Air Safety to Combat Global Catastrophic Biorisks
By Gavriel Kleinwaks, Alastair Fraser-Urquhart, Jam Kraprayoon, and Josh Morrison.
Last updated on December 26, 2022.
The problem: airborne pathogens 2
How to fix indoor air contamination 3
How can we accelerate the deployment of IAQ-related interventions? 4
Pandemic respiratory disease 7
How important is risk from respiratory pathogens? 8
Mechanical Interventions to Improve Air Quality 12
Ultraviolet germicidal irradiation (UVGI) 15
Cost and cost-effectiveness of different mechanical interventions 16
Modeling the efficacy of interventions 20
Integration of room- and city-scale models 22
How could models be improved? 22
Rough estimate of impact from improvements in built environments 23
Bottlenecks and Funding Opportunities 25
What can new funding accomplish? 26
Possibilities for immediate action 29
Infectious diseases pose a global catastrophic risk if they lead to serious pandemics. This is especially true if it involves bioengineered pathogens. Out of the various methods of pathogen transmission, airborne pathogens, particularly viruses, seem especially dangerous as they are easy to spread and difficult to combat. Airborne pathogens are significantly more likely to spread indoors than outdoors, so reducing indoor respiratory pathogen transmission could substantially reduce global catastrophic biorisk by
Current indoor air standards do not consider infectious disease risk, whereas waterborne and foodborne pathogen deaths have been largely eliminated in wealthy nations due to improved water and food sanitation. Indoor air quality, especially concerning infectious diseases, should be protected, like fire safety and water safety in high-income countries.
Known effective interventions to reduce indoor air contamination include increased outdoor air ventilation, high-efficiency particulate air (HEPA) filtering, and ultraviolet germicidal radiation (UVGI). Of these, UVGI technology is the most promising because it can reach considerably higher levels of eACH by directly inactivating pathogens (20+ eACH), it is more energy efficient, and produces no noise pollution. Filtration is a viable option for high levels of eACH up to CDC hospital standards (8-12 eACH), where it is still relatively cost-effective. It also helps to reduce particulate and chemical pollution, which is relevant for immediate health concerns, such as chronic respiratory health and everyday cognitive functioning.
Within UVGI, there are two relevant types: upper-room UVC and low-wavelength light (also called far UVC).
We estimate that the mass deployment of indoor air quality interventions, like ventilation, filtration, and ultraviolet germicidal irradiation (UVGI), would reduce transmission of a measles-like pathogen by 68%. This amounts to ~1/3rd of the total effort needed to prevent a pandemic of any similarly transmissible pathogen and would serve as an important layer of biodefense.
Overall, we can be confident that these interventions effectively reduce pathogen load in the air, but we cannot precisely estimate their impact on population-level transmission. Ideally, to achieve that impact estimate, we would need a detailed model with a wide array of varying inputs built using experimental data from observed transmissions.
Despite the existence of promising technologies, several bottlenecks are preventing the mass deployment of IAQ interventions. Some significant ones include
However, significant opportunities exist to accelerate deployment via advocacy, cost and manufacturing improvements, and research.
We roughly estimate that the total cost of upgrading the air quality systems in all the commercial buildings in the US would come to ~$214 billion at current technology costs.
We give a conservative estimate that reducing the risk of a future pandemic that is twice as bad as COVID by 1% would be worth $200 billion and it seems highly likely that this program would reduce risk or severity of a pandemic by more than 1%.
We think significant action to accelerate deployment of IAQ interventions to reduce biorisk would benefit from funding in the range of $25,000-200M:
Indoor air quality has a wide range of health impacts yet is relatively ignored compared with other health interventions. COVID-19 has created a sea change in scientific attitudes towards aerosol transmission of respiratory disease, and the harmful impact of chemical and particulate air pollution continues to be documented in greater and greater detail. In this shallow investigation-style report, we explore the case for the funders, founders, researchers, and existing organizations to reduce respiratory pathogen burden by improving indoor air quality. We focus on the United States for a few reasons:
We do not discuss the benefits to future generations of reducing truly global catastrophic biothreats, but those may be substantial.
Existing IAQ Policy and Regulation: The majority of air quality guidance is aimed at chemical/molecular pollutants, with little if any focus on infectious disease.
Federal efforts have been proposed to increase US air quality: the American Pandemic Preparedness Plan (AP3) proposed allocating $3.1B for “next-gen PPE and built environment improvements” (with no indication of the split between the two), but AP3 did not pass. Despite this setback, the Biden administration released a plan to advance indoor air quality nationwide by upgrading the filtration and ventilation of federally owned buildings, funding air quality research and identifying gaps, and providing resources and incentives for upgrades in schools and residential buildings. Additionally, organizations that want to upgrade their ventilation and air purifying systems are encouraged to use funds from the American Rescue Plan and Bipartisan Infrastructure Law to do so.
