Quantifying proximity, confinement, and interventions in disease outbreaks: a decision support framework for air-transported pathogens

S.1 RBV calculations

Methods of calculating rebreathed volume (RBV) are described first (Sections S.1-S.4). Then, modifications that account for deposition losses are discussed (Section S.5).

Exhaled volume is treated as a conserved tracer (Bex in the equations that follow). All equations for transport of a contaminant in fluids can be used to predict the exhaled volume per volume of air (b_ex) as a function of location and time: b_ex(t, x). Although exhaled volume can be modeled in three dimensions, only a single spatial coordinate is used in the following discussion for simplicity.

The recipient breathes at a rate p, and the volume of formerly exhaled air that is inhaled by the recipient during an infinitesimal time dt is p b_ex (t,x)dt. Total rebreathed volume during residence time Δt is thus

Units of volume per volume for b_ex can be difficult to conceptualize and the dimensionless quantity b_ex might allow human errors during the creation of a new model. One alternative is modeling a mass concentration, which has a fixed value Cex(kg/m³) in the emitter’s exhaled air. The inhaled total mass can then be divided by C_ex to obtain the rebreathed volume.

We use a value of 8 L min⁻¹ for the human breathing rate, p. This value is common throughout exposure literature (1). This value is greater than the “minute ventilation” used in the medical field (typically 6 L min⁻¹), which is a resting rate that does not include activity.

Some studies report steady-state (unchanging) concentration values (c_ss(x)) that result from a constant emission rate, or source, S. The ratio c_ss(x)/S at any location is a transport function with units of s m⁻³, which quantifies the response of concentration to a unit increase in source strength, assuming that loss processes are linear in concentration. Rebreathed volume during time Δt is where one of the p values accounts for exhalation by the emitter (the source) and one for inhalation by the recipient. The ratio c_ss/S allows the use of measured, modeled, and analytically predicted concentrations. Concentrations could also be simulated with computational fluid dynamic models, as suggested by Sze To and Chao(2). When concentration is changing rather than steady, the calculation of RBV requires an integral formulation.

S.2 RBV outdoors, from plume equations

For outdoor plumes, concentration values at any distance downwind follow a normal probability distribution, with the peak directly downwind from the emitter. Dispersion is described by plume widths that increase with distance, depend on atmospheric stability, and are fit to time-averaged, measured data (3, 4). The standard steady-state Gaussian plume equation is (3):

In this equation, x is the downwind distance of the recipient, and the dependence on downwind distance is expressed by σ_y(x) and σ_z(x), which can be thought of as is the crosswind distance, and emitter and recipient are assumed to be at the same height for the most conservative case. When the recipient is directly downwind, the exponential term becomes unity.

Literature on parameterizing σ_y and σ_z under different atmospheric conditions is extensive. In stable conditions, there is less dispersion, σ-values are smaller, and concentration is greater. We use the σ-values recommended by Briggs, fit to curves known as the Pasquill-Gifford method, as tabulated by Hanna (3). Plume concentrations were calculated for each distance over the entire stability range (classes A through F) for urban conditions and for most open-country conditions. “Open-country” dispersion differs because surface obstacles are not as high and turbulence is relatively lower. We excluded open-country slightly- and moderately-stable conditions (classes E and F), reasoning that the emitter and recipient themselves would enhance the turbulence. We also explored other dispersion relationships (5) as a function of boundary-layer height and vertical velocity; these gave rebreathed volume values in the same order of magnitude.

There is a concern about using Gaussian plume models to describe transport over distances of only a few meters. Values of σ_y and σ_z are generally provided for distances of 100–10,000 meters downwind of the emission source. The term “near-source,” in dispersion literature, is associated with distances of 50–100 m. We found very little support for estimating what we have come to call “source-proximate” dispersion: transport occurring over distances of less than 10 m, relevant to person-to-person transport and to exposures of individuals very near a hazardous material release. Contaminants travel in irregular packets, so the normal or Gaussian shape is found only after averaging (6, 7), typically over several minutes; the Gaussian plume is usually applied more than 50 m from emitters. However, average concentration values do follow a standard plume shape (8, 9), and the concentrations resulting from sporadic transport can be described probabilistically (10). We found only one data set of source-proximate concentrations measured by Jones, at 2, 5 and 15 m downwind (7). Average values of c_ss/S at 2 m were 0.26 s m⁻³, or 0.65 s m⁻³ when adjusted for electrostatic effects that affect those particular measurements. These values lie in the range for plumes that are neutral or slightly unstable, extrapolating the Briggs fits for σ back to 2 m. Furthermore, dispersion principles are the basis for modeling plumes as a series of filaments for odor transmission in the source-proximate range (9), and those simulation results compare well with the Jones measurements. We conclude that standard plume equations can be used to estimate source-proximate concentrations, with caution. The proximity curves in Figure 2 use the classical plume equations for a receptor directly downwind.

