The impact of heating, ventilation, and air conditioning design features on the transmission of viruses, including the 2019 novel coronavirus: a systematic review of filtration

Search Strategy

From inception to June 2020, three electronic databases (Ovid MEDLINE, Compendex, Web of Science Core) were searched by a research librarian (GMT) using concepts related to virus, transmission, and HVAC. The Ovid MEDLINE search strategy is detailed in Table 1. Prior to implementing the searches, the strategies were peer-reviewed by two librarians (TL, AH). In January 2021, the search was updated. Reference lists of all relevant papers were screened, in addition to relevant review articles. Conference abstracts were identified through Compendex and Web of Science and were not included, but literature was searched to see whether any potentially relevant abstracts had been published as complete papers. Limits by year or language of publication were not placed on the search; however, only English-language studies were included due to the volume of available literature and resource constraints. EndNote was used to manage references with duplicate records removed prior to screening.

Study Selection

Study selection had two stages. First, two reviewers independently screened titles and abstracts of references identified by electronic database searches. Relevance of each record was classified as Yes, Maybe, or No. Any conflicts between Yes/Maybe and No were resolved by one reviewer. To ensure consistency among the review team, pilot testing was conducted with three sets of studies (n=199 each). The review team met to discuss discrepancies and develop decision rules after each set of pilot screenings. Second, two reviewers independently reviewed full text articles and applied inclusion and exclusion criteria. Reviewers classified studies as Include or Exclude. Any conflicts between Include and Exclude were resolved by consensus of the review team. One reviewer resolved conflicts between different exclusion reasons. The second stage of screening was pilot tested with three sets of studies (n=30 each). The review team met to resolve discrepancies after each pilot round. The review team conducted screening using Covidence software.

Inclusion and Exclusion Criteria

Inclusion and exclusion criteria are listed in Table 2. This systematic review was part of a larger effort to examine different HVAC design features and virus transmission. While all design features were included in the search and screening process at once, only studies evaluating filtration were synthesized here. A variety of agents were included in the search; however, studies of viruses or agents that simulated viruses were prioritized. Other agents (e.g., bacteria, fungi) would be included only if studies were not available that were specific to viruses. Studies using bacteriophages, which are viruses that infect bacterial cells,²⁰ were included. This review is interested in studies of the indoor built environment (e.g., office, public, residential buildings) which had mechanical ventilation. Primary research that provided quantitative results of the correlation or association between filtration and virus transmission was included. No restrictions were placed on year of publication and only English-language, peer-reviewed publications were included.

Risk of Bias Assessment

For experimental studies, risk of bias was assessed based on three key domains: selection bias, information bias and confounding.^21-22 Reviewers assessed each domain as low, unclear, or high risk of bias using signalling questions¹⁷ from guidance documents for the different study types that were included; e.g., animal studies, laboratory experiments, epidemiological studies.^21-23 For modelling studies, risk of bias was assessed using the following three key domains: definition, assumption, and validation.^23-24 The domain definition considers model complexity and data sources, assumption considers the description and explanation of model assumptions, and validation considered model validation and sensitivity analysis.²⁴ Reviewers assessed each domain as low, unclear, or high risk of bias based on signalling questions.^23-25 Risk of bias items were pilot tested among three review authors. Then, two reviewers (GMT, BAF) applied the criteria independently to each relevant study and met to resolve discrepancies.

