key: cord-0284971-ut9jqfka authors: Zia, K. title: Why a Globally Fair COVID-19 Vaccination? An Analysis based on Agent-Based Simulation date: 2021-10-04 journal: nan DOI: 10.1101/2021.10.03.21264494 sha: 032f23a14c5f426231d4ffbd6030cbb12ac17567 doc_id: 284971 cord_uid: ut9jqfka In this paper, an Agent-Based Model (ABM) is proposed to evaluate the impact of COVID-19 vaccination drive in different settings. The main focus is to evaluate the counter-effectiveness of disparity in vaccination drive among different regions/countries. The model proposed is simple yet novel in the sense that it captures the spatial transmission-induced activity into consideration, through which we are able to relate the transmission model to the mutated variations of the virus. Some important what-if questions are asked in terms of the number of deaths, and time required and the percentage of the population needed to be vaccinated before the pandemic is eradicated. The simulation results have revealed that it is necessary to maintain a global (rather than regional or country-oriented) vaccination drive in case of a new pandemic or continual efforts against COVID-19. After the successful provisioning of COVID-19 vaccination, there has been efforts to achieve a global herd immunity as soon as possible [1] . However, disparity in vaccination drive among different countries can be counter-productive. A general evidence of health disparity in COVID-19 treatments if given in [2] . There is a need to analyze the effects of vaccine provisioning disparity at a global scale [3] . A very few countries are involved in the process of vaccine production. Obviously, these countries would vaccinate their own population first before the vaccine is available to other countries. However, particularly in case of such a complex global system, the intuitive logic, quite obvious in a situation may not be the optimal choice. Through this paper, we provide an evidence to establish it in the context of COVID-19 vaccination drive. September 21, 2021 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021. 10 .03.21264494 doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. In this paper, an Agent-Based Model (ABM) is proposed to evaluate the impact of COVID-19 vaccination drive in different settings. The simulation results have revealed that it is necessary to maintain a globally "fair" [4] vaccination drive in case of a new pandemic or continual efforts against COVID-19. Here fair means a vaccination drive which is not self-centered (focused on developed countries producing the vaccine), but global (equally distributed right from day 1) across all the regions / countries. A simplistic model of virus transmission is used consisting of minimal states of susceptible, vaccinated, infected and recovered. A moving agent in one of these states is tightly bound to the underlying space, where the space is divided into regions to evaluate the region-based vs. global vaccination drive. Additionally, the virus gets mutated, where the extent of mutation is directly related to spatial activity representing the transmissions. And the inactivity is directly proportional to the mutated variant at a location. Already, a few agent-based models concerning the vaccination efficiency have been proposed. Authors in [5] have proposed vaccination strategies with a delayed second dose. They have compared different vaccination products and provided a projected number of infections, serious cases and deaths. Authors in [6] proposed a mathematical model to estimate the impact on mortality and total infections of completely lifting the COVID-19 restrictions. A qualitative study on who should be prioritised for COVID-19 vaccination is given in [7] . Another focused attempt is a simulation study to estimate the future infections rates among the children with vaccination [8] . More closer to our paper is [9] , in which the authors emphasize on accelerating the vaccination drive to mitigate high transmissibility resulting in more deadly variants. However, the model proposed is population based without spatial (regional / country-wise) variations. Another similar model is presented by the authors in [10] . But the system dynamic model again does not cater for spatial considerations and mobility. Another sound research is published in [11] , in which the authors combines the real data with the agent-based model to estimate the impact of lockdown and vaccination against COVID-19. But, this model is about one country only and does not take nonavailability of vaccine as an option. As evidenced by the above, the model proposed in this paper is simple, but, novel in the sense that it captures the spatial virus transmission-related activity into consideration. This enable us to do inquiry about the effectiveness of the vaccination drive in different settings. In the next section, we have detailed the model, followed by simulation and results section. The last section provide a 2 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. detailed outlook of this research in progress. The model represents an individual as a mobile agent, which resides on a static cell. The epidemic spread model is based on simple susceptible-infected-recovered/not recovered (SIR) model to keep focus on vaccination aspect. In addition to these three states, the fourth state of being vaccinated is added. A few agents in the population are made infected at the start of the simulation. Causally, a susceptible agent can directly be transited to infected state or it may transit to vaccinated state. An agent may also be transited to infected state from vaccinated state. An infected agent may be recovered of may not (die). Even a recovered agent can still be susceptible. The simulation runs in iterations. At each time stamp, each of the agents