 )
measure of their importance. For example, in Combination 6, the second
0–0.2 0.699 0.528 0.632 objective has the highest priority based on an expert’s preference, and
0.3–0.4 0.672 0.573 0.544 hence it is given the value of 0.7. We see as a result that Inefficiency is the
0.5 0.699 0.528 0.632
0.6 0.672 0.573 0.544
lowest (0.464) compared to those Combinations where this objective is
0.7 0.654 0.573 0.544 not a priority. As a final note, we see two additional points from the
0.8–0.9 0.672 0.573 0.544 analyses conducted and the tables presented. Firstly, we are able to find
1 0.583 0.544 0.544 the optimal solutions. Secondly, all run-times are reasonable, between
13 and 1421 s. This demonstrates that our approach is particularly
satisfaction levels for all objectives approach each other. Conversely, suitable in practical settings.
considering the vertexγ = 0–0.2 (the top point in the figure), the satis­ In this paper, based on the several parameters of the problem under
faction levels of the objective functions (0.69, 0.63, 0.52) are uncertainty, a robust-fuzzy approach was utilized to solve the problem.
unbalanced. The robust approach used in this paper is the scenario-based robust

Fig. 7. Effect of γ on the satisfaction level of each objective.

Table 6
The results of the relative weightings of objectives’ satisfaction levels θ1 , θ2 and θ3 .
Combination θ1 θ2 θ3 Objective Function Time(s)

TH Cost Inefficiency Social Impact

1 0.50 0.25 0.25 0.581 2.416E + 7 1.416 19.45 642.8

2 0.25 0.50 0.25 0.566 2.436E + 7 1.282 19.15 201.5
3 0.25 0.25 0.50 0.574 2.416E + 7 1.416 19.45 196.7
4 0.70 0.20 0.10 0.589 2.416E + 7 1.416 19.45 932.4
5 0.10 0.20 0.70 0.587 2.629E + 7 0.464 20.06 188.2
6 0.20 0.20 0.10 0.575 2.629E + 7 0.464 20.06 254.6

