key: cord-1013297-a5h726ly authors: Huang, Q.; Mondal, A.; Jiang, X.; Horn, M. A.; Fan, F.; Fu, P.; Wang, X.; Zhao, H.; Ndeffo-Mbah, M.; Gurarie, D. title: SARS-CoV-2 transmission and control in a hospital setting: an individual-based modelling study date: 2020-08-25 journal: nan DOI: 10.1101/2020.08.22.20179929 sha: b0fd71656983c4d43f85814460271a29cf1bb9ea doc_id: 1013297 cord_uid: a5h726ly Background: Development of strategies for mitigating the severity of COVID-19 is now a top global public health priority. We sought to assess strategies for mitigating the COVID-19 outbreak in a hospital setting via the use of non-pharmaceutical interventions such as social distancing, self-isolation, tracing and quarantine, wearing facial masks/ personal protective equipment. Methods: We developed an individual-based model for COVID-19 transmission among healthcare workers in a hospital setting. We calibrated the model using data of a COVID-19 outbreak in a hospital unit in Wuhan in a Bayesian framework. The calibrated model was used to simulate different intervention scenarios and estimate the impact of different interventions on outbreak size and workday loss. Results: We estimated that work-related stress increases susceptibility to COVID-19 infection among healthcare workers by 52% (90% Credible Interval (CrI): 16.4% - 93.0%). The use of high efficacy facial masks was shown to be able to reduce infection cases and workday loss by 80% (90% CrI: 73.1% - 85.7%) and 87% (CrI: 80.0% - 92.5%), respectively. The use of social distancing alone, through reduced contacts between healthcare workers, had a marginal impact on the outbreak. A strict quarantine policy with the isolation of symptomatic cases and a high fraction of pre-symptomatic/ asymptomatic cases (via contact tracing or high test rate), could only prolong outbreak duration with minimal impact on the outbreak size. Our results indicated that a quarantine policy should be coupled with other interventions to achieve its effect. The effectiveness of all these interventions was shown to increase with their early implementation. Conclusions: Our analysis shows that a COVID-19 outbreak in a hospital's non-COVID-19 unit can be controlled or mitigated by the use of existing non-pharmaceutical measures. 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 August 25, 2020. Developing strategies for mitigating the severity of COVID-19 is now a top global health priority. 52 The range of containment strategies employed in different countries and regions varies from shelter-53 in-place orders, the shutdown of public events, travel ban [1] , and visitor quarantine, to intermediate 54 steps that involve partial closures (e.g. schools [2] , workplaces, sporting, and cultural events) [3] . 55 While such drastic steps can reduce infection spread, they exact a heavy toll on society and human 56 well-being. At present the only available means of containing COVID-19 spread is via the use of 57 non-pharmaceutical interventions [4, 5] such as social distancing, self-isolation [6] , tracing and 58 quarantine [6, 7] , wearing facial masks/ personal protective equipment (PPE) [8, 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. The copyright holder for this preprint this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint Our focus here is COVID-19 transmission in a hospital setting, where healthcare workers (HCWs) 68 are at high risk to acquire infection through interactions with fellow HCW and with patients [19] [20] [21] [22] . 69 We developed a novel individual-based model (IBM) for COVID-19 transmission among HCWs, 70 and applied it to explore the efficacy of different control/ mitigation strategies via non- The IBM model was calibrated in a Bayesian framework using empirical data from a non-COVID 79 hospital unit. We used our calibrated model to simulate different intervention scenarios, including 80 adaptive behavior (social distancing in the workplace, individual protection, isolation of infected 81 individuals). In each case, we assessed the effect of interventions on outbreak outcomes: outbreak 82 size, outbreak duration, and workday loss. 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 August 25, 2020. we assign positive infectivity levels ( 0 , 1 , 2 ) to all three stages ( , 1 , 2 ). 91 We modelled social mixing patterns by assuming that HCWs and ward patients interact on a daily 92 basis via aggregating in random groups of HCWs, and via patient visitation by HCWs (See SI in 93 details). The net outcome is a contact pool for each HCW-host, which varies randomly on a daily . 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 August 25, 2020. Figure 1 ). The model simulations were run on a daily basis and implemented in the Wolfram Mathematica HCWs and infected patients; 6) daily isolation of symptomatic cases and recovery (See SI in details). . 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 August 25, 2020. 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint stopped in the department, and the hospitalized patient pool was gradually reduced from 200 to 20 153 by the beginning of February. Over the period from January 5th to February 4th, 92 out of the 171 154 HCWs of the department were suspected or confirmed COVID-19 cases and isolated. New patients 155 were only admitted in early March 2020 when the pandemic was declared under control in Wuhan. Intervention strategies 157 We consider three types of interventions: (i) social distancing (reduced contact rates) among HCWs For our baseline case, we assumed 50% and 100% isolation (quarantine) fractions of symptomatic 161 cases ( 1 ; 2 ), respectively, and fixed infection level of the patient pool (see Table S1 ). To account 162 for model uncertainties, we run each control simulations for 100 posterior parameter samples and 5 163 stochastic model realizations for each sample (500 histories altogether), over a six-month period. For social distancing, we considered 50% and 75% contact-rate reduction relative to their baseline symptoms. An increase