Search strategy

We designed two separate search strategies: one to identify epidemiological and microbiological papers and the other focused on fluid mechanics papers (mechanistic studies). We searched PubMed, medRxiv, arXiv, Scopus, WHO COVID-19 database, Compendex & Inspec. Searches were collaboratively developed by two reviewers (MD and LG) and results exported on 20 and 21 May 2020. Full search details are in Appendix S1 in the Online Supplementary Document . The search strategy and screening, data extraction and quality assessment procedures are summarised in Table 1 .

Inclusion and exclusion criteria

We included studies reporting data on any indoor setting except schools, which is addressed elsewhere in a living systematic review [11], any indoor activities and any potential means of transmission. Other screening criteria differed according to study discipline, as follows.

Epidemiological studies: We excluded studies of transmission within clinical settings and studies focusing purely on the clinical characteristics of cases. We also excluded statistical modelling studies aiming to predict future outcomes, as opposed to descriptive studies characterising past events.

Microbiological studies: We included studies involving the testing of swabs taken from “real world” settings for the presence of SARS-CoV-2. As most of these were conducted in hospital settings, we included studies from both clinical and non-clinical settings. However, to maximise the transferability and generalisability of these findings to non-clinical indoor settings, we excluded microbiological studies of samples collected in areas of the hospital such as operating theatres and Intensive Care Units (ICU) where aerosol-generating procedures are routinely carried out. We also included laboratory studies investigating the persistence and viability of the virus under different controlled conditions.

Mechanistic studies: We included articles reporting data on any respiratory virus, numerical simulation studies focusing on the mechanisms of transmission and studies investigating any mechanisms with potential to influence transmission in indoor environments, such as ventilation, air conditioning or plumbing systems.

Screening procedures

Articles were screened by two separate teams (LG, GN, PK, TH, RN, KO for epidemiological and microbiological studies and AA, SBC, EIE, RG, EP, MS, IMV, SZ for mechanistic studies). Title and abstract and full text screening were conducted by one reviewer within each team, with rejections assessed by a second reviewer.

Data extraction and quality assessment

Data extraction and quality assessment for each article was conducted by a single reviewer, as above. A range of critical appraisal tools was employed, according to study design: case series and case reports were evaluated using Joanna Briggs Institute checklists [4 ]. We adapted a quality assessment tool for epidemiological outbreak cluster studies from the Joanna Briggs Institute checklist for critically appraising case series [4]. We adapted existing tools for the quality appraisal of laboratory experimental studies [5-7 ]. Details of adapted tools are provided in Appendix S2 in the Online Supplementary Document . For other epidemiological study designs, we used Critical Appraisal Skills Programme (CASP) checklists [12]. The overall quality of the epidemiological evidence on each research question was assessed by a single reviewer (RM) on the basis of the GRADE system [13]. Observational epidemiological studies were assigned an a priori grading of low, which could be downgraded on the basis of critical appraisal or upgraded on the basis of consistency across different studies and study designs. Mechanistic and numerical simulation studies were appraised by an expert in the field (IMV, SBC, EP, SZ, MS), based on three sources [8-10] (Appendix S2 in the Online Supplementary Document ). Data extraction was limited to a minimal set of required data items: study question addressed by the article, study design and summary of methods, indoor context, outcome measure, relevant results.

Data synthesis

Data heterogeneity was such that results were synthesised narratively, except for the results on secondary attack rates within households, which were meta-analysed using a fixed effect model in R 3.6.3 [14] using the rma.uni() function in the metafor package [15]. I² and Cochrane’s Q were calculated to assess heterogeneity. For consistency, the same function was used to estimate confidence intervals for SAR in individual studies that were not included in pooled estimates. A fixed effects analysis was chosen because the number of studies was relatively small, and thus a simpler underlying model (fewer assumptions/parameters required) was likely to be better estimated: in addition, there is reasonably good theoretical and simulation evidence that fixed effects models are relatively robust to moderate heterogeneity [16]. There was little evidence of heterogeneity in the data but the number of studies was too small for such evidence to accumulate.

