JID: EOR

ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

European Journal of Operational Research xxx (xxxx) xxx

Contents lists available at ScienceDirect

European Journal of Operational Research

journal homepage: [Link]/locate/ejor

Resource planning strategies for healthcare systems during a

pandemic
Mohammad Fattahi a, Esmaeil Keyvanshokooh e, Devika Kannan d, Kannan Govindan b,c,d,1,∗
a
Newcastle Business School, Northumbria University, Newcastle Upon Tyne, United Kingdom
b
China Institute of FTZ Supply Chain, Shanghai Maritime University, Shanghai, 201306, China
c
Yonsei Frontier Lab, Yonsei University, Seoul, South Korea
d
Center for Sustainable Supply Chain Engineering, Department of Technology and Innovation, Danish Institute for Advanced Study, University of Southern
Denmark, Campusvej 55, Odense M, Denmark
e
Department of Information & Operations Management, Mays Business School, Texas A&M University, College Station, TX 77845, USA

a r t i c l e i n f o a b s t r a c t

Article history: We study resource planning strategies, including the integrated healthcare resources’ allocation and shar-
Received 16 February 2021 ing as well as patients’ transfer, to improve the response of health systems to massive increases in de-
Accepted 10 January 2022
mand during epidemics and pandemics. Our study considers various types of patients and resources to
Available online xxx
provide access to patient care with minimum capacity extension. Adding new resources takes time that
Keywords: most patients don’t have during pandemics. The number of patients requiring scarce healthcare resources
OR in health services is uncertain and dependent on the speed of the pandemic’s transmission through a region. We develop a
COVID-19 pandemic, Resource sharing and multi-stage stochastic program to optimize various strategies for planning limited and necessary health-
allocation, Patients’ transfers, Multi-stage care resources. We simulate uncertain parameters by deploying an agent-based continuous-time stochas-
stochastic programming tic model, and then capture the uncertainty by a forward scenario tree construction approach. Finally, we
Data-driven rolling horizon propose a data-driven rolling horizon procedure to facilitate decision-making in real-time, which miti-
gates some critical limitations of stochastic programming approaches and makes the resulting strategies
implementable in practice. We use two different case studies related to COVID-19 to examine our opti-
mization and simulation tools by extensive computational results. The results highlight these strategies
can significantly improve patient access to care during pandemics; their significance will vary under dif-
ferent situations. Our methodology is not limited to the presented setting and can be employed in other
service industries where urgent access matters.
© 2022 The Author(s). Published by Elsevier B.V.
This is an open access article under the CC BY license ([Link]

1. Introduction established widespread closures and stay-at-home orders to inter-

vene in this issue (Govindan, Mina & Alavi, 2020; Mervosh, Lu &
COVID-19 was first identified in Wuhan, China in December Swales, 2020; Ferreira et al., 2022). However, the accelerated num-
2019 and it has since become a global pandemic (Ferreira, Kan- ber of COVID-19 cases forced many hospitals to discharge exist-
nan, Meidutė-Kavaliauskienė & Vale, 2022). As of December 2021, ing patients earlier to preserve care capacity for COVID-19 patients
there have been more than 250 million reported COVID-19 cases (Parker, Sawczuk, Ganjkhanloo, Ahmadi & Ghobadi, 2020; Tonna et
worldwide. As the result of the COVID-19 pandemic, the world al., 2020) as part of the guidelines from the Centers for Disease
has seen more than five million deaths until now; most health- Control and Prevention (CDC, 2020). Although these approaches
care systems have faced extraordinary challenges. As one of the were effective to treat a greater number of COVID-19 patients, they
most important challenges, outbreaks of the SARS-CoV-2 infec- resulted in poor outcomes for non-COVID-19 patients and a sub-
tion in local communities yield a massive increase in demand for stantial financial loss for healthcare systems.
limited resources such as intensive care unit (ICU) beds, health- Extending healthcare resources’ capacity is impossible for many
care personnel, and mechanical ventilators. Several governments countries in a short time, according to Adelman (2020). Ramp-
ing up production of complex medical equipment, such as ven-
∗
tilators, in facilities configured for other products will require
Corresponding author at: China Institute of FTZ Supply Chain, Shanghai Mar-
itime University, Shanghai 201306, China.
time that several COVID-19 patients don’t have. In such critical
E-mail address: kgov@[Link] (K. Govindan). situations, two other primary strategies are resource sharing and
1
The author order is based on the academic seniority. demand redistribution, which can minimize shortages in response

[Link]
0377-2217/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license ([Link]

Please cite this article as: M. Fattahi, E. Keyvanshokooh, D. Kannan et al., Resource planning strategies for healthcare systems during a
pandemic, European Journal of Operational Research, [Link]
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

to massive hospitalization demand with the minimum extension • We extend our model into a data-driven resource planning ap-
of resources’ capacity. Some facts demonstrate the applicability of proach by developing a rolling horizon procedure, which will
these strategies in countries or local communities. Indeed, the in- help decision-makers make real-time decisions.
fection spreads at varying rates in different regions, and COVID- • We deploy an agent-based continuous-time stochastic model
19 cases peak at different times in different regions. This vari- for modeling the COVID-19 transmission and then a scenario
ance provides an opportunity for sharing some scarce resources tree construction approach to capture the stochasticity of the
such as ventilators; these devices can be transported over large number of infected individuals requiring hospitalization.
distances or within regions to reduce capacity shortfalls caused • We investigate two different case studies by our proposed tools.
by an epidemic surge in a particular area. In addition, regarding
demand redistribution, patients tend to select hospitals/healthcare The organization of the paper is as follows: Section 2 pro-
centers in accordance with their reputation or distance on a lo- vides the literature review. In Section 3, the problem definition and
cal level, which leads to unbalanced patient loads across hospi- MSSP are described. The data-driven decision-making approach
tals/healthcare centers and a decrease in the overall quality of pa- based on the rolling horizon approach is explained in Section 4.
tient care (Drevs, 2013; Varkevisser, van der Geest & Schut, 2012). The agent-based simulation approach and scenario tree construc-
Parker et al. (2020) demonstrated that the operationally feasible tion approaches are presented in Section 5. Computational results
redistribution of newly admitted patients through the network of based on two case studies are provided in Section 6. Several man-
healthcare systems can reduce the patients’ overflow. Healthcare agerial insights derived from our computational results are pre-
systems can consider system-level interventions and patient trans- sented in Section 7. Finally, Section 8 concludes the paper.
fers to maximize the utilization of available resources.
From the medical literature, qualitative studies propose some 2. Literature review
strategies to deal with the capacity shortage under a potential
surge in demand (Mills, Helm & Wang, 2020). This study is mo- Regarding disaster management and humanitarian operations,
tivated by the healthcare capacity concerns created during the abundant literature exists (see e.g., Altay & Green III, 2006; Gupta,
COVID-19 pandemic. Its theory and practice are directly relevant to Starr, Farahani & Matinrad, 2016; Rodríguez-Espíndola, Albores &
the capacity planning through healthcare systems, and the objec- Brewster, 2018; Wex, Schryen, Feuerriegel & Neumann, 2014) and
tive is to optimize the use of resources during a pandemic by vari- several papers investigate the healthcare operations for emergency
ous strategies. We develop a novel data-driven multi-stage stochas- situations (see e.g., Adan, Bekkers, Dellaert, Jeunet & Vissers, 2011;
tic programming approach for managing healthcare resources as Chi, Li, Shao & Gao, 2017; Luscombe & Kozan, 2016; Sung & Lee,
well as demand redistribution to provide care for patients during 2016). However, the nature of epidemic outbreaks and pandemics
a pandemic. In particular, we provide an answer to the question, are meaningfully different from other disasters in terms of their
“what could have been done to mitigate shortages due to a massive dynamic nature, resources’ demand, global scale, and length. The
increase in the demand for limited resources during a pandemic such World Health Organization has defined practical phases in epi-
as COVID-19?” demics and pandemics management: Anticipation, Early detection,
Our decision-making framework considers two types of health- Containment, Control and mitigation, and Elimination or eradication
care resources: 1) direct, in use resources (such as ICU beds and (World Health Organization, 2018). Literature focusing specifically
ventilators) and 2) service resources (such as personnel and lab- on the allocation of healthcare resources during emergencies of-
oratories). We also then explore various possible capacity plan- ten concentrates on the distribution of life-saving medical devices
ning options for healthcare resources, including capacity exten- (Dasaklis, Pappis & Rachaniotis, 2012). Resource allocation helps
sion and relocation, as well as demand redistribution strategies in Containment and Control and mitigation phases when it comes
that are modeled to provide hospitalization services to patients. to healthcare resources planning. Generally, the resource alloca-
This methodology considers uncertainty in the disease spread and, tion problem has been deeply investigated since the 1990s by the
therefore, demand for healthcare resources in various regions or operations research community (see e.g., Bakuli & Smith, 1996;
hospitals. We deploy a simulation approach based on an agent- Elmaghraby, 1993; Fiedrich, Gehbauer & Rickers, 20 0 0; Hegazy,
based continuous-time stochastic model to capture the COVID-19 1999).
spread providing a daily forecast for the hospitalization demand Mills et al. (2020) investigated possible actions of hospitals
over time, and this model can be adapted to any county or geo- to provide immediate additional healthcare services in the case
graphical region. Our modeling study provides critical insights into of urgent massive demand. They focused on strategies related to
how regions or hospitals could cope with a surge in demand for the Containment and Control and mitigation phases. Further, in the
healthcare resources. medical literature, there are some research studies, such as Hick et
The existing scientific literature suffers from a lack of deci- al. (2004), Kaji, Koenig and Bey (2006), and Rothman, Hsu, Kahn
sion support tools for managing healthcare resources during a and Kelen (2006), which identified response components and de-
pandemic, which simultaneously considers the above-mentioned veloped conceptual frameworks to propose qualitative methods for
strategies and demand uncertainty. In this study, the proposed creating the surge capacity without quantification of capacity allo-
data-driven decision-making tool encompasses a large scope of op- cation and relocation approaches. Practical reports in the health-
erational situations. The proof of concepts will be given for alloca- care systems of the Netherlands indicated that the lack of cooper-
tion and relocation of ventilators among several healthcare regions ation between hospitals is a major cause for trauma patients to be
of the US and, secondly, allocation of resources and demand re- transported outside the region because of shortages in ICU capacity
distribution among hospitals in an area of Iran. A summary of the and their corresponding nurses (Litvak, Van Rijsbergen, Boucherie
contributions of this work is as follows: & van Houdenhoven, 2008). Litvak et al. (2008) addressed the ca-
pacity problem related to ICU beds, in which many hospitals in a
geographical region reserve a small number of ICU beds for the
regional emergency patients. Scheduling of nurse shifts and plan-
• We propose a multi-stage stochastic program (MSSP) for the in- ning of workforce are also addressed by Otegbeye, Scriber, Ducoin
tegrated healthcare resources planning and demand redistribu- and Glasofer (2015) and Willis, Cave and Kunc (2018), respec-
tion during a pandemic. This model accounts for various patient tively. Farley et al. (2013) highlighted that emergency department
types and healthcare resources during pandemics. information systems constitute a unique and important role in