The Gaussian shape of the plume cross-section emerges only after averaging over several minutes. The recipient might intercept whiffs, or plume tendrils higher than the average concentration, over short time scales. We estimated an “intermittency factor:” an increase in the average concentration if the recipient were to encounter only the whiffs. Based on the data of Murlis and Jones (10), this increase is about a factor of five, as the transported material is observed only about 15-20% of the time. This factor is quite conservative as these observations are made with time resolution of less than 1 s. The intermittency factor would decrease rapidly with averaging time and disappear after about 20 minutes (11).

For wind speed, 2.5 m s⁻¹ (5.6 mi h⁻¹) at 2 m height is roughly 3.3 m s⁻¹ (7.5 mi h⁻¹) is similar to the annual-mean 10 m wind speed in many cities (https://www.currentresults.com/Weather/US/wind-speed-city-annual.php). When all other factors are equal, outdoor ERBV is inversely proportional to wind speed, so the estimates in Fig 2 may be scaled to different wind speeds using (2.5 m s⁻¹), where u is the wind speed in m s⁻¹.

S.3 Well-mixed indoor environments

A balance equation for the total exhaled air which is introduced by a human breathing, mixed instantly into a box volume V, and carried out by air exiting the volume at a rate Q, is given by:

In indoor air quality literature, the ratio Q/V is known as an air change rate, with units of inverse time, and is often given the symbol α, as shown in the rightmost representation. Other first-order loss rates, such as deposition (Section S.5), can be added into the final term in the equation.

The relationship for exhaled volume per volume of air (b_ex) can be expressed by dividing by the room volume, V. Balance equations, and their use in modeling indoor environments, have been discussed extensively in the literature, and analytic solutions are not repeated here. Of particular note are solutions for accumulating breath (12), for differing occupancy time of emitter and recipient (1) and for accounting for filtration and personal protective equipment (13, 14).

When multiple, well-mixed zones are connected, the equation for B_ex in each zone i is(15): where Q_ij is the flow from zone j to zone i. In matrix form, each element of vectors and corresponds to a zone:

The square flow matrix consists of off-diagonal, positive flow elements Q_ij from zone j (columns) into zone i(rows). The diagonal elements Q_i,i are negative, representing the total flow out of zone i. Off-diagonal elements in each row or column must sum to the diagonal element, assuming incompressible flow. The volume matrix is diagonal only and contains the volumes of each zone. One zone can have a zero concentration and infinite volume to represent the outdoors.

The matrix formulation was coded in MatLab for this work. Other multi-zone models exist, notably CONTAM (16); our representation relies on specifying interzonal flows rather than changing them due to pressures. and enabled the setup of several similar situations with the same mitigation measures. Central air handlers were represented as separate zones to simulate multiple inlets and outlets to and from the same conditioning system. Filtration was represented as a fractional removal at the inlet to the air handler.

The formulation specifies flows between zones or between indoors and outdoors. An inherent assumption is that flows added to each zone do not affect pressures enough to alter other interzonal flows. This assumption is safe when the added supply and return flows are balanced, or when all fresh air is mechanically driven, as is the case in many commercial situations. The assumption is not true in many residences, where the baseline airflow between indoors and outdoors is caused by pressure gradients driven by natural temperature differences and wind. When mechanical flow Q_mech is imposed under these conditions, the added flow Q_add also depends on the natural ventilation rate Q_nat (17):

S.4 Indoor point-release model

Releases in indoor environments do not immediately mix into the room, so both proximity and confinement effects exist. We simulated indoor dispersion and proximity effects with a point-release model as described in Materials and Methods; additional detail and justification is provided here. The point-release model accounts for the influence of walls using an infinite series of mirror-image sources; this series was calculated until it converged. The use of image sources probably overestimates the retention in air of particles that deposit to surfaces, but we used the model mainly to estimate proximity effects near the emitter, where most of the concentration is due to the original plume and not the reflected plumes.

