key: cord-0997786-4nd5wzrm authors: Xiao, S.; Li, Y.; Sung, M.; Wei, J.; Yang, Z. title: A study of the probable transmission routes of MERS‐CoV during the first hospital outbreak in the Republic of Korea date: 2017-10-23 journal: Indoor Air DOI: 10.1111/ina.12430 sha: 0b4b3ff81ed14d80978af1bd116493a6aea3fe23 doc_id: 997786 cord_uid: 4nd5wzrm Infections caused by the Middle East respiratory syndrome coronavirus (MERS‐CoV) are a serious health issue due to their prevalence and associated mortality. However, the transmission routes of the virus remain unclear, and thus, the current recommended control strategies are not evidence based. In this study, we investigated the transmission routes of MERS‐CoV during the first nosocomial outbreak in the Republic of Korea in May 2015 using a multi‐agent modeling framework. We identified seven hypothesized transmission modes based on the three main transmission routes (long‐range airborne, close contact, and fomite). The infection risks for each hypothesis were estimated using the multi‐agent modeling framework. Least‐squares fitting was conducted to compare the distribution of the predicted infection risk in the various scenarios with that of the reported attack rates and to identify the hypotheses with the best fit. In the scenarios in which the index patient was a super‐spreader, our model simulations suggested that MERS‐CoV probably spread via the long‐range airborne route. However, it is possible that the index patient shed an average viral load comparable to the loads reported in the literature, and that transmission occurred via a combined long‐range airborne and close contact route. As a novel human coronavirus with alarming morbidity and mortality, 1 Similar to other coronaviruses, MERS-CoV is thought to be shed mainly in the respiratory secretions of infected individuals and spread through respiratory activities such as coughing. However, the precise human-to-human transmission routes of the virus remain unclear. 6 As shown in Figure 1 , three major possible transmission routes are known to exist for communicable respiratory infections: the long-range airborne route, the close contact route, and the fomite route. [7] [8] [9] The current dominant view is that MERS-CoV largely spreads via the respiratory close contact route, 10, 11 which is supported by evidence that most of the affected healthcare workers (HCWs) have been nurses who have had prolonged and repeated close contact with MERS-CoV patients. 12 Nevertheless, the possible roles of airborne and fomite transmission cannot be excluded. Research suggests that MERS-CoV may be spread through long-range airborne transmission. For example, the viral RNA has been detected on the entrance to air-ventilating equipment, 13 Healthcare settings have been investigated as breeding grounds for large MERS outbreaks because many cases of infection have not been recognized and isolated in time. 3, 5, 18 Similarly, in the Republic of Korea, almost all cases are suspected to have been hospital acquired, originating from Pyeongtaek St. Mary's Hospital. 19 The index patient for the outbreak in this hospital was a 68-year-old man who had recently travelled to the Middle East and was hospitalized from 15 to May 17, 2015 with no isolation or protection measures implemented. 4 He was thought to have directly infected 26 people, including 11 other patients, 13 visitors or family members, and two HCWs. 4 Compared to previous nosocomial outbreaks on or near the Arabian Peninsula, this outbreak was larger 20 and the distribution of infection showed a clear spatial pattern, although the sample size was not sufficiently large for this pattern to be statistically significant. To investigate the probable transmission routes of MERS-CoV, we carried out a detailed modeling study of the outbreak in Pyeongtaek St. Mary's Hospital. We retrospectively analyzed the spatial pattern of infection and developed a multi-agent model to simulate the possible spread of the virus from the index patient's ward by air flow, close contact, and surface touching. Table 1 lists the seven hypotheses examined in the study: three for single routes, three for two-route combinations, and one for a three-route combination. The dose-response relationship model was used to calculate the possible infection risk for each hypothesis. A major challenge with this approach is the lack of appropriate input data, such as dose-response parameters. Hence, a range of such parameters was considered with various value combinations of important parameters. Least-squares fitting was carried out in 1140 scenarios to compare the distribution of the infection risk with that of the reported attack rates. The results presented below provide probable evidence for the transmission of MERS-CoV in different scenarios. Our analyses allowed us to retrospectively acquire the largest virus-containing droplet sizes, dose-response parameters, and virus loads. As revealed by Google Maps and verified by our field visits, Pyeongtaek St. Mary's Hospital is located in a rather open area and is much taller than all of the surrounding buildings, except one, as shown in Figure 2 . The outbreak occurred on the eighth floor of the hospital, for which