In this report, we use air changes per hour (ACH) and equivalent air changes per hour (eACH) as the primary metric for air quality. Aside from eACH, indoor air standards may also be measured by the clean air delivery rate (CADR) calculated by a filter’s air flow rate in cubic feet per minute, or by parts per million (PPM) in the case of particulate and chemical pollutants. All air cleanliness measurements are imperfect proxies for assessing the safety of a room with respect to disease transmission. First, the relationship between amount of pathogen inhaled and cases of infection is unclear, and probably varies for different pathogens. Second, because some IAQ interventions like ultraviolet and low-wavelength disinfection inactivate pathogens at variable rates, the eACH will vary for different pathogens.
The COVID-19 pandemic has caused significant damage worldwide but was by no means unusually destructive. There have been far more lethal historical pandemics, such as the Black Death and the 1918 Spanish Flu, which have killed 20-100 million people (5-10% of the world population at the time). While the burden of endemic infectious disease has trended downward, it is unclear whether the risk of natural catastrophic pandemics is increasing or decreasing.
While naturally evolved pathogens could lead to globally catastrophic pandemics (i.e. destabilizing enough to threaten the entire future of humanity) , evolution tends to optimize for reproductive fitness, rather than optimal virulence. On the other hand, bioengineered pathogens could be developed that are much more dangerous than any with natural origins. As biotechnology progresses and biotechnological capacity diffuses more widely, the accidental or deliberate release of an engineered pathogen becomes increasingly likely.
Regardless of whether a pathogen is natural or engineered, deliberately or accidentally released, some attributes are likely to be essential components of catastrophic pathogens. A report from the Johns Hopkins Center for Health Security notes that a global catastrophic-risk level pathogen is most likely to be a virus, due to viruses’ higher capacity for genetic mutability, and to have respiratory transmission routes, since this is the mechanism most likely to lead to pandemic spread. Current interventions to interrupt respiratory transmission are more difficult to implement than with vector-borne, sexually transmitted, or fecal-oral routes.
The majority of aerosolized respiratory pathogen transmission occurs indoors; in the COVID-19 pandemic it is estimated that likely more than 90% of transmission has occurred indoors and that the odds of transmission are at least 20 times higher indoors than outdoors.
Given the above, improving indoor air quality, i.e. reducing indoor respiratory pathogen transmission, could substantially reduce global catastrophic biorisk by:
Ideally, improving indoor air quality is only a part of a portfolio for reducing global catastrophic biorisk, alongside other interventions like advanced PPE, improving early pandemic detection, and advocacy to better manage dual-use research of concern.
We estimate that 90-99% of COVID-19 infections come from aerosol sources , between 40-80% of influenza transmission, and 20-90% of the overall disease burden of other common cold viruses. The relative importance of modes of transmission between pathogens is very poorly quantified. For most common respiratory pathogens (aside from COVID-19 and to an extent, influenza) the data required to make meaningful quantitative predictions does not currently exist.
IAQ interventions to prevent disease primarily act on aerosolized particles. The impact of IAQ on disease transmission is dependent on the fraction of pathogen transmission attributable to airborne transmission.
While some diseases, most notably COVID-19, TB, and chickenpox, are widely accepted to be dominantly airborne, most respiratory pathogens have historically been assumed to be primarily driven by large droplet/fomite transmission. This assumption now seems highly uncertain given updated research avenues.
We can assume that possible catastrophic pandemics will be primarily caused by aerosolized pathogens, due to the difficulty of controlling airflow in most settings, the difficulty of enforcing air hygiene relative to other modes of transmission, and the ease with which aerosolized pathogens can fill a space and infect many people at once. Therefore when considering global catastrophic pandemic risk, it makes sense to focus efforts on reducing risks from airborne pathogens.
Covid-19: The vast majority of COVID-19 infection is transmitted via aerosols, mostly indoors. Early studies were mostly observational, with a notable early study showing a clear case of aerosol transmission in a restaurant.
Influenza: Over the last decades, large amounts of research have been conducted on influenza transmission, but consensus is far from clear. Literature reviews provide convincing evidence of both closer-range/fomite transmission, and transmission via aerosols (though not without divergent opinions). Computational models can also predict dominant aerosol transmission of influenza.
Killingley, B. (2012) Investigations into Human Influenza Transmission. [Unpublished PhD thesis]
Various real-world intervention and controlled studies have been completed:
There exists highly convincing evidence of all major transmission routes for influenza. However, it seems reasonable to take as a lower bound 40% airborne transmission (the lowest value in the Bangkok/Hong Kong intervention study), and an upper bound of 80% (based on the success of the Livermore hospital study).
Tuberculosis: TB stands out for having a potentially indefinite incubation period. Only 5-10% of people infected with the bacteria ever develop symptoms, most can carry the disease for life without ever knowing. TB is transmitted through the air by aerosol droplets from people with active symptoms. It is extremely infectious; fewer than ten bacteria may cause an infection, compared to as many as 40,000 bacteria in a single sneeze. As a result, a quarter of the world’s population has been infected. In a 1961 study, a team from Johns Hopkins exposed two groups of guinea pigs to air from a TB ward, the air going to one group having been irradiated with UV light first, in order to demonstrate airborne transmission. Infections only appeared in the group with untreated air, showing in the process that UV light is effective at killing the pathogen.