Data Extraction

Reviewers extracted general information about the study (authors, year of publication, country of corresponding author, study design) and methods (setting, population [as applicable], agent studied, intervention set-up). Details on filtration parameters (where available) were extracted, including any filter efficiency ratings or classifications (such as MERV and EU ratings or HEPA) or filter efficiencies expressed as percentages, where applicable. More recent filter efficiencies are based on ASHRAE Standard 52.2-2017 and represent particle size removal efficiencies (PSE) with particles falling into three size ranges: E1 0.30-1.0 µm; E2 1.0-3.0 µm; and E3 3.0-10.0 µm.¹¹ Efficiency of particle removal is usually represented by conventional filter efficiency ratings, such as MERV, EU, F, and HEPA. Quantitative data were extracted, as well as results of any tests of statistical significance related to filtration features. The primary outcome of interest was quantitative measures of the association between filtration and virus transmission. As such, data on actual transmission were extracted where available (i.e., infections), as well as information regarding viruses, filters, setting and experimental test set-up, and outcomes. The distinction between the efficiency of particle removal described above and the efficiency of virus removal is important. The efficiency of virus removal is calculated as the difference between the virus concentration upstream of the filter and the virus concentration downstream of the filter divided by the virus concentration upstream of the filter expressed as a percentage. Only one aerosolized virus study used terminology that referred explicitly to virus, viral filtration efficiency,²⁶ while the remaining studies on virus removal used a variety of terms typical for particle removal: removal efficiency,²⁷ filtrating efficiency,²⁸ filter efficiency,²⁹ and filter reduction efficiency.^30-31 European ratings were presented with MERV rating conversions.³² Using a data extraction form spreadsheet to ensure comprehensive and consistent capture of data, one reviewer extracted data and a second reviewer verified data for accuracy and completeness. Any discrepancies were discussed by the review team.

Data Synthesis

As anticipated, meta-analysis was not possible due to heterogeneity across studies in terms of study design, filtration features examined, outcomes assessed, and reporting of results. Evidence tables were developed to describe the studies and their results, as well as a narrative synthesis of the results. To allow for meaningful synthesis and comparison across studies, the studies are presented and discussed in three groups: animal studies (Table 3), aerosolized virus studies (Table 4), and modelling studies (Table 5).

Animal Studies (n=7)

Six studies evaluated the influence of filtration on PRRSV in pigs^33-38 and one study examined NDV in chickens.³⁹ Of the seven studies, three were performed using experimental test set-ups ups^33-34,39 and four were performed in the field (i.e., pig barns).^35-38

Five of the seven studies found that filtration was associated with decreased virus transmission (Table 3).^33,35-37,39 Pitkin et al³⁵ also found that filtration was associated with decreased daily risk of infections for PRRSV. One study found the use of filtration was associated with decreased presence of PRRSV downstream of the filter³⁴ and another³⁸ found the use of filtration was associated with decreased relative risk of PRRSV introductions.

Dee et al³³ and Hopkins and Drury³⁹ found that increased filter efficiency was associated with decreasing transmission. Hopkins and Drury³⁹ documented that improper installation of a high efficiency filter allowed transmission to happen. Additionally, Dee et al³⁶ found no difference in the transmission reduction between MERV 14 and MERV 16 with no transmission events in both situations. As well, two studies found that filter performance varied by manufacturer when testing similarly rated filters from different companies.^34,39

Aerosolized Virus Studies (n=7)

Five of seven aerosolized virus studies found that filtration was associated with virus removal using PRRSV,³⁰ equine arteritis virus (EAV),³⁰ bovine enterovirus (BEV),³⁰ human adenovirus,²⁹ and the bacteriophages T7^27-28 and MS2.³¹ One study found that increasing MERV rating was associated with decreasing virus concentration for bacteriophage T4⁴⁰ at different locations within an apartment unit, and another found that increasing MERV rating was associated with increasing viral filtration efficiency using bacteriophage MS2²⁶ in a test duct set-up (Table 4).

Like the animal studies conducted by Dee et al³³ and Hopkins and Drury,³⁹ two aerosolized studies found increasing MERV ratings to be associated with increased virus removal efficiencies.^26,40 Zhang et al²⁶ noted that the filter efficiency associated with E1 (i.e., 0.30-1.0 µm; the smallest particle size range for MERV classification) was a good approximation of the viral filtration efficiency. In addition, Washam et al,²⁷ prior to the use of MERV ratings as developed in 1999, identified cost differences between filters of similar efficiency.

