Face masks to prevent community transmission of viral respiratory infections: A rapid evidence review using Bayesian analysis

Background: Face masks have been proposed as an important way of reducing transmission of viral respiratory infections, including SARS-CoV-2. Objective: T o assess the likelihood that wearing face masks in community settings reduces transmission of viral respiratory infections. Methods: We conducted a rapid evidence review and used a Bayesian statistical approach to analysing experimental and observational studies conducted in community-dwelling children and adults that assessed the effectiveness of face mask wearing (vs. no face masks) on self-reported, laboratory-confirmed, or clinically diagnosed viral respiratory infections. Results: Eleven RCT s and 10 observational studies met the inclusion criteria. T he calculation of Bayes factors and cumulative posterior odds from the RCT s showed a moderate likelihood of a small effect of wearing surgical face masks in community settings in reducing self-reported influenza-like illness (ILI) (cumulative posterior odds = 3.61). However, the risk of reporting bias was high and evidence of reduction of clinicallyor laboratory-confirmed infection was equivocal (cumulative posterior odds = 1.07 and 1.22, respectively). Observational studies yielded evidence of a negative association between face mask wearing and ILI but with high risk of confounding and reporting bias. Conclusions: Available evidence from RCT s is equivocal as to whether or not wearing face masks in community settings results in a reduction in clinicallyor laboratoryQeios, CC-BY 4.0 · Article, May 1, 2020 Qeios ID: 1SC5L4 · https://doi.org/10.32388/1SC5L4 1/19 confirmed viral respiratory infections. No relevant studies concerned SARS-CoV-2 or were undertaken in community settings in the UK.


Introduction Introduction
On March 11 2020, the global outbreak of the respiratory virus SARS-CoV-2, which causes COVID-19 (1), was declared a pandemic by the World Health Organisation (2). T he primary route into the body for respiratory viruses such as SARS-CoV-2 is through the nose, eyes and mouth (the 'T -Zone') (3). Multipronged approaches involving both pharmacological (e.g. vaccination) and behavioural measures (e.g. hand washing, social distancing) are required to bring the reproductive number below 1 during respiratory virus epidemics (4)(5)(6)(7). In public health interventions where certainty cannot be assured, it is often necessary to judge the benefits of interventions on their likelihood of benefit versus harm. T his paper reports a rapid evidence review of studies evaluating the wearing of face masks in community settings on the likelihood of their leading to a reduction in the transmission of viral respiratory infections.
SARS-CoV-2 is spread through airborne droplets, and possibly in some cases aerosol, containing virions (8). Face masks of various types (e.g. surgical masks) filter droplets containing virus. However, they may not reduce transmission of the virus in community settings if they are not used correctly and may even increase transmission if they act as fomites or prompt other behaviours that transmit the virus such as face touching. For example, a face mask that has been worn for several hours becomes moist and acts as a potential source of contamination. Studies show that people touch their faces [15][16][17][18][19][20][21][22][23] times per hour on average (9,10), and this may mean that eyes and contaminated face masks are touched, spreading the virus. Several reviews have been undertaken on whether wearing face masks confers net benefit or harm (11)(12)(13)(14)(15)(16)(17)(18)(19). T his rapid review aims to draw together the key evidence to date to try to establish the most comprehensive picture available. Given that policy has to be made on the basis of the likelihood of benefits versus harms rather than necessarily a high degree of confidence that a given policy will have the desired effect, it is important to focus on this likelihood. T herefore, this review includes the use of a Bayesian analysis to calculate cumulative posterior odds of the benefit of face mask wearing. It also widens the scope to consider issues such as adherence and adverse unintended consequences. We aimed to address the following research questions: 1. What is the likelihood that wearing face masks in community settings reduces

