Practical Indicators for Risk of Airborne Transmission in Shared Indoor Environments and their Application to COVID-19 Outbreaks

Box model of infection

The box model considers a single enclosed space, in which virus-containing aerosols are assumed to be rapidly uniformly mixed compared with the time spent by the occupants in the space. This is approximately applicable in many situations, but there are some exceptions such as outbreaks where clear directional flow occurred in a room.^18,20 The mathematical notation used in the paper is summarized in Table S1. The mass balance equation is first written in terms of c, the concentration of infectious quanta in the air in the enclosed space (units of quanta m^-3). Compared with a model written in terms of aerosol or viral particle concentrations, c has the advantage of implicitly including effects such as the deposition efficiency of the aerosol particles in the lungs of a susceptible person, as well as the efficiency with which such deposited particles may cause infection, the multiplicity of infection, etc. The balance of quanta in the space can be written as: where E_p is the emission rate of quanta into the indoor air from an infected person present in the space (quanta / h); f_e is the penetration efficiency of virus-carrying particles through masks or face coverings for exhalation (which takes into account the impact of whether the infector wears a face covering); V is the volume of the space; λ₀ is the first-order rate of removal of quanta by ventilation with outdoor air (h^-1); λ_cle is the removal of quanta by air cleaning devices (e.g. recirculated air with filtering, germicidal UV, portable air cleaners, etc.); λ_dec is the infectivity decay rate of the virus; λ_dep is the deposition rate of airborne virus-containing particles onto surfaces. E_p is a critical parameter that depends strongly on the disease, and it can be estimated with a forward model based on aerosol emission rates and pathogen concentration in saliva and respiratory fluid^31,35 or by fitting a model such as the one used here to real transmission events.^11,31

This equation can be solved analytically or numerically for specific situations. Given the enormous number of possible situations, and given the prevalence of outbreaks resulting from longer events where air is shared for a substantial period of time, we consider a sufficiently long event so that a steady state condition (dc/dt ∼ 0) is a reasonable approximation. Writing λ = λ₀ + λ_cle + λ_dec + λ_dep for simplicity leads to a steady-state infectious quanta concentration of:

Under the assumption of no infectious quanta at the beginning of the event, a multiplicative factor, r_ss, can be applied for events too short to approximately reach steady state (see Section S1 for detail) to correct the deviation of the quanta concentration averaged over the event (c_avg) from that at steady state:

Since the goal is to analyze outbreaks, we assume that only a single infectious person is present in the space, which is thought to be applicable to the outbreaks analyzed below. This allows calculation of the probability of infection, conditional to one infectious person being present. The model can also be formulated to calculate the absolute probability of infection, if we assume that the probability of an infectious person being present reflects the prevalence of a disease at a given location and time (e.g. refs ^30,36).

The dose expressed in infectious quanta (n) inhaled by each of the susceptible persons present in the space (N_sus) is then: where f_i is the penetration efficiency of virus-carrying particles through masks or face coverings for inhalation (which takes into account the effect of the fraction of occupants wearing face coverings); B is the breathing volumetric flow rate of susceptible persons; D is the duration of exposure, assumed to be the same for all the susceptible persons. Substituting:

The number of expected secondary infections increases monotonically with increasing n. For an individual susceptible person, by definition of an infectious quantum, the probability of infection is ²⁸:

For low values of n, the use of the Taylor expansion for an exponential allows approximating P as:

The total number of secondary infections expected, which may also be regarded as the effective reproduction number (R_e) in a given situation, is then: where N_sus is the number of susceptible individuals present, which may be lower than the number of occupants due to vaccination or immunity due to past infection. All the outbreaks studied here occurred before COVID-19 vaccines became available. Thus, the number of secondary infections increases linearly with n at lower values, and non-linearly at higher values. We retain the simplified form to define and calculate the risk parameters, but use equation (6) for fitting the outbreak results in Figure 1b below.

• Download figure
• Open in new tab

Figure 1:

(a) Number of secondary cases vs. the risk parameter H and (b) attack rate vs. the relative risk parameter H_r for outbreaks of COVID-19, tuberculosis, influenza, and measles reported in the literature. The fitted trend line of attack rate as a function of H_r and its estimated uncertainty range (5^th and 95^th percentiles) are also shown in (b). All of the outbreaks investigated here involve the original variants of the virus. A variant twice as contagious should result in moving the results to the left by a factor of two or so.