16
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

approach first introduced by Kouvelis, Kurawarwala, and Gutierrez third (optimistic value ψ 3 ) values of each fuzzy number are generated
(1992). Additionally, the PCCP approach was used to deal with fuzzy based on uniform distribution in the range [lower limit, upper limit].
parameters that can be applied to possibilistic data and provide a min­ Experts kindly provided the lower and upper limits based on their
imum satisfaction level for decision-makers. By using this approach and experience and knowledge. Two values (ψ 1 : most pessimistic value and
applying it to the proposed model, its effectiveness was evaluated. The ψ 4 : most optimistic value) of each fuzzy number are obtained based on
model is also measured by real case study data and the results are pre­ the method proposed by Jiménez (1996). In other words, we suppose
sented in the relevant tables. that ψ̃ is a trapezoidal fuzzy number that is ψ ̃ = (ψ 1 , ψ 2 , ψ 3 , ψ 4 ), where
All the results obtained are exact, where the gap is always 0.00 %. ψ 1 < ψ 2 < ψ 3 < ψ 4 . First, the value of the pessimistic (ψ 2 ) and optimistic
Hence, it can be concluded that the answers are accurate, and the GAMS points (ψ 3 ) are generated randomly, followed by the value of the most
software achieves optimal results within reasonable computational pessimistic (ψ 1 ) and most optimistic points (ψ 4 ). These are calculated
times, thereby requiring no need for heuristic and metaheuristic algo­ using the following equations (Jiménez, 1996):
rithms for the real case. It should be noted that, due to the circumstances
ψ 1 = (1 − δ1 )ψ 2 (51)
of the ambiguous parameters of the problem and their inherent nature,
this approach was proposed and proved effective.
ψ 4 = (1 − δ2 )ψ 3 (52)
The proposed robust-fuzzy approach allows uncertain parameters to
be adjusted as the parameter values are realized. This approach has the where (δ1 , δ2 ) are considered in a range of 0.1 to 0.3 by experts (Uniform
flexibility of adjusting the level of conservativeness of solutions while (0.1,0.3)).
preserving the computational complexity of the nominal problem. This
method offers full control of the degree of conservation for every 5.2.2. Results
constraint. Additionally, this approach protects against the violation of Table 8 shows the results of the proposed approach on the problem
constraints deterministically when only a pre-specified number of the instances. The table shows the TH objective value, satisfaction levels,
coefficient changes. This advantage can be observed in the results of the processing time, and gaps for each value of γ considering different
solution to the model, in which all the results from the proposed problem sizes. We see that the TH function is not very sensitive to
approach are solved efficiently. Moreover, the results demonstrate that changes of γ. For example, in Instance 3 with γ between 0.6 and 0.9, an
the performance of the developed approach is similar to the result of appropriately unique balanced solution is produced. This is further
Torabi and Hassini (2008). Furthermore, this method is more robust and demonstrated in Fig. 8, which shows how increasing γ impacts three
reliable because it is able to set balanced and unbalanced solutions based different satisfaction levels. By increasing γ, we see that all satisfaction
on the preferences of decision-makers, and the solutions of this method levels approach the value of 0.55. The important point is that Instances 5
are consistent with the decision preferences (i.e., there is consistency and 6 are considerably large, requiring significant processing times, so
between the satisfaction level μ and weight vector θh ). In addition, the we consider a gap of 0.5 % to obtain solutions in reasonable times.
TH method is more flexible as it can find various efficient solutions for Nonetheless, we still see that solutions of a reasonable quality cannot be
instances with a certain weight vector θh through changing the γ, and it found for the largest Instance (6). Hence, the current solution approach
is particularly suited to solving multi-objective MILP models. is limited in its ability to scale to deal with a large number of patients,
nurses, and centers. Additionally, the optimistic–pessimistic parameter
5.2. Investigating problem characteristics (λ) can be considered as one of the more important parameters, so we
carry out a sensitivity analysis of it, the results of which are presented in
In the previous section, we demonstrated that the proposed mathe­ Table 9. Similar to the case study, when the robust-fuzzy model is solved
matical model is effective at solving a real-world problem. We also see with the lower value of λ, Cost and Social Responsibility are also low. On
that there is sufficient flexibility to allow different solutions to be found the other hand, the opposite is true when increasing the value of λ.
depending on the specific requirements of decision-makers. However, a Moreover, the best value of Inefficiency will remain fixed because λ does
key aspect that warrants further investigation is how the proposed not directly affect this objective. For example, in Instance 3, when λ
approach works when varying different characteristics of the DPHHC approaches 0.5, the values of Cost and Social Impact increase while the
problem. Hence, this section considers six numerical examples with value of Inefficiency stays fixed. In the other instances, Inefficiency
different sizes and parameter settings, and details of how the problems changes only slightly due to changes in some of the variables.
are generated. Then, we conduct an experimental evaluation of the As mentioned previously, the TH approach is used for the MOLP
performance of the mathematical model in solving these problem model to find a trade-off between the Cost, Inefficiency, and Social Impact
instances. objective functions. In this approach, the weightings of the Cost, In­
efficiency, and Social Impact satisfaction levels are shown by θ1 , θ2 , and
5.2.1. Generating problem instances θ3 , respectively. As seen in Table 10, we consider three combinations of
The characteristics of the problem instances, including the number of the relative weights of objectives and prioritize them in each Combi­
patients (M), nurses (V), center candidates (N), input factors (R), output nation. For example, if decision-makers want to have an efficient HHC
factors (G), and scenarios (S), are shown in Table 7. For each instance, network, θ2 should be given a higher weight (similar to Combination 2
the social parameters, service time, and cost parameters are the trape­ of Instance 1, where the value of Inefficiency is 0). Therefore, the results
zoidal fuzzy numbers from which the second (pessimistic value ψ 2 ) and

Table 7
The numerical instances.
Instance Node(I) Patient(M) Nurse(V) Center Candidate(N) Opened Center One1 Service Two2 Services Input Factor(R) Output Factor(G) Scenario

1 15 9 3 5 2 6–10 5,11 2 2 2
2 18 12 3 5 3 6–11 12–14
3 25 15 4 9 3 10–22 23,24
4 30 18 4 11 4 14–29 12,13
5 45 28 6 16 6 23–44 17–22
6 60 35 7 24 7 25–56 57–59
1
The set of patients who need one medical service.
2
The set of patients who need two medical services.

17
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Table 8
Sensitivity analysis on γ value in the TH method. A “-” implies that no solution was obtained and the time limit expired. For instances with ranges of γ, we report the
average values for the TH objective, time, and gap measures.
Instance γ TH Function Satisfaction Time (s) Gap%

