Concern with the Middle East respiratory syndrome (MERS) remains high among public health authorities worldwide due to the failure to stop the spread of the disease three years after its first description [1]. As of February 2016, 26 countries reported cases, for a total of 1,626 laboratory-confirmed cases of infection with MERS [2]. Cases outside the Middle East region were either travellers or had close contact with imported cases.

The majority of case importation events led to little or no secondary transmission, as expected due to the low human-to-human transmission risk [3–7]. Yet, in May 2015, a case imported to South Korea triggered the largest outbreak outside the Middle East, with a total of 186 confirmed cases [2]. This large epidemic was quite unexpected and calls for a better understanding of the potential for MERS dissemination worldwide and risk of onward transmission.

It has been proposed that countries having a high risk of MERS importation were those receiving the most air passengers from the Middle East [5, 8]. Joint analysis of incidence data in the Middle East region with international travel flows allowed estimating the expected number of MERS introductions in countries outside the affected area [9, 10], for example on coming back from pilgrimage in Mecca [9] or distinguishing between residents and visitors [10]. In the latter, predictions were based on cumulative attack rates per country in the affected area [10], thus disregarding the strong temporal nature of MERS epidemic [11].

In case of importation of a case, experience has shown that household and nosocomial transmission was possible [12–14] with infrequent large outbreaks as in South Korea [2]. A marked heterogeneity in the epidemic outcome following importation has been observed and characterized by transmission trees in South Korea [15] and theoretical results regarding the potential for multiple generations [16] and role of super-spreading events [16, 17]. Yet a comprehensive understanding of the risk factors for local transmission once a MERS case is imported is still missing.

Here we aimed at providing a comprehensive analysis of the risk of MERS importation and subsequent onward transmission through a systematic analysis of all known MERS importation events. We produced predictions based on air traffic data for the risk of MERS importation and estimated the expected number of symptomatic cases imported in countries outside the affected area – here defined as Saudi Arabia, Qatar, Oman, Kuwait, Jordan, United Arab Emirates, Yemen, Bahrain, and also referred to as Middle East. The analysis accounted for seasonal variations of air traffic flows and for temporal and spatial variations of MERS incidence in the Middle East. Predictions were validated against confirmed importations and lack thereof. Focusing on onward transmission, we analysed the risk for local transmission according to sources of heterogeneity in terms of individual awareness (time to hospital admission, declaration of history of travel), country’s reaction (baseline hospital protocols, heightening of infection control strategies), cultural aspects and local customs (e.g. health-seeking behaviour) and phase of the outbreak. We modelled the outcome of importation events according to how cases were identified and managed to provide quantitative estimates on the expected number of secondary cases in the community and in the hospital setting. We finally explored whether MERS awareness in different communities at the time of importation affected case management.

The large number of cases reported in the South Korean outbreak following a MERS case importation generated substantial concern on the risks that countries face towards importation and possible onward transmission. Our quantitative analysis provide important information that can help preparedness and response to such events: MERS international dissemination is strongly associated to air travel flows combined with cases incidence in the source area; transmission is more likely to occur in hospitals; large outbreaks are possible though rare; and high attention to MERS epidemic in the public and among professionals lead to efficient case management.

The role of air travel as driver for pathogen international dissemination has been considered in several prospective studies [28–37], with few retrospective validations [33, 38–42]. For the case of MERS [5, 8–10], qualitative comparisons between predictions and observed importations were presented [9, 10]. Our analysis produces the first quantitative assessment of the accuracy of model predictions for the risk of importation worldwide accounting for all countries with and without reported importations. In addition, importation risk and expected number of imported cases is evaluated across time. The temporal factor is often neglected when dealing with subcritical epidemic as in the case of MERS and overall attack rates are generally considered [10]. MERS epidemic was shown however to display a strong temporal component, both in the zoonotic and human-to-human transmission [11]. Here we found indeed that such temporal variation also affects the expected number of exportations from the Middle East region. The largest concentration of exportations (nine episodes from March to May 2014) occurred during the outbreaks affecting the provinces of Riyadh and Makkah in Saudi Arabia in Spring 2014, causing a 18-fold increase in the expected number of exportations. Spatial resolution is also important, as already highlighted in [10] with a country-level description. Here we considered a total of 20 regions in the affected area, including the countries of Saudi Arabia, Qatar, Oman, Kuwait, Jordan, United Arab Emirates, Yemen, Bahrain similarly to [10], and additionally disaggregating Saudi Arabia into its 13 provinces [11].