2
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

hospitals’ electronic health records, and the information system category is service resources (SER) such as personnel or laborato-
performance affects physician clinician workflow, decision-making, ries that a hospital utilizes as needed.
communication, overall patient safety, and quality of care. Most re- Given the ongoing capacity concerns, based on the types of re-
search works in resource planning in healthcare systems deal with sources, three main strategies are used by various countries during
standard and forecastable leadwork of a single healthcare center. pandemics (especially COVID-19), as follows:
Some other studies address mass casualty incidents that demon- Demand redistribution: to address the balance of loads, redistri-
strate short-term effects on hospitals, but those works contrast bution of patients between hospitals is implemented,
with the long-lasting effects of a pandemic outbreak. Resource allocation (extension): the government or policy mak-
During the influenza pandemic preparedness, Toner and Wald- ers provide more external resources for regions or hospitals such
horn (2006) emphasized the significance of cooperation between as increasing capacity through calling in additional personnel and
different healthcare centers to decrease the extreme healthcare creating new suitable beds.
system stress. Bertsimas et al. (2020) showed the advantages of Resource relocation (sharing): the sharing of capacity between
inter-regional collaboration in sharing ventilators across states in regions (hospitals) is applicable for portable resources, such as
the U.S. Assuming a perfect demand forecast, their deterministic healthcare personnel and ventilators.
optimization model allocates the federal stockpile of ventilators In this section, we propose a multi-stage stochastic program
and determines how many ventilators to transfer between states (MSSP) to determine the optimal demand redistribution, resource
to minimize ventilator shortage costs. Considering the same prob- allocation, and sharing decisions to minimize shortages, medical
lem, but with stochastic demand, Mehrotra et al. (2020) presented treatment refusals or delays, and resource extension. Following
a two-stage stochastic model for allocating and sharing ventilators. practical requirements, the foremost priority is to minimize short-
They demonstrated that sharing ventilators across states could re- ages (non-accepted patients), and the second goal is to minimize
duce shortages. Parker et al. (2020) studied the problem of finding the allocation of new resources to healthcare regions or hospitals.
demand and healthcare resource transfers between hospitals dur- By using an MSSP, the optimization problem has several deci-
ing the COVID-19 pandemic to minimize the required new capacity sion layers, where random parameters are progressively realized,
and shortage for healthcare resources. They used robust optimiza- and decisions should be adapted to this process. Typically, an N-
tion to address demand uncertainty. The concern of this study, re- stage stochastic program includes a sequence of stochastic param-
source planning strategies for healthcare systems during epidemics eters ξ1 , ξ2 , . . . , ξN−1 with a discrete support. A scenario is a re-
and pandemics, is recently highlighted by the operations research alization of these stochastic parameters over the problem’s stages
community to improve the response to pandemics, especially to (periods), and a scenario tree represents the progressive observa-
COVID-19. Further, modeling the effects of pandemics on a region’s tion of these parameters. During a pandemic such as COVID-19,
individuals and their progression is also investigated; some of the a healthcare system faces various patient types in different re-
studies related to COVID-19 are Lewnard et al. (2020), Parker et gions or hospitals requiring treatment (unlike healthcare equip-
al. (2020), Levin et al. (2020), Nabi (2020), Silva et al. (2020), and ment). The number of patients in various types at different re-
Reddy et al. (2020). gions or hospitals is a stochastic parameter in our study, which is
Methodologically, in the healthcare operations management thought of as the healthcare system’s demand.
area, the two-stage stochastic programming approach is commonly The decisions in each stage of an MSSP can be categorized into
employed to formulate various problems that incorporate uncer- two groups: (i) the decisions that are made before the uncertainty
tainty (see e.g., Mehrotra et al., 2020). However, the uncertainty in realization at that stage, (ii) the decisions that are made based
stochastic parameters such as the number of patients is usually re- on the uncertainty realization. In our problem setting, both re-
alized progressively and the decisions at each period or stage are source sharing and extension belong to the first group of deci-
a function of uncertainty observations, previous decisions, and ob- sions, and other decisions, including the acceptance or refusal of
served feedback outcomes up to that stage (Erdogan et al., 2013; patients and demand redistribution, belong to the second group.
Govindan, Fattahi & Keyvanshokooh, 2017). Therefore, multi-stage Fig. 1 illustrates these decisions in our problem in period t ∈ T
stochastic programming will be a more suitable optimization tool where T = {t0 , t0 + 1, . . . , te } is the set of time periods.
that we utilize in this work. A policy should be non-anticipative in an MSSP, which means
There are key differences between the above papers and ours. the decisions made at each stage must not be dependent on the
First, the focus of most studies in resource planning during a pan- future realization of stochastic parameters. There are two com-
demic is a single mitigation strategy to improve the healthcare sys- mon ways to formulate an MSSP (Dupačová, 1995; Kall & Wallace,
tems’ response; however, our model with realistic features con- 1994). In the first, an MSSP is formulated as a sequence of nested
siders various patient types, demand redistribution, and different two-stage stochastic programs in which non-anticipativity is im-
capacity planning options aligned with the type of healthcare re- plicitly imposed. In the second (used in this paper), a set of non-
sources. Second, an MSSP is developed to incorporate the uncertain anticipativity constraints (NAC) is explicitly modeled and these
number of patients requiring treatment during a pandemic. Third, constraints should be considered for the decisions that are deter-
the decisions made by MSSPs are not implementable in practice, mined before uncertainty realization (Dupačová, 1995; Erdogan &
and a data-driven decision-making approach with the help of a Denton, 2013; Kall & Wallace, 1994).
rolling horizon procedure is developed to deal with this issue and To model stochasticity related to various patient types in
to determine real-time decisions. healthcare regions or hospitals as a scenario tree, a set of scenarios
S with countable size |S| is taken into account. The corresponding
3. Problem formulation scenarios’ probabilities are π1 , π2 , . . . , π|S| . If we denote a realiza-
tion for patient number of type k ∈ K at region i ∈ I on period t ∈ T
During a pandemic such as COVID-19, surges in demand for the under scenario s ∈ S by Dsikt and ξts = (Dsikt : i ∈ I, k ∈ K ), then the
healthcare system often occur; it’s common for the healthcare re- realization of stochastic parameters in scenario s ∈ S from period
sources of a hospital to be lower than the required capacity. We t0 to period te is (ξts , ξts +1 , . . . , ξtse ). Fig. 2a shows an example of
0 0
categorize resources in terms of patients’ usage types. The first is a scenario tree with three periods and five scenarios for our prob-
direct in use resources (DUR), which a hospital assigns them to lem with three regions. As an example, for scenario s and period t,
a patient as long as he/she is hospitalized. For example, suitable (|Ds11t |, |Ds21t |, |Ds31t |) is a realization related to the number of pa-
beds and ventilators are in this category for COVID-19. The second tients in type 1 and |Ds11t |, |Ds21t | and |Ds31t | are corresponding to

3
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Fig. 1. Different decision types and the decision-making process over |T | period.

Fig. 2a. A scenario tree example.

region 1, 2, and 3, respectively. Fig. 2b is an alternative represen- Fig. 2b. The corresponding scenario fan.
tation of the scenario tree, which is called scenario fan, where the
individual scenarios observed in the particular stages are disaggre-
gated to form five scenarios. The MSSP is presented as follows:
In this section, the proposed MSSP is presented. However, the   
solution obtained from solving the MSSP is scenario-dependent   s  
Min : πs α wikt + (|T | + 1 − t )nsirt
and, hence, it is not implementable in the real-world practice and
s∈S i∈I t∈T k∈K i∈I t∈T r∈RDUR
does not allow the information attained over time to be used. In-  
deed, the critical limitation of scenario-based stochastic programs   
is that their optimal policy is only valid for a limited set of sce- + (|T | + 1 − t ) s
cirt +β qsi irt
narios. To resolve this issue, we develop a new data-driven Rolling i∈I t∈T r∈RSER i ∈I i∈I t∈T r∈R
 
Horizon Procedure (RHP). Our approach, presented in Section 4, 
addresses this issue and provides real-time day-to-day sharing pol- +γ psi ikt (1)
icy and demand redistribution in a rolling horizon manner. The re- i ∈I i∈I t∈T k∈K
quired notations for presenting the mathematical model are de-
fined in Table 1. kbsirt0 = Ēirt0 ∀ i ∈ I, ∀ r ∈ RDUR , ∀ s ∈ S (2-1)

4
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Table 1
Notations.