Modelling Studies (n=9)

Nine of the filtration studies were modelling, employing the following models: one-dimensional analytical model,⁴¹ Wells-Riley model,^42-45 CONTAM multi-zone IAQ model,⁴⁶ Susceptible-Exposed-Infectious-Recovered (SEIR) model,⁴³ an indoor contact model,⁴³ a control volume conservation model,⁴⁷ Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model,⁴⁸ School Building Archetype (SBA) model using both the Wells-Riley and Monte-Carlo approaches,⁴⁴ the multi-region FATE model of Fauske & Associates,⁴⁵ and dose-response model.⁴⁹ Viruses investigated include SARS-CoV-2,^45,49 measles with influenza size distributions,⁴⁴ influenza,^42-43,46 rhinovirus,⁴⁶ PRRSV,⁴⁷ avian influenza,⁴⁸ and virus modelled as gas.⁴¹ Augenbraun et al⁴⁹ sought to examine SARS-CoV-2 but the relevant infectivity parameters were set to that of influenza. A wide variety of settings were analyzed including a twin aisle airliner cabin,⁴¹ an office building,⁴² a cruise ship,⁴³ a school building,⁴⁴ a detached home,⁴⁶ sow barn,⁴⁷ a farm,⁴⁸ and a quantum physics lab.⁴⁹

From the nine modelling studies, five studies found that increasing filter efficiency associated with decreased virus concentration^41,46 and decreased risk, including absolute risk and relative risk of infection,⁴² transmission risk,⁴⁴ and infection risk⁴⁸ (see Table 5). One study found that the use of filtration was associated with decreased attack rate.⁴³ Two studies found HEPA filtration was associated with decreased risk of SARS-CoV-2 infection⁴⁵ and decreased probability of SARS-CoV-2 infection.⁴⁹ Finally, Janni et al⁴⁷ found that a combination of MERV 8 prefilter and MERV 16 was associated with a lower concentration of PRRSV compared to a combination of MERV 8 prefilter and MERV 14.

Of the nine modelling studies, four evaluated the use of a single filter^43,45,48-49; all agreed that installing a filter is more beneficial than not installing a filter. Like the experimental studies of Dee et al³⁶ and Spronk et al,³⁷ and three modelling studies comparing single filters^41-42,46 agreed that there is a threshold beyond which increasing filter efficiency may not affect the probability of infection. One study using a MERV 8 prefilter was able to model a difference in virus concentration comparing MERV 14 and MERV 16.⁴⁷ Mazumdar and Chen⁴¹ compared HEPA filters of different efficiencies and determined that improvement was not evident when increasing filter efficiencies past 99.90%. Azimi and Stephens⁴² and Brown et al⁴⁶ specifically found that there were diminishing returns for MERV 13 or higher filters when using influenza, while Dee et al³⁶ and Spronk et al³⁷ found no difference between MERV 14 and MERV 16 filters for PRRSV. Azimi and Stephens,⁴² using influenza, compared varying MERV filters and a HEPA filter and found that, for filters MERV 13 or greater, the risk reduction was not limited by the filter efficiency; instead by the amount of recirculating air through the HVAC system. Similar to Washam et al,²⁷ Azimi and Stephens⁴² considered the cost of filtration. Compared with the cost of an equivalent amount of outdoor air ventilation, Azimi and Stephens⁴² found filtration was the less expensive option for all MERV ratings except MERV 4.