We simulated a range of room sizes, air change rates, and emitter locations for 1um, 10um, and 100 µm particles, as summarized in Table S.1 (27 simulations for each particle size). Dispersion parameters depended on ventilation rates as derived from measurements by Cheng et al. in still rooms (18). The proximity effect was greatest when dispersion was slowest, when the emitter was near a wall, and when the room was small. In real situations, dispersion is probably more rapid than the still-room situations shown here, so the proximity effect would be lower. Ranges of rebreathing from all the simulations are summarized and compared with simple indoor and outdoor models in Figure 2 (1 µm) and S1 (10 µm and 100 µm). The rebreathing rates shown are taken at the same height as the emitter, and thus represent the worst case at each distance.

To illustrate the interaction between proximity and confinement further, Figure S.2 shows the modeled rate of rebreathing as a function of residence time and distance from the emitter for a single room size (6m × 6m) and ventilation rate (0.3 ac h⁻¹). The gradient from left to right at less than 1 m distance shows the proximity effect; the gradient from top to bottom at 3 m distance shows the confinement effect; and the contours show the presence of both proximity and confinement effects near the emitter.

The simple point-release model does not capture all air currents and, hence, concentration variations caused by human figures, ventilation flows, and furnishings. Computational fluid dynamic models are more suited to simulate those details. We use simplified models to cover a range of conditions, to increase the generality of conclusions.

We compare our range of breathing rates with those of two studies designed to estimate the proximity effect indoors. Neither of those studies gave RBV, and our interpretation from their results is described here. Licina et al. (19) measured intake fraction for particle releases at different locations around a thermal manikin in a test chamber. The release occurred over 10 minutes, after which the source was turned off, so only the 10-minute concentrations are comparable to the point-release simulations. This study provided a total intake fraction for all 10 minutes. To translate this value to an instantaneous rebreathing rate for Figure 2B, we used our well-mixed room model to determine the ratio between ERBV₁ for time 9-10 min, and ERBV₁ for time 0-10 min, and applied this to the published 10-minute intake fraction. We used only those measurements with the release 1 m from the manikin, at different heights, but not at the floor. See Section S.10.4 for relationship between ERBV and intake fraction.

Liu et al. explored indoor proximity effects using measurements of NO₂. They defined an exposure index as the ratio between the concentrations at a particular point and that averaged throughout the room. We used our well-mixed model to determine steady-state RBV per minute, and multiplied it by the maximum exposure index reported at each distance from the emitter (Fig. 5 in Liu et al.). Like the results of the point-source model, the data points representing the Liu study in Fig. 2 are the worst case. Liu et al. (20) also used computational fluid dynamic simulations to explore the transport of different particle sizes, but did not provide enough information to calculate ERBV. We encourage future studies to provide quantitative values of source strengths and concentrations, from which ERBV can be interpreted.

S.5 Size-dependent deposition losses

RBV differs from ERBV by accounting for size-dependent losses due to particle deposition and removal strategies. In this section, we summarize deposition loss rates that modify the outdoor and indoor values of RBV.

Particle deposition, or attachment of particles to surfaces, combines several processes that depend on particle size:

1. gravitational settling, or direct interaction between a particle and surface due to gravitational forces

2. impaction, or direct interaction between a particle and surface driven by turbulent air flows that enable the particle to contact the surface

3. interception, or indirect interactions in which turbulent air flows bring a particle close to a surface, enabling particle removal, and

4. Brownian diffusion, in which particle interactions with a surface are driven by random motion of the particle in air.

The magnitude of each loss rate depends on particle size. Particles below 0.1 μm in diameter are dominated by Brownian diffusion losses, while the loss of particles above 10- μm diameter is dominated by gravitational settling. The total deposition loss rate is a combination of these factors, and is often described by a parameter known as deposition velocity vd: where m is the mass of particles of a certain size, A is the area to which they deposit, and V is the volume from which they are lost. The quantity (V/A) is the effective height (heff) of the volume; the ratio v_d / h_eff is a loss rate that has units of inverse time; and the product (v_dA) can be treated as a volume flow exiting the space. These terms are included in the mass balance equations.