Sets and indices

T The set of periods indexed by t , t  ∈ T.

S The set of scenarios s, s ∈ S.
I The set of regions i, i ∈ I (it is possible to consider hospitals set instead of regions based on the problem setting).
R The set of healthcare resources r ∈ R. RDUR and RSER denote the direct in use and service resources, respectively. REX and RT R denote the
set of resources with the possibility of capacity extension and sharing, respectively. Finally, RcEX and RcT R are the complement of set REX
and RT R , respectively.
K The set of patient types, k ∈ K.
R (k ) The set of required healthcare resources for patient type k.
I (i, r ) The set of regions, which can receive (forward) resource r ∈ R from (to) region i ∈ I.
I (i, k ) The set of regions, which can receive (forward) patient type k ∈ K from (to) region i ∈ I. It is assumed the patient transfer should be
done by a lead time of less than one day (period).

Parameters

Dsikt The number of patients’ arrival in type k ∈ K at region i ∈ I in period t ∈ T under scenario s ∈ S.
k Average discharge time of an accepted patient in k ∈ K.
Lrii Lead time for transshipment of resource r ∈ RT R between regions i and i ∈ I, Lrii ≥ 1, ∀i, i ∈ I.
λri Lead time of adding resource r ∈ REX at region i, ∀i ∈ I
X̄ikt0 The number of hospitalized patients in type k ∈ K at region i ∈ I at the beginning of planning horizon.
X˜ikt The number of accepted patients in type k ∈ K in region i ∈ I in periods before t0 , which they will be discharged at period t based on
k . This parameter is zero for t ≥ t0 + k .
Ēirt0 The number of available resources r ∈ RDUR in region i ∈ I at the beginning of planning horizon.
C̄irt0 The capacity of resource r ∈ RSER in region i ∈ I at the beginning of planning horizon.
ϕkr The capacity usage coefficient of patient type k ∈ K for resource r ∈ RSER .

Decisions

nsirt The number of added healthcare resources r ∈ (RDUR ∩ REX ) at region i ∈ I in period t ∈ T under scenario s ∈ S.
s
cirt The amount of new capacity for healthcare resource r ∈ (RSER ∩ REX ) added at region i ∈ I in period t ∈ T under scenario s ∈ S.
qsii rt The number of healthcare resource r ∈ RT R , which are transshipped from region i ∈ I to i ∈ I at the beginning of period t ∈ T under
scenario s ∈ S.
vsirt The capacity of healthcare resource r ∈ RSER at region i ∈ I in period t ∈ T under scenario s ∈ S.
xsikt The number of accepted patients in type k ∈ K in region i ∈ I during period t ∈ T under scenario s ∈ S.
wsikt The number of non-accepted patients in type k ∈ K in region i ∈ I in period t under scenario s ∈ S.
s
wikt The number of non-accepted patients in type k ∈ K in region i ∈ I in period t under scenario s ∈ S that cannot be met even with
transferring them to other regions, so sent to the next period.
kbsirt The number of available healthcare resource r ∈ RDUR in region i ∈ I at the beginning of period t ∈ T under scenario s ∈ S.
kesirt Secondary variable that represents the number of healthcare resource r ∈ RDUR in region i ∈ I at the end of period t ∈ T under scenario
s ∈ S.
psii kt The number of patients in type k ∈ K, which are sent from region i ∈ I to i ∈ I at period t ∈ T under scenario s ∈ S.
asikt Auxiliary binary variable that represents the refusal of patients is occurred in region i ∈ I at period t ∈ T under scenario s ∈ S.

kbsirt = kesir (t−1) ∀ i ∈ I, ∀ t ∈ T \{t0 }, ∀ r ∈ RDUR , ∀ s ∈ S (2-2)   

xsikt + wsikt − psii kt = Dsikt + psi ikt + wiks (t−1)
i ∈I (i,k ) i ∈I (i,k )
 
kesirt = kbsirt + qsi ir t−L nsir (t−λir ) xsikt ∀ i ∈ I, ∀k ∈ K, ∀ t ∈ T , ∀ s ∈ S (6-1)
( ri i )+ −
i ∈I (i,r ) k∈K:r∈R(k )
  
+ X˜ikt + xsik(t−k ) − qsii rt  
k∈K:r∈R(k ) k∈K:r∈R(k ) i ∈I (i,r )
s
wikt ≥ wsikt − psii kt ∀ i ∈ I, ∀k ∈ K, ∀ t ∈ T , ∀ s ∈ S
i ∈I (i,k )
∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RDUR , ∀ s ∈ S (2-3)
(6-2)

nsirt = 0 ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RcEX , ∀ s ∈ S (2-4)   

wsikt ≤ max Dsikt × asikt ∀ i ∈ I, ∀k ∈ K, ∀ t ∈ T , ∀ s ∈ S (6-3)
s ∈S

qsii rt = 0 ∀ i ∈ I, ∀ i ∈ I, ∀t ∈ T , ∀ r ∈ RcT R , ∀ s ∈ S (3)    

psi ikt ≤ max Dsikt × 1 − asikt ∀ i ∈ I, ∀k ∈ K, ∀ t ∈ T , ∀ s ∈ S (6-4)
s ∈S
i ∈I (i,k )
  
vsirt = C̄irt0 + cirs (t  −λir ) + qsi ir
(t  −Lri i ) 
t  ≤t t  ≤t i ∈I (i,r ) nsirt = nsirt ∀ i ∈ I, ∀ r ∈ (RDUR ∩ REX ), ∀ t ∈ T ,
    s
− qsii rt  ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RSER , ∀ s ∈ S (4-1) ∀ s, s ∈ S : ξts0 , ξts0 +1 , . . . , ξt−1
s
= ξts0 , ξts0 +1 , . . . , ξt−1 (7-1)
t  ≤t i ∈I (i,r )

s
cirt =0 ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RcEX , ∀ s ∈ S (4-2) s
cirt s
= cirt

∀ i ∈ I, ∀ r ∈ (RSER ∩ REX ), ∀ t ∈ T ,
  s
∀ s, s ∈ S : ξts0 , ξts0 +1 , . . . , ξt−1
s
= ξts0 , ξts0 +1 , . . . , ξt−1 (7-2)
 
   
ϕkr X̄iks t0 + xsikt  − xsik(t  −k ) − X˜ikt  
k∈K:r∈R(k ) t  ≤t t0 +k ≤t  ≤t t  ≤t vsirt = vsirt ∀ i ∈ I, ∀ r ∈ RSER , ∀ t ∈ T ,
  s
≤ vsirt ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RSER , ∀ s ∈ S (5) ∀ s, s ∈ S : ξts0 , ξts0 +1 , . . . , ξt−1
s
= ξts0 , ξts0 +1 , . . . , ξt−1 (7-3)

5
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

and, then, dynamically update the uncertainty for the following pe-
s riods and input parameters of the MSSP.
kbsirt = kbirt ∀ i ∈ I, ∀ r ∈ RDUR , ∀ t ∈ T , To explain how the proposed approach works for the real data
  s
∀ s, s ∈ S : ξts0 , ξts0 +1 , . . . , ξt−1
s
= ξts0 , ξts0 +1 , . . . , ξt−1 (7-4) over a horizon of |T | periods (days), we consider a sample path,
denoted by ω, as real data. The sample path includes the realized
number of patients and discharge of accepted patients over |T | pe-
riods. In planning horizon = {t0 , t0 + 1, . . . , te }, to obtain a policy

qsii rt = qsii rt ∀ i ∈ I, ∀ i ∈ I, ∀ r ∈ R, ∀ t ∈ T ,
  s for t ∗ ∈ T , we solve the MSSP with a scenario tree for the number
∀ s, s ∈ S : ξts0 , ξts0 +1 , . . . , ξt−1
s
= ξts0 , ξts0 +1 , . . . , ξt−1 (7-5)
of patients over periods t ∗ , t ∗ + 1, . . . , t ∗ + |T | − 1. Then for the
implementation of the obtained policy, we solve the MSSP with a
horizon of |T | periods in which for the first period t ∗ , the uncer-
n, c , v, kb, ke, q, w, w , p, x, q ≥ 0 and a ∈ {0, 1}|I|×|K |×|T |×|S| . tain parameters are known (based on sample path ω) and the op-
(8) timal decisions are fixed. For the next period, we update some in-
put parameters of the stochastic model and in addition to model’s
Our main goal in relation (1) is to minimize the amount of cu- parameters that should be updated in each period, we repeatedly
mulative non-accepted patients in the considered healthcare sys- update the uncertainty after a predetermined number of periods
tem over the planning horizon, the total number of added DUR by calibrating parameters of our simulation model based on our
resources, the total number of added SER resources, and the to- observations (See the RHP in Fig. 3).
tal amount of patient transfers and resource sharing. According to The explained procedure should be repeated in each period, and
the practical aspects, the weights α , β , and γ are set such that we roll the patients’ arrival planning horizon forward one day by
the minimization of non-accepted patients has the highest prior- adding a new period to the calendar at every step. In obtaining
ity. The second priority is to minimize the total number of new re- policy in each specific period t ∗ , we should update some parame-
sources. Since the supply of new resources is time-consuming for ters of the stochastic model and consider some modification in the
policy makers, it is desirable to have as much time as possible be- presented model to capture the impacts of previous realized data
fore new supplies of healthcare resources. In order to account for and decisions. In Table 2, we provide the definition of some param-
this aspect in our formulation, the weight |T | + 1 − t is considered eters, which are used for the model’s modifications, and we entitle
for new resources in each period t. Furthermore, this weight can be this model the rolling horizon model. It is worth noting that period
interpreted as some rental cost per day in the objective function. t ∗ is the first period of the horizon in the rolling horizon model.
Finally, we minimize the amount of resource sharing and patients’ In the rolling horizon model, parameters Ēirt0 and C̄irt0 should
transfer, by considering coefficients β and γ , respectively, with the be changed to Ēirt ∗ and C̄irt ∗ , respectively, and their values are
lowest priority in comparison with other goals. In practice, policy based on the available information at the beginning of each period.
makers can set coefficients β and γ based on their necessities. It Constraints (2-3) should be updated as follows:
is worth noting that in the results, the mentioned parameters are 
kesirt = kbsirt + qsi ir
tuned so that the amount of shortages be reasonable and manage- (t−Lri i ):(t−Lri i )≥t ∗
able for the healthcare system. i ∈I (i,r )

