key: cord-1038631-gxmm6ex5 authors: Gilani, Hani; Sahebi, Hadi title: A Data-Driven Robust Optimization Model by Cutting Hyperplanes on Vaccine Access Uncertainty in COVID-19 Vaccine Supply Chain date: 2022-03-11 journal: Omega DOI: 10.1016/j.omega.2022.102637 sha: 19ca9b8f75ac35d1524f31a7ce4f9752c79873b2 doc_id: 1038631 cord_uid: gxmm6ex5 The worldwide COVID-19 pandemic sparked such a wave of concern that made access to vaccines more necessary than before. As the vaccine inaccessibility in developing countries has made pandemic eradication more difficult, this study has presented a mathematical model of a sustainable SC for the COVID-19 vaccine that covers the economic, environmental and social aspects and provides vaccine both domestically and internationally. It has also proposed a robust data-driven model based on a polyhedral uncertainty set to address the unjust worldwide vaccine distribution as an uncertain parameter. It is acceptably robust and is also less conservative than its classical counterparts. For validation, the model has been implemented in a real case in Iran, and the results have shown that it is 21% less conservative than its classical rivals (Box and Polyhedral convex uncertainty sets) in facing the uncertain parameter. As a result, the model proposes the construction of two domestic vaccine production centers, including Pasteur Institute and Razi Institute, and five foreign distributors in Tehran, Isfahan, Ahvaz, Kermanshah, and Bandar Abbas strategically. In the stochastic approach, the key role is played by the distribution function, but in the robust method [22, 23] , the "uncertainty set" plays the main role in finding more appropriate solutions; the better is the set selected, the better will be the solutions; hence, set selection is vital to the quality of the solutions of the mathematical model. An appropriate "uncertainty set" is that obtained based on various hypotheses and empirical validations. Recent years' information explosion and expansion of new data-collection technologies have revolutionized dealing with uncertain parameters. The great access to the information collected from different industries is quite helpful in health systems which are usually affected more due to the uncertain nature of such parameters as the accessibility, demand and system efficiency. Data can make a significant change in robust optimization, which deals with a set of uncertain parameters, and enable, more than before, the implementation of realistic and sustainable cases against the fluctuations of these parameters. This study that presents a multi-level, multi-period COVID-19 vaccine distribution supply chain MILP model, involves 2 innovations: 1) operational, where the economic, environmental and social dimensions are considered in objective functions 1, 2 and 3, respectively, to create a sustainable model, and the chain addresses both the internal and external supplies, and 2) theoretical that deals with the uncertain accessibility of foreign vaccines. It presents a data-driven robust optimization (DDRO)-Cutting planes approach in agreement with uncertain parameters where the "uncertainty set" is polyhedral to provide enough flexibility for the readers and stakeholders in the intended space. The rest of the paper has been so organized as to review the literature in Section 2 (focusing on the robust optimization of the vaccine distribution SCND model under pandemic conditions), implement the sustainable SC in Section 3, describe various forms of the proposed approach, using cutting planes, in Section 4 (where the uncertainty set is modeled with a polyhedral set) and describe the case study in Section 5. A review of major studies on the robust optimization and vaccine distribution SC models under pandemic conditions shows that reliable SCND facility locating models (with disorder confrontation approach) were first studied by [24] and, then, [25] conducted a review study on the robust SCND and categorized the uncertain nature of the SC parameters. During these years, there were other studies [26, 27] that examined other SC features. The current study has proposed a robust SCND model for a sustainable vaccine distribution that considers the economic, environmental and social dimensions. Here, the important point is how the model deals with uncertain parameters because their nature has further strengthened the use of robust data-driven optimization approaches discussed next. Robust optimization was first proposed by [28] in two dimensions: 1) optimality robustness which means the model will be close to optimal and 2) feasibility robustness which means the model will stay close to the feasible space in almost all cases of the realized solutions; "close" depends highly on how the modeler defines it. While stochastic approaches are impractical in uncertain environments where the distribution function of uncertain parameters is not known, robust optimization methods are quite practical because uncertain parameters can be modeled continuously or as discrete scenarios. Robust optimization models in the literature are classified as: 1) scenario-based, 2) fuzzy programming-based and 3) uncertainty interval-based. In group 1, robustness measuring criteria minimize both the maximum cost and maximum regret (values farthest from the optimum). Some studies have minimized: 1) the maximum regret in the RSCND (closed-loop design) [29, 30] , 2) the average regret [31] and 3) the maximum cost [32, 33] , and some [34, 35, 36] have used constraints to ensure that the regret may not exceed a certain limit. Percent risk is another criterion used to measure the robustness of SC models [37, 38, 39] . In group 2, robust models are based on fuzzy sets. In SCNDs with uncertain parameters, the literature has repeatedly made use of the fuzzy programming based either on "flexibility" in operators, especially constraints, or "possibility" of the distribution of uncertain parameters. All studies (e.g., [41, 42, 43] ) on these two methods have been solved in a general structure that converts fuzzy models to crisp models [40] . In group 3 that relates to uncertainty interval-based models, when uncertainty is associated with intervals, robust optimizations generally involve a high degree of conservatism; in other words, the problem infeasibility is highly protected. Study [44] was the first in this group that was aimed to present a model that kept the problem feasible for almost each feasible solution in a predefined set for all uncertain parameter realizations, but its problem was its high conservatism level. Studies [45] and [46] were proposed to reduce this level, but their approaches involved highly nonlinear formulations. For a better performance, Bertsimas [47] proposed a different approach where conservatism was controllable and, most importantly, its formulations were linear. While all previous models made real-world problem-decisions in only one step, subsequent studies [48] showed that they were usually made in several steps. "Distributionally robust optimization" is the title recent studies [49, 50] have given to the uncertain distribution function in the robust optimization. The vaccine SC in pandemic conditions has severe operational and time constraints. Uscher et al. [51] studied the influenza vaccine in 48 countries in 28 of which the high-risk vulnerable groups had the injection priority to reduce large-scale pandemic risks. To distribute the H1N1 vaccine, Araz et al. [52] divided the state of Arizona into 15 districts prioritizing 4 based on the estimates of the pandemic peak periods, population-based distribution sequence and a number of other factors. Considering communications and ethics, Medlock and Galvani [53] developed an influenza vaccine distribution model that based allocations on such criteria as death, infection, age, and economic costs. Believing that