Evidence synthesis Evidence synthesis
Results from individual comparisons and outcomes in individual studies were tabulated in terms of adjusted odds ratios and 95% confidence intervals with the control group as the reference.
Following inspection of the results, it was decided to undertake Bayesian analyses to quantify the likelihood that face masks were effective. T his involved calculating Bayes factors for each comparison and each outcome in each study, and then combining these Bayes factors to calculate cumulative posterior odds of a reduction in respiratory viral infections (23). Bayes factors represent the ratio of the likelihood that a given hypothesis (H1) is true versus another hypothesis (H0). In this case, H0 was that there was no difference between intervention and control conditions. T wo different H1s were tested: 1) a small effect of a reduction of up to 10% in the odds of infection (adjusted odds ratio of 0.90), and 2) a large effect of up to a 50% reduction in the odds of infection (adjusted odds ratio of 0.50). H1s were specified using a half-normal distribution starting on 0 with a standard deviation of the expected effects size (i.e. 10% or 50%) (24

Study description Study description
A total of 486 records were identified in the 10 literature reviews, 29 full texts were assessed, 21 of which met the inclusion criteria (see Figure 1).  Korea (45) and T hailand (46). No studies were found that were conducted in the UK.

Outcomes Outcomes
Adherence to mask use Studies operationalised self-reported adherence as hours/day of mask use  (26,37,38,43,46), the proportion of participants reporting mask use always or most of the time (vs. sometimes or never) (27,29,34,35,42), the proportion of participants reporting mask use as instructed (44) and the proportion of participants who reported mask use within 48 hours of symptom onset (39). Of studies explicitly commenting on the level of adherence to mask use, three study authors stated that adherence was 'good' (26,42,43) and three stated that it was 'poor' (35,39,44). Six studies did not report adherence to mask use (28,31,33,40,41,45).
Effectiveness of mask use RCT s T he outcomes of included studies are reported in T able 2. One study found lower rates of self-reported symptoms of influenza-like illness (ILI) in the intervention compared with the control arm; however, in secondary analyses with laboratory-confirmed ILI, the rate of infection was less in the control arm than the intervention arm (26). T en studies, two of which were pilot studies, found no statistically significant reduction in the rate of laboratory-confirmed or self-reported symptoms of ILI with face mask use in their primary analyses (33,(35)(36)(37)(38)(39)(42)(43)(44)46). In post-hoc (underpowered) analyses, however, significant reductions in rates of ILI were reported in six studies. T wo studies found reduced rates of ILI in weeks 3-6 of the study period (totaling 6 weeks) (37,38). One study found significantly reduced odds of a household contact developing laboratoryconfirmed ILI when the analysis was confined to participants who were allocated to the intervention or control arms within the first 36 hours of symptom onset in the index patient (36). One study found a significant reduction in the number of viral symptom episodes in a multivariable analysis following adjustment for age of the index case, education level of the caretaker and home crowding index (but not in a univariable analysis) (39). T wo studies found a significant reduction in the rate of ILI in household contacts when, in a post-hoc analysis, they restricted the analysis to participants who received the face masks within 36 hours or two days of index case diagnosis or symptom onset, respectively (42,44). One study found a significant reduction in respiratory infections when restricting the analysis to the less stringent end-point of clinical respiratory illness (as compared with laboratory-confirmed infections or ILI) (33). T he calculation of Bayes factors and cumulative posterior odds indicated that data showed a moderate likelihood of a small effect for the wearing of face masks on self-reported symptoms but evidence on clinically-or laboratory-confirmed ILI was equivocal (see T able 3). Observational studies Six observational studies found a significant reduction in self-reported respiratory virus symptoms in individuals who reported the use of face masks (as compared with no face mask use) (27,29,34,40,41,45). Four studies found no significant reduction in respiratory virus symptoms in individuals who reported the use of face masks (28,(30)(31)(32).