Table S.2 summarizes deposition velocities used in calculation. Outdoor values of v_d are the gravitational settling velocity, assuming a density of 1 g cm⁻³ for particles that are mostly water. The fraction lost is estimated in the plume model as:

We assumed an effective height of 1.5 m, the approximate mouth height. Deposition losses over source-proximate distances are small; even for 10-um particles, less than 0.2% of particles are lost.

Deposition velocities in the indoor environment have been parameterized theoretically (21), modeled with computational fluid dynamics (22) and measured (23, 24). The theoretical model by Lai and Nazaroff (21) is well-established in the indoor air literature, and its predictions are used as central values. Measured deposition is often faster (i.e. a larger deposition velocity) than model predictions, especially for particles smaller than 1 μm and in occupied houses (25). The assumption of deposition to upward-facing surfaces only is likely acceptable for 10-µm and greater particles, but not for 1-µm and smaller particles.

Table S.2 summarizes uncertainties in deposition velocity based on observations (26, 27) and measurements from the HOMEChem experiment. Computational fluid dynamic models have simulated lower values (22), which are not included in the uncertainty ranges because they do not improve agreement with measured values.

Indoor deposition velocities are similar for particles of 0.1µm and 1µm. Although the 0.1-μm particles settle more slowly than 1-μm particles, they have greater losses by Brownian motion. For this reason, we do not report 0.1-μm particles separately in the mitigation measures; the calculation of rebreathed volume would be nearly identical. Unlike outdoor deposition losses, indoor losses over the period of confinement are significant, equal to or greater than typical ventilation rates, as shown in Table S.2. For 10-μ the 0.1-μm particles settle more slowly than 1-µm particles, they have greater losses by Brownian motion. For this reason, we do not report 0.1-μm particles, these losses are particularly important in determining indoor concentrations. The effect of deposition uncertainties on indoor rebreathing rates is shown in Fig 2C in the main text and further discussed in Section S.11.

S.6 Particle removal by face coverings

Reduction of ERBV by face coverings (masks) is included in Figure 3 for comparison with other mitigation measures. Mask efficiency ranges are estimates for the wide variety of face coverings in public use. Low-cost mask testing is becoming widespread (28) but counts particles rather than assessing removal of different particle sizes. Official mask-testing protocols recommend measurement of particles at 0.3 μm. For 1-μm particles, we used the tabulated range of efficiencies from a laboratory that maintained an active and on-line database of tests at 0.3μm size, covering many common materials (29). In our figures, the range of particle penetration for a single mask was 70–95%. Impaction and other mechanisms should stop many 10-µm particles, but they can escape by leaking around the material of an unsealed mask. Oberg and Brosseau (30) reported surgical mask penetration of less 1% to 67% for 3.1-μm particles, with most of the penetration values below than 1%. For 10-μm particles, we used penetration values of 1–30% to account for cloth masks and poor fits. We found no data for 100-μm particles and used penetration of 1%-10%; the highest value accounts for possible breakup of very large particles through the mask. This range probably overestimates penetration but does not affect any of the comparisons.

S.7 Residential situations

Inputs for residential simulations are summarized in Table S.3. In the baseline residential case, we assume that the central air handler does not operate, corresponding to mild weather conditions. We also assume that bathroom and kitchen fans are not used continuously.

The floor area covers the most common housing sizes in the United States, representing approximately two-thirds of the U.S. housing stock (31). The ceiling height is a typical height for U.S. residences. If a portion of the home were closed off, the smaller end of floor area and volume, associated with higher ERBV would be appropriate. However, people who might be concerned with infecting each other are most likely to interact in rooms whose connectivity comprises a large fraction of the home, such as a connected living room, kitchen, and dining room.

Recommended ventilation depends on the number of occupants and the floor area of the home. In the most recent guidance, the lowest air flow normalized by volume corresponds to an air-change rate of about 0.3 ac h⁻¹ (32), and we use this as a central value.