the 6-12 year-old children were the cause of the inter-individual virus transmission because of their high activity/contact, in spring 2009 in Mexico, Lee et al. [54] considered the 20-39 age range eligible to receive H1N1 vaccine and developed a nonlinear dynamic model for its optimal allocation. Prioritizing the helpless and homeless for the H1N1 virus vaccination during a pandemic, Buccieri and Guetz [55] set up a case in Toronto, Canada, and vaccinated 38% of them in the existing clinics. Identifying the vaccine supply-distribution system as the most important factor, Davila-Payan et al. [56] developed a multivariate linear regression model to vaccinate the children and high-risk people against the H1N1 virus pandemic. Using a push-based strategy to distribute the vaccine in different areas, Huang at al. [57] conducted a study in Texas, USA, to categorize pregnant women, infants, and high-risk people as the priority groups. Chen et al. [58] developed an age-based simulation model to find the optimal COVID-19 vaccine allocation strategy by dividing the population into seven parts each with five age groups. They examined both dynamic and static policies and showed that first the older and then the younger groups had to be vaccinated. An optimal vaccine SCND requires establishing an effective supplier-health department relationship [59] . In this regard, Marcello et al. [60] conducted a study in NY, USA, in 2009 on the vaccine distribution in an H1N1 pandemic using the data of medical centers to estimate the number of the distributed vaccines and showed that 96% of the suppliers were satisfied with the information recorded for the vaccine SC management. Covering about 89% of the drug stores in their study area for vaccine distribution, Fitzgerald et al. [61] conducted a study on the coordination/cooperation of suppliers, drug stores and public health centers across the US. In a study aimed to examine the vaccine supply-distribution methods used by governments in developed countries, Turner [62] showed that the traditional influenza vaccine distribution policies were weak and worrying tools and involved significant delays. To analyze different vaccination scenarios in the USA to prevent people from being hospitalized and die, Biggerstaff et al. [63] implemented a spreadsheet-base model the variables of which were the number of patients, different vaccination times, number of doses prescribed for various age groups and the hospitalization rate. They found that the starting time of the pandemic and vaccination were very influential factors in the vaccination program. In a recent study, Govindan et al. [64] used the Fuzzy Inference System (FIS) for the demand management in healthcare systems to help reduce the devastating effects of COVID-19 disease by dividing the community into four groups based on the immunity system -age and underlying diseasesand showed that their classification helped fighting the disease in health systems. Tavana et al. [65] presented a mixed-integer linear programming model for equitable COVID-19 vaccine distribution in developing countries, Shim [66] , Roy et al. [67] and Abbasi et al. [68] optimized the allocation of limited COVID-19 vaccine supply, Santini [69] assigned swabs and reagent for PCR testing, and other studies [70, 71] considered the service level for herd immunity and capacity planning in the COVID-19 vaccine SC; however, none investigated the uncertain vaccine accessibility in developing countries. On the pandemic crisis, recently, several review studies have been conducted based on challenges of the COVID-19 pandemic situation in vaccine resilience supply chain and ripple effect in supply chains, including [72] , [73, 74] , and [75] . They categorized challenges in the literature and via vaccine supply chain (VSC) that the highest-importance are as follows:  Lack of vaccine monitoring bodies  Limited number of vaccine manufacturing companies  Unavailability of vaccines for developing countries  Lack of proper planning and scheduling  Vaccination cost and lack of financial support for vaccine purchase  Lack of correspondence between the VSC members Considering supply chain challenges before a vaccine is administered to the general population can help vaccines successfully distribute vaccines. Therefore, solving critical challenges to the COVID-19 VSC is customary for a sustainable VSC [76] that could help the countries around the world to get out of the pandemic. So, research papers must follow to solve these challenges in the vaccine supply chain as this paper is organized to this goal. Studies on the vaccine SC under worldwide pandemic conditions (supposed to continue till 2022), as well as the world vaccine suppliers, suggest that COVID-19 vaccine SCNDs are necessary, more than ever, to vaccinate global communities. But, supplying vaccines for less developed communities have an uncertain nature because the demand is high and suppliers are limited [77] . Here, robust models that yield robust solutions against uncertain vaccine-accessibility parameters seem appropriate because they do not need distribution functions for uncertain parameters; however, their high conservatism that imposes high costs on the system is a main weakness that may hinder their implementation, but being in the age of data and using the merits of robust data-driven optimization models can be a key to the solution of this problem. As COVID-19 has long epidemic periods, sustainability is an issue worth considering in its vaccine SC, and since the literature is rather poor in robust studies that discuss the SC's economic, environmental and social dimensions, the present study has considered the mentioned dimensions and addressed the COVID-19 vaccine SCND at both the strategic level (construction site and warehouse capacity decisions) and operational level (inventory and product flow decisions). Its main innovations are as follows:  Addressing robust models' high conservatism through a robust data-driven model.  Using cutting hyperplanes to implement more accurate and realistic data in the mentioned model.  Comparing the proposed DDRO approach with the box and polyhedral uncertainty set.  Supplying vaccine from several different domestic/foreign suppliers.  Addressing sustainability in three vaccine SC indicators by analyzing the environmental and social effects.  Implementing the social dimension in three indicators (No. of created jobs, days lost due to damage, potential hazards)  Implementing a case study in Iran to validate the proposed approach. The proposed problem addresses the network design of an integrated producer-to-consumer SC that sends the vaccine to the end customer while packed and stored under its own specific conditions considering the vaccine flow and right supplier selection. One way to prevent infection to spread in the human body during a COVID-19 pandemic is vaccination by which communities can ensure long life, health and well-being (Pfizer, 2017), but since its outbreak is vast and multiple, the production, distribution and quality control of its vaccine have become a serious issue. As the pandemic is now prevailing, minimizing its related complexities through a producer-distributorconsumer coordination can be effective in controlling the people's health [78] . The vaccine SC complexities are due to the production difficulties (manufacturing, mass production, etc.), globalization (global-level supply-demand balancing leading to adequate distribution), and observation of rules and regulations governing the development and manufacture of the vaccine. Since the international community needs public safety, the vaccine production and distribution trends will move towards higher quality and compliance with the related standards. Hence, to better respond to the complexities of the vaccine distribution network, it is necessary to