Risk of Bias

Twelve of the 14 experimental studies had low risk of bias for all three domains: selection bias, information bias, confounding (Table 6). Two experimental studies had low risk of bias for selection bias and confounding but had unclear risk of bias for information bias because of a lack of clarity in the methodological details.^28-29 Seven of the nine modelling studies had low risk of bias for all three domains: definition, assumption, validation (Table 7). Two modelling studies had low risk of bias for assumption and validation but had unclear risk of bias for definition because of a lack of clarity about the HEPA filter efficiency⁴⁵ and contribution of fresh air.⁴⁹

Implications for Research

The modelling studies spanned from 2009 to 2020 and the aerosolized virus studies spanned from 1966 to 2021; some of the more recent studies were motivated by the COVID-19 pandemic. Many viruses were examined; however, it is particularly interesting that there were only two coronavirus studies (both SARS-CoV-2) for filtration, both modelling, and only one aerosolized study that used MS2 as a potential surrogate for SARS-CoV-2 based on size.²⁶ Additionally, one of the studies⁴⁹ discussed the implications of filtration in relation to SARS-CoV-2 but used influenza to set relevant infectivity parameters. The limited number of coronavirus (and specifically SARS-CoV-2) studies suggests a potential research gap to be addressed in the future. There is a very rich body of literature of engineering or laboratory studies which focuses on collecting physical metrics expected to affect transmission; these were not considered in this search as our focus was on connecting transmission itself with installed equipment. The trade-off between isolating specific parameters related to filtration physics or analysing more combined links in the chain of transmission is an important consideration and we stress that the literature gaps tend to be in studies which consider systems holistically. This points to a need for interdisciplinary studies which combine knowledge of well controlled filter experiments with less well controlled “real world” situations which inherently capture combined synergies in the transmission chain as well as possible “unknown unknowns” in that chain.

Interestingly, the animal studies only spanned from 1971 to 2013. Furthermore, no field studies examining potential mitigation of virus transmission in humans using filtration were found. This suggests a research opportunity to investigate the use of filtration on mitigating virus transmission in “real-world” built environments occupied by humans, as well as epidemiological studies during disease outbreaks. Taken together, these findings identify an opportunity for more research that investigates HVAC filtration mitigation of human viruses and, specifically, human coronaviruses.

More recent aerosolized virus studies use the ASHRAE Standard 52.2 test duct.^26,31 This establishes an important level of consistency for future research. The aerosolized virus studies that state an explicit interest in SARS-CoV-2 used the bacteriophage MS2, which has a smaller size than SARS-CoV-2. Future research could explore the ramifications, if any, of this size differential on virus transmission given the size of droplet or droplet nuclei in which the virus itself is suspended.

Implications for Practice

Some findings are straightforward to address in practice. For example, proper installation is required for a filter to be effective in removing virus and preventing transmission.³⁹ Other findings are challenging to address in practice. Two studies documented variations in the performance of similarly rated filters that were purchased from different manufacturers.^36,39 Variations in performace can occur in filters with the same MERV rating from the same manufacturers, not just different manufacturers. This alone implies that within the broad criteria of MERV ratings, there is enough variation in performance to point to a need for more filter sub-ranges or more stringent testing methods of MERV value.

A practical benchmark emerging from this review was, from the findings of Zhang et al,²⁶ that a conservative estimate of the viral filtration efficiency could be approximated by the E1 efficiency used to determine MERV rating in ASHRAE Standard 52.2-2017. Not surprisingly, increasing filter efficiency, whether based on increasing filter efficiency before MERV ratings were established,³⁹ increasing MERV rating,⁴² or increasing from MERV to HEPA,³⁴ was associated with mitigation of virus transmission. An important practical consideration is that the improvement is limited above the filter efficiency MERV 13. This concept of diminishing returns was documented in both animal studies^36-37 and modelling studies.^41-42,46 Spronk et al³⁷ and Dee et al³⁶ found that both MERV 14 and MERV 16 were associated with no transmission, while Azimi and Stephens⁴² and Brown et al⁴⁶ showed that filters rated MERV 13 or better were associated with limited improvement. One model demonstrated decreased virus concentration with MERV 14 versus MERV 16 but both had MERV 8 prefilters.⁴⁷ While this could suggest an important role of prefilters at higher MERV ratings, the results are difficult to generalize across MERV ratings because only two were modelled. Although one study investigated prefilters, no studies examined filter replacement frequency within the context of viral filtration efficiency.