Air change rates vary widely throughout the housing stock. Characterization of leakiness is often reported as “ACH50”, or flow at 50 Pascals of depressurization by a blower door fan normalized by the volume of the conditioned space. The pressure under testing conditions is much greater than naturally-occurring pressure differences; the measured flow must therefore be adjusted to natural conditions. This adjustment depends on factors such as climate and house height, but a common rule of thumb is dividing ACH50 by 20 to obtain air-change rates under normal conditions (33). A database for the United States (34, 35) found that 3% of homes had ACH50 below 2, corresponding to a natural air-change rate of about 0.1 ac h⁻¹, and we use this value as the lower end of the range. Average ACH50 for homes participating in weatherization programs, and hence older, have been reported as around 20 (36), corresponding to about 1 ac h⁻¹ under natural conditions, and we use this value as the upper end of the range. The range (0.1–1 ac h⁻¹) spans about a factor of three below and above the ASHRAE 2019 target. Values above 1 ac h⁻¹ in normally-ventilated spaces are not supported by observations in the United States. Estimating air-change rates with tabulations of ACH50, without adjustment, would greatly overestimate ventilation.

Table S.4 summarizes mitigation measures for residential cases and provides justification. The “Label” column and order corresponds to the labeling in Figure 3. The “Description” column gives fan or air handler flows and filter ratings. Filter ratings are reported according to the Minimum Efficiency Reporting Value (MERV) from ASHRAE Standard 52.2–2017 (37). We used central values of particle removal for each rating. MERV ratings cover three particle sizes: 0.3–1.0 μm, 1.0-3.0 μm, and 3.0-10 μm. The 1.0-3.0 μm range is more representative of removal efficiency at μm (38), and we use that range for 1-μm particles.

S.8 Commercial situations

Inputs for the commercial baseline simulations are summarized in Table S.5. Central air handlers often run continuously to provide fresh air, unlike in residential situations. The floor area is an open office arrangement, chosen to be similar in size to the residential situation. While there is no requirement in the United States for square footage per person, several real estate websites indicate that open office areas should provide about 100 ft² (10 m²) per person, so this area would house about 20 people. Conditioning systems for this floor area would be sized at about “4 tons” or about 14 kW, and the total airflow for this size would be about 1600 cfm.

A common conditioning and ventilating approach for commercial buildings is to recirculate most of the building air through a system located on the roof (“rooftop unit”), and to introduce fresh air from outdoors (“outdoor air”) into this rooftop unit. Thus, more air is supplied to the conditioned space than is withdrawn from it, and the flow is balanced when building air exits through cracks. The baseline is the minimum required by ASHRAE Standard 62.1–2019 (39). The filter rating of MERV 8 is also the minimum required by ASHRAE 62.1–2019. Because the baseline commercial case has some outdoor ventilation air and some filtration, mitigation strategies that appear identical have different reductions than they do in residential situations.

Table S.6 summarizes mitigation measures for commercial cases, with the “Label” column corresponding to the labeling in Figure 3. The “Description” column provides fan or air handler flows and filter ratings.

S9. Influence of baseline dose on effectiveness of mitigation measures

Decreasing ERBV and hence dose of infectious particles cannot alter risk below the minimum effective dose. Above the dose where 95% or more of individuals are infected (ID₉₅), moderate reductions do not decrease risk either. Further, risk of infection approximately responds to the logarithm of dose between those points. We use two illustrative dose-response curves to demonstrate how the starting point (baseline dose and risk) and the slope of the dose-response curve affect expected risk reductions.

We use the summary of Watanabe et al.(40), who synthesized several studies on SARS-CoV-1, to provide illustrative dose-response curves. SARS-CoV-2 is similar to SARS-CoV-1, but we do not assume that the two viruses have the same dose-response curve. We use only features of the curve’s shape to explore how mitigation measures might reduce risk. Watanabe et al.(40) explored both exponential and beta-Poisson distributions to fit observed risk of infection. They recommended the exponential distribution because the latter, more complex distribution did not have a statistically better fit to the data. However, the exponential distribution has a fixed width in log-space, so for this exploration we used the beta-Poisson distribution.

We created two distributions with identical values of ID₅₀, but different widths (Fig. S.5). The narrower curve (“moderate” in Fig. 4) has a factor of 100 between the doses associated with 5% risk and with 95% risk, informally called “2 logs.” This factor is similar to that of the exponential curve in Watanabe. In the wider curve (“wide, illustrative” in Fig. 4), the distance between 5% risk and 95% risk is a factor of 3100 (“3.5 logs”), similar to curves estimated by Kitajima et al. (41) for H5N1.