optimize the planning and implementation of the SC network structure. As production and distribution processes are two main vaccine SC features that make collective safety and health accessible to communities, they need special attention for design and planning. The SC in Fig. 1 includes domestic and foreign suppliers, international packaging centers, storage and distribution centers and provincial health centers as demand points. The chain, considering accessibility to a limited capacity of domestic vaccines, transfers them from the main producer in a foreign country to international packaging centers in the same country, then sends them to the destination country after packaging, keeps them in the storage and distribution centers in that country and, finally, transfers them to central health centers as points of demand by internal transportation. COVID-19 vaccine can be stored in the production, packaging storage and distribution centers and the model determines its optimum storage level during the planning time horizon. The vaccine storage is possible by refrigerating and freezing and construction of storage and distribution centers can be based on either of them. The SC inter-level goods transportation is both internal (by road/railway) and external (sea/air). In the proposed SC, decisions are strategic and tactical (optimized by the model); the former determines the site/capacity of the storage/distribution/packaging centers and the related technologies and the latter include inventory, rate of product flow, supplier selection (in each time period) and the transportation mode. The proposed mathematical model has been so formulated as to minimize costs and environmental effects and maximize the positive effects of the SC activities as its social responsibility. It uses a robust optimization model to deal with the uncertain nature of supplying foreign vaccines for developing countries, but since its conservatism level is high, a DDRO-cutting planes model has been proposed for its solution. The measuring criteria of the social/environmental effects are examined in subsequent lines. The COVID-19 vaccine SCND sustainability requires the effects of all its activities to be analyzed both environmentally and socially [79, 80, 81] ; the former needs an accurate and standard measurement criterion from the highest to the lowest chain level. For this purpose, the life cycle assessment (LCA) is the most common structure founded based on ISO 14040 and 14044, but since its direct use is both costly and time consuming, use is made of the "ReCiPe" method that calculates the environmental effects based on such activities as transportation, producing various materials and inventory maintenance using the SimaPro commercial software and an Eco-indicator 99 Standard-based updating database [82] ; the same have been used in the current study to assess the environmental effects. Measuring the social responsibility is generally difficult because it has a complex nature and lacks a specific criterion and depends on several stakeholders and disciplines each with its own size and criterion. Eventually, as organizations and companies considered the sustainable SC development an important issue, the International Standard Organization (ISO) recently introduced the ISO2600 measurement standard which is a measuring criterion for the social responsibility and is classified into: 1) organizational governance, 2) human rights, 3) labor practices, 4) environment, 5) fair operating practices, 6) consumer issues and 7) community involvement and development. This study has tried to accurately estimate the chain's social impacts by identifying the related criteria in the COVID-19 vaccine production/distribution SC and use them in the mathematical modeling. According to ISO2600, stakeholder selection is of special importance as it is the first step in identifying these criteria. Table 1 that lists the concerns for categorizing these criteria has three social responsibility measurement criteria to find the social effects of the COVID-19 vaccine SC, which are integrated by weighting coefficients that not only point to the importance of the criteria, but are also used most to help integrate different criteria with different measurement units. Following the Covid-19 pandemic and losses of different jobs worldwide, developing novel SCs for its vaccines can effectively create new jobs to avoid its adverse social effects, but the pandemic's temporary nature makes the job security a critical issue for the employees. Expert human resources working in vaccine production plants are usually exposed to being contaminated with chemical products; these are issues necessary to consider for the SCs' social sustainability. Unique features of the proposed vaccine SC model that make it different from other models are: 1) establishing separate distribution and packaging centers for special vaccine maintenance, 2) showing the importance of the social dimension of a sustainable development, especially in the COVID-19 pandemic situation, 3) paying attention to different transportation modes in a suitable substrate for all routes for special vaccine maintenance, 4) considering uncertainty of the foreign vaccine accessibility (because of inter-buyers competitions), 5) paying attention to supplier selection based on several COVID vaccine suppliers all over the world and 6) supplying vaccines from domestic and foreign suppliers. Notations of the proposed mathematical model are provided in the Appendix and its objective functions are introduced next to develop its structural constraints. Based on the sustainability concept, the objective functions considered for COVID-19 vaccine SCND are economic, environmental and social. • Components of the economic objective function: where is the sum of the fixed costs of constructing: 1) international packaging centers for delivery to other countries, 2) distribution centers in the destination country and 3) domestic production centers. where is the variable costs of providing capacity in each SC facility; internationally, the capacity involves those of the packaging centers, domestic distribution centers and domestic producers in each period. where is the inventory costs for each facility for, respectively, those in the international manufacturers and packaging centers, and those in the domestic producers and distribution centers in the destination country in each period of the planning horizon. and (Eqs. 4, 5 and 6) show, respectively, the vaccine production, distribution and packaging costs in the related international and domestic centers in each period of the planning horizon. shows the shortage-related costs in each international production, packaging and distribution center and in domestic production centers in each period of the planning horizon. where is the capacity expansion cost assignable to each facility if the design requires the expansion of international production, packaging and distribution centers and domestic production centers, respectively. where (a prominent part of the economic objective function of costs associated with transportation) relates, respectively, to the costs of transportation from the international supplier to packaging centers, then to distribution centers in the destination country, then to provincial medical centers and also from domestic suppliers to medical centers in each period of the planning horizon. (10) ∑ ∑ ∑ (11) ∑ where (Eq. 10) is the supply costs of raw materials needed to produce the vaccine in both international and domestic centers in each time period and (Eq. 11) is the fixed costs of purchasing road vehicles. (12) where is the sum of the periodic costs. (13) ∑ ∑ which is the final COVID-19 vaccine SC economic objective function where Z 1 is minimized with the interest rate (IR) parameter and the net present value relation plus