Based on constraints (2-1)-(2-4), the available DUR at the be- + Q̄irt + nsir (t−λir ) − xsikt
ginning and end of each time period are calculated. The possibility k∈K:r∈R(k )
of the extension of DUR resources and resource sharing over the   
planning horizon are considered by constraints (2-4) and (3), re- + xsik(t−k ):(t−k )≥t ∗ + X˜ikt − Xˆikt
spectively. In addition, relations (4-1) show the available capacity k∈K:r∈R(k ) k∈K:r∈R(k ) k∈K:r∈R(k )

of SER resources in each period and under each scenario. The pos- − qsii rt
sibility of adding SER resources over the planning horizon is con- i ∈I (i,r )
sidered by constraints (4-2). For the acceptance of patients’ hospi-
talization, constraints (5) guarantee the available capacity for serv-
∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RDUR , ∀ s ∈ S (9)
ing them in terms of SER resources. Based on constraints (6-1)- By constraints (9) in the rolling horizon model, the impact of pa-
(6-4), the amount of patients’ acceptance, non-acceptance, and re- tients’ acceptance, resource sharing decisions, and realized uncer-
distribution are calculated. Constraints (6-2) obtain the number of tainty will be captured. Furthermore, constraints (4-1) should be
refused patients that cannot be met even by transferring them to modified as follows:
   
other regions (hospitals). Further, constraints (6-3) and (6-4) guar- vsirt = C̄irt ∗ + cirs (t  −λir ) + qsi ir
antee that a region (hospital) can accept the patients from other (t  −Lri i ):(t  −Lri i )≥t ∗ + Q̄irt 
t  ≤t t  ≤t i ∈I (i,r ) t  ≤t
regions (hospitals) in each period if it does not refuse any patients  
at that period. Constraints (7-1)-(7-5) are NACs in our MSSP, which − qsii rt 
are considered for the decisions that are made before uncertainty t  ≤t i ∈I (i,r )

realization at each stage. Variable types and ranges are defined in ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RSER , ∀ s ∈ S (10)
constraints (8). It is worth noting that the proposed model is flex- Constraints (5) is also modified in the rolling horizon model as fol-
ible to be used for a set of regions as well as hospitals. lows:
 
   
4. Data-driven decision-making by the RHP ϕkr X̄iks t ∗ + xsikt  − X˜ikt  − Xˆikt  − xsik(t  −k )
k∈K:r∈R(k ) t  ≤t t  ≤t t ∗ +k ≤t  ≤t
Here, we propose a data-driven resource planning framework
≤ vsirt ∀ i ∈ I, ∀ t ∈ T , ∀ r ∈ RSER , ∀ s ∈ S (11)
under uncertainty by using an RHP to implement our MSSP in real
time. The rolling horizon approach makes the obtained policy im- Finally, the patients which are not accepted in any region before
plementable in practice and evaluates the policy empirically. By period t ∗ should be considered in constraints (6-1) for the first pe-
riod by substituting parameter W  ik instead of w ik(t ∗ −1 ) .
s
this approach, the latest data that is revealed as time progresses
enables us to adjust our decisions over time. In other words, we The RHP has been applied for MSSPs in a few studies (Fattahi
observe the realization of the uncertain parameters in one period & Govindan, 2018, 2020) and one can refer to these studies for

6
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Fig. 3. The RHP in this study.

Table 2
Parameters for updating the rolling horizon model.

Ēirt ∗ The amount of resource ∈ RDUR , which are available in region i ∈ I at the beginning of period t ∗ .
C̄irt ∗ The capacity of resource r ∈ RSER in region i ∈ I at the beginning of period t ∗ .
X̄iks t ∗ The number of hospitalized patients in type k ∈ K at region i ∈ I at the beginning of period t ∗ .
X˜ikt The number of accepted patients in type k ∈ K in region (hospital) i ∈ I in periods before t ∗ , which they will be discharged at period t
based on k . This parameter is zero for t ≥ t ∗ + k .
Xˆikt The number of accepted patients in type k ∈ K in region (hospital) i ∈ I in periods before t ∗ , which is supposed to be discharged in
period t based on k , but they are discharged in periods before t ∗ .
Q̄irt The number of healthcare resources r ∈ RT R , which are transshipped from other regions to region i ∈ I in periods before t ∗ and will be
available at the beginning of period t ∈ T .
W  ik The number of patients in type k ∈ K at region ∈ I, which are not accepted in any regions in period t ∗ − 1 (previous period of the
beginning period).

more information about the estimation of the true objective func- Quarantined (IQ ), Hospitalized at the critical stage (IH ), Recovered
tion in real time by rolling horizon simulation. In other words, by (R), and Deceased (D). In our model, a proportion of the exposed
assuming enough realized sample paths, we can evaluate the poli- individuals in an age subgroup are quarantined as soon as they get
cies from MSSPs by the rolling horizon simulation. Fig. 3 shows the into the Infected stage, which means they are isolated from the
RHP in this study. population and do not infect additional individuals. Our simulation
network related to defined compartments is illustrated in Fig. 4.
5. Scenario tree construction for multivariate stochastic We construct the base model for each group and the disease
parameters spread is modeled through two main parts: 1) the disease progress
for an infected individual, and 2) the spread of the disease be-
In this study, we focus on one category of COVID-19 pandemic tween the members of the population. In other words, we assume
patients, those who get the SARS-CoV-2 virus from infected in- that each individual will lie in one of the compartments in Fig. 4.
dividuals in a cohort. This group includes patients with a critical PQ , PR , and PD are the probabilities of self-quarantine of an in-
state who will die if they remain untreated. These patients should fected individual, recovery of an infected individual without hos-
be hospitalized, and their treatments, including an ICU with me- pitalization, and death of a hospitalized individual, respectively. In
chanical ventilation, can prevent a subset of deaths among them. the model, we assume 80% of 0–19y group, 50% of 20–59y, and
To construct a scenario tree for the MSSP, we follow the approach 90% of ≥60y group quarantine themselves after getting the infec-
presented by Ekici, Keskinocak and Swann (2014) for the simula- tion. Generally, we model a defined cohort of individuals (e.g., pop-
tion of the number of patients who need hospitalization in a re- ulation of a region) for a given number of days (simulation hori-
gion with insights from experts in a medical school. Accordingly, zon). The simulation time unit is one day. Susceptible individuals
an agent-based continuous-time stochastic model is constructed can acquire SARS-CoV-2 infection from infected individuals in the
for the COVID-19 transmission. cohort. Once an individual is infected, he/she progresses through
Firstly, the entire population is divided into three age groups various infection states until either recovery or death. The severity
(0–19y, 20–59y, or ≥60y), that helps model the various types of of disease and the length of stay in each disease state are based
interactions between people in the population. The population in on age-specific transition probabilities of COVID-19 natural history,
a region is classified into Susceptible (S), Exposed (E), Infected (IT ), estimated from historical data and scientific reports (Haridy, R.,

7
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Fig. 4. The simulation network of our model.