in targeted isolation assumes more intensive screening or test sensitivity. We 169 also studied the effect of isolating pre-symptomatic/ asymptomatic cases ( -pool). This task is more 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint quarantine pool over the outbreak duration. The latter gives a simple economic measure of outbreak 175 impact and putative interventions. In each control experiment, we compare the ratio of two outputs 176 (outbreak size and workday loss) to their baseline values, and record these relative values and their 177 distribution. Another important factor in the hospital setting is the in-patient pool. In our case (a non-COVID unit 179 in Wuhan), it varied from the full capacity to zero. The key inputs of the patient pool included (i) 180 infected prevalence, (ii) mean patient infectivity to HCWs. The former is controlled by patient 181 admission and screening/isolation procedures; the latter can be modulated by using PPE. We also 182 explored the effect of different timing of PPE implementation and its efficacy. Model calibration 185 The predictions from the calibrated IBM were very close to the observed data on daily symptomatic 186 and quarantine cases (see Figure 2 ). The fraction of asymptomatic disease-progress pool, A ν , was 194 . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint 195 The baseline scenario showed that almost all HCWs get infected, resulting in significant workday 196 loss, 1050 (90% CrI: 913-1282) over the six-month period (Table 1 Figure 3 illustrates the combined effect of facemask and social contact. We used the same values of 208 facemask efficacy and contact rates as Table 1 . For each value of facemask efficacy, we observed a 209 consistent reduction of the outbreak size with reduced contact rates. It varied from 13% -34% drop 210 for low-efficacy facemask (50% protection), to 30% -60% drop for high-efficacy facemask (95% 211 protection). We observed a similar percentage reduction for the workday loss. So the impact of 212 reduction of contact rates was much greater under the higher efficacy of facemasks. 213 We also explored the effect of timing of intervention by the following three scenarios: (1) at the 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint high-efficacy facemasks and reduced contact rates) were implemented at the start of the outbreak, 218 we observed 80% -90% reduction of the outbreak size (a near-complete control). A later 219 implementation (e.g. after the first identified case), gave 60% -85% reduction. If the timing was 220 delayed to e.g. 10% identified cases, these numbers dropped to 40% -60%. All intermediate cases 221 were shown in Figure. 3. We next looked at the effect of HCWs screening and isolation via two scenarios. The first scenario , f f at (90%, 100%), and varied -fraction from 10% to 60%. We still found the 232 effect of such a strategy was limited, it often prolongs the outbreak duration without affecting its 233 size. Besides, such a strategy can incur an economic burden by increased workday loss, though the 234 effect is subtler, as increased quarantine rate can slower transmission rate, hence fewer hosts would 235 be infected and need isolation. More significant progress was achieved by controlling the patient 236 source, via reduced patient prevalence (screening), or reduced infectivity (PPE) (see Figure. 4A-B). 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint 13 use. We assumed PPE provides 80% protection (via the reduced probability of transmission from an 240 infected patient). We also varied the infected prevalence level of the patient pool, from 0% to 5% 241 (baseline case was 2%). We found the control of patient infection (via e.g. PPE, screening and 242 isolation, particularly for new patients) can reduce outbreak size, even though the bulk of 243 transmission is carried over by inter-staff HCW contacts. We found the combined strategy (enhanced 244 HCW screening/isolation with patient control) could lead to marked improvement both in outbreak The current model setup is subject to some limitations. First, it was designed for a single hospital 302 unit and simplified treatment of the patient pool, as the target group in our study was HCW-pool. More realistic local communities could combine multiple units (e.g. large hospital), with refined 304 population structure (e.g. patients, visitors, staff), and more complex interactions (e.g. 'random' and 305 . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint 'scheduled' contact pools). Empirical data on these interactions will be required to adequately 306 parameterize such models. Second, although we have made an effort to characterize the SARS-CoV- . 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 August 25, 2020. informed consent was required before the data collecting, and participants were informed that they 323 could refuse to answer any question. The questionnaire did not ask about infection status, and no 324 biological samples were collected. 326 Not applicable. 328 The computer codes, including the aggregated data, implemented in the Wolfram Mathematica 332 The authors declare that they have no competing interests. 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 August 25, 2020. The authors would like to appreciate all healthcare workers in this study. XJ and HZ had full access 348 to all the data in the study and took responsibility for the integrity of the data. . 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 August 25, 2020. Lancet Public Health 2020. . 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 August 25, 2020. . 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 August 25, 2020. . 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 August 25, 2020. shows the corresponding predicted daily pre-symptomatic/ asymptomatic cases, , and symptomatic 482 cases, 1 + 2 , respectively. The grey shaded regions are 90% credible intervals. . 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 August 25, 2020. 