Level 1 Level 2 Level 3

1 0.00 0.643 0.999 0.935 0.000 1.044 0.000000

0.05 0.611 1.000 0.935 0.000 2.266 0.000000
0.10 0.585 0.341 1.000 0.494 2.567 0.000000
0.20–0.60 0.504 0.342 1.000 0.494 23.28 0.000000
0.70 0.422 0.341 1.000 0.494 5.743 0.000000
0.80–0.90 0.382 0.342 1.000 0.494 33.79 0.000000
1.00 0.342 0.342 0.342 0.494 11.05 0.000000
2 0.00–0.10 0.701 0.677 0.741 0.693 21.20 0.000000
0.20 0.698 0.676 0.741 0.693 23.77 0.000000
0.30–0.90 0.687 0.677 0.741 0.693 73.58 0.000100
1.00 0.677 0.677 0.677 0.693 19.06 0.000000
3 0.00–0.10 0.750 1.000 0.852 0.398 434.1 0.000000
0.20–0.50 0.636 0.994 0.852 0.436 4691 0.000016
0.60–0.90 0.563 0.540 0.712 0.652 2320 0.000004
1.00 0.539 0.539 0.539 0.652 522.3 0.000000
4 0.00–0.05 0.783 1.000 0.678 0.681 2850 0.000035
0.10 0.775 0.999 0.678 0.681 2459 0.000000
0.20–0.40 0.765 1.000 0.678 0.681 9569 0.000004
0.50 0.732 0.999 0.678 0.681 1070 0.000000
0.60 0.728 0.742 0.736 0.723 859.2 0.000015
0.70–0.90 0.725 0.743 0.736 0.723 5502 0.000005
1.00 – – – – – –
0.00–0.50 0.674 0.795 0. 916 0.312 207.3 0.000233
5 0.10 0.638 0.794 0.916 0.312 1912 0.000538
0.20 0.613 0.469 0.882 0.597 1002 0.000917
0.30–0.40 0.586 0.468 0.882 0.597 990.5 0.011152
0.50 0.559 0.469 0.882 0.597 1165 0.020526
0.60 0.541 0.468 0.882 0.597 1025 0.027894
0.70 0.556 0.552 0.839 0.548 999.4 0.000870
0.80–0.90 0.539 0.523 0.839 0.548 1031 0.000352
1.00 0.480 0.480 0.791 0.496 1036 0.043126
6 0.00–0.30 0.731 0.775 0.781 0.667 629.4 0.000099
0.40 0.657 0.607 0.768 0.695 3173 0.054662
0.50 0.704 0.775 0.781 0.667 1032 0.000911
0.60 0.700 0.688 0.737 0.723 1030 0.000070
0.70–0.80 0.696 0.689 0.737 0.723 1026 0.004729
0.90 0.692 0.690 0.737 0.723 2989 0.000158
1.00 0.689 0.689 0.689 0.723 1063 0.000421

Fig. 8. The effect of γ on the satisfaction level of each objective function in Instance 3.

show that each objective function with a larger value of θh has a greater 6. Conclusion
improvement (the value of the improved function is highlighted in each
Combination) compared to other functions. In this section, we could find Home health care centres are increasingly playing an important role
optimal solutions in most cases, particularly in small and medium sizes in improving health care systems. The sustainable and efficient design of
of instances. For some large-scale problems where the solutions are not an HHC can significantly impact health care services by reducing costs
provably optimal, the gaps are very close to 0. In future research, ap­ and inefficiency and increasing the positive social impact. This paper
proaches like metaheuristics and hybrid approaches such as matheur­ presents a comprehensive multi-objective efficient and sustainable
istics can be of great potential for very large size problems. optimization model to tackle this problem. Two main measures, DEA
and CSR, are included in this study. The DEA measure inserted in the
HHC network makes it possible to reduce the inefficiency of centres. In

18
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Table 9
Sensitivity analysis on λ value in the robust-fuzzy method. A “-” implies that no solution was obtained and the time limit expired.
Instance λ TH Function Objective Function Time (s) Gap%

Cost Inefficiency Social Impact

1 0.1 0.634 2.341E + 8 0.240 11.57 3.117 0.000000

0.2 0.477 2.736E + 8 0.000 11.93 7.419 0.000000
0.3 0.658 3.212E + 8 0.240 12.38 0.920 0.000000
0.5 0.518 3.986E + 8 0.000 13.10 2.412 0.000000
2 0.1 0.692 2.572E + 8 0.517 11.93 46.70 0.000000
0.2 0.690 3.002E + 8 0.517 12.35 13.01 0.000000
0.3 0.697 3.433E + 8 0.517 12.77 27.45 0.000000
0.5 0.648 4.002E + 8 0.794 13.53 587.7 0.000067
3 0.1 0.702 3.538E + 8 0.442 17.80 1437 0.000155
0.2 0.590 4.084E + 8 0.442 18.12 3320 0.000081
0.3 0.635 4.741E + 8 0.442 18.75 175.1 0.000000
0.5 0.697 6.055E + 8 0.442 20.00 439.3 0.000003
4 0.1 0.590 4.084E + 8 0.442 18.12 8139 0.000070
0.2 0.732 5.506E + 8 1.284 24.52 1074 0.000014
0.3 0.734 6.384E + 8 1.284 25.28 3467 0.000000
0.5 – – – – – –
5 0.1 0.648 7.428E + 8 0.984 35.71 1112 0.000221
0.2 0.559 8.698E + 8 0.883 37.00 1165 0.020526
0.3 – – – – – –
0.5 – – – – – –
6 0.1 0.707 8.607E + 8 1.999 41.68 1070 0.000120
0.2 0.704 1.003E + 9 1.705 42.92 1032 0.000911
0.3 0.687 1.152E + 9 1.705 44.26 1078 0.000188
0.5 0.732 1.420E + 9 1.270 46.95 1045 0.000055