What evidence is there for aerosolised transmission?

Table 2 summarises the evidence we identified on aerosolised transmission. The discipline of fluid mechanics provides important insights into the physical behaviour of respiratory droplets in the air. Respiratory droplets range in size from <10 μm to >1000 μm. Larger droplets follow a ballistic trajectory, falling to the ground within a few metres, the exact distance depending on the force with which they are ejected [17 ,18,36-38]. Smaller droplets (diameters of the order of 10 μm or smaller) fall so slowly through the air that they have time to evaporate [ 38]. These very light, desiccated particles, or aerosols, can then remain suspended in the air, potentially for several hours [18 ,19], and can travel long distances on air flows before eventually landing [17,19,20]. Studies conducted following the 2003 SARS outbreak provided evidence consistent with aerosolised transmission within buildings, influenced by the effects of ventilation and plumbing systems [21 -25]. In order to ascertain whether aerosolised transmission of SARS-CoV-2 is possible, it is first necessary to establish whether and for how long it is able to persist in the air. In a laboratory-based study, van Doremalen et al found that aerosolised particles of SARS-CoV-2 remained viable for 3 hours (median half-life 1.09 hours, 95% credible interval 0.64, 2.64) [26]. Taken together, the fluid mechanics and microbiological studies demonstrate that aerosolised transmission of SARS-CoV-2 is theoretically possible.

To investigate whether aerosolisation of viral particles might actually be occurring, we found six studies which collected and analysed air samples [28-32,39]. Four of the six studies detected SARS-CoV-2 RNA [28,29,32,39 ]. Whilst the presence of viral RNA can indicate the presence of live virus, it can equally, however, simply indicate the presence of fragmented dead virus, which does not pose an infectivity risk: laboratory culturing methods are required to establish the presence of live virus [40].

We found one study which used tracer gas measurements and computational fluid dynamics simulations to predict the spread of droplets exhaled by the index case in an outbreak linked to a restaurant in Guangzhou, China [35]. The researchers found evidence consistent with aerosolised transmission over short distances within a crowded and poorly ventilated space. We found two observational epidemiological studies reporting evidence relevant to the question of aerosolised transmission. One was an epidemiological investigation report describing a large outbreak in Washington State, USA, linked to a choir practice, which was consistent with aerosolised transmission [33]. However, descriptive epidemiological studies of other outbreaks have failed to find evidence consistent with aerosolised transmission. For example, in an analysis of the outbreak on the Diamond Princess cruise ship researchers argued that the absence of any cross-room transmission once passengers had been quarantined in their cabins supports the hypothesis that transmission was via droplets/fomites and not airborne via the air conditioning system [34].

Taken together, we found evidence that although aerosolised transmission is theoretically possible, we found no conclusive epidemiological evidence of this actually occurring. Table 3

What evidence is there for faecal-oral transmission?

Other human coronaviruses can be transmitted via the faeces of infected individuals [25,49,50], so it is important to establish whether SARS-CoV-2 can be transmitted in this way. We reviewed five case series [42, 43,45-47], two case reports [41 ,44], one non-systematic review article [48] and three surface and air sampling studies [29,31,39] ( Table 3 ). Emerging evidence suggests that gastro-intestinal (GI) symptoms in SARS-CoV-2 may be the result of viral invasion of ACE2 expressing enterocytes of ileum and colon, as seen with SARS-CoV [ 51]. However, GI symptoms are less common in SARS-CoV-2 than in SARS-CoV or MERS [48]. All eight articles we reviewed reported detection of SARS-CoV-2 viral RNA in faecal samples using RT-PCR. However, study quality was poor: studies were small and lacked detail and results are difficult to compare because of the different parameters and time frames used, such that estimates of the proportion of adult cases with viral RNA detectable in faeces varied widely. Several studies reported evidence that SARS-CoV-2 faecal samples still tested positive after throat swabs had turned negative [42-44,47 ]. Several studies reported that the presence of SARS-CoV-2 viral RNA or live virus in faecal samples was unrelated to the presence of gastro-intestinal symptoms [42,44,47 ]. We also reviewed three studies which collected environmental samples, two in clinical settings [29,39] and one in a cruise ship [31], which suggest that aerosolisation of viral particles may occur through toilet flushing. Two studies highlighted the detection of SARS-CoV-2 RNA on the floor surrounding toilets used by confirmed cases, which is consistent with aerosolisation of virus particles through toilet flushing [29 ,31]. The highest concentration of SARS-CoV-2 RNA detected in air samples by [39] was in a patient toilet cubicle. However, despite widespread confirmation that viral RNA can be detected in faecal samples, we found no evidence for transmission of the virus by this route. The detection of viral RNA does not mean that live virus is present or that patients are infectious. The only study we found which attempted to isolate live virus was able to isolate infectious virus from samples taken from patients’ throats and lungs, but not from faecal samples, even though these samples had high concentrations of viral RNA [45]. This was a very small study and results require replication. In summary, although viral RNA can be detected in the faeces of cases, we found no evidence of transmission via this route, either through the contamination of surfaces or through aerosolisation.