the floor plan is shown in Figure 3 . The floor plan, excluding • To the best of our knowledge, this study is the first mechanism-based investigation of the transmission route of MERS-CoV during the first hospital outbreak in the Republic of Korea. Our findings on the possible roles played by the long-range airborne and close contact routes suggest the importance of ventilation systems and droplet precautions in healthcare environments. Our study also reveals the importance of obtaining more accurate data on dose-response parameters, the transport of bio-aerosols in air and on surfaces, and virus survival in air and on surfaces. F I G U R E 1 The three major transmission routes: close contact, fomite, and long-range airborne routes. The person in red is the index patient the elevators and storerooms, was used as the computational domain. The index patient stayed in the hospital from 15 to 17 May 2015, and thus, this period was taken as the suspected exposure period. The symptoms of the index patient included fever, cough, dyspnoea, and myalgia. 21 The eighth floor contained one nurses' station and 32 wards, comprising 19 single rooms (Rooms 8201-8219), six double rooms (Rooms 8101-8106), and seven large rooms for five (Room 8109) or seven persons (Rooms 8107 and 8108, and Rooms 8110-8113). According to the hospital staff, nurses and cleaners visited patients on the floor from Room 8101 in a clockwise direction to Room 8219. A mechanical ventilation system, which included a supply and an exhaust, was installed in all rooms except the index patient's room, Room 8104. 22 The mechanical ventilation rate was assumed to be 10 L/s per person. Each room had a separate ceiling fan coil unit to provide cooling and heating, and each toilet had a separate exhaust system. The toilet exhaust air flow rate was assumed to be 6 ACH. There were two small windows (each 35 × 35 cm) in the single rooms and double rooms, four in the large rooms, and 10 in the corridor. The doors and windows in the regular rooms were usually closed. Because there was no ventilation system in the index patient's room, we assumed that the door and windows in this room were slightly (15%) open during the day and completely closed at night. Mary's Hospital. We studied the infection patterns of the inpatients because they stayed on the floor longer and more data are available. During the period in which the index patient was hospitalized, 11 MERS cases of 67 normal patients were reported in the first generation on the eighth floor. 4 We assumed that all of the exposure doses Because we evaluated the seven hypotheses based on the differences between the attack rate and predicted infection risk in different regions, the spatial distribution characteristics of the different hypotheses were important factors in our study. However, the existing models, such as the discrete-time Markov chain model 8, 23 and the differential equation model, 24, 25 are commonly used to estimate overall infection risk, and thus predict uniform distributions. Therefore, we developed a multi-agent model, 26 which enabled us to model the individual heterogeneities caused by the agents' behavior and geographical environment. 27, 28 Figure 4 shows the system architecture of our model. The system is modular based and consists of four basic components: the initialization generator, simulation engine, global database, and data processing module. Appendix S1: Fig. S1 shows the basic procedures included in the program for the model. The initialization generator has two branches, namely the geometric generator and agents generator. The geometric generator is used to T A B L E 1 Hypotheses on the transmission modes of MERS-CoV based on the three basic transmission routes in Figure 1 . A transmission mode can be either a single route or combined routes Table S2) and virus inactivation rates (Appendix S2: Table S3 ). The agents generator was used to create representative individuals during the outbreak. Five representative "resident" roles were identified as study objects, that is, inpatients, visitors, family members, nurses, and cleaners. The numbers of agents are shown in Appendix S2: Table S8 . Each autonomous agent directly corresponds to one human individual in the real world. The simulation engine, which comprised behavior models of the representative roles, was used to simulate the behavior of the agents. In the behavior models, rules were set to regulate the time sequences For the close contact route, the normal distribution model 36, 37 was used to acquire the movement patterns in the corridor and the contact probability of any pair of two agents was derived. Based on a respiratory jet assumption, 8 Based on the above parametric model, we fitted the parameters (c a , c c , c f , η r L 0 , η m L 0 , d g ) to the attack rates during the outbreak using a standard approach, namely least-squares fitting. 