Common cold viruses: From the 1970s to the 1980s, two teams carried out human challenge trials on rhinovirus, where volunteers challenged with rhinovirus interacted with healthy volunteers. The first, in 1978 in Virginia, found that hand-to-hand transmission is an efficient way to transfer rhinovirus infection, while attempts to cause large droplet and aerosol spread mostly failed. Then in a three-study series running through 1987, a Wisconsin team built a challenge model, found virucidal treated tissues were effective in preventing transmission of rhinovirus, and found that inducing infection via a fomite and large droplet route was ineffective, while measures designed to specifically induce aerosol routes of transmission maintained high attack rates.
The two teams came to two separate determinations of the importance of aerosol transmission. However, the Wisconsin studies were more likely to generate accurate results. For example, efforts (not described in the paper) were made to reduce air leakage in the Wisconsin study. Despite these controlled natural exposure studies being some of the highest-quality research ever performed on pathogen transmission, the results have not caused significant change.
Another study in an Army barracks demonstrated a newer building with a lower ventilation rate was associated with an average of 45% higher risk of common cold infection (typically adenovirus), providing strong evidence that aerosol transmission is important for other common cold pathogens.
In general, despite the more limited range of studies, convincing evidence shows a significantly higher fraction of transmission might be via the aerosol route than historically acknowledged. We think it is reasonable to say that between 20% and 90% of common cold transmission occurs through the aerosol route.
More research into airborne disease transmission is highly warranted. Given existing pandemic history and pandemic potential stemming from characteristics of the diseases listed above, less dangerous pathogens such as flu strains used in challenge trials or common cold strains might be used to analyze airborne disease transmission in great detail. In performing such analysis, it would be vital to limit the possibility of dual-use research (ie, research that could be used both maliciously and benevolently) by ensuring that research was focused on limiting the spread of existing or strictly weakened forms of pathogens rather than engineering pathogens to be more transmissible.
The body of work on this topic is of limited size and quality. Simply isolating a mode of transmission (even before attempting to quantify importance) outside of a highly controlled environment is difficult, and many observational studies suffer from confounding variables. Methods exist to retrospectively model the importance of various transmission routes based on previous data, but suffer from significant gaps limiting use.
Another challenge for both interpretation and usability of data are the development of pervasive errors in the medical literature, significant enough to obviate the results of some studies. Some of the most common are:
Known effective interventions to improve indoor air quality include increased outdoor air ventilation, high-efficiency particulate air (HEPA) filtration, and ultraviolet germicidal radiation (UVGI). Within UVGI, there are two relevant interventions: upper-room UVC and low-wavelength light (also called far UVC). Air quality standards are typically set in terms of air changes per hour (ACH) or clean air delivery rate (CADR), measured as units of clean air produced per unit time. These interventions are valuable because they are relatively pathogen agnostic and can act as a layer of passive biodefense (compared to developing a specific vaccine).
Ventilation exchanging outdoor air with indoor directly achieves true air turnover, whereas filtering and UVGI impact is measured in equivalent air changes per hour (eACH). The resultant eACH in a room outfitted with filters or UVGI lights depends on several different factors, including the quality of the filter or power of the lights, the number and placement of lights or filters, and air mixing. In addition, for UVGI application, the pathogen in question is relevant since the eACH calculation depends on the percentage of pathogens deactivated over the course of the time frame; tuberculosis and COVID-19 are the primary pathogens typically used to calculate eACH.
Ventilation: Strong evidence exists that increased ventilation has a marked effect on infection rates (and health more generally), supporting the efficacy of ventilation as a general method to reduce prevalence of pathogens. However, in order for ventilation to be effective, the air entering a room must be of higher quality than the air leaving. Outdoor air is an easy source, provided it is of reasonable quality, meaning this strategy can be less effective in highly polluted areas. In addition, outdoor air often requires temperature changes to be acceptable indoors, which means expending significant amounts of energy on climate control. Ventilation is therefore likely to be a less attractive option for organizations attempting to provide building occupants with cleaner air, given the permanent and continuous added cost of energy use and the fact that increased energy use would work against LEED standard compliance.
In general, ventilation can follow one of three overall strategies: mixing, displacement, and personalized. Mixing ventilation simply adds clean air to dirty air, where displacement ventilation aims to take advantage of the effect caused by pathogens being emitted in a warm plume, combined with the natural heating effect of humans, which likely creates a thermally stratified layer above head height with a higher concentration of pathogens. This process theoretically results in a risk reduction factor of 1.2-2 for influenza. Personalized ventilation gives each occupant a designated ventilation flow, creating a similar risk reduction factor. These strategies, or similar variants, may be effective at reducing pathogen transmission, but data is limited.
Filtration: Filtration involves passing air through a filter, designed to remove some proportion of particles from the air. This method is effective in reducing both pathogen transmission and some indoor air pollution, including PM2.5. In addition, outdoor air can be filtered before being introduced indoors to improve quality. Filters cause a pressure drop; the higher the filtration rate, the higher the pressure drop, meaning more energy is required to move the same amount of air through a building.