What evidence is there regarding the role of ventilation systems in indoor transmission?

We found six experimental and numerical simulation fluid mechanics studies addressing the role of ventilation systems in indoor transmission ( Table 4 ). These demonstrate that air currents are responsible for the dispersal of both aerosols and large droplets within buildings, between different rooms and even between different floors [22,52]. Studies show that this dispersal can be amplified by a variety of factors, including ventilation and air conditioning systems [35], differences of temperature between rooms [53] and air currents entering through open windows [ 54]. However, ventilation systems are also likely to dilute the concentration of viral particles in the air and thereby to play a potential role in decreasing transmission [22,55]. Ventilation systems thus have the potential to decrease virus transmission risk near the source but to increase virus transmission risk further away from the source. However, we found only one study which investigated this question specifically in relation to SARS-CoV-2 [35]. This study used tracer gas experiments and fluid dynamics numerical modelling to predict the location of cases within a poorly ventilated restaurant. Based on this one study alone, which was subject to modelling assumptions and results which were case specific and not clearly generalizable to other indoor environments, the overall quality of the evidence on the role of ventilation systems in indoor transmission of SARS-CoV-2 was judged to be low.

What evidence is there regarding the role of plumbing systems in indoor transmission?

There is no direct evidence that SARS-CoV-2 is transmissible via infected faeces; however until this is demonstrated definitively, it is important to understand the potential role of defective plumbing systems. Investigations following the SARS-CoV pandemic provided evidence that defective U-traps played a role in the transmission of SARS-CoV in a large outbreak in the Amoy Gardens residential complex in Hong Kong in 2003. During this outbreak, 321 cases in the apartment complex were linked to faecal-oral transmission [50]. Subsequent simulations have demonstrated that aerosols can be generated in vertical soil stack pipes when toilets are flushed and, if U-traps are defective, can enter a room due to the suction generated by the ventilation system [25,49,56,57]. In this context, contaminated aerosols originating from breath or sewage are more likely to be warmer than the surrounding air, and so are more likely to travel from the lowest to the highest floors of a building than vice versa. The lower the environmental air temperature, the more significant the aerosol transmission from the lowest floors to the highest floors [ 58]. Evidence is summarised in Table 5 . In summary, for infectious viruses present in faeces, there is strong real-scale experimental evidence demonstrating the potential for defective plumbing systems to amplify transmission within high-rise buildings, and this is consistent with observational epidemiological evidence. However, as outlined above, we found no evidence for the presence of infectious SARS-CoV-2 in faeces, nor for covid-19 outbreaks amplified through plumbing systems.

What evidence is there regarding transmission via different indoor surfaces (materials and specific objects)?