2020; Hu et al., 2020; Liu, Gayle, Wilder-Smith & Rocklöv, 2020; (ξt ∗ , ξt ∗ +1 , . . . , ξt ∗ +|T |−1 ), depends on uncertainty at periods before
Mizumoto, Kagaya, Zarebski & Chowell, 2020; Wang et al., 2020; time period t ∗ . If we consider the realized uncertainty before t ∗
Yang et al., 2020; Zhou et al., 2020; Perc, Gorišek Miksić, Slavinec as ξ̄[t ∗ −1] , the dependency of ξt ∗ to ξ̄[t ∗ −1] can be presented as
& Stožer, 2020). In this study, we considered that the individuals ξt ∗ (ξ̄[t ∗ −1] ). During any pandemic, many parameters can affect dis-
who are in the critical disease state should be managed in an ICU
ease spread and transmission and we capture ξt ∗ (ξ¯[t ∗ −1] ) by our
with mechanical ventilation.
simulation model and update the scenario trees in predetermined
Effective reproduction number (R0 ) is another important param-
time periods over the planning horizon.
eter of our model which is the average number of secondary cases
caused by an infectious individual; it determines the infectivity of
the virus. We update the input parameters each week to account 6. Computational results
for the variation of effective reproduction number. By consultation
from partner medical schools, we have obtained the other param- We will give a proof of concept for our data-driven MSSP
eters and constructed the described disease spread model. methodology using two case studies during COVID-19 pandemic:
By running the simulation model, we obtain the number of pa- 1) sharing mechanical ventilators among a subset of regions in the
tients in critical state at each region/hospital over a planning hori- U.S., 2) sharing nurses and ventilators and demand transfers be-
zon. The results of the simulation model depend on the values of tween hospitals in a geographical area of Iran.
PQ , PR , PD , and R0 , which can change within some small intervals
based on the literature’s data. Therefore, by running the simula- 6.1. Case study 1: sharing ventilators among a subset of U.S.
tion model several times, we can obtain a set of discrete scenar- healthcare regions
ios for the stochastic parameter as a scenario fan. We then con-
struct a scenario tree based on the generated scenario fan and In this case study, we address sharing of ventilators among
reduce the number of scenarios in order to avoid computation- healthcare regions in an area of the USA based on Keyvashokooh,
ally intractable stochastic programs. To do so, we deploy a for- Fattahi, Zokaeinikoo, Freedberg and Kazemian (2020). During each
ward scenario tree construction method proposed by Heitsch and peak of the pandemic, additional ventilators were obtained from
Romisch (2005) based on the proposed heuristics by Dupačová, the government to cope with the surge in ventilator demand. Our
Gröwe-Kuska and Römisch (2003). They proposed two approaches data-driven optimization model informs an optimal ventilator al-
to transform a scenario fan into a scenario tree called as the for- location and relocation policy so that the uncertain demand can
ward and backward constructions. The generated scenario fan fol- be satisfied with the fewest possible ventilators. This ensures that
lows a probability distribution F and if we transform it into a sce- hospitals can better serve non-COVID patients and potentially can-
nario tree with probability distribution F T , the Kantorovich dis- cel fewer procedures by accommodating the needs of patients with
tance (DK ) between F and F T should be less than a predetermined as few ventilators as possible. It should be mentioned, in this case
value ε . In other words, the reduction algorithms apply maximal study, we address one type of critical patients needing ventilators,
reduction strategy such that DK (F , F T ) < ε . and patient transfer between regions is not reasonable and appli-
We use forward scenario tree construction approach in this pa- cable since we have not considered hospitals in this case study.
per and bundle the scenarios for each period t ∈ T . For detailed ex- In solving the MSSP, 150 scenarios in the form of a scenario fan
planations related to the scenario construction approach, one can are simulated by using the agent-based continuous-time stochas-
refer to Fattahi and Govindan (2018), Fattahi, Govindan and Key- tic model for COVID-19 transmissions, and then the scenarios are
vanshokooh (2018), and Fattahi and Govindan (2020). Further, pa- reduced and converted into a scenario tree by the forward sce-
rameter ε is considered as εr × εmax where εr is a constant value nario construction approach. In our implementations, parameter
between zero and one representing a scale for the amount of re- εr is set to 0.7. It is worth noting for setting the value of εr , we
duction in the initial scenario fan and εmax is the minimum dis- have done stability analysis based on the approaches proposed by
tance between F and one of its scenarios with probability one. It is Fattahi and Govindan (2018), and the in-sample and out-of-sample
worth noting that by increasing the reduction scale εr , the number stability error are 2.2% and 1.8%, respectively. Regarding computa-
of obtained scenarios decreases, so the information loss increases. tional tractability, in examined case studies, our model is solvable
However, as the number of scenarios decreases, we have a better with various settings by the CPLEX solver in less than 5 min.
computational tractability for solving the MSSP. Therefore, there is The impact of sharing ventilators strategy. In order to investi-
a trade-off between the number of scenarios and computational gate the importance of sharing strategies, 20 sample paths are gen-
tractability. erated based on our simulation model that represent the realized
The RHP enables us to deal with the uncertainty realization number of new patients in need of ventilators on each day over
over time. The uncertainty vector at each time period t ∗ , ξt ∗ = the planning horizon (90 days) and the length of ventilator use for

8
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Table 3
Ventilator needs and sharing outcomes under two strategies.

Total new ventilators Total transshipments Total shortages Maximum number of shortages
required, Mean (SD) between regions, Mean (non-accepted patients), (non-accepted patients) over
(SD) Mean (SD) scenario paths

No sharing 1574.7 (114.4) 0 (0) 16.1 (8.1) 34

Sharing 662.5 (21.2) 2898 (132.1) 14.8 (7.9) 27

Fig. 7. Average total number of required nurses in case study 2.

Fig. 5. Average number of required ventilators in each day in Case study 1.

though the impact of sharing strategy on the non-accepted pa-
tients’ number is not meaningful based on our mathematical mod-
eling, the total number of needed ventilators in the case of no
sharing increases significantly. In other words, in our optimization
problem, we have assumed the needed ventilators will be supplied
by the government in the US and because of this issue, we have
not any significant increase in the value of shortages.

6.2. Case study 2: demand redistribution and sharing resources

among a subset of Iranian hospitals

In this case study, main hospitals in some cities of Iran are con-
sidered during the COVID-19 pandemic that contains 20 hospitals
in a healthcare region. Each hospital has an initial capacity of ven-
Fig. 6. Average cumulative number of required ventilators in Case study 1. tilators, ICU beds, and medical personnel (nurses). The ICU beds
and ventilators correspond to the DUR and medical personnel re-
late to the SER. Here, sharing of ventilators and medical personnel
each patient. Considering this set of samples, we implement our and patients’ transfer are possible to provide services to infected
data-driven approach to obtain the optimal sharing policy corre- patients requiring the hospitalization. The horizon of three months
sponding to each sample path. The average number of new venti- in the third peak of COVID-19 is considered for this case study. Our
lators required to cope with the demand and its cumulative value data-driven model optimizes various strategies such that demand
under two strategies (sharing and no sharing strategies) are shown can be satisfied with fewest resources possible.
in Figs. 5 and 6, respectively.
As seen in Fig. 5, when resource sharing is not done, a new 6.2.1. The impact of sharing resources and patients transfer strategies
supply of ventilators is needed earlier than when ventilators are We investigate the importance of our strategies in this section.
shared. Under the sharing strategy, some of the early shortages Considering a set of samples as realized uncertainty, we implement
can be eliminated by moving ventilators from regions with excess our data-driven approach to obtain the optimal resource planning
ventilators to those experiencing a shortfall. More importantly, we corresponding to each sample path. The average total number of
can see in Fig. 6 that the no sharing strategy requires significantly new nurses and ventilators required to cope with the demand un-
higher number of additional ventilators to cope with the demand. der four strategies (no sharing and no patient transfer, sharing and
While 1574.7 additional ventilators are needed in average to avoid no patient transfer, no sharing and patients transfer, sharing and pa-
refusal of patients’ care as much as possible across this US area tients transfer strategies) are shown in Figs. 7 and 8, respectively.
under the no sharing strategy, this area under the sharing strat- More details of information obtained from simulating case study 2
egy requires only 662.5 additional ventilators to achieve the same are reported in Table 4. It is worth noting that through solving case
outcome. The more details of information obtained from simulat- study 2 by the RHP over 90 days, the average number of scenarios
ing case study 1 are reported in Table 3. Through solving our case in the constructed scenario trees by εr = 0.7 is 24.1.
study by the RHP over 90 days, we have constructed a scenario Presented results in Table 4 show that we can improve the re-
tree for each day and updated the input parameters of our agent- quired ventilators and nurses by about 20% and 14%, respectively,
based simulation model, weekly. It is worth noting that the average by using both sharing and patients’ transfer strategies. Further, the
number of scenarios in the constructed scenario trees by εr = 0.7 sharing strategy in this case study is more effective than patients
is 22.8. transfer strategy in terms of required new resources. However, pa-
From Table 3, we can see the amount of sharing between the tients transfer strategy has a better impact on the reduction of
regions is significant under the sharing strategy. In addition, al- non-accepted patients in compared to the sharing strategy. Fig. 9

9
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Table 4
Ventilators and nurses needed under four strategies.

Total new ventilators Total new nurses required, Total shortages Maximum number of shortages
required, Mean (SD) Mean (SD) (non-accepted patients) (non-accepted patients) over
Mean (SD) scenario paths

No sharing, No patients 1877.1 (129.2) 397.8 (28.9) 18.8 (9.9) 27

transfer
Sharing, Patients transfer 1498.1 (112.7) 342.5 (26.1) 2.1 (1.01) 5
Sharing, No patients 1558.5 (113.1) 359.3 (27.9) 9.3 (3.7) 14
transfer
No sharing, Patients 1661 (115.4) 366.2 (25.0) 7.4 (2.8) 10
transfer

6.2.2. Demand redistribution vs resource sharing

The presented results in the previous section highlight the
importance both of resource sharing and demand redistribution
strategies. However, some policy makers may have different pri-
orities in using these strategies based on the existing healthcare
infrastructure. In this sub-section, we investigate how different pri-
orities can be embedded in our model.
In our results in the previous sub-section, the same weights are
considered for β and γ in the sharing and patients transfer strategy.
Here, we report the sensitivity of the average amount of ventilators
transshipments, nurses’ transshipments, and patients transfers in
this strategy to γβ in Fig. 10.
Fig. 8. Average total number of required new ventilators in case study 2.
As shown in Fig. 10, the sensitivity of nurses and ventilators
transshipments to γβ value are relatively the same. On the other
shows the main patients transfer through the considered region in hand, we can obtain various policies in terms of using resource
the sharing and patients transfer strategy. sharing and patients transfer by setting parameters β and γ .