time-independent costs minus road vehicles' scrap costs. • Components of the environmental objective function: where is the environmental emissions due to vaccine transportation from the foreign supplier to packaging centers, then to distribution centers, then to medical centers and, finally, from the domestic producer to medical centers in each time period. where , EFFR and (Eqs. 15, 16, 17, respectively) are the environmental effects due to the production process in both domestic and foreign plants (Eq. 15), the construction of international packaging facilities, distribution centers and domestic production centers (Eq. 16) and the packaging process in international centers (Eq. 17) in each time period. where is the environmental effects of the vaccine maintenance in international production, packaging and distribution centers and domestic production centers in each time period. (19) where is the sum of the environmental effects of the time-dependent variables. (20) ∑ which is the final environmental objective function. • Components of the social objective function: where is the number of jobs created by transportation from the foreign manufacturer to the packaging center, then to the domestic storage and distribution centers, and then to provincial health centers, and also among domestic manufacturers and health centers. where is the number of jobs created in each domestic/foreign vaccine production center. is the number of jobs created based on the activities of international packaging centers and (Eq. 24) is that created in domestic distribution centers. is the first term of the social objective function and shows the sum of the jobs created in all SC processes. where is the second term of the social objective function and is related to the people's number of lost days due to injuries in the workplace. where is the last term of the social objective function and is related to the workplace risks calculated only in the production process in foreign and domestic centers. Now, all the current SC social effects are formulated as follows by giving weight to each component; maximizing number of jobs due to SC activities with weight (with +ive sign), number of days lost due to injuries with weight (with -ive sign for minimization) and environmental risks with a weight (with -ive sign for minimization): This section presents the model constraints on the capacity (facility and inventory), transportation, construction, and capacity expansion and set of constraints on vaccine flow and inventory balance.• This section deals with the constraints on the capacity to ensure that the capacity of each domestic/foreign production center is less than or equal to its designed capacity in each period; this also applies to international packaging and domestic distribution centers that depend on different storage technologies. Equations (29) (30) (31) (32) formulate these constraints for, respectively, foreign production centers, domestic production centers and packaging centers. The above formulae limit the production of each facility based on its designed capacity, but here, effort is made to rationally determine this capacity by determining the related upper and lower limits in the construction and economic dimensions. Eq. (33) specifies the design capacity of a foreign production center ( ) with two upper and lower limits ( ، ); this trend is repeated for domestic production centers, foreign packaging centers and distribution centers in Equations (34) (35) (36) . Here, capacity-related constraints are considered for centers capable of keeping inventory. Eq. (37) limits the foreign supplier inventory ( ) by maximizing the amount storable in that center ( ). These constraints have also been formulated for domestic packaging, distribution and production centers in Eqs. (38) (39) (40) . Constraints related to the flow of raw materials to domestic and foreign production centers are next. It should be ensured that the raw material required for vaccine production should be less than or equal to the maximum amount available in each time period. Eq. (41) is related to the raw material supply for foreign producers and Eq. (42) is related to that for domestic producers. Capacity expansion is an important concept aimed to create a progressive capability for the capacity of the designed facility to be available when needed to better meet the demand. To formulate this concept, use is made of the designed capacity in each period which equals that of the previous period plus the capacity expansion in that period. This definition has been formulated for all four types of facilities (domestic and foreign production centers, packaging centers and distribution centers) in Eqs. (43) (44) (45) (46) . As mentioned before, international transportation is by sea and air and the domestic transportation is by railway and road. Constraints related the capacity of the transportation fleet are ( ) and those on accessibility to a sufficient number of vehicles are ( ). Eqs (47) (48) relate to the vaccine transportation within each country from production centers to packaging centers and from distribution centers to provincial medical centers, Eq. (49) is used for the same fleet for domestically produced vaccine delivery and Eq. (50) is for international transportation by sea and air. Constraints discussed here try to balance the vaccine flow in different COVID-19 vaccine SC layers; formulas (51-55) try to match the demand and supply by balancing the flow and correct SC modeling. Eq. (51) uses the -type rate ( ) to calculate the vaccine production in foreign centers based on the available raw materials, Eq. (52) uses the vaccine packaging coefficient ( ) to determine the packages prepared for external shipment, Eq. (53) uses the distribution coefficient to convert part of the shipment to distribution packages ( ) after the international shipment and arrival of packages in the destination country and Eq. (54) uses the relevant factor ( ) to prepare the raw materials for vaccine production in a conversion process in domestic centers. Obviously, vaccines produced and sent from both foreign and domestic production centers to provincial medical centers should meet the demand of each center in each period ( ), but since this parameter is uncertain, the accessibility of each vaccine in the competitive world market ( ̃ ), is calculated in Eq. (55) . Equations (56) (57) (58) (59) have been formulated to create balance in centers that keep inventory of goods. These values are calculated assuming that foreign and domestic production, packaging and distribution centers store inventories. Eq. (56) shows the inventory balance in foreign production centers, which should be equal to that center's outflow, that is, production in that period plus the reduced inventory minus the shortage. Equations (57) (58) (59) show the same concept for the packaging, distribution and domestic production centers. Eq. (60) shows the rational constraint on the selection of a vaccine storage technology in a distribution center and Eq. (61) re-determines the status of each variable. In dealing with uncertain parameters, the robust optimization has been generally popular with researchers in recent years and used more than previous uncertain stochastic programming approaches. This tendency has been greater in SCNDs due to high investments and information inaccessibility at the beginning of the design. Selecting robust optimization approaches is due, mainly, to their independence from the distribution function of uncertain parameters because they can provide an accurate estimate of such parameters by only knowing their variation interval; access to their distribution is not necessary. The robust optimization method proposed in this study has higher conservatism than its rivals; it defines a closed convex set for the problem's uncertain parameters and