We identified 13 studies investigating the transmission potential of different materials, surfaces and objects in indoor environments, summarised in Table 6 [26,27,29-32,59-65]. The length of time SARS-CoV-2 remains viable on surfaces depends on the type of surface and the environmental conditions. Experimental evidence from tightly controlled laboratory studies indicates that the virus survives better on smooth, non-porous surfaces, at low temperatures and in damp conditions [ 26,59-61]. It can also survive under acidic conditions, such as the stomach [61]. Although there is general agreement among studies that the virus survives for longer on smooth surfaces and at lower temperatures, estimates of precisely how long it can survive on different surfaces vary considerably among studies, likely because of differences in experimental conditions. Furthermore, these studies are silent on the infectious dose and do not quantify the risk of transmission associated with touching different objects and surfaces. It is also important to note that studies conducted under strict laboratory conditions are not directly applicable to real-world contexts, so these findings must be triangulated with studies collecting and analysing environmental samples. Several studies reported detecting viral RNA on a wide range of high-touch objects; however in low quantities [27,29-32,62,63]. As highlighted above, viral RNA can be either live virus, which poses an infectivity risk, but equally it can be fragmented dead virus which does not have the ability to cause infection. We found three studies which attempted to culture live virus from environmental samples, all with inconclusive or negative results [29 ,31,32]. Epidemiological evidence on this question is inconclusive because it is difficult to distinguish from descriptive epidemiological data alone between fomite and droplet transmission. A contact tracing report on a church outbreak in Singapore found that one of the three secondary cases had no direct contact with the presumed index cases (a couple visiting from China), but occupied the seat that one of them had vacated [65]. However, this was a small, very low quality study and whilst this evidence is consistent with transmission via touching a contaminated object, it is also consistent with airborne transmission.

What evidence is there for the transmission of COVID-19 in indoor residential settings?

Eight studies included data on transmission in residential settings. Four of these reported on household transmission [66-69 ], providing data on secondary attack rates (SAR, defined as the probability that an infection occurs among susceptible people within a specific group, such as a household or close contacts [70] (SARs)) ( Table 7 ). We conducted a meta-analysis of the SARs for these four studies. The pooled SAR for people living in the same household was 11% (95% CI = 9, 13) ( Figure 2 ). We found four studies reporting data for estimating SARs amongst residents in communal living environments [34,71 -73]. SARs for residents in these settings are shown in Table 8 . SARs for staff working in these settings are shown separately in Table 9. These studies involved very different types of population (elderly nursing home residents, passengers on a cruise ship and people experiencing homelessness), so it was not appropriate to conduct a meta-analysis. The SARs for people living in communal settings were significantly higher than the SARs for households. The quality of epidemiological evidence for transmission in residential and communal settings was poor.

Figure 2.  Forest plot – pooled estimate of household secondary attack rate (SAR). I2 = 0.00%, Q(df = 3) = 1.72, P  = 0.63.

What evidence is there for the transmission of COVID-19 in indoor workplaces?

Six studies reported on transmission among workers or at workplaces, where details were provided about the nature of the work and workplace. These were: care home workers [71], cruise ship crew [34], staff at a shelter for people experiencing homelessness [73], staff at an assisted and independent living community for the elderly [72 ], workers at meat/poultry processing plants [74] and shop workers [65]. Four of these studies provided data for the estimation of SARs among staff ( Table 9 ). All four of the workplaces shown in Table 9 are also places of residence (SARs for residents are shown in Table 8 ). SARs for staff and residents were not significantly different in the assisted and independent living community or in the shelter; however, SARs were significantly higher for residents than for staff on the cruise ship (P = 0.000017) and in the care home (P < 0.00001). We found two workplace studies which did not present sufficient data to estimate SARs but nevertheless provide insight into workplace transmission ( Table 10 ). A CDC paper reporting on outbreaks in meat processing plants across the USA [74] identified a range of key drivers. These included difficulty in maintaining the 2 m social distance on the production line at break times and while entering/exiting the facility; difficulty implementing covid-19-specific disinfection guidelines; socioeconomic challenges related to poverty, such as people continuing to work whilst ill, especially where attendance is incentivised and workers living in overcrowded, multigenerational households; communication challenges such as the inaccessibility of health and safety training to non-English speakers and to non-literate workers; sharing of transportation to work; and adherence to correct usage of face coverings. Another driver may be that factories are noisy environments, where people may have to shout, thus transmitting droplets over longer distances. Results of a small contact tracing study of an outbreak in Singapore connected with a shopping trip of a group of tourists from China points to close and prolonged interactions with a case as a driver of transmission [65]. Overall, the quality of the epidemiological evidence on workplace transmission was poor. There is considerable variability in workplace contexts, making it difficult to synthesise conclusions across settings, and detail is often lacking as to potential transmission mechanisms.