Standalone filtration units (graded using the MERV scale, with HEPA filters being the most efficient, removing >99.9% of small particles) have been shown to reduce the exposure to pathogenic aerosols under controlled conditions, with 5 eACH HEPA filtration in classrooms enough to cause a 4-5 fold drop in pathogen dose. In a model of a 30 person restaurant, with baseline US prevalence, increasing ACH of 0.8 to 12 eACH using HEPA filters averted an estimated 54 COVID-19 infections per year, with a gain of 1.35 QALYs.
Addition of filters to existing ventilation systems in a typical model scenario has been shown to reduce relative risk of infection of influenza by up to 47%, at a total annual cost of $352 for HEPA filters, with MERV 13/14 filters (removing a lower fraction of particles), shown to be nearly as effective at considerably lower costs (total annual cost $156, $119 per unit risk reduction compared with $232 for HEPA).
In addition to reducing pathogen transmission, filtration has benefits for respiratory health and cognition, due to its ability to remove harmful particulate, gaseous, and chemical pollutants. Given these benefits, widely investing in improved filtration in built environments is likely to help the population even in non-pandemic years.
Light in the UVC band (up to 280 nm wavelength) can be used to inactivate pathogens via protein/DNA damage, but does not reduce particulate pollution. This method is known as ultraviolet germicidal irradiation, or UVGI. Wavelengths at the higher end of the UVC spectrum are easier to produce via lamps, but are harmful to humans, causing corneal damage within hours, as well as significant skin damage.
Upper-Room UVC: This strategy directs UVC light to the top of the room, so harmful UVC is not passed to humans below. Studies have shown 80% efficacy in TB transmission reduction with guinea pigs exposed to hospital air, with strong evidence demonstrating reduction of various pathogens concentration under laboratory conditions. Models predict between a 1.6 and 3.4-fold decrease of TB infection in a hospital waiting room using lighting with eACH 7.5 and 31.7, respectively. Strong knowledge of the mechanisms of UVC allow creation of predictive models for inactivation ability by pathogen.
Low-wavelength light/far UVC: Recently, significant interest has grown in a narrow band of UVC light of 200-230 nm, which is ionizing enough to inactivate pathogens, but not to penetrate the outer layers of human skin or the corneal layer. This band is frequently referred to as “far UVC”, but in order to avoid confusion with upper-room UVC, we’ll refer to it as low-wavelength light. In theory, low-wavelength light can be used much more easily in many environments to inactivate pathogens, without harming humans. It can be used to interrupt surface, short-range aerosol, and droplet transmission, which is difficult to prevent via other mechanical interventions, making it potentially the most effective intervention for reducing existential biorisk Low-wavelength lamps have been so recently developed that this end of the spectrum is generally not included in analysis of current interventions, and lacks long-term human safety data.
Low-wavelength light has broad germicidal activity, with low doses (permitted under current regulations) sufficient to inactivate 90% aerosolized coronaviruses in 8 minutes, and 99.9% in 25 minutes. Efficacy can vary from pathogen to pathogen, but low-wavelength light causes no currently known significant damage (1, 2) to human skin and cell models even at doses significantly higher than required germicidal doses.
Long-term exposure studies in humans and adjustment of regulations could be required for widespread acceptance, and further studies are warranted. Given that low-wavelength light (under test conditions) provides 184 eACH at an irradiation level already permitted in the US for 8 continuous hours, no evidence has yet raised concrete safety concerns. Low-wavelength lamps also generate some ozone, but safe levels/limits of ozone exposure are already regulated by a number of bodies. No standard procedure exists for testing purposes and estimates for production vary widely, so quantification is difficult.
Low-wavelength light is difficult to produce, strictly limiting the consumer base for low-wavelength lamps so far. Widespread commercialization will likely require the development of LEDs in the correct light band, but it is difficult to estimate when these will become commercially available. Low-wavelength LED technology requires further fundamental research into materials and manufacturing techniques in order to improve efficiency and cost. For an extremely rough estimate, we can say that the requisite research might be part of a PhD program, taking about 5 years (assuming multiple streams of research/multiple PhD programs running in parallel). Then assuming that some of those research pathways are successful, demonstrating that efficient, cost-effective low-wavelength LEDs are feasible, it might take another 5 years to achieve full commercialization.
While the potential and expected impact on airborne pathogen transmission matter more for assessing the attractiveness of different mechanical interventions from an x-risk perspective, cost and cost-effectiveness matters to government and corporate adopters since these potential adopters are more likely to adopt these interventions at particular price points.