40 In this approach, the residual sum of squares (RSS) is calculated as a measure of fit, with a small RSS indicating a good fit of the model to the data. Therefore, we formulated the following minimum mean squares error problem: where N is the number of simulations, n is the number of divided regions, λ j is the number of inpatients in the jth region, I i,j is the predicted infection risk in the jth region generated by the ith simulation, and A j is the reported attack rate in the jth region during the outbreak. Because this problem involves the nonlinear function I i,j and integer constraints, it is generally NP hard. Therefore, we applied a greedy search to solve the problem. In this study, a scenario was defined as a value combination of (η r L 0 , Because the agents' behavior retained partial randomness Therefore, for wards of the same size, more aerosols spread to the downstream wards than to the adjacent wards. However, because the toilet exhaust fans were run at the same ventilation rates for all rooms, the negative pressures in the small rooms were higher than those in the large rooms, and thus, the infection risk in the small rooms in the F I G U R E 5 Spatial distribution of the predicted average infection risk (for 1000 simulations) via three transmission routes at 24:00 on May 17, the end of the computational period. A, Long-range airborne route. B, Close contact route. C, Fomite route. The largest virus-containing droplet size d g = 100 μm, dose-response parameters in respiratory tracts η r = 3.2/mRNA copy and on mucous membranes η m = 3.2 × 10 −3 / mRNA copy, and the viral load L 0 = 10 10 mRNA copies/mL. Room numbers and the average infection risk in the room are marked in blue and red, respectively. Empty beds are marked in gray and occupied beds are red. Different levels of infection risk are represented by the intensity of red shading adjacent wards was higher than that in the large rooms in the downstream wards ( Figure 5A ). The infection risk was lowest in the remote wards ( Figure 5A ). The remote wards had at least two zones, Zones 34 and 36, between them and the source ward (Appendix S1: Fig. S2 ), so the aerosol concentration was further diluted. Furthermore, during the day, the air flow could transport aerosols from Zone 34 to Zone 36 only when the wind was from the northwest (Appendix S3: Fig. S6 ), which meant that most of the time there were no virus sources for Rooms 8113 and 8201-8219. Similarly, in the remote areas, the infection risk in Rooms 8213-8217 was even lower than that in Rooms 8218 and 8219 ( Figure 5A ). For the close contact route, with the assumed four parameters, the transmission was very effective compared to the other two routes. Thus, the infection risk for a susceptible person via this route was mainly determined by whether a person met with the index patient. Under the normal distribution assumption, people would be more likely to stay in nearby areas than in remote areas so the index patient and people nearby would share a similarly high chance of appearing in the neighborhood areas and have more opportunities for close contact, resulting in a sequentially decreasing risk of infection in the source, adjacent, downstream, and remote wards ( Figure 5B ). For the fomite route, the infection risk was mainly induced by the nurses' routine rounds and contaminated environmental surfaces. We assumed that the nurses conducted routine rounds five times a day at 07:00, 11:00, 15:00, 18:00, and 21:00 (Appendix S2: Table S3 ). In most cases, patients have the same opportunities to come in contact with common surfaces, and thus, these surfaces contribute to a uniform spatial distribution. However, in this outbreak, the environmental surfaces common to all patients were not a significant factor. The private toilets in the rooms were preferentially chosen by patients and family members (an average of 21 times/day for each room), while the common toilets in the corridor were used at a low frequency (an average of 5 times/ day for each). Therefore, the overall distribution mostly complied with the direction of the routine round ( Figure 5C ). . In these scenarios, the parameters were relatively small, so all seven hypotheses predicted very low infection risks, which deviated greatly from the reported attack rates. As d g increased, the predicted infection risks became higher and thus the numbers of these kinds of scenarios decreased. The scenarios with a d g of 20 μm are shown in Figure 6A . In these scenarios, our assumed d g was smaller than the largest initial diameter for the airborne droplets (30 μm), meaning that all of the virus-laden bio-aerosols were airborne droplets and could remain suspended in the air for a long time. Therefore, the viruses on the surfaces mainly originated from the deposition of airborne droplets, which caused negligible infection risk due to the fomite route. In the scenarios in which the products of the viral load and dose-response parameters in respiratory tracts η r L 0 were no more than 10 10.50 /ml, the infection risk caused by any single-route mode was small, so the probable transmission mode was Hypothesis 4 [Long air + Close] (green dots). In scenarios with η r L 0 ranging from 10 11 to 10 11.25 /ml, the infection risk caused by any single route was quantitatively comparable with the attack rates, so the probable mode was Hypothesis 1 [Long air] (red dots), which is also qualitatively consistent with the reported attack rates. In the remaining scenarios, the infection risks caused by the combined routes