The Impact of Community Masking on COVID-19: A Cluster-Randomized Trial in Bangladesh
Our take — This was a very large and well-designed cluster-randomized controlled trial of a multi-pronged intervention program to encourage mask-wearing in rural and peri-urban Bangladesh from November 2020 to April 2021; it was available as a preprint and is thus not yet peer reviewed. The study found that the intervention package (which included mask distribution, public role-modeling, and encouragement to non-mask-wearers in public settings) more than tripled public mask usage behaviors (from 13% to 42%) without diminishing observed physical distancing. Most importantly, villages with the intervention had a 10% relative reduction in symptomatic seroprevalence (reported COVID-19 symptoms with presence of SARS-CoV-2 antibodies) compared to control villages. Surgical masks were apparently more effective than cloth masks. This study did not measure whether mask wearing was associated with COVID-19 at an individual level, and there are potential problems with the outcome measure of symptomatic seroprevalence. However, the study provides evidence that interventions promoting mask use can reduce community SARS-CoV-2 transmission even when fewer than half of community members wear masks. If the results are valid, they imply that near-universal mask wearing would be associated with much larger reductions in transmission.
Randomized Controlled Trial
Study population and setting
This was a cluster-randomized trial of community-level promotion of mask-wearing, designed to estimate the impact of interventions on SARS-CoV-2 symptomatic seroprevalence in Bangladesh from November 2020 to April 2021. Secondary outcomes included mask-wearing behavior, physical distancing behavior, and COVID-19 symptom prevalence. The study was conducted in 600 rural and peri-urban villages containing 341,830 adults throughout Bangladesh. From an initial sampling frame of 1,000 rural and peri-urban unions (a union is a rural administrative unit in Bangladesh, consisting of ~80,000 people), the authors used pairwise randomization to select 300 intervention and 300 control unions, and selected one village (containing the largest regional market) in each union. The intervention was rolled out in waves from November 2020 to January 2021, and lasted for 8 weeks. Control group villages received no intervention. Intervention group villages received a multi-pronged mask promotion strategy, designed in conjunction with local leaders. The intervention had five elements: 1) one-time mask distribution and promotion to households, 2) mask distribution in markets 3-6 days per week, 3) mask distribution at mosques on three Fridays, 4) mask promotion in public spaces and markets, during which non-mask-wearers were encouraged to wear masks, and 5) role-modeling and active promotion of masks by religious and other community leaders. In addition, intervention villages were cross-randomized to four additional interventions: 1) cloth vs. surgical masks, 2) village-level incentives if a given level of mask-wearing was achieved vs. no incentives, 3) public commitments via provision of signs to households vs. no public commitments, and 4) additional text message reminders vs. no reminders. There were also three additional household-level cross randomizations: 1) messages focused on community protection vs. self-protection, 2) twice-weekly text message reminders, and 3) asking households to make a verbal commitment to mask-wearing. The households received differently colored masks depending on the sub-arm to which they were randomized, and the color schemes themselves were randomized village-by-village. The behavioral outcomes for this study were measured by an enumerator who recorded behaviors in a public space in each village. The enumerator noted whether people were wearing masks, the colors of those masks, whether both mouth and nose were covered, and whether people were keeping physical distance from each other. Behavioral data was measured up to 10 weeks following the start of the intervention, and additionally 20-27 weeks after baseline. One member per household completed phone surveys at weeks 5 and 9 to assess SARS-CoV-2 symptom prevalence (consistent with the WHO COVID-19 case definition for suspected or probable cases with an epidemiologic link) in the previous month. Participants who reported symptoms were tested for IgG antibodies against SARS-COV-2 during weeks 10-12.
Summary of Main Findings
Symptomatic seroprevalence: Symptom data was collected from 98% of the study population; 8.6% of participants in control villages reported having COVID-19 symptoms, while the figure for intervention villages was 7.6%. Of all 13,893 symptomatic participants, 40% consented to providing blood, and 37% were tested for SARS-CoV-2 antibodies. The adjusted prevalence ratio for symptomatic seroprevalence was 0.907 (95% CI: 0.817 to 0.997), implying a 9% relative reduction in symptomatic seroprevalence comparing intervention villages to control villages. When considering only symptoms, the adjusted prevalence ratio was 0.88 (0.83 to 0.93), implying a 12% reduction in self-reported COVID-19 symptoms during the study period for intervention villages relative to control villages. Subgroup analyses suggested greater reductions in symptomatic seroprevalence for surgical masks compared to cloth masks, and for older participants relative to younger participants. Among participants aged 60 years and older, the adjusted prevalence ratio was 0.65 (0.46 to 0.85), implying a 35% reduction in symptomatic seroprevalence in intervention villages.
Mask-wearing behavior: The estimated prevalence of mask-wearing behavior was 13% in the control group and 42% in the intervention villages, implying that the intervention increased mask-wearing behavior by 29 percentage points. Models controlling for baseline characteristics of villages (baseline mask-wearing, baseline SARS-CoV-2 symptom prevalence) produced the same result. Physical distancing behavior also increased from 24% in the control arms to 29% in the intervention arms, suggesting that increased mask wearing did not cause net risk-compensating behaviors. Mask wearing behavior effects persisted for the full 10-week period of the study; however, by weeks 20-27, mask-wearing in intervention villages declined to 22% while mask-wearing in control villages declined to 14%. None of the cross-randomized interventions (e.g. changes in messaging, monetary incentives, etc.) had any substantial effects on mask-wearing behavior. Evidence from pilot studies indicated that the primary mechanism of the intervention’s effectiveness was promotion of mask-wearing in public spaces.
This study was a very large and well-designed cluster randomized controlled trial of a realistic intervention. Outcome measures included directly observed behavior, rather than self-reported behavior. Symptom data was measured in a very high proportion of participating households. Analyses were pre-specified.
The outcome measure– the proportion of individuals who reported symptoms and who tested positive for antibodies– is a poor proxy for actual SARS-CoV-2 infections during the study period. Moreover, only 40% of those reporting SARS-CoV-2 symptoms consented to providing a blood sample for antibody testing. If there were systematic differences between intervention and control villages in the proportion of antibody-positive individuals among the >60% of symptomatic individuals who were not tested, the results would be biased. As this was a cluster randomized trial, the study could not link individual mask-wearing behavior with incident SARS-CoV-2 infection, and could only measure putative effects on community-level transmission. It is unclear whether the serological assay used to assess the primary outcome was validated in the population under study, or when serological measurements were taken relative to symptom onset. Furthermore, using symptomatic seroprevalence as an outcome means that authors could not distinguish between effects on infections and on symptom severity. Another key limitation is that the persons recording the behavior data could not be blinded to whether the village was a control or intervention arm, leaving open the possibility that data recorders could have been influenced by knowledge of the study arm. People from control villages may have acquired masks or encountered mask promotion in nearby intervention villages, which would dilute the apparent effectiveness of the intervention on behavior. Finally, it is unclear whether or how the intervention would need to be modified to obtain similar results outside the setting of Bangladesh.
This study is the largest and best-designed randomized controlled trial to date of a realistic non-pharmaceutical intervention on SARS-CoV-2 transmission.