As a case study in the cost of upgrading ventilation for a large public space, the Center for Health Security report on school ventilation (Appendix F) focuses on a direct comparison between the cost-effectiveness of ventilation versus the early CDC surface cleaning guidelines. A more comprehensive analysis was prevented by the knowledge gaps in aerosol transmission discussed below. Based on expert interviews, the report estimates that a school would need $6,000 per classroom for upgrading HVAC systems to provide air quality equivalent to about 5 to 7 air changes per hour (ACH). At an estimate of 2.5 million public school classrooms nationwide, the cost of upgrading all schools would be $15 billion (although students in each upgraded school would benefit before all schools were upgraded, so the total outlay is not needed for intermediate benefits). Rothamer et. al. show that increasing airflow in a schoolroom from their measured baseline of 1.34 ACH to 5 ACH reduces the probability of infection by about half.
This report only includes the cost for upgrading systems and running new systems at a basic level, but does not include the costs of post-upgrade energy consumption. If upgrades are done primarily through ventilation, there is substantial added energy consumption, estimated by one expert at a 15-20% overall energy cost increase for 10 eACH throughout a school.
In order to estimate the cost of upgrading the general stock of public buildings in the US, we’ll use the published estimate that educational space uses 14% of commercial floorspace in the US. We’ll additionally estimate that public K-12 buildings are about half of that floorspace, so 7% of commercial floorspace. Then the total cost of upgrading the air quality systems in all the commercial buildings in the US would come to about $214 billion.
This figure is prohibitively expensive for rapid implementation. However, it assumes the cost of upgrading systems stays fixed, but this field is getting increased attention and investment so costs might come down considerably over the next decade. Also, more targeted programs addressing high-priority public spaces as an intermediate step would be less expensive and still reduce pandemic risk and improve everyday health. For example, building on the estimate of $15 billion to upgrade public primary education facilities, we can produce the following upgrade cost estimates using the percentage breakdown of commercial building stock:
If upgrades to public buildings were to be implemented across a decade in the US, $21 billion a year would be spent on a complete air quality upgrade program. For comparison, in 2021 alone, the US Department of Defense spent $10 billion on facilities maintenance and construction and $141 billion on weapons and systems procurement. We use the comparison with defense spending because biosecurity is an important component of national security and these figures demonstrate what people are willing to spend on defense, not because we would expect government spending to fully fund this program.
Researchers from the Institute for Progress and the Johns Hopkins Center for Health Security demonstrate that the COVID pandemic cost the US at least $10 trillion in combined economic and health losses. Using their lower-bound numbers and lenient assumptions for a future pandemic (half as destructive as COVID), they estimate that it would be worth $50 billion to reduce the risk of a future pandemic by 1%. Naturally, given the optimism of these assumptions, pandemic reduction efforts are potentially worth much more.
We give a conservative estimate that reducing the risk of a future pandemic that is twice as bad as COVID by 1% would be worth $200 billion and it seems highly likely that this program would reduce risk or severity of a pandemic by more than 1%
The below studies give some indications of cost and cost-effectiveness (in terms of ACH/eACH) for different IAQ interventions. While actual implementation costs vary somewhat depending on the installation, the following points broadly summarizes cost and cost-effectiveness:
Upper-bound of effectiveness (ACH/eACH)
Installation cost per ~70m2 room
Relative operational cost
($ USD/ACH or eACH)
$6000 (modern HVAC system, to provide air quality equivalent to 5-7 ACH)
High, due to large amounts of energy spent on climate control
~$135 per ACH
$1000-1500 (multiple HEPA purifiers equivalent to 4-6 ACH)
High, as a higher number of filters reduces air pressure, so more energy is used to move air through a building
~$110 per eACH
$1500-2500 (8-12 incremental eACH)
Low, and costs are likely to be stable annually since there is no need for climate control
~$14 per eACH
$2500-5000 (10 incremental eACH or 30 ‘breathing zone’ eACH)
Similar to upper-room, but higher due to current bulbs being less efficient
$15-46 per eACH
Table: Summary of different mechanical IAQ interventions in terms of upper-bound effectiveness, installation cost, relative operating costs, and cost-effectiveness [various sources]
Nardell (2021) compares the cost-effectiveness of different mechanical interventions, determining that upper-room UV is the best option when comparing it against mechanical ventilation and filtration. Upper-room UV was calculated to produce up to 24 eACH under standard air mixing conditions (ie, air mixing resulting from convection currents and people moving through the room), and was estimated to cost roughly $14 per eACH in a hospital room, making it over nine times more cost-effective than mechanical ventilation. By contrast, three air filters that were compared against mechanical ventilation and upper-room UVC were estimated to cost $100-$300 per eACH, ranging from about half as cost-effective as mechanical ventilation to about the same cost-effectiveness. As a baseline, the model estimates that mechanical ventilation alone provides about one air change for about $135.
Currently, models to predict the efficacy of air safety interventions exist. However, a lack of real-world data means that many lack real-world validation, and most models are unsuitable for predicting the efficacy of airborne interventions for averting pandemics.
This is because:
Ideally, to be more confident about estimating the efficacy of any air safety intervention, we would need a detailed model with a wide array of varying inputs that was built using experimental data from observed transmissions.
Room scale models predict the efficacy of interventions on a small scale. Specifically, they assess the probability of infection based on the mix of susceptible and infectious people occupying a space, and the rate at which an infectious person is able to infect susceptibles.