make a causal change in its state depending upon its surrounding. The model is implemented through three sequential modules (procedures), namely transmission, recovery and vaccination, followed by running the module which let the agents move randomly. This procedure is executed by all the agents (randomly ordered) which are infected now. The procedure is dependent on the following variables: CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint • transmission rate of agent a (one of the infected agent) pinned onto the location transmissibility. The location transmissibility depends on how transmission intensive the neighborhood of agent a is. This in turn is handled by cell (patch) variable activity, which is increased as soon as a transmission happens due to the resident agent. The sample activity maps comparing transmission activity of two cases (Figure 1a vs. Figure 1b) . • A neighboring agent n which is susceptible. • Neighborhood-based variable ratio vaccinated, which is the ratio of neighborhood agents of agent n that are in vaccinated state. If an agent a has a susceptible agent n in its neighborhood, agent n becomes infected with a probability defined by transmission rate (TR [n]) that is been encountered. TR [n] is equal to difference of transmission rate of the source agent (TR [a]) and the ratio of agents already vaccinated (ratio vaccinated) in the neighborhood of agent n. Hence, the intensity by which the infection is induced by a is depleted based on vaccination. TR [a] is incremented before, depending on how transmission active the neighborhood of a has been. If n is infected, the activity of underlying patch is also increased by a factor mutation index, which represents how effectively mutation is changing the local transmission. The pseudo code given in Snippet 2.1 describes the transmission mechanism run by each infected agent: This procedure applies to those agents which are infected. Each of these agents either remains in the infected state, gets recovered, or dies, depending on the number of days it has been infected and chances (controlled by simulation parameters of mortality and efficacy). The self-explanatory pseudo code given as Snippet 2.2 describes the mechanism run by each infected agent: According to a vaccination rate, the agents which are not yet vaccinated are vaccinated. This is a simple procedure just setting state of an agent to true. The self-explanatory pseudo code given as Snippet 2.3 describes the vaccination process: CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint A cell space of size 160 × 120 is used as the simulation world, wrapped around horizontally as well as vertically. Therefore a population of a maximum of 19200 non-overlapping agents can be generated, each occupying a single patch. We opted to take 50% of the population (9600 agents). Each patch is first initialized with activity 0. As the simulation progresses, all the agents move on patches with a designated speed. They also change their health states (based on the model above), and may end up as dead agents, thus no more part of the simulation world. In Table 1 , the list of global simulation parameters are given. The other parameters such as transmission range (radius of neighborhood), mortality rate, and efficacy of vaccine are also used, however, these acquire static values of 1.5 patch distance, 0.2, and 0.8, respectively. Whereas, the last two parameters define the four cases we have considered. These cases are given in Table 2 . What do these cases mean? Since, in the simulation, the vaccination happens from day 1, the vaccination rate equal to 0% means that the transmission of virus happens along with the vaccination drive. Obviously, this did not happen in case of COVID-19. Hence, 0% vaccination rate 7 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. when the active cases were quite low and a substantial population was vaccinated. Therefore, the prior represents an ideal and the later represents a real scenario (but an intermediate one). Since, the vaccination drive (as we have seen in case of COVID-19) was / has been countrycentered, where the (mostly) developed countries vaccinated their own population first, followed by provision of the vaccine to other countries. We term this scenario as self-centered, and it is implemented in the simulation by introduction of regional blocks (blocks? = true). Therefore, some blocks have more vaccines in comparison to others. An alternate strategy would have been a globally balanced vaccination drive irrespective of country of manufacturing. Thus, that would have been a global drive, in which, all the blocks would have been the same (hence, block? = false). In fact, there are 12 regional blocks as shown in Figure 5 . Although, a clear differentiation is enough for modeling purposes, however, the blocks/regions are sorted from left to right and from bottom to top, in terms of availability of vaccine. So for example, region 1 is the worst, followed by region 2, 3, and so on. The last column of Table 2 summarizes the scenarios in terms of the above two dimensions. So the idea is to measure the performance of the vaccination drive in all four scenarios and do a comparative analysis. The simulations are average of sufficient number of runs. The results are compared based on the following outcomes. The results shown in graph in Figure 6 represent the number of agents in different states after the simulation ends (there is no more infected agent in the population). The graph in Figure 9 shows the number of people dead and time when the simulation 8 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint ends, with particular focus on blocks situation. • Starting with the self-centered real scenario (case 4), it can be seen that, we end up with almost 10% of the