Fig. 9. Main patients’ transfers through the considered region in case study 2 (more than 10 over the planning horizon in average).
10
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Fig. 10. The sensitivity of the average amount of nurses’ transshipments, ventilators Fig. 12. Average number of required ventilators in case study 1 and in the case of
transshipments, and patients transfers to γβ . no update, which is assumed that ventilators would be released after 14 days.

Fig. 11. The release time of ventilators after usage for the COVID-19 patients.
Fig. 13. Average total number of required ventilators in case study 1 for risk-neutral
and risk-averse policies.

6.3. The significance of data-driven decisions

One of the main advantages of our proposed data-driven 6.4. Risk-averse decisions
methodology is the progressive information update based on un-
certainty realization over time. In this sub-section, we assess the Here, we determine a risk-management policy by using the
impact of data-driven decision-making in terms of one aspect. The Conditional Value at Risk (CVaR) as a well-defined risk mea-
discharge time of a patient, denoted by  , has a significant im- sure. Linear programming techniques are used for formulating the
pact on the optimal decisions. In this paper, we have assumed that problem by this risk measure (Ahmed, 2006). By assuming the
the discharge time of patients in critical state and the resources’ cumulative distribution function of random variable Q as FQ (. ),
release are the same. Using published data on the duration of ven- the Value at Risk at the confidence level c (V aRc ) is V aRc (F ) =
tilator use for patients with COVID-19, we used a lognormal distri- inf {θ ∈ R| FQ (θ ) ≥ c} and hence CV aRc (F ) = E(Q |Q ≥ V aRc (F )).
bution for  (Ludwig, Jacob, Basedow, Andersohn & Walker, 2021). Additionally, based on the formulation presented by Rockafellar
+
Based on our historical data, the value of  follows log-normal dis- and Uryasev (2002), CV aRc (F ) = inf{z + 1−c
1
E[(F − z ) ]}.
z∈R
tribution and ln( ) has normal distribution with mean 2.5 and One important issue related to the risk-averse MSSPs is the
standard deviation 0.93. Fig. 11 shows the release time of ventila- time consistency. Recently, it has been highlighted by several stud-
tors in our historical data. ies as a desirable property of a problem. Informally, in order to
In the optimization model, we set the value of  to the mean preserve the time consistency in MSSPs, by given the available in-
duration of ventilator use (14 days). Then, by implementation of formation at the time when a policy is determined, the optimal-
our data-driven approach, we account for the released ventilators ity of the policy should only be with respect to possible future
in each region on each day. In order to investigate the importance realizations (Homem-de-Mello & Pagnoncelli, 2016). Ruszczyński
of this data-driven approach, we assume that all ventilators will (2010) defined the time consistency in dealing with sequences
be released after 14 days and use our model without updating re- of random variables in the dynamic programming approach, and
leased ventilators on each day. As shown in Fig. 12, without a data- Shapiro (2009) focused on the stability of decision variables at
driven approach in case study 1, 764.5 additional ventilators are each stage in risk-averse MSSPs.
called on to ensure demand is met as much as possible, whereas We consider CV aRc of total non-accepted patients instead of its
with a data-driven approach, the same is achieved with only 662.5 expected value and, we have used the approach of Yin and Büyük-
ventilators. Further, in case study 2, the average amount of re- tahtakin (2021)) for modeling the risk-averse MSSP. Yin and Büyük-
quired ventilators and nurses increase about 11% and 9%, respec- tahtakin (2021)) confirmed that their modeling approach preserves
tively, if we assume a constant value for  and do not update the the time consistency. In other words, our formulation enforces the
input parameters based on the obtained information. time consistency by non-anticipativity constraints.
Note we have only examined the importance of parameter  In case study 1, we examine the total number of new venti-
in this sub-section, and in our data-driven model, we update the lators over 90 days for risk-averse decisions where c is equal to
discharge of patients as well as the uncertainty through our data- 0.8 and 0.95. In Fig. 13, the needed ventilators for risk neutral and
driven RHP. risk-averse decisions are illustrated.

11
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Table 5
New ventilators required and the number of non-accepted patients with the risk-neutral and risk-averse policies.

New ventilators required, Mean (SD) Total non-accepted patients, Mean (SD) Maximum of non-accepted patients
over scenarios

Risk-averse Policy with c =0.95 710.8 (16.7) 3.2 (2.7) 9

Risk-averse Policy with c = 0.8 699.8 (17.2) 4.3 (2.2) 11
Risk-neutral Policy 662.5 (21.2) 14.8 (7.9) 27

Table 6
New ventilators required and non-accepted patients with the risk-averse, stochastic, and deterministic policies in case study 1.

New Ventilators Required, Mean (SD) Total non-accepted patients, Mean (SD) Maximum non-accepted patients over
scenarios

Risk-averse Policy with c = 0.95 710.8 (16.7) 3.2 (2.7) 9

Risk-neutral policy by MSSP approach 662.5 (21.2) 14.8 (7.9) 27
Risk neutral policy by two-stage 677.2 (23.7) 15.1 (9.2) 27
stochastic programming approach
Deterministic Policy 619.3 (24.9) 26.2 (15.3) 53

Table 7
New ventilators and nurses required and non-accepted patients with the risk-averse, stochastic, and deterministic policies in case study 2.

New ventilators required, Mean New nurses required, Mean Total non-accepted patients, Maximum non-accepted
(SD) (SD) Mean (SD) patients over scenarios

Risk-averse Policy with 1541.7 (78.1) 355.1 (14.8) 0.81 (0.22) 2

c = 0.95
Risk-neutral policy by MSSP 1498.1 (112.7) 342.5 (26.1) 2.1 (1.01) 5
approach
Risk neutral policy by 1589.4 (128.1) 353.8 (32.9) 2.9 (0.92) 6
two-stage stochastic
programming approach
Deterministic Policy 1382 (189.7) 307.9 (48.1) 16.7 (2.9) 24

From Table 6, we can see about 2.4% improvement of the MSSP

in compared with the two-stage stochastic program in terms of
the total number of new ventilators in case study 1. In addition,
as shown in Table 7 for case study 2, the improvement of required
ventilators and nurses are 5.7% and 3.2%, respectively. In the two-
stage stochastic model, all allocation and sharing decisions should
be made at the beginning of the planning horizon. Additionally,
the poor performance of the deterministic policy is highlighted in
terms of the amount of non-accepted patients in which the aver-
age number of patients is considered instead of stochastic patients’
number.
The importance of lead times. Our extensive computational
experiments indicate that the number of non-accepted patients
is mainly dependent on lead-times in both of risk-neutral and
Fig. 14a. Sensitivity of non-accepted patients to lead-times. risk-averse policies. For case study 1, Fig. 14a presents the sen-
sitivity of the average non-accepted patients on lead time val-
ues related to the risk-neutral and risk-averse policies, and
As shown in Fig. 13, the risk management policies call for about Fig. 14b illustrates the sensitivity of the average cumulative new
8% more ventilators. However, this comes with the benefit of fewer ventilators’ requirement on lead time values. In the sensitivity
non-accepted patients and standard deviation of new ventilators in analysis various multiplier coefficients are considered for lead
practice. CV aRc quantifies the expected value of the worst (1 − α )% times.
non-accepted patients. If we increase the value of parameter c, As shown in Fig 14, the lead time values has a main neg-
CV aRc accounts for the risk of higher number of non-accepted pa- ative impact on the output of resource planning during a pan-
tients. As a consequence, larger values for parameter c result in demic in both of risk-averse and risk-neutral policies. However,
larger values for CV aRc . Here, we have investigated two values for presented results show that the risk-averse policy has a more sta-
c, including 0.8 and 0.95, in case study 1, and the corresponding bility against the increase of lead times in terms of non-accepted
results are reported in Table 5. patients amount. If policy makers are able to decrease the lead
Here, we report the new resources allocation as well as the time values, the responsiveness of healthcare systems would be
number of non-accepted patients in both of case study 1 and 2 improved during a pandemic.
under risk-averse and risk-neutral policies. Furthermore, we com-
pare our results by the obtained policies from two-stage stochas- 6.5. The impacts of interventions on resources need
tic programming and deterministic models to highlight the impor-
tance of the uncertainty consideration and multi-stage stochastic Regarding case study 2, on several occasions, Iranian gover-
programming approach. In the two-stage stochastic model, all al- nors announced stay-at-home orders aimed to slow the spread of
location and sharing decisions should be made at the beginning of COVID-19. The policy makers’ orders were effective in reducing
the planning horizon. In Tables 6 and 7, the results related to case transmissions and can be captured by the reduction in estimated
study 1 and 2 are reported, respectively. effective reproduction number of the disease.