case study of the Kermanshah province in Iran, and the second is to

Table 10 understand how the approach scales and deals with varying problem
The results of the relative weightings of objectives in three Combinations: (I)
characteristics.
θ1 =θ2 < θ3 , (II) θ1 =θ3 < θ2 , and (III) θ2 =θ3 < θ1 for all instances.
The results demonstrate that practically viable solutions can be
Instance Combination Objective Function Time Gap found by using the proposed approach, and in reasonable time frames.
(s) %
Cost Inefficiency Social That is, considering several characteristics of centres (e.g., population,
Impact living costs, pollution levels), the proposed model is able to identify a
1 I 2.777E 0.240 11.977 7 0 subset of potential locations which are typically the best locations for
+8 centres. Moreover, the decision-makers are able to test different sce­
II 2.736E 0.000 11.938 9 0 narios (e.g., prioritizing inefficiency over cost or social impact, for
+8
III 2.656E 0.129 11.900 8 0
example), leading to different choices for the opening of centres. A study
+8 of the different problem characteristics, particularly increasing
2 I 3.002E 0.517 12.355 26 0 numbers, shows that our approach scales well.
+8 The managerial implications for health networks in terms of
II 2.893E 0.517 11.979 16 0
improving collaboration and coordination with home health care ser­
+8
III 2.893E 0.517 11.979 16 0 vices should be complemented by parallel actions by public authorities
+8 in order to be effective during an epidemic. Furthermore, if govern­
3 I 4.359E 2.991 18.516 30 0 mental organizations want to decide how to allocate their workers to
+8 patients and distribute drugs in future health crises, they will need to
II 4.084E 0.442 18.126 43 0
+8
develop their health care services and logistics abilities vastly. This
III 4.068E 0.442 18.100 92 0 needs expertise in proper medicine techniques, logistics planning, po­
+8 tential suppliers, and quality guarantees. Creating these abilities can
4 I 5.915E 3.990 24.934 1587 0 either be performed internally or, more efficiently, by depending on
+8
stand-by professional groups, including logistics services, and home
II – – – – –
III 5.506E 1.284 24.527 25.109 0 health care systems set up in response to a severe crisis. On the other
+8 hand, owing to the dynamic nature of the problem, most of the infor­
5 I,II,III – – – – – mation on aspects such as the established cost of health centres and the
6 I 1.014E 1.999 43.022 1876 0 times for services exists in an uncertain environment that is considered
+9
II 1.021E 1.270 42.526 2020 0
here. This shows that this design and proposed approach are useful for
+9 policymakers and governments in a real situation. These teams need to
III 1.003E 1.705 42.929 5376 0 support operating performance at regional levels in addition to strategic
+9 decision-making.
While we see that the proposed mathematical modeling is effective,
it is just an initial step towards tackling HHC problems, and there are a
CSR, the opening of a new centre in each district not only creates
number of potential limitations. For example, one important factor is the
employment opportunities for nurses as human capital, but also in­
approaches to scale in dealing with very large-scale problems, where
creases the economic development in less developed districts. Therefore,
time considerations also become important (e.g., solving multiple sce­
the motivation is to provide a feasible approach that can assist stake­
narios quickly). Incomplete approaches like metaheuristics and hybrid
holders in identifying the best locations for opening centres. To this end,
approaches such as matheuristics may have great potential in such
we conduct a numerical study consisting of two parts. The first is a real

19
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

situations. Furthermore, when increasing the number of scenarios, sce­ costs increase, but at the same time lead to increased patient satisfaction
nario reduction approaches can usefully be used. Given the uncertain and help to reduce death rates.
aspects of the problem, other methods tailored to deal with uncertainty,
such as stochastic programming, can also be of great benefit. Finally, the CRediT authorship contribution statement
problem itself can be extended in several ways. For instance, disruptions
to the HHC network can pose problems if some locations are affected Mahdyeh Shiri: Conceptualization, Writing – original draft, Meth­
(such as in earthquakes), leading to increased demands and closure of odology, Software. Fardin Ahmadizar: Conceptualization, Formal
routes that require additional resiliency measures. Another example is analysis, Writing – review & editing, Supervision. Dhananjay Thir­
where matching a nurse’s skills levels to the requirements of patients uvady: Conceptualization, Formal analysis, Writing – review & editing,
may be of significant interest. Supervision. Hamid Farvaresh: Conceptualization, Formal analysis,
The proposed model can be implemented to care for patients with Writing – review & editing.
Covid-19. However, if some patients have tested positive for Covid-19,
to reduce the transmission rate of the virus, the model can be appro­ Declaration of Competing Interest
priately modified as part of a future study. Some aspects to consider in
this context are whether Covid-19 patients might need to be visited only The authors declare that they have no known competing financial
by certain nurses who are not allowed to treat other patients, in which interests or personal relationships that could have appeared to influence
case nurses need to be categorized. In such circumstances, the network the work reported in this paper.