A further element of novelty of the present study is its ability to provide projections on the expected number of importations by country with no prior knowledge on either MERS epidemiology or under-reporting ratio. This is possible by relying on three main assumptions: (i) 100 % detection accuracy of MERS surveillance in countries of European Union and North America; (ii) uniform under-reporting of cases in the affected area in time and space [11]; (iii) homogeneous mixing between travellers and local residents.

The third assumption may be responsible for some of the limitations of the model. The major discrepancy we found between predictions and observations is the large risk of importation predicted for India, Pakistan, and Indonesia, though these countries did not report any MERS case. This was also observed in [10]. It may suggest that cases were imported in these countries but went undetected. However it may also be due to non-homogeneity in the mixing or travel behaviour of classes of individuals, namely travellers to the affected area vs. region residents, that is known to impact the conditions for pathogen dissemination [43]. Non-homogeneous mixing may affect travellers’ probability to be infected by the virus because of altered risk of contact with zoonotic sources or of exposure to nosocomial outbreaks. This may vary among travellers and be related for instance to different purpose of visit (e.g. short-visit tourists vs. seasonal workers). The similarity of findings regarding high-risk countries between our study and previous work [10], notwithstanding different modelling assumptions, suggests that country-specific aspects may be determinant for the observed discrepancy. On one side there is a surveillance system whose accuracy may vary by country. On the other, purpose of visit and associated mixing may be country-related. India, Pakistan, and Indonesia are indeed in the top 4 nationalities in the expat community in Saudi Arabia [44].

The detailed analysis of the history following importation provided important findings on the risk of a local outbreak generated by an infected traveller. While no secondary cases were reported in the majority of observed importation events, four events led to one (in two instances), two, or 31 secondary cases. This heterogeneity was captured by our models through large overdispersion in transmission, in agreement with previous works [15–17]. Our findings suggest that the probability to observe a future MERS importation leading to a number of secondary cases larger than the South Korean one is of the order of 1-5 %, consistent with prior work on the full nosocomial South Korean outbreak [15]. In line with another study, the risk of observing a secondary case is predicted to be larger than 20 % if isolation in hospital is not fast (>one week), independently of the time spent in the community [16].

The observed variability in epidemic outcomes may result from at least four characteristics. First, the duration of the transmission risk period in the destination country, from date of importation or symptoms onset to isolation or death, is clearly an important risk factor for local transmission. The longer the transmission risk period, the larger are the opportunities for susceptible individuals to be exposed to the infectious case, both in the community and after hospitalization.

Second, nosocomial transmission of MERS appeared more efficient than transmission in the community. This points to the need for focusing interventions on rapid case identification and effective isolation and for improving infection control protocols in hospitals to prevent transmission. The large outbreak observed in South Korea was indeed attributed to sub-optimal infection prevention and control measures in hospitals [45]. These findings are in line with previous analyses on MERS transmission in healthcare settings and with modelling studies evaluating the impact of mechanisms to control SARS spread in 2003 [15, 46–48].

The third aspect pertains to heterogeneity in case finding and management. For example, imported cases could visit one to four health-care facilities before getting a diagnosis. This behaviour may be based on individuals’ decisions but it may also be induced by country-specific regulations of the national health-care system that determines how patients can access professionals or it may be influenced by local customs. This behaviour was found to be associated with a higher probability of secondary transmission in our study, indicating that the local practice of seeking care in multiple health-care facilities may have contributed to the initial spread in South Korea, in agreement with the findings of Ref. [45].

The fourth aspect is represented by lack of individual awareness regarding the potential risk that was found to be a contributing factor increasing transmission risk. Indeed, not reporting a history of travel to the health practitioners increased the probability of having secondary cases (although not statistically significant). For emerging diseases whose non-specific symptoms preclude a fast diagnosis, travel history is a critical element for patient assessment. Moreover, cases who experienced onset of symptoms before importation were less likely to generate secondary cases. Traveling while ill from a country where a MERS outbreak is ongoing increased individual awareness on the infection risk and induced a more precautionary individual behaviour. Clearly, these two aspects are correlated, as 70 % of the cases who declared history of travel also travelled while ill.

We explored three digital indicators for collective attention and awareness corresponding to different communities: general population, professionals and health authorities. Digital sources have been recently used in the context of early detection and surveillance [49, 50]. They have also been used as a source to study public concern in response to an ongoing epidemic threat [49, 51].

Our analysis showed that general public’s attention measured by Google Trends is highly correlated with the amount of news circulating amongst the international infectious disease community (ProMED-mail) and of official reports published by the WHO. Its curve in time shows the presence of high peaks followed by a quick decline. This is a common behaviour generally found in the social media response following specific triggering events [49, 52, 53]. In our study, events triggering collective attention can be identified with unexpected episodes of case importations or resurgence of the epidemic in the affected area.