presents, on its basis, a closed formula of its robust counterpart. Here, the COVID-19 vaccine SC is executed by first introducing and formulating a robust optimization method based on closed uncertainty sets and then presenting the proposed approach. To see how to address set-oriented robust uncertain parameters, consider the following linear problem: where is a random variable belonging to set that has a symmetric distribution in the range [-1,1]. If this variable is multiplied by the variation rate of each uncertain parameter interval and its nominal value is added, the result will be a mathematical representation of the uncertain parameter based on its changes. Substituting the closed form obtained above in Eq. (63) for ̃ and ̃ will yield: where and belong to polyhedral set (Fig. 2) where for the values on the right. Using a random variable in the form " ", this set is defined with control parameter Γ as follows to limit the uncertainty space: ∑ Γ ̂ ̂ The approach discussed in the previous section belongs to the category of static robust optimization methods that have a fixed routine throughout the planning horizon. The main weakness of such methods is their high conservatism and paying special attention to the occurrence of the worst case of the uncertain parameter. In SCNDs, this weakness imposes high costs on the system mostly because of lying in a safe point preventing the occurrence of the predicted state with high probability. To address the high conservatism of static robust optimization approaches, researchers have, in recent years, developed new dynamic robust optimization methods that highly reduce conservatism in classical models by realizing some data of the uncertain parameters over time and limiting, on this basis, their set [84] . Fig (3) shows the decision-making structure in robust dynamic programming by a numerical example. Dynamism in the DDRO modeling based on closed convex sets reduces the conservatism level. Among conservatism reduction approaches, optimization methods, e.g. cutting hyperplanes, limit the space of the set of uncertain parameters based on the data realization [85] . After the realization of some of the data, these planes are plotted on the set of uncertain parameters, in the next time period, based on these data (Fig. 4) to not only limit its high conservatism, but also improve the solution robustness. Given the number of cutting hyperplanes (C) with gradient vector and intercept , hyperplane is defined with random vector as follows: where is the number of uncertain parameters in the i th constraint. The considered cutting planes consist of a coefficient ( ) and a constant ( ) the equation of which is formulated by a random variable . In this study, the DDRO approach has been implemented based on a polyhedral packaging uncertainty set where cutting planes reduce conservatism and increase robustness by limiting these areas. Next, effort is made to obtain a linear closed form of the common area between the initial sets and the cutting planes. the inner maximization equation of which is as follows: The final linear form of which with auxiliary variables and is: ̂ ̂ To keep the convexity of the uncertainty space, use is made of linear cutting planes with the following general form: where is the number of cutting planes and is the cardinality of the set of uncertain parameters. The optimization model of achieving the components of these planes is as follows: where is a sample of uncertainty space ̃ ̂, specifies its size, is the sample distance to cutting plane, ́, is a normal vector with states to create different cutting planes and is a threshold to be specified each time. The final equation of cutting planes is as follows: In Dec. 2019, authorities in Wuhan City, Hubei Province, China, reported some unknown virus-caused pneumonia cases related to a seafood wholesale market [86] which was first referred to as a Mysterious Respiratory Disease in China and then renamed as COVID-19 by the WHO [87] . COVID-19 is caused by an emerging bat-origin corona virus (SARS-CoV-2) not previously reported in humans [88] . It is a highly acute and contagious disease began in China in mid-December and spread rapidly during the New Year holidays because information was none and trips were not restricted. Prioritizing prevention over treatment has long been an issue in the world health system and the sudden onset of the COVID-19 crisis reaffirmed this priority. Inability to control this disease has put much pressure on the countries' treatment systems and consequences of the related imposed costs have had such profound effects on their economies that early vaccine accessibility has become their first priority to appropriately respond to and prevent the crisis. Vaccines, approved by the WHO till April 2021, include Sputnik V, Pfizer, Astrazeneka, Modrena and Sinopharm. COVIran Barakat is an Iranian vaccine that has passed its third clinical phase, entered the treatment system for general vaccination and used first to vaccinate the leader of Iran to show his respect for young Iranian scientists. The COVID-19 vaccine SC consists of such complicated systems as processing, equipment, means of transportation and afflicted places that need proper temperature for their inventory. This SC is aimed mainly to provide all humans with a better vaccine access and distribution and since failure to understand its components and interactions can make the vaccine accessibility difficult around the world, a robust SCND can further enhance the vaccine accessibility, especially in developing countries. The present study has implemented its model in a practical case in Iran. With 2,823,887 affected and 77,994 deaths as of May 2021 [89], Iran has been one of the twenty countries with the highest COVID-19 afflicted and present in all the related peaks. Various disease outbreaks made the government try to not only control it with such policies as closing various centers nationwide and enacting quarantine-related laws at different intervals, but also consider a comprehensive nationwide vaccination plan on its agenda. The present case study has addressed the Russian Sputnik V, Indian Astrazeneka, and Chinese Sinopharm vaccines ordered by the government of Iran, and also the Iranian COVIran Barakat that has passed its third clinical phase. It has entered the treatment system to supply part of the vaccine needed for a general nationwide vaccination (there are still other domestic vaccines underway to be dealt with later). This section presents the data-collection sources for the input parameters of the COVID-19 vaccine SC to fruitfully conduct the case study by obtaining valid information. As the temperature for the vaccine storage/transportation was ─ 80 o C, the fleet used in this study operated accordingly. This final section is aimed to address the vaccine demand and injection priorities. According to the strategic vaccination document approved by the Iranian government, groups prioritized for vaccination are (Fig. 5 ): 1) medical staff, 2) the elderly and those with underlying diseases, 3) those with sensitive jobs and high public contacts and 4) the public; the related information has been collected from the "Ministry of Health" and "Statistics Center" of Iran [89, 95] . It is worth noting that since all vaccines are injected in two doses, demands in periods after the first will be the total of the first dose for new people and the second dose for those already received the first injection. The model presented in this study considers a 12-month planning time horizon for this project. In consultation with Iranian social pathology experts, weight coefficients were determined for critical social effects in the Covid-19 vaccine SC resulting in 0.41 for occupational justice, 0.3 for environmental vulnerability and, finally, 0.29 for job security. The proposed multi-objective