Appendix A

arsi − arsj + capv xsijv ≤ capv − dj ∀v ∈ V, i, j ∈ I, s ∈ S (A1)

∑∑
xsmjv = βm ∀m ∈ M, s ∈ S (A2)
j∈I v∈V

∑ ∑
xsimv − xsmjv = 0 ∀v ∈ V, m ∈ M, s ∈ S (A3)
i∈I j∈I

∑∑
xsnmv = 1 ∀v ∈ V, s ∈ S (A4)
n∈N m∈M

∑
m+n+1
xsihv = 1 ∀v ∈ V, h ∈ H, s ∈ S (A5)
i=n+1

am ≤ stsiv ≤ bm ∀i ∈ I, m ∈ M, v ∈ V, s ∈ S, i = m (A6)
∑ ∑
dm xsmjv ≤ capv ∀v ∈ V, s ∈ S (A7)
m∈M j∈I

∑∑
xsniv ≤ M yn ∀n ∈ N, s ∈ S (A8)
i∈I v∈V

xsijv , yj ∈ {0, 1}, stsiv ≥ 0 ∀i, j ∈ N, v ∈ V, s ∈ S (A9)

References Carello, G., & Lanzarone, E. (2021). An implementor-adversary approach for uncertain
and time-correlated service times in the nurse-to-patient assignment problem.
Computers & Operations Research, 105378.
Addis, B., Carello, G., Grosso, A., Lanzarone, E., Mattia, S., & Tànfani, E. (2015).
Carello, G., Lanzarone, E., & Mattia, S. (2018). Trade-off between stakeholders’ goals in
Handling uncertainty in health care management using the cardinality-constrained
the home care nurse-to-patient assignment problem. Operations Research for Health
approach: Advantages and remarks. Operations Research for Health Care, 4, 1–4.
Care, 16, 29–40.
Allaoua, H., Borne, S., Létocart, L., & Calvo, R. W. (2013). A matheuristic approach for
Chen, Y., Liang, L., & Yang, F. (2006). A DEA game model approach to supply chain
solving a home health care problem. Electronic Notes in Discrete Mathematics, 41,
efficiency. Annals of Operations Research, 145(1), 5–13.
471–478.
Cinar, A., Salman, F. S., & Bozkaya, B. (2021). Prioritized single nurse routing and
Bahadori-Chinibelagh, S., Fathollahi-Fard, A. M., & Hajiaghaei-Keshteli, M. (2019). Two
scheduling for home healthcare services. European Journal of Operational Research,
constructive algorithms to address a multi-depot home healthcare routing problem.
289(3), 867–878.
IETE Journal of Research, 1–7.
Cook, W. D., Roll, Y., & Kazakov, A. (1990). A dea model for measuring the relative
Barma, P. S., Dutta, J., & Mukherjee, A. (2019). A 2-opt guided discrete antlion
eeficiency of highway maintenance patrols. INFOR: Information Systems and
optimization algorithm for multi-depot vehicle routing problem. Decision Making:
Operational Research, 28(2), 113–124.
Applications in Management and Engineering, 2(2), 112–125.
Decerle, J., Grunder, O., El Hassani, A. H., & Barakat, O. (2016). A two-phases
Braekers, K., Hartl, R. F., Parragh, S. N., & Tricoire, F. (2016). A bi-objective home care
matheuristic for the home care routing and scheduling problem. IFAC-PapersOnLine,
scheduling problem: Analyzing the trade-off between costs and client inconvenience.
49(12), 1484–1489.
European Journal of Operational Research, 248(2), 428–443.
Decerle, J., Grunder, O., El Hassani, A. H., & Barakat, O. (2018). A memetic algorithm for
Cappanera, P., Scutellà, M. G., Nervi, F., & Galli, L. (2018). Demand uncertainty in robust
a home health care routing and scheduling problem. Operations Research for Health
Home Care optimization. Omega, 80, 95–110.
Care, 16, 59–71.
Carello, G., & Lanzarone, E. (2014). A cardinality-constrained robust model for the
Entezari, Z., & Mahootchi, M. (2021). Developing a mathematical model for staff routing
assignment problem in home care services. European Journal of Operational Research,
and scheduling in home health care industries: Genetic algorithm-based solution
236(2), 748–762.
scheme. Scientia Iranica, 28(6), 3692–3718.
Erdem, M., & Koç, Ç. (2019). Analysis of electric vehicles in home health care routing
problem. Journal of Cleaner Production, 234, 1471–1483.