What evidence is there for the transmission of COVID-19 in other indoor settings (social, community, leisure, religious, public transport)?

We found three epidemiological studies reporting on transmission related to social, religious, community or leisure settings and providing sufficient data to estimate SARs ( Table 11 ). Two studies report on a total of three outbreaks related to religious gatherings or churches [65,68]. One study investigated evidence for transmission in a clinic waiting room [67].

Estimated SARs ranged from 2.1% at a church service in Singapore [65 ] to 25.3% at an extended, overnight religious gathering in Malaysia [68]. The study investigating transmission in clinic waiting rooms followed up 95 people who spent time in clinic waiting rooms with affected individuals in USA. No cases were detected [67]. The quality of this observational evidence on transmission in social/community settings was very poor and there was considerable heterogeneity of contexts and variability in the results.

Do particular activities convey greater risk (eg, shouting, singing, eating together, sharing bedrooms)?

Different activities involve the emission of different numbers of respiratory droplets. Evidence from fluid mechanics experiments shows that the number of droplets ejected increases in the order: breathing, heavy breathing, speaking, singing, coughing, sneezing. There is a very significant (orders-of-magnitude) difference in the numbers of droplets emitted between each of these levels and the next [75-79]. There is also evidence that pronouncing some vowel sounds results in the emission of more droplets than others; however these risk differences are relatively small compared to the risks between, for example, coughing and singing [80]. Although the physical properties and behaviour of droplets emitted via different mechanisms are well characterised, it is not possible directly to compare the risks of transmission associated with heavy breathing with those associated with coughing or sneezing. This is because whilst breathing is a continuous activity, coughing and sneezing are discrete events and are thus not directly comparable in terms of risk level. Different activities result in the emission of droplets of different sizes (for example, small droplets are emitted during breathing, and large droplets when sneezing). Thus droplets emitted by these different activities will be associated with different transmission mechanisms. A final point to consider is that droplets emitted through these different mechanisms are generated in different parts of the respiratory system, and thus, are likely to have different viral loads.

Table 12 details four descriptive epidemiological studies which describe transmission via daily living activities among people living together in households [66 -69]. The results of these studies are consistent with the hypothesis that close and prolonged contact through activities such as sharing beds, bathrooms, eating together, face to face contact and spending time in the car together are likely to increase the risk of transmission. Again, however, the quality of individual studies was poor or very poor and there is insufficient evidence to evaluate the relative risk of specific activities or behaviours from these studies.

The four studies we found which report on transmission in communal contexts are consistent with the conveyance of risk through close contact daily living activities. It is striking that the SAR reported in the care home [71] is an order of magnitude higher than that reported in the senior assisted and independent living community [72], a much less communal setting, where elderly residents lived largely independently in separate apartments. It is important to note, however, that although the age profile in the two settings is likely to be similar, the residents of the nursing home were likely frailer. Also, ascertainment of the denominator in the care home study was not precise, so these results are uncertain.

What evidence is there for the appropriate length of distancing between people?

Our findings are consistent with the hypothesis that the main route of CoV-2 transmission is through person-to-person short-range transmission, which occurs through large respiratory droplets ejected while speaking, coughing and sneezing. The distance that these respiratory droplets travel before falling to the ground depends on (among other factors) how they were generated. The physical behaviour of droplets is well characterised: those generated by speaking fall to the ground within 1 m or closer to the speaker [38]; droplets generated by coughing travel about 2 m [17] and those generated by sneezing can travel 8 m before falling to the ground [36]. On the basis of this evidence, our review finds no evidence to support a relaxation of the 2 m social distancing recommendation ( Table 13 ).