Risk parameters for airborne infection

We define the relative risk parameter, H_r, and the risk parameter, H, for airborne infection in a shared space. The purpose of H_r and H is to capture the dependency of P and N_si, respectively, on the parameters that define an event in a given space, in particular those parameters that can be controlled to reduce risk of shared-room airborne transmission. To better capture the controllable actions, E_p and B were split into factors that can and cannot be controlled. E_p can be expressed as the product E_p0 x r_E where E_p0 and r_E are, respectively, the quanta shedding rate of an infectious person resting and only orally breathing (no vocalization); and the shedding rate enhancement factor relative to E_p0 for an activity with a certain degree of vocalization and physical intensity (see Table S2a for detail). B can be expressed as B₀ x r_B, where B₀ and r_B are the volumetric breathing rate of a sedentary susceptible person in the age group of 41-<51 years (numerically also the average for all age groups) and the relative breathing rate enhancement factor (vs. B₀) for an activity with a certain physical intensity and for a certain age group (see Table S2b for detail). E_p0 is uncertain, likely highly variable across the population, and variable over time during the period of infectiousness.^31,35,37 It may also increase due to new virus variants such as the SARS-CoV-2 Delta variant, that is more contagious, assuming that the increased contagiousness is due to increased viral emission or reduced infectious dose (both of which would increase the quanta emission rate).^38,39 We note that some variants could in principle also increase transmissibility by lengthening the period of infectiousness for a given person, which by itself would not increase the quanta emission rate in a given situation. B₀ is relatively well known, and varies with a susceptible person’s age, sex, and body weight, in addition to physical activity level. r_E and r_B are less uncertain than E_p0 and are functions of the specific physical and vocalization activities.^31,35,40 Thus, they are useful in capturing the quantitative impact of specific controllable factors. There could be factors beyond those considered here that lead to variation of viral emissions, such as the respiratory effort of patients with breathing disorders, such as emphysema or asthma.⁴¹ Such factors can be incorporated into updated Tables in the future.

Then P can be expressed as a function of E_p0, B₀ and the product of the other controllable factors as:

Where H_r is the relative risk parameter:

It has a unit of h² m^-3, which indicates the increase of risk with duration (h), the inverse of the ventilation rate expressed as air changes per hour (1/h^-1 = h), and the inverse of the volume of the space (1/m³). The other parameters such as the effect of masking are dimensionless. H_r includes the relative increase of the emission with activity (r_E), but not the quanta emission rate by a resting and orally breathing infector (E_p0). This allows using the same risk parameters for different diseases, which will naturally separate in the graphs according to their transmissibility. The four terms that make up λ may vary in relative importance for different diseases and conditions. λ_dec ∼ 1.1 h^{-1 42} has been reported for COVID-19. λ_dec depends on temperature and relative humidity.^43,44 λ_dep depends on particle size and the geometry and airflow in a given space. Respiratory particle sizes in the range from 1-5 μm are thought to play a role in aerosol transmission of COVID-19, due to a combination of high emission rates by activities such as talking⁴⁵ and low deposition rates. λ_dep for a typical furnished indoor space span 0.2-2 h^-1 over this size range, with faster deposition for larger particles.⁴⁶ λ₀ varies from ∼12 h^-1 for airborne infection isolation rooms,⁴⁷ ∼6 h^-1 for laboratories, ∼0.5 h^-1 for residences,⁴⁸ and ∼1 h^-1 for offices.^1,49 Very little ventilation data is available for many semi-public spaces such as shops, restaurants and bars or transportation. λ_cle can vary from 0, if such systems are not in use, to several h^-1 for adequately sized systems. Ventilation with clean outdoor air will be important in most situations, while virus decay and deposition likely contribute but are more uncertain for COVID-19, based on current information. In particular, the size distribution of aerosols containing infectious viruses is uncertain.

We consider a worst-case scenario where rates of deposition and infectivity decay are small compared with ventilation and air cleaning and can therefore be neglected. This also allows using the same relative risk parameter to compare different airborne diseases. This yields:

H_r can be recast as: where:

Assuming that all of the people present in a space are susceptible to infection, L is equivalent to the ventilation plus air cleaning rate per person present in the space, (typically expressed in liters s^-1 person^-1 in standards and guidelines such as from refs ^50–52. If some fraction of the people present are immune to the disease, then L is larger than the corresponding personal ventilation rate in the guidance. While this recasting will be useful to persons familiar with ventilation guidelines, we keep the form in equation (11) for most further analyses, since the number of people allowed in a space is one of the critical variables that can be examined with this relative risk parameter.