linear mathematical optimization model with three conflicting objective functions that face uncertainty in the vaccine accessibility parameter, has been solved with the GAMS 24.1.2 software using a quad-core, 2GHz frequency, 8 GB RAM processor and the epsilon constraint approach. Among the set of obtained Pareto solutions, the authors have selected the one that yields efficient and same-level values in all objectives. The linear cutting planes, consisting of coefficients and fixed values, were formed by the mathematical DDRO model implemented based on closed uncertainty sets using data sampling and models presented in Eqs (76) (77) . As the foreign vaccine accessibility parameter ( ̃ ) (3 in this study) is uncertain, planes are required (Table 3) to proceed. Applying these planes on the multidimensional uncertain parameter set yielded the final model results. Next, the SC model was implemented after applying the cutting planes formed in the set of uncertain accessibility parameters of each vaccine type. Fig (6) shows the SC network that supplied the COVID-19 vaccine for Iran in an international dimension. Packaging centers for distribution purposes were selected, from among the potential locations, to be in Moscow, Beijing and Delhi based on the mathematical model solution results ( ), and 8000 of the Russian, 9000 of the Chinese and 10000 of the Indian vaccine doses were considered in each package for international shipments. After the international packaging centers were determined and the optimal vaccine flow outside Iran found its final form, the SC form was addressed inside the country. Out of nine potential locations for foreign-to-domestic vaccine distribution centers, the optimal solution selected only 5 -Tehran, Isfahan, Ahvaz, Kermanshah and Bandar Abbas -to receive the vaccines bought from foreign manufacturers and ship them to the desired distribution center ( ); on the other side of the defined SC lay the domestic (COVIran Barakat vaccine) production centers: 1) Pasteur Institute of Iran, in Tehran and 2) Razi Vaccine and Serum Research Institute, in Alborz ( Fig. 7) ( ). Regardless of the type, all four available vaccines types -three imported and one domestic -can be used to meet the nationwide demand. The previous section located the construction sites of 5 distribution centers of the imported vaccines all of which used all three vaccine types to meet the demand of the study population. Table (4) shows the rate of the vaccine distribution flow to provincial medical centers in each period for different centers and vaccine type (∑ ). As shown, since the vaccine production and transportation costs are different, the model has used some vaccines more in certain periods to reduce costs. Table (5) lists the vaccines produced in domestic centers (Fig. 7) in each time period separately for the related production center (∑ ). Tables (4) and (5) show an appropriate foreign-domestic production balance in the optimal solution. It is obvious in Table (5) that the domestic vaccine accessibility in the early period is far from expectation considering the commencement of the production process. Regarding road transportation, the model has determined the need for 108 truck units for all SC displacements. t=1 t=2 t=3 t=4 t=5 t=6 t=7 t=8 t=9 t=10 t=11 t=12 Razi technology Incubator -2249429 804219 2435860 2238037 --2943422 2943422 2943422 2943422 4143422 Pasteur institute of Iran -3793579 4658872 2956487 1665899 613472 281078 4671100 3000000 3000000 3000000 6815479 Table 5 -Domestic vaccine flow to provincial medical centers Fig (8) shows which foreign or domestic vaccine producer covers the provincial medical centers to meet the demand for a nationwide vaccination. Supplying by internal vaccines shows that not all medical centers have been supplied with even a minimum amount due, maybe, to the limited domestic production capacity in the study period. In the view related to foreign vaccines, each medical center is often covered by several distributors reducing the transportation costs. Meeting the demand for COVID-19 vaccine through both domestic and foreign producers requires careful planning of their capacities. Fig (9) shows the production capacities ( ) of the Russian Sputnik V, Chinese Sinopharm and Indian Astrazeneka designed for uncertain parameters of accessibility and ancillary costs with the maximum capacity relating first to the Russian and then to the Indian vaccines to meet the demand in the early years. Fig (10) shows the design capacity variable ( ) for domestic production centers where the capacity specified for "Pasteur" is somewhat more than that for "Razi". This can also be concluded from Fig (8) where the coverage of this center is more than that of the Razi Institute. Next, the proposed mathematical DDRO model is compared with the deterministic programming and box and polyhedral uncertainty set-based approaches to evaluate its performance. The rapid corona disease epidemic boosted the demands for its vaccine in developed countries after it was produced and made its access uncertain for developing countries. The proposed data-driven approach has addressed this uncertainty and is validated next for each foreign vaccine's accessibility. To this end results of the proposed DDRO approach are considered as the solutions of the mathematical model and compared with the box and polyhedral set in two uncertainty modes for the accessibility parameter based on the classical convex programming; in other words, solutions of other methods are ̂ ̂], has been changed ( is the distance between the real and realized data). (11) shows the comparisons made in the present study based on the economic cost objective function. Performance of the three approaches facing uncertainty is shown by changing the real-realized data difference parameter in its uncertainty interval. In the box uncertainty set-based approach, solutions are much worse than others due to the method's high conservatism that ensures solutions have high robustness against the variations of the uncertain parameter, but in the polyhedral uncertainty set-based approach, the objective function shows lower costs and solutions are improved and have less conservatism, but still impose high costs on the system for cases that are less likely to occur. In short, the proposed DDRO method greatly reduces the conservatism of classical, uncertainty set-based approaches and the solutions, while quite robust, are much better than those of other approaches. Fig (12) compares the solutions in different approaches based on the economic objective function and the difference between the real and realized values. As shown, the classical box/polyhedral uncertainty set-based approaches have created high conservatism in solutions, causing the environmental objective function to get worse, but the proposed approach has shown acceptable performance in the environmental dimension, reduced the high conservatism of classical approaches significantly and yielded a much more robust solution. This event is clearly obvious in Fig (13) in the behavior of the three-part social objective function. In the vaccines distribution context, a pandemic VSC is different than that of a traditional VSC because governments are directly procuring vaccines from the manufacturers bypassing the traditional chains of wholesalers and distributors. Hence, healthcare experts and VSC analysts are looking for proper policies and adequate strategies for appropriate vaccine manufacturing and distribution to fight against the COVID-19 pandemic. Investigating the existing literature shows the vaccine supply chain has not been updated according to the pandemic situation. The main challenges that are not considered in COVID-19 vaccine supply chain literature and can affect supplying vaccines performance are as follow: • Integrating import and manufacture decisions