The most common method is based on the Wells-Riley equation, which expresses pathogen emission from infectors in terms of quanta, a single quanta being the average amount of pathogens required to cause an infection. The standard equation assumes a perfectly-well mixed room, meaning that each emitted quantum has a 63% chance to cause infection in a susceptible individual, assuming no removal of pathogens from the air.
The difficulty with this model is estimating the quantum generation rate, which is calculated backward from epidemiological studies, but involves a significant amount of inherent uncertainty. In general, the rest of the model (assuming a well-mixed environment) follows numerically from this quantum generation rate, meaning most of the uncertainty in the model (such as pathogen emission and infectivity) is included in this number.
Additionally, assuming a well-mixed environment can cause significant errors. However, there are efforts to improve upon the basic model, by estimating quantum emission rate by what fraction of air is rebreathed, based on real-world examination of CO2 levels.
Most population models use the SEIR model, which stands for “susceptible, exposed, infected, recovered.” This is a set of differential equations that dictates the chance of a susceptible person becoming infected based upon number of exposures to infectors in a simulated population, and adjusts this chance based on the proportion of the population currently infectious.
However, to model the effect of the deployment of air safety interventions, assuming a homogeneous population is not appropriate. Unlike a universal masking policy, or vaccination, which can be modeled to have more general effects, air safety interventions may have very different impacts on the progression of a pandemic depending on the environment of deployment (e.g. installations in a restaurant may have very different effects to installation in schools).
To avoid this problem, an SEIR model can be combined with a simulation of a population, where the population is split up into different classes of people (such as children, teachers, workers, stay-at-homes etc), and different environments (such as supermarkets, schools, offices, homes etc). By modeling SEIRs through this population, the effect of reducing spread in particular areas can be estimated. Constraints on computing power are a problem for more complex models that involve modeling of different types of population, such as the variance between large cities in different countries or the difference between a city and a rural environment.
Currently literature using such models is sparse. One simplified model of Hong Kong predicts that increasing ventilation rates by 5 ACH in all public buildings reduces attack rate of smallpox by ~80% and total infection by ~97% in a medium transmission scenario. The same model showed a similar increase in ventilation rates had significant effects on reducing peak and total infections in simulated influenza outbreaks even with varying proportions of airborne transmission. A later, more detailed Wells-Riley/human behavior integrated model predicted that increasing ventilation in all buildings threefold (to 3-6 ACH depending on building) would suppress a smallpox outbreak. These papers provide specific examples of an end-to-end generated model, but they use pathogen and intervention strategies different enough from a catastrophic outbreak that the results are not particularly generalizable.
Studies using a Wells-Riley equation to model the effect of ventilation on population transmission predicted ventilation rates up to 12 ACH brought a hypothetical airborne virus with a quantum emission rate of ~26 from an R0 of ~10 to <1. A second Wells-Riley/SEIR model predicted a 60-80% reduction in R0 for a hypothetical airborne pathogen with 15 additional ACH. These models, aiming to provide a more general estimation technique, do not simulate a population, instead integrating the Wells-Riley with a standard SEIR model. In addition, unlike many models focused on current endemic pathogens, these do not account for cost-effectiveness.
Ideally, a model would be able to predict the efficacy of air safety interventions, taking into account:
A model should then be able to predict:
Existing efficacy studies: Studies investigating various interventions are largely unhelpful for validating models or understanding the effect of those interventions at a population level. For example, recent studies in laboratory chambers clearly demonstrates how low-wavelength light greatly reduces pathogen load in the air, but that reduction in pathogen load cannot be directly connected to precise reduction in transmission. There are studies that estimate the infectious dose of various pathogens, but they provide an imperfect bridge to population-level intervention efficacy models, due to the wide range of estimates and the variation among individuals. On the other hand, several studies intended to directly investigate the effect of interventions have serious issues with methodology and practicality that limit their usefulness.
While there currently is not a robust model for predicting the efficacy of indoor air safety interventions, in this section we hope to provide a reasonable sketch of how much air quality improvements in public spaces can reduce disease transmission. When considering air quality upgrades in built environments, we focus on public spaces where superspreader events are more likely and improvements are easier to confirm. The following calculation is extremely rough and a more informative, detailed model is sorely needed for a full analysis of possible public health measures.
Say that 95% of transmission occurs indoors, that 80% of transmission occurs in public spaces (including offices, schools, gyms, theaters, eateries, etc), and that ideal adoption of current pathogen mitigation measures in public environments (including ventilation, filtration, and use of germicidal UV light, especially upper-room UV) can reduce transmission by 90%. Each of these factors is independent of the others, so by using the described program of air quality interventions to address transmission in public spaces, overall transmission in the population can be reduced by 68%.
The most transmissible pathogen we know of is measles, which is estimated to have an R0 of around 20. Say that the goal of a pandemic prevention program is to reduce pathogen transmission by 98%, bringing R0 of measles to 0.4, thereby preventing a pandemic of any similarly transmissible pathogen. In order to achieve this goal, various types of interventions will need to be used in conjunction, such as combining indoor air cleaning with widespread use of personal protective equipment. In particular, there is room for significant further development of PPE and UV lights, which could make both interventions more effective and easier to adopt at large scales.