We insert equation (9) into equation (8) and obtain where we define the risk parameter

When there is only one infector present and n is small, H is proportional to N_si, i.e., the outbreak size. When it is unknown whether an infector is present, H is an approximate indicator of the absolute probability of infection (P_a), since the expected value of number of infectors (N_i) is the product of number of occupants (N) and probability of an occupant being infectious, a measure of prevalence of infectious people in local population (η_I), as shown below:

The precise estimation of η_I is complex for several reasons. Most transmission may be associated with those infected with high viral loads,^53,54 as many infected people may not shed virus into the air.³⁷ Also much spread is by people with few or no symptoms who may not know that they are infected,⁵⁵ however a fraction of symptomatic infectious people are typically in isolation. There is also significant variation in viral load and infectivity during the course of the disease.⁵⁶ For these reasons, it is very difficult to determine η_I precisely based on test data. For a situation with multiple potential infectors present (e.g., a COVID ward in a hospital), the risk parameters should be multiplied by the number of infectors.

For the analysis later in the paper that does not involve activity type or face covering choice (and thus does not involve r_E, r_B, f_e, or f_i), we define another parameter (H’), which is closely related to H:

This parameter only captures the characteristics of susceptible people’s presence in the indoor space, not of their behavior.

Value of the risk parameters for documented outbreaks of COVID-19

An important advantage of the simple risk parameters is that their values can be calculated for outbreaks that are documented in the scientific literature. Values for documented COVID-19 outbreaks are shown in Table 1 (r_E and r_B are estimated based on the likely types of activities in each case,^31,35,40 see Table S2 for typical values). Also included are values for outbreaks documented in the literature for tuberculosis and measles, which are widely accepted to transmit through the air, and an influenza outbreak that was clearly due to airborne transmission. We have included all the outbreaks known to us for which sufficient information was available to estimate airborne infection risk. Most public health investigations so far in this pandemic have neglected to report ventilation rates, the volume of the space, filter and air cleaner efficiencies, and other building science details, and thus their airborne risk cannot be estimated. It is important that future outbreak reports include this information, to allow expanding our knowledge of the circumstances conducive to airborne transmission of different diseases.

We see that the COVID-19 outbreaks that have been documented span ∼2.5 orders of magnitude range of the risk parameter H ∼ 0.09-30 persons h² m^-3. Numbers of secondary cases of these outbreaks generally increase with H (Figure 1a). Aiming to maintain values far below the threshold of 0.05 persons h² m^-3 should help reduce outbreaks.

H_r correlates well with the attack rates for the outbreaks reported in Table 1 (Figure 1b, H_r can be calculated for more literature outbreaks than H). COVID-19 outbreaks are observed for H_r > 0.001 h² m^-3, and thus indoor activities should be limited to conditions below this value during the pandemic whenever possible.

A trend line can be fitted to the attack rate vs. H_r dataset with the Box/Wells-Riley model (Figure 1b), with the fitting parameter being E_p0, i.e. the basic quanta shedding rate (when breathing only, no vocalization). A best-fit E_p0 of 18.6 quanta h-1 was obtained (with B₀ = 0.288 m³ h^-1 assumed for all occupants for simplicity). When the uncertainties of H_r and attack rates are considered, the uncertainty of E_p0 can also be estimated through Monte Carlo uncertainty propagation, described in detail in Section S2. The 5^th and 95^th percentiles of the E_p0 values are 8.7 and 45.4 quanta h^-1, respectively. This range is higher than that suggested by Buonanno et al. (2 quanta h^-1),^31,35 but overlapping with the uncertainty range provided by those authors. Note that outbreaks are typically only observed for individuals with higher quanta emission rates; many infected individuals have low emission rates and the risk in those situations will be lower than estimated here.^31,35 The attack rates estimated according to this trend line have a high correlation with the actual attack rates (r² = 0.90; Figure S1). Given the small size of the dataset of H_r and attack rates, we cross-validate this fitting by the leave-one-out method.⁶⁷ The 12 values of E_p0 obtained in the cross validation have maximum and mean relative absolute deviations of 20% and 6%, respectively, from the best-fit value, showing the robustness of the fitting.