were too large and those caused by the long-range airborne route were either too large or too small, while those caused by the close contact route were relatively stable, so the probable transmission mode was Hypothesis 2 [Close] (orange dots). The scenarios with d g values larger than 30 μm are shown in Figure 6B -D. When the values of the products of viral load and T A B L E 3 Hypotheses with the best fit (the minimum RSS) in the eight more-likely scenarios. The eight scenarios are indicated in Figure 6 with Roman numerals. The dose-response parameter in the respiratory tract η r is 3.2/mRNA copy, and that on the mucous membranes η m is 3.2 × 10 −3 /mRNA copy dose-response parameters in respiratory tracts η r L 0 and on mucous membranes η m L 0 were relatively small, the exposure doses due to any single-route or double-route mode were low, so only the combination of three routes (purple dots) caused an infection risk quantitatively comparable to the attack rates, although the fit was not good. As d g increased, the infection risks due to the single-route and double-route modes increased, and thus, the number of scenarios with the threeroute mode (purple dots) decreased. In this study, to evaluate the transmission hypotheses, we estimated the spatial distribution of the infection risk under each hypothesis and then compared these results with the reported outbreak data. Therefore, a good estimation model was needed to reflect the underlying mechanisms of the hypotheses. Nevertheless, the models, such as the discrete-time Markov chain model 8, 23 and the differential equation model, 24, 25, 43 were not applicable due to their system-based feature, which removes the individual heterogeneity of infection risk. 28 In fact, the human behavior in those models is merely described by behavioral frequency. However, in practice, human behavior is also dependent on behavioral rules and geometry constraints. To address the above problems, we used a multi-agent model 26 to analyze the multi-route transmission. Different from the existing system-based models, the agent-based model is able to characterize the diversity of individuals. More specifically, the behavior of individuals is simulated according to behavioral rules and geometry constraints, which take the sequence of human actions into account. Furthermore, the multiple surfaces in our system were identified as different units in contrast to the homogeneous systems in previous works. As shown in Figure 5 , our results reflect the individual heterogeneity in the spatial distribution of infection risk, which could not be obtained from other models. Although the ranges of (η r L 0 , η m L 0 , d g ) for the outbreak were un- saliva has been found to spread MERS-CoV to humans, 11 and its classification into disorder group C and predicted hard outer shell suggest that MERS-CoV is likely to be transmitted by saliva. 46 Therefore, we investigated the largest virus-containing droplet size d g in the range from 20 to 200 μm. The range for the viral load was very large, from 10 2 to 10 11 mRNA copies/mL in the respiratory tract, with the average value of viral load during the first week after diagnosis being 5 × 10 7 mRNA copies/mL in fatal cases and 3.9 × 10 6 mRNA copies/ mL in survivors. 47 According to a study by Gryphon Scientific 48 based on datasets from animal models of MHV and SARS-CoV, we estimated the dose-response parameter of MERS-CoV on mucous membranes as 3.2 × 10 −3 /mRNA copy. Due to the absence of data, the doseresponse parameter in the respiratory tract was set as 10 3 -fold higher than that on the mucous membranes, similar to the influenza A virus. 24 The eight more-likely scenarios (Scenarios I-VIII) based on the above analyses are summarized in Table 3 and are shown in Figure 6 with Roman numerals. With a viral load L 0 of 3.9 × 10 6 mRNA copies/mL (Scenarios I, II, III and IV), none of the hypotheses had good fit ( Figure 6 and Table 3 , the contribution made by the long-range airborne route to the infection risk was low, and that made by the close contact route was predominant, especially by the mechanism of inhalation of inspirable droplets. If the index patient was a super spreader, the viral loads might be even higher than the average value in fatal cases (5 × 10 7 mRNA copies/mL). When the two dose-response parameters were fixed (η r = 3.2/ mRNA copy; η m = 3.2 × 10 −3 /mRNA copy) and the viral load increased from 10 6 to 10 11 mRNA copies/mL, the overall best fit was achieved with a high viral load (the largest dots: L 0 = 5.6 × 10 10 mRNA copies/ mL in Figure 6A ; L 0 = 1.8 × 10 10 mRNA copies/mL in Figure such as the transfer rates between hands and surfaces (Appendix S2: Table S2 ) and first-order inactivation rates (Appendix S2: Table S3 ). These parameters were estimated or surrogated with data on other coronaviruses, influenza viruses, bacteriophages, or even bacteria, which could introduce errors into the results. Third, there is a lack of detailed information on individual behavior during the outbreak, especially that of the index patient. Individual differences in the behavioral modes of the nurses when visiting patients were also not considered in this study. These data are crucial