The value of the described program (reducing transmission by 68%) relative to the absolute goal (reducing transmission by 98%) can then be considered as logarithmic progress toward the overall goal, with further investment to be made. The intervention program is then about 29% of the total investment needed to achieve a transmission reduction of 98%. In estimating that proportion, we assume that other intervention programs, such as development and distribution of improved personal protective equipment, could be “stacked” with this indoor air intervention program for roughly the same cost. We prioritize this program not because of obvious cost favorability, but because of its capacity to address superspreader events and international spread (eg, by greatly reducing transmission in airports), and because it is a program of “passive” interventions, which do not rely on individuals’ actions to achieve the majority of the gains. (Contrast this with the “active” interventions described in Kevin Esvelt’s “Delay, Detect, Defend” Geneva Paper, such as equipment, resilient production, and diagnostics.)
This transmission sketch is extremely rough on many counts. Further technological development of interventions could reduce transmission even further; conversely, we cannot expect ideal adoption of mitigation measures, so changes to public built environments can only be one tool used in fighting pandemics. In addition, this value estimate does not address the value of reducing R0, even if R0 does not fall below 1. An intervention program that fell short of averting a pandemic could still be enormously valuable by buying extra time to develop and distribute medical countermeasures such as vaccines and PPE. Also, in this estimate, we assume there is no self-correcting behavior in the population as a result of IAQ interventions.
Overall, we can be confident that these interventions effectively reduce pathogen load in the air, but we cannot precisely estimate their impact on population-level transmission without a detailed model (see recommendation further down).
Each of the recommendations below will be associated by number with the bottlenecks we believe are addressed through that recommendation.
Funding opportunities exist in advocacy, changing costs and manufacturing, and research.
IAQ policymaking occurs along an adoption curve that includes implementation by businesses, voluntary certification codes, retrofitting government buildings (like schools), subsidies for private renovations, building code requirements for new construction, and codes requiring implementation in existing buildings. We see three advocacy opportunities as immediately attractive:
Other potentially attractive opportunities are:
An advanced market commitment (AMC), funded by government, philanthropy, or business, could spur development of a product by committing to a purchase once technology meets certain specifications. AMCs already have a track record, such as Operation Warp Speed, which incentivized COVID vaccine development and acquired COVID vaccines in the US. One likely use-case could be development of LED-style low-wavelength lamps to replace KrCl lamps (currently expensive and produced by only a few manufacturers). There may also be relatively expensive products in the upper-room UV and filtration arenae where cost could be reduced by guaranteeing purchases at greater scale, possibly as a way of subsidizing early pilot experimentation.
As of right now, it appears that it is possible to create low-wavelength LEDs, but it is unclear whether they can be manufactured in a reliable and cost-effective way. The blue LED was only developed in the 1990s and was considered a major breakthrough at the time; even lower-wavelength LEDs are likely to require the development of new semiconductors or new manufacturing methods. However, once low-wavelength LED manufacturing can be made reliable at a high quality, the demand is likely to be high (especially if supported by an AMC, as mentioned above). Another option for supporting this LED development is direct investment or funding for fundamental materials and manufacturing research. Such funding could take the form, for example, of support for PhD students or other researchers working in the field, which is well within the normal activities of several philanthropic or governmental organizations.
Low-wavelength safety testing: Low-wavelength light is a potentially transformative intervention, and studies to develop a safety record sufficient for wide use in humans should be a high priority. Studies have already been successfully conducted on realistic 3D skin models, with intense monitoring for damage, and some longer-term studies on mice made deliberately susceptible to tumors. Interventions of a similar risk have been proposed based on the evidence of models. In-human longer-term studies could be feasible on a dedicated population (possibly an office block), with monitoring for early signs of damage, combined with an early efficacy study.
Valid efficacy models: Creating a way to experimentally test the efficacy of various IAQ interventions will be a necessary component of engendering and optimizing implementation over the long-term. One model to do so is the idea described above of randomizing experimental pilots in early adopters.
CNE studies: Another approach is to utilize controlled natural exposure (CNE) studies, which are a version of human challenge studies where uninfected “recipient” volunteers are exposed to infected “donor” volunteers. Despite their ability to provide some of the highest-quality, cleanest quantitative data of aerosols, transmission routes, and interventions, they are uncommon, with only two large-scale studies in the last two decades - one in 2010, finishing with an attack rate too low to be of use and one additional study planned over the next five years.
We think that exploration of CNE studies stands to be a valuable research contribution requiring a high level of cooperation between fields.That said, these studies are still in their infancy; using them to experimentally test intervention efficacy may require significant investment on the order of tens of millions of dollars, which may not be cost-justified from an EA perspective.
To provide a rough estimate of the impact of a widespread air quality campaign in the US, we made some basic assumptions about the timeline and possible impact of a campaign, and compared the result against a counterfactual baseline. You can see our calculation here and input new assumptions to see how they affect a campaign’s impact.