Close alignment supports the dominant shared-room airborne character of these COVID-19 outbreaks. If the outbreaks had major components of other transmission modes (e.g., fomite or close-range transmission), we would expect a dependence on other parameters not considered here and overall much lower correlation with the risk parameters. Nevertheless, it can be observed that the attack rates of the outbreaks at low H_r (∼0.01 h² m^-3 or lower) are higher than the fitted curve. This can be due to several factors, including i) other transmission routes (e.g., short-range airborne transmission, for which the risk cannot be well captured by the model in this study and can still be significant at low H_r) or ii) the detection of only outlier cases of long-range airborne transmission resulting from the variability of E_p0 (e.g., cases where the infector’s actual quanta emission rate was extremely high), as the other cases may have too few secondary infections to be documented in the literature.

Effect of building parameters vs. human activities

The type of activity performed in each case (captured by the product of r_E and r_B) contributes substantially to the difference in H between these cases. When human activities are not taken into account, the parameter H’ only spans a narrow range of 0.09-0.56 person h² m^-3. This is probably due to similar per-person ventilation rates in many public indoor spaces (on the order of a few liter s^-1 person^-1)⁵⁰ and similar lengths of common events (in hours).

Similar to H, the variation in the values of H_r for the outbreaks in Table 1 is also largely determined by r_E and r_B. If they are not taken into account, H_r for all outbreaks would vary in the narrow range 0.001 to ∼0.01 h² m^-3, as V and λ₀ are building characteristics and D, as discussed above, is usually in hours. This implies that, in the presence of a single infector, reducing vocalization and/or physical intensity levels of the indoor activity is a very effective way to lower the infection risk of susceptible individuals. Reducing event length can also help, while reducing occupancy cannot in this case, as shown by equation (10).

It is possible that some of the most visible outbreaks are associated with super-emitter individuals, who shed virus particles at higher rates than others.^41,68–70 If that is the case, the actual H_r values at which significant transmission starts to appear in the presence of individuals that are not super emitters may be higher than those determined here. However, if super-emitters are important, so will be their contribution to total spread, and thus one should try to reduce the risk to reduce the probability of such events occurring.

Values of the risk parameters for outbreaks of other airborne diseases

In Table 1 and Figure 1 we also include a few reported indoor outbreaks of three other diseases with significant airborne transmission, i.e., tuberculosis, influenza, and measles. For outbreaks to have a similar number of secondary cases or attack rate, H or H_r needs to be higher for tuberculosis and influenza and lower for measles than that for COVID-19 (Figure 1). Note that many of the children present in the measles outbreak were vaccinated, but the risk parameter framework can still be applied by considering the number of susceptible children present. This difference is mainly due to differences in E_p0 (lower for tuberculosis and influenza and higher for measles; Figure 1b). A higher E_p0 for measles may indicate a larger amount of airborne measles virus in breath or a steeper dose-response curve for the measles virus than SARS-CoV-2, or both. A novel disease as contagious as measles would make almost any indoor situation prone to superspreading. On the other hand, tuberculosis and influenza are less contagious.

Tuberculosis transmission is propagated because untreated infected people remain contagious for years.⁷¹ The influenza outbreak occurred in an airplane without ventilation with the index case constantly coughing, and represents an extreme for this disease.⁶⁴ Most influenza patients emit significantly less virus.⁷² More discussions about quanta emission rates of different diseases can be found in refs ^35,73.

Given that both the measles and tuberculosis pathogens are widely accepted as airborne, the intermediate risk profile for COVID-19 in Figure 1b is not inconsistent with airborne transmission, contrary to frequently made arguments.^74,75 Contagiousness of a disease does not necessarily indicate the transmission route.⁷⁶ Airborne diseases can vary in their contagiousness depending on parameters such as the amount of virus shed, the survival of the virus in the air, the dose-response relationship for infection, and other parameters. The only fundamental requirement is that transmission needs to be sufficient for the disease to survive as such, something COVID-19 has had no trouble with so far.