for building multi-agent models and influence the exposure doses via the three transmission routes. In this study of the first nosocomial MERS outbreak in the Republic of Korea, the scenarios with a super-spreader or a super-spreading event corresponded with unusually high viral loads or sources. Our modeling suggested that long-range airborne transmission of MERS-CoV was the most credible hypothesis in explaining the observed data. Our long-range airborne route model predicted the spatial infection pattern well, that is, the downstream wards had a higher infection risk than the adjacent wards. However, in the scenarios in which the index patient shed a more "typical" amount of MERS-CoV, as param- Middle East respiratory syndrome coronavirus (MERS-CoV): announcement of the Coronavirus Study Group Middle East respiratory syndrome coronavirus (MERS-CoV) MERS-CoV outbreak in Jeddah: a link to health care facilities Epidemiologic features of the first MERS outbreak in Korea: focus on Pyeongtaek St. Mary's Hospital World Health Organization. WHO statement on the tenth meeting of the IHR Emergency Committee regarding MERS Information about Middle East respiratory syndrome (MERS) Transmission of influenza A in human beings Relative contributions of four exposure pathways to influenza infection risk Transmission of SARS and MERS coronaviruses and influenza virus in healthcare settings: the possible role of dry surface contamination Summary and risk assessment of current situation in Republic of Korea and China Middle East respiratory syndrome epidemiological findings from a retrospective investigation Environmental contamination and viral shedding in MERS patients during MERS-CoV outbreak in South Korea Extensive viable Middle East respiratory syndrome (MERS) coronavirus contamination in air and surrounding environment in MERS isolation wards Stability of Middle East respiratory syndrome coronavirus (MERS-CoV) under different environmental conditions Middle East respiratory syndrome coronavirus (MERS-CoV): update Interim infection prevention and control recommendations for hospitalized patients with Middle East respiratory syndrome Coronavirus (MERS-CoV) Transmission characteristics of MERS and SARS in the healthcare setting: a comparative study MERS outbreak in Korea: hospital-to-hospital transmission Middle East respiratory syndrome coronavirus (MERS-CoV): summary of current situation, literature update and risk assessment The first case of the 2015 Korean Middle East respiratory syndrome outbreak 30 cases linked to Pyeongtaek St. Mary's Hospital An integrated model of infection risk in a healthcare environment Quantifying the routes of transmission for pandemic influenza Relative Effectiveness of Ventilation in Community Indoor Environments for Controlling Infection. Hong Kong: HKU Theses Online (HKUTO) Role of fomites in SARS transmission during the largest hospital outbreak in Hong Kong A multi-agent based framework for the simulation of human and social behaviors during emergency evacuations Agent-Based Models. London: Sage Fundamentals of Nursing, Concepts, Process and Practice. Chesmut Hill, MA: School of Nursing Boston College A quantitative approach to defining 'high-touch' surfaces in hospitals Multi-zone modeling of probable SARS virus transmission by airflow between flats in Block E The Sverdlovsk anthrax outbreak of 1979 Modeling the atmospheric dispersion of foot-and-mouth disease virus for emergency preparedness Role of two-way airflow owing to temperature difference in severe acute respiratory syndrome transmission: revisiting the largest nosocomial severe acute respiratory syndrome outbreak in Hong Kong Prediction of natural ventilation in buildings with large openings The scaling laws of human travel Understanding individual human mobility patterns How far droplets can move in indoor environments: revisiting the wells evaporation-falling curve Airborne spread of measles in a suburban elementary school Applied Regression Analysis An evaluation of hospital hand hygiene practice and glove use in Hong Kong Bacterial contamination of the hands of hospital staff during routine patient care Assessing infection control measures for pandemic influenza Modality of human expired aerosol size distributions An appropriate lower respiratory tract specimen is essential for diagnosis of Middle East respiratory syndrome (MERS) Prediction of intrinsic disorder in MERS-CoV/HCoV-EMC supports a high oral-fecal transmission Viral shedding and antibody response in 37 patients with Middle East respiratory syndrome coronavirus infection Supplemental information: dose response parameters for gain of function pathogens Hospital-associated outbreak of Middle East respiratory syndrome coronavirus: a serologic, epidemiologic, and clinical description Preliminary epidemiologic assessment of MERS-CoV outbreak in South Korea the role of super-spreaders in infectious disease Predicting super spreading events during the 2003 SARS epidemics in Hong Kong and Singapore Middle East respiratory syndrome coronavirus: another zoonotic betacoronavirus causing SARS-like disease Additional Supporting Information may be found online in the supporting information tab for this article.