Diminishing returns are not limited to MERV ratings as Washam et al,²⁷ prior to the use of MERV ratings, found that similar efficiency can be achieved using a lower cost option. In addition, the cost increases with increasing MERV rating and HEPA filtration relative to the cost for MERV 13; however, the cost of filtration for these filters remains less than the cost of outdoor air ventilation with the equivalent transmission reduction.⁴² This finding is consistent with current ASHRAE recommendations for reducing airborne infectious aerosol exposure.¹² Interestingly, ASHRAE indicates that MERV 13 is recommended but MERV 14 is preferred.¹³ The use of MERV 14 is supported by WHO in the context of COVID-19.¹⁴

Installing or upgrading an air filtration system can sometimes be an expensive intervention. However, illness costs money as well and there is considerable research available showing simply how improved indoor air quality (IAQ) infrastructure investment is profitable when building owners and users interests are aligned⁵⁰. Installing air filtration on large sow farms is estimated to cost $150-200 per sow⁵¹. Alonso et al³⁸ postulates that air filtration like this is quite common now due to the major economic impact of PRRSV on swine production. When considering poultry, Hopkins and Drury³⁹ agreed that the additional cost of high efficiency filters would be justified by the investment return of the flock. A single influenza case results in an approximate economic loss of $375 in the United States⁵². Azimi and Stephens⁴² found that upgrading a MERV 7 filter to MERV 13 would only be $17 in labour cost annually in a typical office environment while reducing the number of infections by 1. Not only does air filtration reduce the number of viral infections, it improves the overall air quality by reducing the amount of allergens and asthma triggers in the air. An extra $50 per year for high efficiency air filters in a family residence is seen as a small additional cost against the enormous financial and health burden of asthma⁴⁶.

Some studies have shown that there can be a plateau in effectiveness for filters past MERV 13 in typical situations^42,46. Mazumdar and Chen⁴¹ found that a 94% efficiency filter would not be able to protect passengers sitting far away from the infectious source in an airliner cabin and that a 99.9% filter was needed for their application. They go on to state that further increase in filter efficiency might be difficult and its effectiveness not clear. All filters with the same rating are not equal either. Differences in the filter media or frame design across companies and inconsistency in the filter rating systems across countries can contribute to discrepancies in filter efficiency³⁴. The source of the raw material of the filter media can also impact filter quality³⁵. Zhang et al²⁶ found that the viral filtration efficiency (VFE) is generally correlated with the MERV rating but they are not the same. They found that the VFE is always higher than E1, but lower than E2 or E3 efficiencies. This is all to say that choosing the right filter for the right application is important. The location, use case, existing system, budget, and acceptable level of risk should be taken into account when considering air filtration systems^33,36,37.

When considering scientific testing of filters, the size of the aerosol challenge should be considered very carefully. Many studies have stated that the level of contamination in their experiments was probably much larger than what would be expected in their respective typical situations^27,34,39,40. This could lead to certain interventions being wrongfully deemed inadequate.

Filtration should not be seen as the only approach to infection mitigation. Filtration should be paired with other infection mitigation strategies such as vaccination⁴⁴, distancing shared space users, and implementing a wait time before different users access the space⁴⁹, and other strict biosecurity measures³⁰. Installation of a high efficiency filter increases the pressure drop which can lead to greater energy use if the system has to run more often to service the same volume of air. If installing or upgrading the air filtration system is not possible, Bandaly et al²⁹ suggests switching to 100% outdoor air because some viral particles that pass through an inferior filter will still be infectious.

From our review of the literature, it is clear that upgrading or installing high efficiency air filtration is cost effective in the long run and reduces virus transmission. It is recognized that not all spaces are the same and it might not be feasible to implement high efficiency air filtration. If this is the case, then other infection mitigation measures such as UVGI, increasing ventilation rate, or controlling the relative humidity should be considered.