20
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Errarhout, A., Kharraja, S., & Corbier, C. (2016). Two-stage stochastic assignment Malagodi, L., Lanzarone, E., & Matta, A. (2021). Home care vehicle routing problem with
problem in the home health care. IFAC-PapersOnLine, 49(12), 1152–1157. chargeable overtime and strict and soft preference matching. Health Care
Fathollahi-Fard, A. M., Ahmadi, A., Goodarzian, F., & Cheikhrouhou, N. (2020). A bi- Management Science, 1–20.
objective home healthcare routing and scheduling problem considering patients’ Manavizadeh, N., Farrokhi-Asl, H., & Beiraghdar, P. (2020). Using a metaheuristic
satisfaction in a fuzzy environment. Applied Soft Computing, 93, Article 106385. algorithm for solving a home health care routing and scheduling problem. Journal of
Fathollahi-Fard, A. M., Ahmadi, A., & Karimi, B. (2020). A Robust Optimization for a Project Management, 5(1), 27–40.
Home Healthcare Routing and Scheduling Problem Considering Greenhouse Gas Mankowska, D. S., Meisel, F., & Bierwirth, C. (2014). The home health care routing and
Emissions and Stochastic Travel and Service Times. In Green Transportation and New scheduling problem with interdependent services. Health Care Management Science,
Advances in Vehicle Routing Problems (pp. 43–73). Cham: Springer. 17(1), 15–30.
Fathollahi-Fard, A. M., Ahmadi, A., & Karimi, B. (2021). Multi-Objective Optimization of Milburn, A. B., & Spicer, J. (2013). Multi-objective home health nurse routing with
Home Healthcare with Working-Time Balancing and Care Continuity. Sustainability, remote monitoring devices. International Journal of Planning and Scheduling, 1(4),
13(22), 12431. 242–263.
Fathollahi-Fard, A. M., Govindan, K., Hajiaghaei-Keshteli, M., & Ahmadi, A. (2019). Mutingi, M., & Mbohwa, C. (2014). Multi-objective homecare worker scheduling: A fuzzy
A green home health care supply chain: New modified simulated annealing simulated evolution algorithm approach. IIE Transactions on Healthcare Systems
algorithms. Journal of Cleaner Production, 240, Article 118200. Engineering, 4(4), 209–216.
Fathollahi-Fard, A. M., Hajiaghaei-Keshteli, M., & Tavakkoli-Moghaddam, R. (2018). Nasir, J. A., & Kuo, Y. H. (2020). A decision support framework for home health care
A bi-objective green home health care routing problem. Journal of Cleaner transportation with simultaneous multi-vehicle routing and staff scheduling
Production, 200, 423–443. synchronization. Decision Support Systems, 138, Article 113361.
Fikar, C., & Hirsch, P. (2015). A matheuristic for routing real-world home service Nickel, S., Schröder, M., & Steeg, J. (2012). Mid-term and short-term planning support
transport systems facilitating walking. Journal of Cleaner Production, 105, 300–310. for home health care services. European Journal of Operational Research, 219(3),
Fikar, C., & Hirsch, P. (2017). Home health care routing and scheduling: A review. 574–587.
Computers & Operations Research, 77, 86–95. Nikzad, E., Bashiri, M., & Abbasi, B. (2021). A matheuristic algorithm for stochastic
Fiksel, J. (2006). Sustainability and resilience: Toward a systems approach. Sustainability: home health care planning. European Journal of Operational Research, 288(3),
Science, Practice and Policy, 2(2), 14–21. 753–774.
Ghannadpour, S. F., & Zarrabi, A. (2019). Multi-objective heterogeneous vehicle routing Omrani, H., Adabi, F., & Adabi, N. (2017). Designing an efficient supply chain network
and scheduling problem with energy minimizing. Swarm and Evolutionary with uncertain data: A robust optimization—data envelopment analysis approach.
Computation, 44, 728–747. Journal of the Operational Research Society, 68(7), 816–828.
Gong, X., Geng, N., Zhu, Y., Matta, A., & Lanzarone, E. (2020). A Matheuristic Approach Parmigiani, A., Klassen, R. D., & Russo, M. V. (2011). Efficiency meets accountability:
for the Home Care Scheduling Problem With Chargeable Overtime and Preference Performance implications of supply chain configuration, control, and capabilities.
Matching. IEEE Transactions on Automation Science and Engineering, 18(1), 282–298. Journal of Operations Management, 29(3), 212–223.
Goodarzian, F., Abraham, A., & Fathollahi-Fard, A. M. (2021). A biobjective home health Pishvaee, M. S., Torabi, S. A., & Razmi, J. (2012). Credibility-based fuzzy mathematical
care logistics considering the working time and route balancing: A self-adaptive programming model for green logistics design under uncertainty. Computers &
social engineering optimizer. Journal of Computational Design and Engineering, 8(1), Industrial Engineering, 62(2), 624–632.
452–474. Rahbari, A., Nasiri, M. M., Werner, F., Musavi, M., & Jolai, F. (2019). The vehicle routing
Grenouilleau, F., Legrain, A., Lahrichi, N., & Rousseau, L. M. (2019). A set partitioning and scheduling problem with cross-docking for perishable products under
heuristic for the home health care routing and scheduling problem. European Journal uncertainty: Two robust bi-objective models. Applied Mathematical Modelling, 70,
of Operational Research, 275(1), 295–303. 605–625.
Habibnejad-Ledari, H., Rabbani, M., & Ghorbani-Kutenaie, N. (2019). Solving a multi- Regis-Hernández, F., Carello, G., & Lanzarone, E. (2020). An optimization tool to
objective model toward home care staff planning considering cross-training and dimension innovative home health care services with devices and disposable
staff’s preferences by NSGA-II and NRGA. Scientia Iranica, 26(5), 2919–2935. materials. Flexible Services & Manufacturing Journal, 32(3).
Haeri, A., Hosseini-Motlagh, S. M., Ghatreh Samani, M. R., & Rezaei, M. (2020). A mixed Rest, K. D., & Hirsch, P. (2016). Daily scheduling of home health care services using time-
resilient-efficient approach toward blood supply chain network design. International dependent public transport. Flexible Services and Manufacturing Journal, 28(3),