12
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

prove the required resources about 21% while the improvement

related to the usage of the single resource sharing and patients’
transfer strategy are about 17% and 12%, respectively.
The proposed data-driven decision-making framework can help
decision makers adjust their decisions in real-time based on the
past observation of uncertain parameters and their prior decisions.
We highlight the importance of the information update related the
release time of healthcare resources. In the accessible historical
data, the release time of ventilators used for the COVID-19 pa-
tients follows a log-normal distribution and varies largely in dif-
ferent cases; our data-driven decision-making framework improves
the additional capacity allocation in both case study 1 and 2 by
real-time decisions.
We highlight the significance of the MSSP in compared with de-
terministic and two-stage stochastic programming model. Further,
Fig. 14b. Sensitivity of resource allocation to lead-times.
by developing the risk-averse MSSP with the CVaR of non-accepted
In this part, we investigate the effect of these interventions on patients instead of its expected value, we investigate the risk-
optimal policy and resources need. In particular, we model two neutral and risk-averse decision-making in healthcare resources
scenarios in which the transmission rate from the considered hori- planning. Our experimental results show that the risk-averse deci-
zon is 50% higher or lower compared to the observed rate. If in- sions make the expected allocation of additional resources worse,
terventions were less effective, ventilator and nurse needs derived its standard deviation lower, and reduces the amount of non-
from our optimal policy would increase 18% and 12%, respectively. accepted patients in our case studies. Therefore, in many practi-
This is because less effective interventions would result in more in- cal situations, we can increase the robustness of our decisions by
fections and, subsequently, a higher demand for resources. On the employing a risk-averse objective function.
other hand, if interventions were more effective, the need for ven- Finally, interventions like closure of non-essential business,
tilators and nurses from our optimal policy would decrease 7% and mask wearing, and social distancing protocols are very effective in
5%, respectively. reducing transmissions. Such mandates are captured by consider-
ing different values for the effective reproduction number in our
7. Managerial insights simulator, and we show these strategies can meaningfully reduce
the required additional resources in hospitals.
To meet potential surges in healthcare resources’ demand un-
der pandemics, our optimization model considers various types of
patients as well as resources (DUR and SER) simultaneously for the 8. Conclusion
first time in the literature. Further, the model considers the shar-
ing strategy and patients’ transfers to avoid non-accepted patients We introduce a new integrated resource sharing and de-
as much as possible by using the fewest additional resources. The mand redistribution problem during pandemics. Our optimiza-
presented optimization setting is flexible and applicable for differ- tion is applicable for various patient types and required health-
ent healthcare resources planning problems, which is confirmed by care resources. Under a multi-period setting, an MSSP with non-
our computational results for two real examples. anticipativity constraints is developed to obtain the optimal shar-
Our empirical results provide insight into how hospitals in dif- ing, patients’ transfer, and capacity allocation decisions.
ferent regions could cope with the increase in demand for health- Methodologically, we formulate our problem as a mixed-integer
care resources, which results from local surges in infections during linear programming model, which is solvable by the CPLEX as a
COVID-19. Based on the computational results, by sharing ventila- commercial solver. The real-world applicability of the proposed
tors among regions of an area of the USA, the average number of MSSP is deeply investigated by two real case studies. A new data-
ventilators allocation to regions decreases about 58%. In the sec- driven decision-making approach is developed to implement the
ond case study (hospitals of a healthcare region of Iran), it de- decisions made by the MSSP in real-time. This approach enables
creases about 17%. The main reason for different significance of decision-makers to employ the data that is realized over time and
the ventilators sharing strategy between these two cases is that to adjust the corresponding decisions in a rolling horizon frame-
the infection spreads at varying rates in different regions in case work.
study 1. Therefore, this provides an opportunity for sharing scarce In the computational results, we illustrate the validity of our
resources such as ventilators, which can be transported over large model and its importance in resource planning during COVID-19
distances within regions to alleviate capacity shortfalls caused by pandemic. Further, the significance of the MSSP is compared with
an epidemic surge in a region. More importantly, governors and both deterministic and two-stage stochastic programming models,
decision makers have to ‘find’ significantly more ventilators to cope data-driven decisions, the sharing of healthcare resources, the de-
with the demand in the absence of sharing, which will take time; mand redistribution, and risk-averse decisions are discussed and
several patients don’t have time during pandemics. analyzed. Our decision-making framework showcases its capabil-
Based on our results from solving case study 2, we can find ities and flexibility with its exceptional performance in reducing
patients’ transfer as another efficient strategy in reducing the to- required new healthcare resources during pandemics.
tal required capacity in hospitals. In order to facilitate patients’ To capture the demand uncertainty and create an efficient sce-
transfer between hospitals to manage the healthcare systems’ ca- nario tree in our optimization problem, a simulation approach
pacity, an infrastructure is necessary for transshipping patients be- based on an agent-based continuous-time stochastic model is used
tween hospitals in a short time. Since the patients’ transfer has not to model the disease spread. Next, by applying the forward sce-
any lead time compared with resource transshipments in our case nario tree construction technique, we reduce the scenarios’ num-
study, it is more impactful in terms of the minimization of non- ber and convert them into a scenario tree. The efficiency of this
accepted patients (See Table 4). On the other hand, the integrated method is confirmed by in-sample and out-of-sample stability
use of sharing resources and demand redistribution strategies im- analysis.