population dead, whereas, recovered, vaccinated and susceptible follow the same order. If we compare these results with global real scenario (case 2), the dead % drops down to less than 1 (see Figure 7 ). It should be noted that a very small fraction of infected agents (taken value of 0.005-0.010%) dispersed randomly across the space at the start of the simulation, does not really represents the reality. However, it can be considered as cancelling the fact that vaccinations are also done at random, not according to the severity of the regional infectiousness. Nevertheless, it does not take away the fact that a global vaccination drive 9 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. would have helped a lot. • For the ideal scenarios, case 3 represents the self-centered situation. The dead % increases to 22. Whereas, for a global situation (case 1), it could be reduced to 9.5%. However, case 3 is much more diversified in terms of distribution of deaths (see Figure 8 ). It is clear from the graph shown in Figure 6 that a global vaccination drive, right from the start (in case of availability of vaccine) or whenever it was available, would have resulted in much less deaths. Now the question is do these deaths happen only in regions which are not self-centered. For this we consult the graph shown in Figure 9 . As expected, in case 4, the number of people dead follow almost the same order from 1 to 12. But, if we compare it with case 2, case 2 gives a much improved global picture, in fact 10 times better than case 4. The gain among various blocks range from 5 times (for blocks with availability) to 15 times (for blocks with nonavailability). However, this disparity (between blocks) becomes really negligible if we compare case 3 with case 1. But, the global gain of case 1 is almost 2 times that of case 3. So, in future, in case of a pandemic (with availability of vaccine), it will be strategically beneficial to go for a global vaccination drive rather than the self-centered one. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. A closure happening at a early time is absolutely beneficial as the World is tired and seeking a normal social and economic activity. At the time scale of the simulation, it is evident in the graph shown in Figure 9 that worst case time is for case 4. We gain at least 20% in case 2. When comparing case 3 with case 1, the gain is 10%. So global vaccination drive is also beneficial in terms of time needed to eradicate the pandemic. The timing dynamics aspect is visually shown in Figure 10 where we compare case 3 with case 1 in terms of time to vaccinate 20% of the population. In the sample run, it took 110 iterations to do it in case 3 as compared to 45 iterations in case 1. Finally, we have results of how much population is needed to be vaccinated in each case in Figure 11 . Obviously, if the vaccine is invented late (case 2 and 4), it will be required more, almost 85% of the population. And, there is no difference between these two cases. However, the time to achieve this number is lesser in case 4. Whereas, the time to achieve the required percentage between cases representing availability of vaccine right from the start is not much different among case 3 and 1, but, the required % decreases to almost 30% in case 3 when compared to more than 60% in case of case 1. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint 12 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint There are several challenges in fair distribution of COVID-19 vaccine [12, 13] . In this paper, an Agent-Based Model is simulated to evaluate the effectiveness of COVID-19 vaccination drive based on its provision to different regions of the world. The model proposed is simple yet novel in the sense that it captures the spatial transmission-induced activity into consideration, through which we are able to relate transmission model to the mutated variations of the virus. The results of the simulation suggest that it is necessary to maintain a global (rather than regional or country oriented) vaccination drive in case of a new pandemic or continual efforts against COVID-19. It results in lesser number of deaths, time and quantity of vaccination required. The model can be extended in following ways: • A more sophisticated virus transmission model can be used. • The relationship of mortality and efficacy is dependent on variant of the virus. We have taken only static values, which can be replaced by evidenced values of mortality and efficacy based on the infectiousness of the virus. • Real demographic, vaccination, population and mobility data can be used to gain a more insightful analysis. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint Covid-19: Evidenced health disparity Multivalue ethical framework for fair global allocation of a covid-19 vaccine Covid-19, equity and fair priority vaccination Evaluation of covid-19 vaccination strategies with a delayed second dose Quantifying the impact of lifting community nonpharmaceutical interventions for covid-19 during vaccination rollout in the united states Who should be prioritised for covid-19 vaccination? Simulated identification of silent covid-19 infections among children and estimated future infection rates with vaccination Accelerated vaccine rollout is imperative to mitigate highly transmissible covid-19 variants Fortaleza, Vaccination efforts in brazil: scenarios and perspectives under a mathematical modeling approach Estimating the impact of interventions against covid-19: from lockdown to vaccination, medRxiv Challenges in ensuring global access to covid-19 vaccines: production, affordability, allocation, and deployment International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 4, 2021. ; https://doi.org/10.1101/2021.10.03.21264494 doi: medRxiv preprint