Transactions in Operational Research, 27(4), 1962–2001. 495–525.
Inuiguchi, M., & Ramı́k, J. (2000). Possibilistic linear programming: A brief review of Rodriguez, C., Garaix, T., Xie, X., & Augusto, V. (2015). Staff dimensioning in homecare
fuzzy mathematical programming and a comparison with stochastic programming in services with uncertain demands. International Journal of Production Research, 53
portfolio selection problem. Fuzzy Sets and Systems, 111(1), 3–28. (24), 7396–7410.
Jiménez, M. (1996). Ranking fuzzy numbers through the comparison of its expected Rodriguez-Verjan, C., Augusto, V., & Xie, X. (2018). Home health-care network design:
intervals. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, Location and configuration of home health-care centers. Operations Research for
4(04), 379–388. Health Care, 17, 28–41.
Jones, D., & Tamiz, M. (2016). A review of goal programming. Multiple Criteria Decision Samani, M. R. G., Hosseini-Motlagh, S. M., & Homaei, S. (2020). A reactive phase against
Analysis, 903–926. disruptions for designing a proactive platelet supply network. Transportation
Khodaparasti, S., Bruni, M. E., Beraldi, P., Maleki, H. R., & Jahedi, S. (2018). A multi- Research Part E: Logistics and Transportation Review, 140, Article 102008.
period location-allocation model for nursing home network planning under Shahnejat-Bushehri, S., Tavakkoli-Moghaddam, R., Boronoos, M., & Ghasemkhani, A.
uncertainty. Operations Research for Health Care, 18, 4–15. (2021). A robust home health care routing-scheduling problem with temporal
Kisialiou, Y., Gribkovskaia, I., & Laporte, G. (2019). Supply vessel routing and scheduling dependencies under uncertainty. Expert Systems with Applications, 115209.
under uncertain demand. Transportation Research Part C: Emerging Technologies, 104, Sharifi, M., Hosseini-Motlagh, S. M., Samani, M. R. G., & Kalhor, T. (2020). Novel
305–316. resilient-sustainable strategies for second-generation biofuel network design
Klimberg, R. K., & Ratick, S. J. (2008). Modeling data envelopment analysis (DEA) considering Neem and Eruca Sativa under hybrid stochastic fuzzy robust approach.
efficient location/allocation decisions. Computers & Operations Research, 35(2), Computers & Chemical Engineering, 143, Article 107073.
457–474. Shi, Y., Boudouh, T., & Grunder, O. (2017). A hybrid genetic algorithm for a home health
Kouvelis, P., Kurawarwala, A. A., & Gutierrez, G. J. (1992). Algorithms for robust single care routing problem with time window and fuzzy demand. Expert Systems with
and multiple period layout planning for manufacturing systems. European Journal of Applications, 72, 160–176.
Operational Research, 63(2), 287–303. Shi, Y., Boudouh, T., & Grunder, O. (2019). A robust optimization for a home health care
Lanzarone, E., & Matta, A. (2012). A cost assignment policy for home care patients. routing and scheduling problem with consideration of uncertain travel and service
Flexible Services and Manufacturing Journal, 24(4), 465–495. times. Transportation Research Part E: Logistics and Transportation Review, 128, 52–95.
Lanzarone, E., & Matta, A. (2014). Robust nurse-to-patient assignment in home care Shi, Y., Boudouh, T., Grunder, O., & Wang, D. (2018). Modeling and solving
services to minimize overtimes under continuity of care. Operations Research for simultaneous delivery and pick-up problem with stochastic travel and service times
Health Care, 3(2), 48–58. in home health care. Expert Systems with Applications, 102, 218–233.
Lanzarone, E., Matta, A., & Sahin, E. (2012). Operations management applied to home Shiri, M., Ahmadizar, F., & Mahmoudzadeh, H. (2021). A three-phase methodology for
care services: The problem of assigning human resources to patients. IEEE home healthcare routing and scheduling under uncertainty. Computers & Industrial
Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 42(6), Engineering, 158, Article 107416.
1346–1363. Shiri, M., Ahmadizar, F., Mahmoudzadeh, H., & Bashiri, M. (2019). A Two-Phase Method
Liang, L., Yang, F., Cook, W. D., & Zhu, J. (2006). DEA models for supply chain efficiency for Robust Home Healthcare Problem: A Case Study. In International Conference on
evaluation. Annals of Operations Research, 145(1), 35–49. Health Care Systems Engineering (pp. 193–203). Cham: Springer.
Lin, M., Ma, L., & Ying, C. (2021). Matching daily home health-care demands with supply Snyder, L. V., & Daskin, M. S. (2006). Stochastic p-robust location problems. Iie
in service-sharing platforms. Transportation Research Part E: Logistics and Transactions, 38(11), 971–985.
Transportation Review, 145, Article 102177. Span, P. (2016). Wages for home health care lag as demand grows. New York Times,
Liu, W., Dridi, M., Fei, H., & El Hassani, A. H. (2021). Hybrid Metaheuristics for Solving a (September 23).
Home Health Care Routing and Scheduling Problem with Time Windows, Torabi, S. A., & Hassini, E. (2008). An interactive possibilistic programming approach for
Synchronized Visits and Lunch Breaks. Expert Systems with Applications, 115307. multiple objective supply chain master planning. Fuzzy Sets and Systems, 159(2),
Liu, B., & Iwamura, K. (1998). Chance constrained programming with fuzzy parameters. 193–214.
Fuzzy Sets and Systems, 94(2), 227–237. Veenstra, M., Roodbergen, K. J., Coelho, L. C., & Zhu, S. X. (2018). A simultaneous
Liu, R., Yuan, B., & Jiang, Z. (2020). The large-scale periodic home health care server facility location and vehicle routing problem arising in health care logistics in the
assignment problem: A region-partition-based algorithm. IEEE Transactions on Netherlands. European Journal of Operational Research, 268(2), 703–715.
Automation Science and Engineering, 17(3), 1543–1554.