simultaneously in countries with vaccine production knowledge Planning vaccine supply chain in a sustainable form (especially the social aspect) • Addressing Covid-19 vaccine access uncertainty in the competitive world market In this regard, the present study has been aimed to fill these gaps so that governments can remove the barriers to the public access to appropriate vaccines and according to the results there is considerably good integration to solve the problems. As the proposed study has been implemented in a real case study in Iran, so far among the first ten most affected countries, the managerial insights have been discussed for the results of the study in each section. For countries that have the vaccine production knowledge, the proposed model enables an integrated approach that illustrates, in its network, the levels associated with domestic and foreign suppliers for the vaccine production and importation, respectively. In the COVID-19 vaccine issue, governments, which are generally the upstream-level decision makers, can benefit from the integrated decisionmaking by precise vaccine production plans. Benefits of this concurrence are: 1) reduced costs of supplying the Covid-19 vaccine, especially in a situation where the world is facing an economic crisis caused by the pandemic, 2) careful planning of the vaccine importation based on the production capacity of each case due to the high complexity of the vaccine importation supply chain, 3) establishing proper coordination with the world's main producers of the Covid-19 vaccine -Sputnik, Sinopharm and AstraZeneca -and 4) precise production planning for countries that have the vaccine production knowledge. Using a real case study in Iran, the proposed model depicts the relations among the main foreign vaccine producers for importation in Fig. 6 , locating the domestic production facilities and foreign vaccine distribution centers in Fig. 7 , and allocating vaccine producers to meet the demand in provincial medical centers (Fig. 8) . Tables 4 and 5 specify the optimal flow rate of the domestic and foreign vaccines based on the production and importation capacities of each vaccine over the horizon, and formulate the economic objective function to minimize costs and optimally balance the importation and production. The environmental and social activists are nowadays seriously concerned about how wrongly the humans behave with the nature and believe it is a threat for the next generation's access to a healthy environment and community. This situation has worsened with the COVID-19 disease pandemic, loss of many jobs and increased environmental effects. The proposed COVID-19 vaccine model considers the environmental and social dimensions of a sustainable SC network and formulates its second and third objective functions accordingly. The second objective function minimizes environmental effects of the entire network processes and ensures the related activists that the emission of greenhouse gases is controlled to a desirable level based on the ReCipe Standard. The third objective function addresses such social concerns as the number of created job opportunities, days lost due to the damage, and contaminating products in a situation where the manpower has lost many jobs due to the pandemic throughout the world. A comprehensive consideration of this dimension of the sustainable SC network based on the ISO2600 Standard ensures the human resources activists that, under the pandemic, the desired effects of the Job security, occupational justice and environmental vulnerability on the network's manpower will be maximized; results of these objective functions are shown in Figures 12 and 13 . A challenge facing a vaccine SC in a disease pandemic is the competitive markets where governments around the world try to provide the vaccine. Such markets can be due to the governments' varying behavior such as providing beyond their needs, production limitations in producing countries, and so on that lead to uncertainties to access and purchase the vaccine. This study has used modern, robust, data-driven optimization methods and implemented the model under uncertain conditions using cutting planes to present a realistic view of the uncertain nature of the vaccine accessibility in pandemic situations because such planes are extracted by realizing the vaccine accessibility data for purchasing. By controlling the conservatism, the proposed approach allows Govt. managers to make vaccine production-/importation-related decisions that conform more to reality and avoid high-conservatism decisions that impose high costs and adverse environmental/social effects on the natural system. Results obtained from this merit of the proposed approach are shown in Figures 11, 12 and 13 in each of the related objective functions. The present study has addressed a robust COVID-19 vaccine SCND using a mathematical multi-period mix integer linear programming model implemented for a 1-year time horizon for a nationwide vaccination in Iran. It is a robust SC network with economic, environmental and social dimensions; the latter considers the jobs created and job days lost by workers due to risks and injuries in the workplace. Against the worldwide vaccine-access injustice, a robust data-driven model has been developed to address the uncertainty of the vaccine accessibility based on closed convex uncertainty sets. Number of purchased vehicle mod l for carrying materials (just trucks) Equal to 1 if packaging center j is opened; 0 otherwise Equal to 1 if distribution center k with inventory tech m is opened; 0 otherwise Equal to 1 if internal manufacture center v is opened; 0 otherwise The designed capacity of foreign vaccine type p at foreign vaccine manufacturing center i in time period t The designed capacity of foreign vaccine type p at packaging center j in time period t The designed capacity of foreign vaccine type p at foreign vaccine manufacturing distribution center k with inventory tech m in time period t The designed capacity of internal vaccine at internal vaccine manufacturing center v in time period t Amount foreign vaccine type p produced at foreign vaccine manufacturing center i in period t Amount of internal vaccine produced at internal vaccine manufacturing center v in period t Amount of foreign vaccine type p pre-distributed at distribution center k with inventory tech m in period t Amount of foreign vaccine type p packaged at packaging center j in period t Amount of raw material foreign vaccine type p purchased at foreign vaccine manufacturing center i in period t Amount of raw material internal vaccine purchased at internal vaccine manufacturing center v in period t The amount of foreign vaccine type p transported from foreign vaccine manufacturing center i to packaging center j with transportation mod l at time phase t The amount of foreign vaccine type p transported from packaging center j to distribution center k with transportation mod l at time phase t The amount of foreign vaccine type p transported from distribution center k to provincial medical center r with transportation mod l at time phase t The amount of internal vaccine transported from internal vaccine manufacturing center v to provincial medical center r with transportation mod l at time phase t The inventory of foreign vaccine type p at foreign vaccine manufacturing center i in period t The inventory of foreign vaccine type p at packaging center j in period t The inventory of foreign vaccine type p at distribution center k with inventory tech m in period t The inventory of internal vaccine at internal vaccine manufacturing center v in period t The capacity expansion of foreign vaccine type p at foreign vaccine manufacturing center i in period t The capacity expansion of foreign vaccine type p at packaging center j in period t The capacity expansion of foreign