Here we found that MERS confirmation and isolation during peaks of collective attention were considerably faster. This suggests that an increased collective awareness may induce changes in the management of the patient allowing the health system to successfully reduce the time from admission to isolation. Being this time critical for transmission risk, increased awareness is identified as a key factor to control the generation of cases after MERS importation. Similar results were also obtained for the indicators measuring awareness in the public health and infectious diseases community. The case importation in South Korea occurred when attention was at its baseline level (as measured by all indicators), and indeed it was reported that MERS appearance was “unfamiliar to most physicians” in the country [54].

No correlation was found instead between collective attention and time to admission. This suggests that collective attention may preferably act on the case management by health authorities, whereas individual awareness may more likely be responsible for individual change of health-seeking behaviour, as discussed previously.

The impact of collective awareness and public health mobilization strategies in response to an epidemic has already been observed in past outbreak. As SARS outbreak progressed, reduced duration from onset to hospital admission and reduced number of hospital visits per patient were observed due to overall increased awareness and public health recommendations [47, 55–57]. Similarly, during the South Korean outbreak of MERS, infection control measures strengthened, and the delay from illness onset to confirmation shortened as the epidemic progressed [58, 59].

There are however two important differences between our findings and the above. First, awareness and concerns are known to be stronger for the population experiencing the outbreak (as in the above cases) than for yet unaffected populations (as in our study) [53]. Nonetheless, our findings show that higher awareness, even when induced by non-local events, allow countries to better manage an importation episode. Second, here we were able to explicitly measure awareness in time and relate it to a quantifiable change of behaviour (case management), further illustrating its impact on transmission. From the use of surveys to digital data, quantifying awareness has become an important aspect of epidemic surveillance and control [49, 51, 60–62]. Few studies have however measured how variation in awareness affects an epidemic during its course [50, 63]. Moreover, MERS poses additional challenges as public attention is known to wane rapidly in response to external events [64] unless it is prompted by additional triggering events or mounting concern, as it happened with the rapidly increasing number of cases of Ebola epidemic [5]. The subcritical nature of MERS epidemic, with an incidence characterized by subcritical spread and sporadic peaks [11], induces a similarly fragmented timeline of attention. For example, the large peak of Spring 2014 increased the risk of MERS dissemination. The management of imported cases during that period however benefited from an increased attention to the disease epidemic, and no secondary transmission was observed. Conversely, despite the risk of importation was much lower during the beginning of 2012 and Spring 2015, the few importations observed during that period were able to initiate local transmission and threaten the health security of the destination countries. By analysing the various aspects characterizing the epidemic, our analysis is able to identify the factors that may help improving the reaction of countries at risk of importation.

Other limitations of our study need to be mentioned. We assumed that no importation events had been missed in Europe and estimated under-reporting in the Middle-East accordingly. However, if 1 importation case had been missed in Europe, underreporting would be 25 % rather than 18 %, and the model-predicted number of cases worldwide increase by 10 % over the period. If one further assumes that secondary transmission was unlikely for these unrecognized cases, the probability of transmission following importation would be overestimated. For example, taking the model-predicted number of importation cases worldwide for the period as exact, the probability of local transmission would have been 6 % (4 among 70) rather than the current 18 % (4 among 22).

Furthermore, the second best-fitting model among the ones considered for transmission following importation, with only marginally larger AIC, did not support setting-specific transmission. Our main result is however in line with previous findings on nosocomial transmission of MERS in the Middle East [15]. In addition, there were two secondary cases in the Italian episode that were not confirmed by official sources. Including these two cases in the analysis, we still found that longer time to isolation was associated with more secondary transmission, although evidence on setting-specific differences in transmission was less marked (see Additional file 1).

We evaluated the risk of MERS exportation worldwide integrating seasonal air traffic flows and time-varying incidence of cases in the affected area, and accounting for under-reporting of cases. We conducted a comprehensive analysis of all reported MERS imported cases to validate modelled importation risk and assess onward transmission. Our findings confirm the critical role of air travel in the risk of international dissemination. In case of MERS introduction, prompt identification of the infection in patients seeking medical attention, strict infection control measures, and effective isolation of the patient in the nosocomial setting are key to efficient prevention and control of the outbreak. Increasing awareness at collective and public health levels worldwide is found to be associated with higher local preparedness, prompter and strengthened precautionary measures and isolation procedures to prevent further spread. Our findings provide a quantitative assessment for public heath authorities to face the variability associated to importation risk and potential for local transmission, to inform preparedness plans, and identify the critical measures that should be considered to reduce the likelihood of future outbreaks.