13
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

Although our work is motivated by healthcare operations man- Hegazy, T. (1999). Optimization of resource allocation and leveling using ge-
agement under pandemics, our models and insights can also be netic algorithms. Journal of Construction Engineering and Management, 125(3),
167–175.
applied to other service industries under massive increases in de- Heitsch, H., & Romisch, W. (2005). Generation of multivariate scenario trees to
mand. Moreover, this study has a few limitations, which can be model stochasticity in power management. In 2005 IEEE Power Tech Russia Con-
addressed by future researches. In our model, we do not consider ference Proceedings (pp. 1–7). IEEE.
Hick, J. L., Hanfling, D., Burstein, J. L., DeAtley, C., Barbisch, D., Bogdan, G. M.,
priorities in responding to patients in the case of shortfalls and pa- et al. (2004). Health care facility and community strategies for patient care
tients’ preferences in selecting hospitals. surge capacity. Annals of Emergency Medicine, 44(3), 253–261.
Homem-de-Mello, T., & Pagnoncelli, B. K. (2016). Risk aversion in multistage
References stochastic programming: A modeling and algorithmic perspective. European
Journal of Operational Research, 249(1), 188–199.
Adan, I., Bekkers, J., Dellaert, N., Jeunet, J., & Vissers, J. (2011). Improving operational Hu, Z., Song, C., Xu, C., Jin, G., Chen, Y., Xu, X., et al. (2020). Clinical characteristics
effectiveness of tactical master plans for emergency and elective patients under of 24 asymptomatic infections with COVID-19 screened among close contacts in
stochastic demand and capacitated resources. European Journal of Operational Nanjing, China. Science China Life Sciences, 63(5), 706–711.
Research, 213(1), 290–308. Kaji, A., Koenig, K. L., & Bey, T. (2006). Surge capacity for healthcare systems: A
Adelman, D. (2020). Thousands Of Lives Could Be Saved In The US During The conceptual framework. Academic Emergency Medicine, 13(11), 1157–1159.
COVID-19 Pandemic If States Exchanged Ventilators: Study examines how lives Kall, P., & Wallace, S. W. (1994). Stochastic programming p. 307. Chichester: Wiley.
could be saved by allowing US states to exchange ventilators during the Keyvanshokooh, E., Fattahi, M., Zokaeinikoo, M., Freedberg, K. A., &
COVID-19 pandemic. Health Affairs, 39(7), 1247–1252. Kazemian, P. (2020). Mitigating the COVID-19 Pandemic through Data-Driven
Ahmed, S. (2006). Convexity and decomposition of mean-risk stochastic programs. Resource Sharing (working paper) [Link]
Mathematical Programming, 106(3), 433–446. abstract_id=3703181.
Altay, N., & Green, W. G., III (2006). OR/MS research in disaster operations manage- Levin, A. T., Hanage, W. P., Owusu-Boaitey, N., Cochran, K. B., Walsh, S. P., &
ment. European Journal of Operational Research, 175(1), 475–493. Meyerowitz-Katz, G. (2020). Assessing the age specificity of infection fatality
Bakuli, D. L., & Smith, J. M. (1996). Resource allocation in state-dependent emer- rates for COVID-19: Systematic review, meta-analysis, and public policy impli-
gency evacuation networks. European Journal of Operational Research, 89(3), cations. European Journal of Epidemiology, 1–16.
543–555. Lewnard, J. A., Liu, V. X., Jackson, M. L., Schmidt, M. A., Jewell, B. L., Flores, J. P.,
Bertsimas, D., Boussioux, L., Cory-Wright, R., Delarue, A., Digalakis, V., Jacquillat, A., et al. (2020). Incidence, clinical outcomes, and transmission dynamics of se-
Kitane, D. L., Lukin, G., Li, M., Mingardi, L., & Nohadani, O. (2021). From predic- vere coronavirus disease 2019 in California and Washington: Prospective cohort
tions to prescriptions: A data-driven response to COVID-19. Health care manage- study. BMJ (Clinical research ed.), 369.
ment science, 1–20. Litvak, N., Van Rijsbergen, M., Boucherie, R. J., & van Houdenhoven, M. (2008). Man-
Centers for Disease Control and Prevention National Center for Immunization and aging the overflow of intensive care patients. European Journal of Operational
Respiratory Diseases. (2020). Overall Percentages of Visits for ILI and Percentage Research, 185(3), 998–1010.
of Visits for ILI by Age Group Reported by A Subset of ILINet Providers. Cent. Dis. Liu, Y., Gayle, A. A., Wilder-Smith, A., & Rocklöv, J. (2020). The reproductive number
Control Prev.. [Link] of COVID-19 is higher compared to SARS coronavirus. Journal of Travel Medicine.
Chi, H., Li, J., Shao, X., & Gao, M. (2017). Timeliness evaluation of emergency re- Ludwig, M., Jacob, J., Basedow, F., Andersohn, F., & Walker, J. (2021). Clinical out-
source scheduling. European Journal of Operational Research, 258(3), 1022–1032. comes and characteristics of patients hospitalized for Influenza or COVID-19 in
Dasaklis, T. K., Pappis, C. P., & Rachaniotis, N. P. (2012). Epidemics control and logis- Germany. International Journal of Infectious Diseases, 103, 316–322.
tics operations: A review. International Journal of Production Economics, 139(2), Luscombe, R., & Kozan, E. (2016). Dynamic resource allocation to improve emer-
393–410. gency department efficiency in real time. European Journal of Operational Re-
Drevs, F. (2013). How patients choose hospitals: Using the stereotypic content search, 255(2), 593–603.
model to model trustworthiness, warmth and competence. Health Services Man- Mehrotra, S., Rahimian, H., Barah, M., Luo, F., & Schantz, K. (2020). A model of sup-
agement Research, 26(2–3), 95–101. ply-chain decisions for resource sharing with an application to ventilator allo-
Dupačová, J. (1995). Multistage stochastic programs: The state-of-the-art and se- cation to combat COVID?19. Naval Research Logistics (NRL), 67(5), 303–320.
lected bibliography. Kybernetika, 31(2), 151–174. Mervosh, S., Lu, D., & Swales, V. (2020). See which states and cities have told resi-
Dupačová, J., Gröwe-Kuska, N., & Römisch, W. (2003). Scenario reduction in stochas- dents to stay at home. The New York Times, 3.
tic programming. Mathematical Programming, 95(3), 493–511. Mills, A. F., Helm, J. E., & Wang, Y. (2020). Surge capacity deployment in hospitals:
Ekici, A., Keskinocak, P., & Swann, J. L. (2014). Modeling influenza pandemic and Effectiveness of response and mitigation strategies. Manufacturing & Service Op-
planning food distribution. Manufacturing & Service Operations Management, erations Management.
16(1), 11–27. Mizumoto, K., Kagaya, K., Zarebski, A., & Chowell, G. (2020). Estimating the asymp-
Elmaghraby, S. E. (1993). Resource allocation via dynamic programming in activity tomatic proportion of coronavirus disease 2019 (COVID-19) cases on board the
networks. European Journal of Operational Research, 64(2), 199–215. Diamond Princess cruise ship, Yokohama, Japan, 2020. Eurosurveillance, 25(10),
Erdogan, S. A., & Denton, B. (2013). Dynamic appointment scheduling of a stochastic Article 20 0 0180.
server with uncertain demand. INFORMS Journal on Computing, 25(1), 116–132. Nabi, K. N. (2020). Forecasting COVID-19 pandemic: A data-driven analysis. Chaos,
Farley, H. L., Baumlin, K. M., Hamedani, A. G., Cheung, D. S., Edwards, M. R., Solitons & Fractals, 139, Article 110046.
Fuller, D. C., et al. (2013). Quality and safety implications of emergency depart- Otegbeye, M., Scriber, R., Ducoin, D., & Glasofer, A. (2015). Designing a data-driven
ment information systems. Annals of Emergency Medicine, 62(4), 399–407. decision support tool for nurse scheduling in the emergency department: A case
Fattahi, M., & Govindan, K. (2018). A multi-stage stochastic program for the sustain- study of a southern New Jersey emergency department. Journal of Emergency
able design of biofuel supply chain networks under biomass supply uncertainty Nursing, 41(1), 30–35.
and disruption risk: A real-life case study. Transportation Research Part E: Logis- Parker, F., Sawczuk, H., Ganjkhanloo, F., Ahmadi, F., & Ghobadi, K. (2020). Optimal
tics and Transportation Review, 118, 534–567. Resource and Demand Redistribution for Healthcare Systems Under Stress from
Fattahi, M., & Govindan, K. (2020). Data-driven rolling horizon approach for dynamic COVID-19. [Link]
design of supply chain distribution networks under disruption and demand un- Perc, M., Gorišek Miksić, N., Slavinec, M., & Stožer, A. (2020). Forecasting covid-19.
certainty. Decision Sciences In press. Frontiers in Physics, 8, 127.
Fattahi, M., Govindan, K., & Keyvanshokooh, E. (2018). A multi-stage stochastic pro- Reddy, K. P., Shebl, F. M., Foote, J. H., Harling, G., Scott, J. A., Panella, C., et al. (2020).
gram for supply chain network redesign problem with price-dependent uncer- Cost-effectiveness of public health strategies for COVID-19 epidemic control in
tain demands. Computers & Operations Research, 100, 314–332. South Africa: A microsimulation modelling study. The Lancet Global Health.
Ferreira, F. A., Kannan, D., Meidutė-Kavaliauskienė, I., & Vale, I. M. T. (2022). A so- Rockafellar, R. T., & Uryasev, S. (2002). Conditional value-at-risk for general loss dis-
ciotechnical approach to vaccine manufacturer selection as part of a global im- tributions. Journal of Banking & Finance, 26(7), 1443–1471.
munization strategy against epidemics and pandemics. Annals of Operations Re- Rodríguez-Espíndola, O., Albores, P., & Brewster, C. (2018). Disaster preparedness in
search (in press). humanitarian logistics: A collaborative approach for resource management in
Fiedrich, F., Gehbauer, F., & Rickers, U. (20 0 0). Optimized resource allocation for floods. European Journal of Operational Research, 264(3), 978–993.
emergency response after earthquake disasters. Safety Science, 35(1–3), 41–57. Rothman, R. E., Hsu, E. B., Kahn, C. A., & Kelen, G. D. (2006). Research priorities for
Govindan, K., Fattahi, M., & Keyvanshokooh, E. (2017). Supply chain network design surge capacity. Academic Emergency Medicine, 13(11), 1160–1168.
under uncertainty: A comprehensive review and future research directions. Eu- Ruszczyński, A. (2010). Risk-averse dynamic programming for Markov decision pro-
ropean Journal of Operational Research, 263(1), 108–141. cesses. Mathematical Programming, 125(2), 235–261.
Govindan, K., Mina, H., & Alavi, B. (2020). A decision support system for demand Shapiro, A. (2009). On a time consistency concept in risk averse multistage stochas-
management in healthcare supply chains considering the epidemic outbreaks: tic programming. Operations Research Letters, 37(3), 143–147.
A case study of coronavirus disease 2019 (COVID-19). Transportation Research Silva, P. C., Batista, P. V., Lima, H. S., Alves, M. A., Guimarães, F. G., &
Part E: Logistics and Transportation Review, 138, Article 101967. Silva, R. C. (2020). COVID-ABS: An agent-based model of COVID-19 epidemic to
Gupta, S., Starr, M. K., Farahani, R. Z., & Matinrad, N. (2016). Disaster management simulate health and economic effects of social distancing interventions. Chaos,
from a POM perspective: Mapping a new domain. Production and Operations Solitons & Fractals, 139, Article 110088.
Management, 25(10), 1611–1637. Sung, I., & Lee, T. (2016). Optimal allocation of emergency medical resources in a
Haridy, R. (2020). CDC director warns 25 percent of COVID-19 cases may mass casualty incident: Patient prioritization by column generation. European
present no symptoms. [Link] wellbeing/covid- 19- cases- Journal of Operational Research, 252(2), 623–634.
contagious-asymptomatic-presymptomatic-cdc-director/. Toner, E., & Waldhorn, R. (2006). What hospitals should do to prepare for an in-

14
JID: EOR
ARTICLE IN PRESS [m5G;February 28, 2022;16:56]

M. Fattahi, E. Keyvanshokooh, D. Kannan et al. European Journal of Operational Research xxx (xxxx) xxx

fluenza pandemic. Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Willis, G., Cave, S., & Kunc, M. (2018). Strategic workforce planning in healthcare: A
Science, 4(4), 397–402. multi-methodology approach. European Journal of Operational Research, 267(1),
Tonna, J. E., Hanson, H. A., Cohan, J. N., McCrum, M. L., Horns, J. J., Brooke, B. S., 250–263.
et al. (2020). Balancing revenue generation with capacity generation: Case dis- World Health Organization. (2018). Managing epidemics: Key facts about major deadly
tribution, financial impact and hospital capacity changes from cancelling or re- diseases. World Health Organization.
suming elective surgeries in the US during COVID-19. BMC Health Services Re- Yang, Y., Yang, M., Shen, C., Wang, F., Yuan, J., Li, J., et al. (2020). Laboratory diagnosis
search, 20(1), 1–7. and monitoring the viral shedding of 2019-nCoV infections. MedRxiv.
Varkevisser, M., van der Geest, S. A., & Schut, F. T. (2012). Do patients choose hos- Yin, X., & Büyüktahtakin, İ. E. (2021). Risk-averse multi-stage stochastic program-
pitals with high quality ratings? Empirical evidence from the market for angio- ming to optimizing vaccine allocation and treatment logistics for effective
plasty in the Netherlands. Journal of Health Economics, 31(2), 371–378. epidemic response. IISE Transactions on Healthcare Systems Engineering, 1–52
Wang, W., Xu, Y., Gao, R., Lu, R., Han, K., Wu, G., et al. (2020). Detection of SARS– (just-accepted).
CoV-2 in different types of clinical specimens. JAMA, 323(18), 1843–1844. Zhou, F., Yu, T., Du, R., Fan, G., Liu, Y., Liu, Z., et al. (2020). Clinical course and risk
Wex, F., Schryen, G., Feuerriegel, S., & Neumann, D. (2014). Emergency response in factors for mortality of adult inpatients with COVID-19 in Wuhan, China: A ret-
natural disaster management: Allocation and scheduling of rescue units. Euro- rospective cohort study. The Lancet, 395(10229), 1054–1062.
pean Journal of Operational Research, 235(3), 697–708.

15