Many indoor air quality projects could build on each other and create momentum for further efforts, and a dedicated funding pathway could coordinate several complementary projects. For example, a useful long-term path might start by funding a set of scientific studies. As research produces further data on interventions and optimal programs, funding could be used for dedicated advocacy and deployment in partnership with early organizational adopters. This implementation would in turn lead to iterative research and wider deployment.
Projects in these areas could absorb significant amounts of funding along a wide range. For example:
There are a few reasons ways IAQ interventions could fail and could end up being harmful:
Many of these risk factors can be mitigated by the activities recommended in this report (e.g., developing better models and metrics, real-world efficacy studies, robust safety studies, monitoring public attitudes, and advocacy efforts).
Appendix 1: Report for Open Philanthropy Cause Exploration Prize, which formed the first draft of this report, although less focused on catastrophic pandemic risk and pilot programs.
Appendix 2: Summaries of EA organizations’ work on indoor air quality.
Appendix 3: Notes from Henna Dattani on low-wavelength UV.
Appendix 4: Convergent Research’s executive summary on germicidal UV
Appendix 5: Sketch of possible UV-C pilot program for a given office space.
 Measles has the highest transmissibility of any known pathogen.
 R0, pronounced “R naught,” is a mathematical term that indicates how contagious an infectious disease is. It tells you the average number of people who will contract a contagious disease from one person with that disease. If R0 is less than 1, each existing infection causes less than one new infection, and the disease will decline and eventually die out.
 Air quality standards are typically set in terms of air changes per hour (ACH) and equivalent air changes per hour (eACH).
 ASHRAE refers to the American Society of Heating, Refrigerating, and Air-Conditioning Engineers. It is a professional association and regulations on indoor workplace settings are primarily based on ASHRAE guidelines.
 8 out of 10 top airports for passenger traffic are in the US.
 Factors seem to point in both directions, with the development of vaccinations and therapeutics and greater understanding of disease transmission reducing the risk. On the other hand, increased air travel and larger domestic animal reservoirs point to increased risks of natural GCBRs.
 Of the “environmental” pathogens, waterborne and foodborne pathogen deaths have been largely eliminated in wealthy nations due to improvements in sanitation and broad access to treatment. However, air sanitation has yet to reach the same standards as water and food sanitation, even in wealthy nations.
 We received this information from an expert in the field who had personal knowledge of the study design; he did not explicitly agree to be cited by name.
 This total cost estimate also uses the $15 billion estimate for upgrading school HVAC systems. However, if air quality improvements included UVGI to achieve target standards, rather than relying on HVAC alone, the cost could be significantly lower due to the higher cost-effectiveness of UVGI.
 Order-of-magnitude check: In the US, about 1 million people have died of COVID. Government agencies typically use $1-10 million for the value of a statistical life, ie, how much should be spent to save a life. These figures would place the cost of COVID at $1-10 trillion in life loss alone, or that hypothetically the US government should be willing to spend up to $10 trillion to fully avert another COVID-size pandemic.
 See estimate in “Rough estimate of impact” section
 The dollar cost of one equivalent ACH, considering the amortized cost of installing and running mechanical ventilation.
 Most HVAC systems in public buildings in the US do not have the duct or blower capacity to be increased to 6 ACH [link]
 This is the preferred ACH level recommended by the CDC for an airborne isolation rooms in hospitals and is achievable via filtration, but cost and noise become prohibitive closer to this level [link]
 Source is same as above.
 Upper-room UVC, with good air mixing, has been shown in the real world to achieve 24 eACH and studies suggest it’s possible to achieve >100 eACH when paired with adequate ventilation.
 322 eACH is the top end of the estimated ACH in a ‘high’ exposure scenario using five lamps, though this did not exceed the ACGIH threshold limit value for the skin. This could potentially be even higher.
 Price range from conversation with a low-wavelength light vendor.
 Very rough calculation based on upper-bound eACH and installation cost ranging from $5000-15,000, does not include operating costs.
 These costs are very rough and actual installation and operational costs are highly variable, depending on room size, electricity prices, outside temperature.
 It is difficult to know how much transmission generally occurs in public spaces. However, early in the pandemic, it was estimated that around 80% of new infections were generated by about 10% of cases (Nature, Science), implying public settings and especially superspreader events had a very large role in spread. With no better estimate of spread in public settings, I used the 80% figure directly.
 Ie, in order to find overall transmission reduction, I just multiplied the listed factors.
 We have not done a cost comparison with other programs.
 Self-correcting refers to situations like during the COVID-19 pandemic, people appeared to dynamically adjust their behavior based on apparent COVID-19 prevalence.
 Some known manufacturers include: Ushio, Eden Park Illumination, and Sterilray. Ushio is the current leader in terms of lamp efficacy/lifetime/filter quality, and was on the scene earliest.
 We were recently informed that Prof. Ernest P. Blatchley III of Purdue University is working on something similar, although we have not spoken with him.