Graphical representation of relative risks of different situations

When it is not known whether infectors are present at an indoor event, all occupants must be considered possible infectors. We assume that the probability of an occupant being infectious is the same as the fraction of infectious people in the local population (η_I). H indicates the risk of an outbreak. Consequently, the risk also depends on the number of occupants in addition to vocalization level, event duration, ventilation, and mask wearing. Jones et al.²⁵ estimated the dependency of the infection risk on these factors and tabulated it in a manner similar to Table 2. However, they only did so qualitatively. Having defined H as a risk parameter, we can assess the risk more quantitatively based on H values (as well as contact times allowed until outbreak risk is significant (H = 0.05 persons h² m^-3) and attack rates) under different conditions (Table 2). Although the actual risk also depends on η_I and the choice of the threshold for high risk (red cells in Table 2) is subjective, the risk parameter (H value) in Table 2 seems to vary in a smaller range than the corresponding table in Jones et al..²⁵ We also show that being outdoors (with much better ventilation than indoors) has a greater expected benefit than they estimated.

Note that, although occupancy has no impact on attack rate if an infector is present, occupancy affects the risk in two ways when the presence of infector(s) is unknown, i.e., i) the probability of the presence of an infector in a certain locality and ii) the size (number of secondary cases) of the outbreak if it occurs. Therefore, lowering occupancy has double benefits.

Risk evaluation for indoor spaces with pre-pandemic and mitigation scenarios

Values of the risk parameter H for some typical public spaces under pre-pandemic conditions are tabulated in Table S4 and shown in Figure 2. H in all pre-pandemic settings is on the order of 0.05 persons h² m^-3 or higher, implying significant risk of outbreak during the pandemic. Often, ventilation rates may be lower than official guidance due to e.g. malfunction, lack of maintenance, or attempts to save energy. Substandard ventilation, coupled with poor air distribution, substantially increases the risk of an outbreak and its size. This is consistent with observations that indicate that COVID-19 outbreaks are disproportionately observed in poorly-ventilated environments, which are specifically spaces with little to no added outdoor air or adequately filtered air.⁷⁷However, H and H_r of all of the pre-pandemic spaces are in a regime highly sensitive to mitigation efforts. Therefore mitigation measures such as increasing ventilation or air cleaning, reducing voice volume when speaking, reducing occupancy, shortening duration of occupancy and mask wearing are required to reduce the risk of transmission in similar settings. With mitigation measures implemented, H in these settings can be lowered to the order of 0.01 persons h² m^-3, low enough to avoid major outbreaks. Particularly, for hospital general examination room, a high-risk setting where there can be coughing infectors emitting quanta at a high rate, a combination of an improvement of ventilation rate to 6 h^-1, application of higher efficiency air filters, a halved duration, and a requirement of fit-tested N95 respirators can lower the attack rate from ∼90% to negligible (Figure 2c). Use of high-quality masks (e.g., N95/FFP3) is highly effective among the mitigation measures.

• Download figure
• Open in new tab

Figure 2:

(a, b) same format as Figure 1, but for COVID-19 only. Also shown are the H and H_r values for several common indoor situations listed in Table S4. The H values for the cases with pre-pandemic settings except for a lower occupancy and the H and H_r values for the ASHRAE standard cases⁵⁰ with pre-pandemic settings except for a lower ventilation rate are shown for comparison. The standalone legend box is for (a) and (b) only. (c) The approximately multiplicative effects of various mitigation measures for the hospital general examination room case are also shown as an example.

Calculation of risk parameters for specific situations

The calculation of H, H’, and H_r for specific situations of interest has been implemented in the COVID-19 aerosol transmission estimator, which is freely available online.³⁰ The estimator is a series of spreadsheets that implement the same aerosol transmission model described by Miller et al.¹¹ It allows the user to make a copy into an online Google spreadsheet or download it as a Microsoft Excel file for adaptation to the situations of interest to each user. The model can be used to estimate the risk of specific situations, to explore the reduction in transmission due to different control measures (e.g. increased ventilation, masking etc.), and to understand aerosol transmission modeling for incorporation into more complex models. The model also allows the estimation of the average CO₂ concentration during an activity, as an additional indicator of indoor risk, and to facilitate the investigation of the relationship between infection risk and CO₂ concentrations.^36,78 A screenshot of the estimator is shown in Figure S2.

In addition a sheet allows recalculating Table 2 in this paper for sets of parameters different from those used here (and shown in Table S2).