21
M. Shiri et al. Expert Systems With Applications 211 (2023) 118185

Wang, F., Liao, F., Li, Y., Yan, X., & Chen, X. (2021). An ensemble learning based multi- Yang, M., Ni, Y., & Yang, L. (2021). A multi-objective consistent home healthcare routing
objective evolutionary algorithm for the dynamic vehicle routing problem with time and scheduling problem in an uncertain environment. Computers & Industrial
windows. Computers & Industrial Engineering, 154, Article 107131. Engineering, 160, Article 107560.
Weiszer, M., Burke, E. K., & Chen, J. (2020). Multi-objective routing and scheduling for Zahiri, B., Zhuang, J., & Mohammadi, M. (2017). Toward an integrated sustainable-
airport ground movement. Transportation Research Part C: Emerging Technologies, 119, resilient supply chain: A pharmaceutical case study. Transportation Research Part E:
Article 102734. Logistics and Transportation Review, 103, 109–142.
Xu, J., & Zhou, X. (2013). Approximation based fuzzy multi-objective models with Zhalechian, M., Tavakkoli-Moghaddam, R., Zahiri, B., & Mohammadi, M. (2016).
expected objectives and chance constraints: Application to earth-rock work Sustainable design of a closed-loop location-routing-inventory supply chain network
allocation. Information Sciences, 238, 75–95. under mixed uncertainty. Transportation Research Part E: Logistics and Transportation
Yalçındağ, S., & Lanzarone, E. (2021). Merging short-term and long-term planning Review, 89, 182–214.
problems in home health care under continuity of care and patterns for visits. Journal Zhang, T., Yang, X., Chen, Q., Bai, L., & Chen, W. (2018). Modified ACO for home health
of Industrial & Management Optimization. care scheduling and routing problem in Chinese communities. In 2018 IEEE 15th
Yalçındağ, S., Matta, A., Şahin, E., & Shanthikumar, J. G. (2016). The patient assignment International Conference on Networking, Sensing and Control (ICNSC) (pp. 1–6). IEEE.
problem in home health care: Using a data-driven method to estimate the travel Zheng, C., Wang, S., Li, N., & Wu, Y. (2021). Stochastic joint homecare service and
times of care givers. Flexible Services and Manufacturing Journal, 28(1–2), 304–335. capacity planning with nested decomposition approaches. European Journal of
Operational Research.

22