Predictors of clinical outcomes Predictors of clinical outcomes
Four studies assessed whether self-reported adherence to mask use was a predictor of clinical outcomes, three of which observed a positive association (26,34,44) and one did not (43). T wo studies found reduced rates of infection when participants had been allocated to wear face masks within 36 hours of symptom onset (36,42). One study found that when the number of protective behaviours (e.g. hand washing, face mask use) was considered as a continuous variable, those engaging in a greater number of protective behaviours experienced shorter duration of respiratory illness (31).
Adverse unintended consequences Adverse unintended consequences T he majority of included studies did not report on whether there were unintended consequences. T wo studies found that 50-75% of participants in the face mask arm reported pain/discomfort with mask use (43,44). One study found that those allocated to the face mask arm (as compared with those allocated to the face mask plus hand sanitiser or control arms) reported significantly less use of hand sanitiser (38). Four studies reported no significant differences in hand hygiene across study arms (35,36,39,46).

Quality appraisal Quality appraisal
T he quality ratings for each study are reported in T able 4.

Study limitations
Participant blinding to group allocation was not possible. Some studies reported contamination as participants in control arms decided to use face masks of their own accord (26,33,35,36). Use of self-reported (as opposed to laboratory-confirmed) respiratory virus symptoms or illness was commonplace. Overall, adherence to face mask use was poorly recorded.

Inconsistency of results
Only one of the 11 higher-quality studies employing RCT designs found a significantly reduced rate of ILI in their primary analyses. Both of the two higher-quality observational studies found a significantly reduced rate of clinically-or laboratory-confirmed ILI (41,45).
Hence, the results are inconsistent across study designs and outcome assessments, with those employing more robust designs finding a non-significant effect of face mask use. Indirectness of evidence Only four of the included studies were conducted during an ongoing epidemic (31,34,42,45) and none was conducted during the SARS-CoV-2 pandemic. Only one of the 11 RCT s assessed transmission in the wider community (39); the remaining studies assessed viral spread to contacts who shared accommodation. A key concern during respiratory virus pandemics is transmission outside the household of index patients.

Imprecision
T he RCT that found a significant effect of face mask use did not provide a confidence interval for the point estimate (26). One of the two higher-quality observational studies reported a narrow confidence interval, likely due to the large sample size (41). T he remaining five observational studies with positive results reported wide confidence intervals (27,29,34,40,45), thus indicating poor precision of the effect of face mask use.

Reporting bias
Most analyses were not pre-registered, opening the possibility (especially in secondary analyses) of 'cherry picking' of findings.
T able 4. GRADE quality ratings for the included studies. Implications for policy and practice Implications for policy and practice While the potentially biased self-reported outcomes from RCT s suggest a small benefit of face mask wearing, findings on clinically-and laboratory-confirmed infection remain equivocal. In addition, none of the studies concerned SARS-CoV-2 and none were conducted in the UK. All were in community settings that were different in many respects from the situation pertaining to SARS-CoV-2 in the UK. In light of this, judgements about the benefits or harms of wearing face masks will have to be made using a priori arguments rather than the data reviewed here: the scientific evidence should be considered equivocal. Such arguments should pay special attention to specific settings where the risk of infection is high and the opportunity for physical distancing is low (e.g. on crowded public transport), and to the need for education and training to maximise the potential benefits of wearing masks and mitigate the risk that they will transmit infection by acting as fomites.
Future research priorities Future research priorities A standard protocol needs to be established for evaluating the benefits or harms of specific approaches to promoting face mask wearing in defined settings and populations.
T hese protocols need to use objective measures of infection and take special precautions to minimise the risk of bias. T hey also need to include specific information on what was done to promote the appropriate use of face masks and collect data on spillover effects. Such a protocol is urgently needed for the COVID-19 pandemic but will continue to be relevant for future epidemics.

Conclusions Conclusions
Evidence from RCT s is equivocal on whether face mask wearing in community settings reduces the transmission of clinically-or laboratory-confirmed viral respiratory infections.
RCT s and observational studies have found an effect on self-reported symptoms, but this may be the result of reporting bias and confounding. No relevant studies concerned SARS-CoV-2 or were undertaken in community settings in the UK.

Conflicts of interest Conflicts of interest
None declared.

Ref erences
Ref erences