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint B). We used the following marking: (pink) baseline patient infection level, (red) reduced patient 499 infection by 80% via PPE use by HCWs. In column A, the quarantine fraction of moderate/severe 500 cases ( 2 ) was fixed at 100%, and the quarantine fraction of mild cases ( 1 ) was varied from 60% to 501 100%. In column B, we fixed symptomatic ( 1 ; 2 ) quarantine fractions at (90%, 100%), and varied 502 the quarantine fraction of pre-symptomatic/ asymptomatic -pool from 10% to 60%. ii) after the first identified HCW infection, iii) no PPE use. We considered different levels of 505 prevalence of the infected patient pool: 0%, 1%, 2% (baseline value), 3%, 4%, 5%. . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint 1 Tables Table 1: Effects of implementing social distancing through reduction of contacts alone and wearing face masks alone, from the start of the outbreak. We simulated a six-month intervention-regimen for the calibrated model. The progress was measured in terms of outbreak size, workday loss, and cumulative quarantine incidence. Reasonable (50% and 75%) reduction of contact rates and levels of efficacy of facial masks (50%, 67%, 75%, 85%, 95%) were chosen. The results shown are predicted median (90% Credible Interval The 2) Initial infection state. We assume that only one HCW was in the latent E stage at the beginning of the simulation, while the rest 170 HCWs were susceptible; 3) Stage-specific infectivity levels ( 0 , 1 , 2 ) for ( , 1 , 2 ) . We assume 1 = 2 , for symptomatic groups ( 1 , 2 ) , while asymptomatic infectivity 0 = 0.55 * 1 [2] . Then we calibrated 1 to be 0.2245 (90% CrI: 0.202 -0.262) using the Bayesian method, consequently, 0 = 0.124 (90% CrI: 0.111 -0.144); 4) Individual susceptibility level ( ). In general, may depend on host health /immune status, individual behavior e.g. use of facial masks or PPE, and environmental conditions, = 0 (fully protected), = 1 (fully susceptible). Work stress is another factor that can affect susceptibility. According to [3] . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint 6) Social mixing patterns between HCW, and external patient sources; Intra-HCW contacts were simulated as daily aggregation in small groups of 2,3, or 4 hosts, randomly drawn from the current working HCW pool. The basic case (full HCW working pool) assumes 60 pair-contacts, 30 triple-contacts, 8 quadruple-contacts per day. It approximately gives a 2.2 contact rate per HCW-host per day from hospital data ( Figure S2 ) [3] . Larger host aggregations are possible, but we ignore them here. Besides internal mixing, each HCW-host visited patients at a prescribed rate = 20/per day. Unlike HCW staff, the patients were not individualized, but a random patient cluster (determined by mean visitation number per day) was drawn from the total patient pool (200) with a prescribed infected fraction (0.02), using a hypergeometric distribution. According to hospital data, no new patients were admitted after January 19th, and their pool was discharged after January 19th, at a rate of 5% /day, and HCWs will wear PPEs in face of patients after Jan 19th, so we assume patient infectivity, , is decreased to be 0.02, i.e., the efficacy of PPE is about decreasing risk of infection by 80%, when we calibrated our IBM. Daily isolation of symptomatic cases and recovery. We assumed that HCWs expressing symptoms are tested, and prescribed fractions ( 1 ; 2 ) of ( 1 ; 2 ) put in isolation. The assumed quarantine fractions (0 < < 1) combine limited test sensitivity, and overlapping 'COVID-like' symptoms, expressed by other (non-COVID) hosts. According to data, the Union Hospital in Wuhan has much more strict quarantine policy after January 19th, when they were aware of the seriousness of COVID-19, so we fixed 1 to be 0·1 and 0·8 before and after January 19th, respectively, and fixed 2 to be 0.15 and 0.85 before and after Jan 19th, respectively, when we calibrated our IBM. . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint ( 1 , 1 ) The average duration of mild symptomatic ( → 1 → ) (5;9) [1, [4] [5] [6] [7] [8] ( 1 , 1 2 , 2 ) The average duration of moderate/severe symptomatic ( → 1 → 2 → ) ( 0 , 0 ) initial susceptible population; initial exposed population (170;1) Data . 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint World Health Organization, Organization WH: Report of the WHO-China joint mission on coronavirus disease 2019 (COVID-19) Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARS-CoV-2) SARS-CoV-2 infection in health care workers: a retrospective analysis and a model study The incubation period of coronavirus disease 2019 (COVID-19) from publicly reported confirmed cases: estimation and application Clinical characteristics of coronavirus disease 2019 in China Characteristics of and important lessons from the coronavirus disease 2019 (COVID-19) outbreak in China: summary of a report of 72 314 cases from the Chinese Center for Disease Control and Prevention Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. The lancet 2020 Figure S1 . A typical daily snapshot of transmission pathways resulting from random mixing. Table at the bottom shows selected contact pools, and the graph arrows indicate which susceptible hosts (grey) were potentially in contact with infective ( − ) hosts (red). In our scheme, a single 'red' can infect multiple susceptibles, and a susceptible (grey) can be linked to multiple infectives.. 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 August 25, 2020. . 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 August 25, 2020. . 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 August 25, 2020. 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint quarantine pool, respectively. Panel (E) shows daily infection incidence ( → ). Panel (F) shows daily quarantine incidence (yellow) and quarantine count (blue), respectively.. 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 August 25, 2020. . https://doi.org/10.1101/2020.08.22.20179929 doi: medRxiv preprint