S.10. Chain of infection, rebreathed volume, and the Wells-Riley equation

The Wells-Riley relationship (42, 43) combines characteristics of a population, infectors, and the environment into a mean number of infections μ: where I is the number of infectious disease carriers in a space, q is the emission rate of infection quanta, p is the human breathing rate (average 8 L min⁻¹), Δt is the time of the recipient’s exposure, and Q is the ventilation rate. This value of μ is then used (S.11) in the Poisson distribution to determine a probability of infection.

The Wells-Riley equation has often been used in situations where the number of carriers and the ventilation rate is known or can be guessed. It has also been related to the reproductive number R0 and the susceptible-infectious-recovered (SIR) structure (44). Here, we suggest a slight reformulation of the mean infection rate to distinguish separate roles within the chain of transmission: reservoir (carriers), environment, and host (recipients).

For a single emitter, the concentration of infectious material in a ventilated room at steady-state is q/Q, and the dose is the inhaled volume times the concentration: (q p t/Q). The emission rate of infection quanta is a characteristic of an individual carrier and is an unknown in an emerging situation. Quantifying a carrier’s emission rate as Θ_c, quanta emitted per volume exhaled, the mean number of infections considering all carriers can be written: where Σ_c indicates a sum over all carriers. When the room is in steady state, meaning that no concentrations change, the term (p² Δt/Q) is the rebreathed volume (RBV). Even in a more complex ventilation situation, the dose obtained from a single carrier is Θ_c multiplied by the rebreathed volume.

The influence of particle size will be discussed later. Until then, we acknowledge that particle size does have an influence by using ERBV instead of RBV in each equation.

Next, we add some terms that account for other variations. Some infectious individuals might exhale more or less volume, and likewise some susceptible individuals might inhale more or less. These factors can be represented by adjustment ratios between the individual’s breathing rate and the average human breathing rate: γ_c for the carrier, γ_s for the susceptible host, so that

A major insight from Wells’ work was the use of an infection quantum. This concept liberated the analyst from quantifying exactly how many virions or other pathogens were emitted, or how many were required in a minimum dose. The only requirement was determining the emission rate of just enough pathogens to cause an illness, which could be retrospectively calculated from epidemiological observations. However, this use of indivisible quanta implies that the minimal infective dose is identical for every recipient. The infectious dose of SARS-CoV-2, or any emerging disease, for humans is unknown and dependent on numerous host factors, including receptor binding and distribution. It is possible that some hosts are susceptible to a lower dose of the virus than others. Again using a ratio φ between the minimal infective dose for the susceptible individual and that for an ‘average’ individual,

The mean infection rate is now separated into three distinct components that parallel the divisions in Figure 1, as summarized in Table S.7.

The purpose of including the ratios γ_i, γ_s and φs is not to call for more information that is difficult to acquire, particularly for a novel disease. When characteristics of individual emitters are not known, equation (S.14) can be used. However, Equation (S.16) makes explicit the roles of carrier characteristics and individual susceptibility when selecting situations for interaction. For example, if breathing rates of both emitters and recipients increase by about a factor of 7 in exercise rooms (γ_i = 7, γ_s = 7), then achieving the mean infection rate found in a room of sedentary people would require decreasing person-to-person ERBV by a factor of 49, even without accounting for changes in exhaled infection quanta. Likewise, if it is found that individuals between ages 70–75 are infected at one-quarter the dose of the average population, ERBV for their situations should decrease by a factor of 4 compared with another situation that has an acceptable mean infection rate for the average population.

S.11. Influence of deposition loss on effectiveness of mitigation measures

When deposition loss rates are significant, they can dominate total removal, and increasing ventilation may have a small incremental effect. This is not a new finding (e.g. (45)), but is so important to the ability to mitigate ERBV₁₀ that it is expanded here. When ventilation is added to a room, the fractional increase in ventilation flow (δQ/Q) produces a fractional decrease in steady-state concentration (δC/C), if ventilation is the only removal. This change in concentration is the expected reduction. For contaminants that have other removal mechanisms, adding ventilation flow produces a lower reduction in concentration. We define ventilation efficiency as: or the ratio of the actual change in contaminant concentration (or ERBV) to the change that would be expected without deposition. Ventilation efficiency depends on the original ventilation rate in addition to deposition and other losses, and is graphed in Fig. S.6 for a range of deposition velocities and three baseline ventilation rates. The range of deposition velocities in Table S.2 is also shown on the figure. For 10-µm particles, ventilation efficiency ranges from 1–25%.