vaccine type p at distribution center k with inventory tech m in period t The capacity expansion of internal vaccine at internal vaccine manufacturing center v in period t The shortage of foreign vaccine type p at foreign vaccine manufacturing center i in period t The shortage of foreign vaccine type p at packaging center j in period t The shortage of foreign vaccine type p at distribution center k with inventory tech m in period t World Health Organization, Immunization, Vaccines and Biologicals Towards effective COVID-19 vaccines: Updates, perspectives and challenges Impacts of epidemic outbreaks on supply chains: mapping a research agenda amid the COVID-19 pandemic through a structured literature review Reducing Supply Risks by Supply Guarantee Deposit Payments in the Fashion Industry in the "New Normal after COVID-19 Pandemic Risk Management in Operations and Finance The effect of COVID-19 on the banking sector Supply-chain management: logistics catches up with strategy Supply chain management: myth or reality? Supply Chain Management Facility location and supply chain management-A review Managing The Supply Chain: Definitive Guide. Tata McGraw-Hill Education Essentials of supply chain management Stochastic Optimization of Supply Chain Resilience under Ripple Effect: A COVID-19 Pandemic Related Study A bibliography for some fundamental problem categories in discrete location science Facility location in supply chain design Stochastic programming Introduction to stochastic programming Stochastic programming models Tractable stochastic analysis in high dimensions via robust optimization Robust optimization Robust solutions of linear programming problems contaminated with uncertain data Data-driven robust strategies for joint optimization of rail renewal and maintenance planning An expanded robust optimization approach for the berth allocation problem considering uncertain operation time Models for reliable supply chain network design The design of robust value-creating supply chain networks: a critical review Competitive supply chain network design: An overview of classifications, models, solution techniques and applications Sustainable supply chain network design: An optimization-oriented review Robust optimization of large-s-cale systems Robust reverse production system design for carpet recycling A robust design for a closed-loop supply chain network under an uncertain environment Robust sustainable bi-directional logistics network design under uncertainty A location-routing problem with disruption risk Investigating risk and robustness measures for supply chain network design under demand uncertainty: A case study of glass supply chain Robust and reliable forward-reverse logistics network design under demand uncertainty and facility disruptions Robust optimization of fourth party logistics network design under disruptions An enhanced possibilistic programming approach for reliable closed-loop supply chain network design An effective heuristic algorithm for robust supply chain network design under uncertainty Capacity planning and warehouse location in supply chains with uncertain demands Dynamic supply chain network design for the supply of blood in disasters: A robust model with real world application Possibilistic linear programming: a brief review of fuzzy mathematical programming and a comparison with stochastic programming in portfolio selection problem Robust possibilistic programming for socially responsible supply chain network design: A new approach A possibilistic programming approach for closed-loop supply chain network design under uncertainty Sustainable sugarcane-to-bioethanol supply chain network design: A robust possibilistic programming model Convex programming with set-inclusive constraints and applications to inexact linear programming Robust solutions to uncertain semidefinite programs Robust convex optimization Robust discrete optimization and network flows Adjustable robust solutions of uncertain linear programs Distributionally robust optimization and its tractable approximations Distributionally robust convex optimization Priority setting for pandemic influenza: an analysis of national preparedness plans Geographic prioritization of distributing pandemic influenza vaccines Optimizing influenza vaccine distribution Modeling optimal age-specific vaccination strategies against pandemic influenza Ethical vaccine distribution planning for pandemic influenza: Prioritizing homeless and hard-to-reach populations System factors to explain 2009 pandemic H1N1 state vaccination rates for children and high-risk adults in US emergency response to pandemic Equalizing access to pandemic influenza vaccines through optimal allocation to public health distribution points Allocation of COVID-19 vaccines under limited supply Optimizing vaccine distribution networks in low and middle-income countries Distribution of pandemic influenza vaccine and reporting of doses administered Integrating pharmacies into public health program planning for pandemic influenza vaccine response Vaccine procurement during an influenza pandemic and the role of advance purchase agreements: Lessons from 2009-h1n1 Estimating the potential effects of a vaccine program against an emerging influenza pandemic-United States A decision support system for demand management in healthcare supply chains considering the epidemic outbreaks: A case study of coronavirus disease 2019 (COVID-19) A mathematical programming approach for equitable COVID-19 vaccine distribution in developing countries Optimal allocation of the limited COVID-19 vaccine supply in South Korea Optimal time-varying vaccine allocation amid pandemics with uncertain immunity ratios Modeling vaccine allocations in the covid-19 pandemic: A case study in australia Optimizing the assignment of swabs and reagent for PCR testing during a viral epidemic Strategies for ensuring required service level for COVID-19 herd immunity in Indian vaccine supply chain A capacity planning approach for sustainable-resilient supply chain network design under uncertainty: A case study of vaccine supply chain OR-methods for coping with the ripple effect in supply chains during COVID-19 pandemic: Managerial insights and research implications Viability of intertwined supply networks: extending the supply chain resilience angles towards survivability. A position paper motivated by COVID-19 outbreak Viable supply chain model: integrating agility, resilience and sustainability perspectives-lessons from and thinking beyond the COVID-19 pandemic Lean resilience: AURA (Active Usage of Resilience Assets) framework for post-COVID-19 supply chain management Supply chain viability and the COVID-19 pandemic: A conceptual and formal generalisation of four major adaptation strategies Vaccine distribution chains in lowand middle-income countries: A literature review Pricing the COVID-19 vaccine: A mathematical approach Evaluating the effect of key performance indicators of vaccine supply chain on sustainable development of mission indradhanush: A structural equation modeling approach A sustainable crossefficiency DEA model for international MSW-to-biofuel supply chain design The convergence of the World Health Organization Member States regarding the United Nations' Sustainable Development Goal 'Good health and well-being Optimal Design and Operation of the green pistachio supply network: A robust possibilistic programming model Convex optimization The analytics edge. Dynamic Ideas LLC Data-driven robust optimization under correlated uncertainty: A case study of production scheduling in ethylene plant Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China Surveillance case definitions for human infection with novel coronavirus (nCoV), interim guidance Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia The authors confirm contribution to the paper as follows: