key: cord-1008153-0iurfr6h authors: Moon, S. a.; Scoglio, C. title: Contact Tracing Evaluation for COVID-19 Transmission during the Reopening Phase in a Rural College Town date: 2020-06-26 journal: nan DOI: 10.1101/2020.06.24.20139204 sha: 1f574c060aa3a0ec1146563444311283340f7e1a doc_id: 1008153 cord_uid: 0iurfr6h Contact tracing can play a vital role in controlling human-to-human transmission of a highly contagious disease such as COVID-19. To investigate the benefits and costs of contact tracing, we develop an individual-based contact-network model and a susceptible-exposed-infected-confirmed (SEIC) epidemic model for the stochastic simulations of COVID-19 transmission. We estimate the unknown parameters (reproductive ratio R0 and confirmed rate {delta}2) by using observed confirmed case data. After a two month-lockdown, states in the USA have started the reopening process. We provide simulations for four different reopening situations: under "stay-at-home" order or no reopening, 25% reopening, 50% reopening, and 75% reopening. We model contact tracing in a two-layer network by modifying the basic SEIC epidemic model. The two-layer network is composed by the contact network in the first layer and the tracing network in the second layer. Since the full contact list of an infected individual patient can be hard to obtain, then we consider different fractions of contacts from 60% to 5%. The goal of this paper is to assess the effectiveness of contact tracing to control the COVID-19 spreading in the reopening process. In terms of benefits, simulation results show that increasing the fraction of traced contacts decreases the size of the epidemic. For example, tracing 20% of the contacts is enough for all four reopening scenarios to reduce the epidemic size by half. Considering the act of quarantining susceptible households as the contact tracing cost, we have observed an interesting phenomenon. When we increase the fraction of traced contacts from 5% to 20%, the number of quarantined susceptible people increases because each individual confirmed case is mentioning more contacts. However, when we increase the fraction of traced contacts from 20% to 60%, the number of quarantined susceptible people decreases because the increment of the mentioned contacts is balanced by a reduced number of confirmed cases. The main contribution of this research lies in the investigation of the effectiveness of contact tracing for the containment of COVID-19 spreading during the initial phase of the reopening process of the USA. COVID-19 has affected the lives of billions of people in 2019-2020. The COVID-19 disease is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and has caused a global health emergency. The world health organization (WHO) declared it as a Public Health Emergency of International 5 Concern on January 30, 2020 [1] . The number of confirmed reported cases by SARS-CoV-2 has been rising. On May 31, 2020, worldwide there were 5, 939, 234 laboratory-confirmed cases with 367, 255 deaths [2]. Many countries issued a pandemic lockdown to slow down the spreading of COVID-19. In the United States, a 'stay-at-home' order was issued in many 10 states. However, those pandemic lockdowns have a massive impact on the economy. All the States of the USA started reopening gradually from early May. Understanding the impact of mitigation strategies on the spreading dynamic of COVID-19 during the reopening phase of the USA is essential. In this work, we assess the impact of contact tracing under four reopening scenarios: 25% 15 reopening, 50% reopening, 75% reopening, and 100% reopening. 2 . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint Individual-based contact-network models are a powerful tool to model COVID- 19 spreading due to its person-to-person spreading nature. In this work, we develop an individual-based network model for a college town, Manhattan, KS, where households represent nodes of the network. We select Manhattan, KS, 20 as our study area, since it is a typical college town in a rural region of Kansas, the home of Kansas State University. There are 20, 439 occupied households in Manhattan, KS, according to census 2018 [3] . The connections between two individual households represent the contact probabilities between the members of the households. The individual-based approach provides the flexibility to 25 observe the local dynamic at the individual level. It also allows us to include in the model a mitigation strategy at the individual level, such as contact tracing. To design an epidemic model for COVID-19 is challenging, as many epidemic features of the disease are yet to be investigated, such as, for example, the transmission rate, the pre-symptomatic transmission rate, and the percentage of the 30 asymptomatic population. These uncertain characteristics make epidemic modeling challenging as the outcomes of the model are sensitive to the assumption made on the uncertainties. Therefore, we use a simple epidemic model with four compartments -susceptible-exposed-infected-confirmed (SEIC)-capable of imitating the COVID-19 transmission and flexible enough to cope with new 35 information. This model has only two unknown parameters: the reproductive ratio R 0 , and the confirmed case rate or reporting rate δ 2 . We use confirmed COVID-19 cases from March 25, 2020 to May 4, 2020 in Manhattan, KS as data, and estimate the unknown parameters from data. We consider that a confirmed COVID-19 patient cannot spread the disease anymore except in his/her 40 own household. In the spreading of COVID-19, there are pre-symptomatic and asymptomatic cases that do not show any sign of illness [4] . Besides, there is a strong possibility that infected cases not detected exist. In our epidemic model, those unreported cases are included indirectly through infected to confirmed transitions. 45 Since a vaccine is not available for COVID-19, contact tracing is a key mitigation strategy to control the spreading of COVID-19. Contact tracing is a mitigation 3 . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint strategy that aims at identifying people who may have come into contact with a patient. This mitigation strategy prevents further spreading by isolation of exposed people. The public health personnel have used contact tracing as a 50 tool to control disease-spreading for a long time [5] . We implement the contact tracing strategy according to CDC guidance [6] through a two-layer network model with a modified SEIC epidemic model. Contact tracing is effective at the early stage of an epidemic when there is a limited number of cases. We choose a college town, Manhattan (KS), for our study. Most college towns have a lim-55 ited number of cases because educational institutes have been closed since early March 2020. Feasibility of contact tracing to control COVID-19 spreading was analyzed using a branching process stochastic simulation for three reproductive ratios R 0 = 1.5, 2.5, and 3.5 [7] . The authors find that sufficient contact tracing with quarantine can control a new outbreak of COVID-19. They mostly focus 60 on the question of how much contacts need to be traced to control an epidemic for the three levels of reproductive ratio. However, this article neither explored the effectiveness of contact tracing for a specific location, nor investigated the cost of contact tracing. In this research, we develop an individual-based network framework to assess the 65 impact of contact-tracing in the reopening process in a college town of Kansas. To analyze the cost of contact-tracing represented by the number of quarantined susceptible people, we develop a contact network and estimate the reproductive ratio R 0 and confirmed rate (infected to laboratory-confirmed transition) from observed confirmed case data in Manhattan KS. We use our individual-based . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint • A rigorous estimation of the reproductive ratio R 0 and confirmed case rate (infected to laboratory-confirmed transition) from observed confirmed case data. • A thorough investigation of costs and benefits of contact-tracing in the reopening process in a college town of Kansas. The individual-based network model is developed to represent the heterogeneity in people mixing. Our individual-based network epidemic model is general and flexible. It can be used to estimate, and model contact-tracing for COVID-19 85 in any location. It can also be used for any other disease that has a similar spreading mechanism like COVID-19. This paper is organized as follows: section 2 proposes an individual-based contact network framework with two networks: the full network and the limited network, to represents the contact situation namely before the reopening pro- This section proposes a method to develop an individual-based contact network model capable of representing heterogeneous social mixing. In this network, occupied households are in the individual node level, a connection be-100 tween two households represents the contact probability between members of these households. The network has N nodes and n people. To develop this network, we consider five age-ranges: under 18, 18 − 24, 25 − 34, 35 − 59, and over 60. Each age-range has n i people, where i ∈ {1, 2, 3, 4, 5}. We distribute the n people randomly into the N occupied households according to five social char-105 acteristics: age, average household sizes, family households, couple, living-alone 5 . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint [3] . We maintain the average household sizes, number of family households, number of couples, and number of living-alone households. Besides, a person under 18 years old is always assigned in a house with at least one adult person. After assigning the people, an age-specific network is developed for each age 110 range and a random mixing network for all ages. Then a combination of the six networks provides the full network. A full network represents a contact network for a typical situation. The configuration network model [8] is used to develop age-specific networks and the random mixing network. The steps to develop age-specific networks are: Step 1: For each person j (here, j ∈ 1, 2, ..., n), contacts c j is assigned from a Gaussian distribution N (µ, σ 2 ). The mean µ of the Gaussian distributions are taken from the daily average number of contacts per person in each age-range [9, 10, 11] . The average daily contacts per person are given in Table 1 In the random-mixing-network, the number of contacts is assigned randomly from the N (2, 1) distribution for a person j. The Gaussian or normal distribution is the distribution of real numbers; therefore, the 130 number from the N (µ, σ 2 ) distribution is rounded to the closest integer. Step 2: For each person j, contacts for its belonging household k is assigned by (c j − h k − 1). Here, c j is the number of contacts for a person j, h k is the household size or number of people of the household k, person j lives in the household k, j = 1, 2, 3......n, and k = 1, 2, 3......N . 135 6 . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint Step 3: From the mixing patterns of different age-ranges, people have a strong tendency to meet people with their same age range (more than 80%) [9, 10, 11] . Therefore, We keep the maximum number of contacts among the same age ranges and a small percentage for the other age ranges. The percentage of contacts in the same age-specific-network for each age-range 140 is given in Table 1 . Degree d ki of a node k in the age-specific network i is s% of (c j − h k − 1), here, s% of average daily contacts of a person happens with the people of his same age-range. Step 4: After assigning degree, d ki for N nodes or households, The configuration network model [8] creates half-edges for each node, then chooses two 145 nodes randomly and connect their half-edges to form a full edge [8] . The population and network characteristics for the five age-specific networks for Manhattan, KS are given in Table 1 . According to census 2018, Manhattan, KS has n = 55, 489 people and N = 20, 439 occupied households [3] . The full network is a combination of five age-specific networks and a random-150 mixing network. Adjacency matrix for the full network A f is a summation of six adjacency matrices: Here, A i is the adjacency matrix for the age-specific network i, and A r is the adjacency matrix for the random mixing network. Age-specific networks and the random mixing network are unweighted and undirected. However, the full network is a weighted and undi- CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint which is closed since early March 2020. Besides, Manhattan, KS was under the "Stay-At-Home" order from March 27, 2020 to May 4, 2020 [12] . To represent this unusual situation, the full network is modified to a limited network version. As the educational institute was closed, we randomly reduce 90% links from the age-specific networks for the age-ranges under 18, and 18 − 24. The Google COVID-19 community mobility reports provide a percentage of movement changes in different places (for example, workplaces, recreational areas, parks) [13] . We reduced 40% links randomly from the age-specific networks for 25 − 34, and 35 − 59 age-ranges for the movement changes in the workplaces [13] . The number of links in the limited network is 155762. The limited network 175 is given in the supplementary materials. . CC-BY-NC 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 June 26, 2020. In this section, we design an epidemic model to simulate the COVID-19 spreading; later, we estimate the unknown parameters (reproductive ratio R 0 180 and confirmed rate δ 2 ) of the epidemic model. We simulate four reopening scenarios using the estimated parameters: under "Stay-At-Home" order, 25% reopening, 50% reopening, and 75% reopening. This model assumes that there is no particular mitigation strategies have applied except general lockdown. This research propose a susceptible-exposed-infected-confirmed (SEIC) epidemic model to simulate the spreading of COVID-19 (Fig. 2) . This model has four compartments: susceptible S, exposed E, infected I, confirmed C. A susceptible node does not introduce to the virus yet, an exposed node introduces 190 to the virus, but the viremia level is not strong enough to infect others, an infected node has strong viremia to infect others, and a confirmed node is a laboratory-confirmed COVID-19 case. The SEIC model has three transitions, which are divided into two categories: edge-based (S → E), and nodal (E → I; 15] . An edge-based transition of a node depends on the state of its contacting nodes or neighbors in the contact network with its own state. A nodal transition of a node only depends on the own state. Each edge-based transition has an influencer compartment. A transition from susceptible to exposed (S → E) of a susceptible node depends on the infected neighbors of that node. Therefore 200 it is an edge-based transition, and the infected compartment is the influencer compartment of this transition. In this work, we are using the term 'neighbors of a node k' for the nodes, which have the shortest path length 1 from the node k. The transition rate of the susceptible to exposed (S → E) transition of a node k is β 1 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint is the number of infected neighbors of the node k. The transition rate for the transition exposed to infected (E → I) is δ 1 . The confirmed rate of an infected person is δ 2 . We assume that a laboratory-confirmed case will be isolated and cannot spread the disease outside of his household anymore. An infected node To do the simulation, we use GEMFsim; it is a stochastic simulator for the 220 generalized epidemic modeling framework (GEMF), which was developed by the Network Science and Engineering (NetSE) group at Kansas State University In GEMF, the joint state of all nodes follows a Markov process that arises from 11 . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint Table 2 : Description of the susceptible-exposed-infected-confirmed (SEIC) epidemic model. type transition transition rate estimated node-level transition. A node can change its state by moving from one compartment to another compartment through a transition. One assumption of the GEMF system is, all the events or transitions are independent Poisson processes with the constant rate; this assumption leads the system to a continuous-time 230 Markov process. Initially, the simulation starts by setting two infected nodes randomly. The SEIC model has two unknown parameters: reproductive ratio R 0 , and 235 confirmed or reporting rate δ 2 . To estimate the R 0 and δ 2 , we have used con- for reporting rate δ 2 is 1 6.54 day −1 (95% confidence interval: 1 7.89 − 1 6.05 day −1 ). . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint These estimated values are specific for Manhattan, KS. We have considered that some people will develop severe symptoms, and they will be reported as a con-245 firmed case of COVID-19 sooner. However, some people will produce deficient symptoms, and may they will not be tested. Therefore, the estimated confirmed rate is an average of all possibilities. We use approximate Bayesian computation based on sequential Monte Carlo sampling (ABS-SMC) approach to estimate the parameters [18, 19] . A sensitivity analysis for R 0 and δ 2 on the mean-squared error between confirmed cases data and simulated results is presented in Fig. 6 . From the sensitivity analysis, the mean-squared error is low when the reporting time is high. It indicates undetected COVID-19 patients in the system. It means that an infected node needs to be infected for a longer time for the better fitting with 255 the data. It also indicates that the testing of COVID-19 is not sufficient. In this subsection, we simulate the confirmed cases (or cumulative new cases per day) for two months: June and July using the SEIC epidemic model with the estimated parameters. To simulate, we assume that there is no change 260 except reopening from pandemic lockdown. We are presenting four reopening situations: Stay-at-home is still there or no reopening, 25% reopening, 50% reopening, and 75% reopening. Kansas has started to reopen step by step after May 4, 2020. We use the limited network to simulate from March 25, 2020 to May 4, 2020; then, we change the network concerning the reopening situation. For example, for a 25% reopening situation, we add 25% missing links randomly (which are present in the full network but not in the limited network). We preserve the states of each node at May 4, 2020 in the network then use it as the initial condition for the simulation for the reopening situation (from May 4, 2020 to July 1, 2020). Fig. 4 is showing the mean (dashed lines) and median 270 (solid lines) of the confirmed cases of the 1000 stochastic realizations of the four reopening situations. The zoom-in window in Fig. 4 shows the time period when data was used to estimate the parameters of the epidemic model. . CC-BY-NC 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 June 26, 2020. . https://doi.org/10.1101/2020.06.24.20139204 doi: medRxiv preprint . The light-colored boxes represent more mse than dark-colored boxes. The color boxes with number "1" means that mse≤ 3, number "2" means that 3