Effectiveness of N95 respirators versus surgical masks against influenza: A systematic review and meta-analysis

2.1 Inclusion and exclusion criteria

Inclusion criteria were (1) study type: RCT (including cluster-randomized trial) and nonrandomized controlled study; (2) participants: humans with influenza (including pandemic strains, seasonal influenza A or B viruses and zoonotic viruses such as swine or avian influenza), and other respiratory viral infections (as a proxy for influenza); (3) intervention and comparator: N95 respirators versus surgical masks; (4) primary outcome: laboratory-confirmed influenza; (5) secondary outcomes: laboratory-confirmed respiratory viral infections, laboratory-confirmed bacterial colonization, laboratory-confirmed respiratory infection, and influenzalike illness; and (6) settings: hospital or community. RCTs were selected due to the potential possibility of high evidence level. Exclusion criteria were (1) theoretical models; (2) human ⁄nonhuman experimental laboratory studies; and (3) conference abstract.

2.2 Search strategy

We searched PubMed, EMBASE, and The Cochrane Library databases from inception to January 27, 2020, to identify published systematic reviews on evaluating the use of masks for preventing influenza. Search strategy in PubMed could be found in Table 1, and the strategy was adequately adjusted to use in other databases. Then, primary RCTs included in the systematic reviews were identified. Additionally, we conducted an additional search to identify RCTs published in the past five years from January 27, 2015, to January 27, 2020, using the databases and search strategies described above. We also searched for ClinicalTrials.gov to obtain unpublished data. There were no publication status and language restrictions on selecting the studies.

2.3 Study selection and data extraction

Two reviewers independently screened the articles based on the titles, abstracts and full texts. Then, two reviewers independently exacted the following data from included studies: first author, publication year, country, disease, details of study population and intervention, study design, sample size, settings, and results. All disagreements were resolved by discussion.

2.4 Risk of bias assessment

Two reviewers independently assessed the risk of bias of the selected RCTs using the Cochrane Risk of Bias tool,¹⁶ which includes domains on random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, and selective reporting. For each RCT, every domain was judged among 3 levels: high risk, unclear risk, and low risk. Disagreements were resolved by discussion.

2.5 Data analysis

All statistical analyses were performed using Review Manager (RevMan) version 5.3. Comparable data from studies with similar interventions and outcomes were pooled using forest plots. Relative risk (RR) with 95% confidence intervals (CIs) for dichotomous data was used as the effect measure. Between-study heterogeneity was assessed using the I² for each pooled estimate.¹⁷ We adopted a random-effects model for heterogeneity P < .10. We performed a subgroup analysis based on the settings (hospital, community) due to the possibility of clinical heterogeneity. A sensitivity analysis was conducted to evaluate the robustness of the results by excluding individual studies for each forest plot. Funnel plots were planned to assessed publication bias. Because of the small number of studies available for each pooled estimate, we failed to assess publication bias.

3.1 Search results and study characteristics

The details on the literature search and screening process can be found in Figure 1. Excluded studies and reasons for exclusion were shown in Table 2. In total, we included six RCTs^{12, 18-22} and found no unpublished data of RCTs from ClinicalTrials.gov. The characteristics of these RCTs were presented in Table 3. The included studies published between 2009 and 2019. A total of 9171 participants in Canada, Australia, China, or America were included, and the number of participants in each RCT ranged from 435 to 5180 patients. The follow-up duration varied from 2 to 15 weeks. Five studies included participants in hospitals,^{12, 18, 20-22} and one in households.¹⁹ Because of different definitions of outcome in included studies, we redefined the laboratory-confirmed respiratory infection as respiratory influenza, other viruses or bacteria infection.

3.2 Risk of bias

The results of the risk of bias assessment can be found in Figure 2. Five studies reported the computer-generated random sequences, while only one mentioned randomization. All studies did not mention allocation concealment. Participants and trial staff were not blinded in two studies, and the other two studies failed to mention the blinding of participants and personnel. Four studies did not report whether the outcome assessors were blinded. All studies had complete outcome data or described comparable numbers and reasons for withdrawal across groups and prespecified outcomes.

3.3 Effectiveness

Five RCTs involving 8444 participants reported laboratory-confirmed influenza.^{12, 18-21} Meta-analysis with fixed-effects model revealed that there was no statistically significant differences in preventing influenza using N95 respirators and surgical masks (RR = 1.09, 95% CI 0.92-1.28, P > .05) (Figure 3). The results of subgroup analyses were consistent with this regardless of the hospital or the community. The results of the sensitivity analysis were not altered after excluding each trial.

Four RCTs^18-21 involving 3264 participants reported laboratory-confirmed respiratory viral infections. Meta-analysis with fixed-effects model revealed that there were no statistically significant differences in preventing respiratory viral infections using N95 respirators and surgical masks (RR = 0.89, 95% CI 0.70-1.11, P > .05) (Figure 4). The results of subgroup analyses were consistent regardless of the hospital or the community. However, the sensitivity analysis after excluding the trial by Loeb et al¹⁸ showed a significant effect of N95 respirators on preventing respiratory viral infections (RR = 0.61, 95% CI 0.39-0.98, P < .05).

Two RCTs^{21, 22} involving 2538 participants reported laboratory-confirmed bacterial colonization. Meta-analysis with fixed-effects model revealed that compared with surgical masks, N95 respirators significantly reduced bacterial colonization in hospitals (RR = 0.58, 95% CI 0.43-0.78, P < .05) (Figure 5). The sensitivity analysis showed that the results did not change after excluding each trial.

Two RCTs^{12, 22} involving 6621 participants reported laboratory-confirmed respiratory infection. Meta-analysis with random-effects model revealed that there were no statistically significant differences in preventing respiratory infection using N95 respirators and surgical masks in hospitals (RR = 0.74, 95% CI 0.42-1.29, P > .05) (Figure 6). However, the sensitivity analysis after excluding the trial by Radonovich et al¹² showed a significant effect of N95 respirators on preventing respiratory infection (RR = 0.53, 95% CI 0.35-0.82, P < .05).

Five RCTs involving 8444 participants reported influenza like illness.^{12, 18-21} Meta-analysis with random-effects model revealed that there were no statistically significant differences in preventing influenza like illness using N95 respirators and surgical masks (RR = 0.61, 95% CI 0.33-1.14, P > .05) (Figure 7). The results of subgroup analyses indicated that statistically significant superiority of N95 respirators over surgical masks against influenza like illness (RR = 0.37, 95% CI 0.20-0.71, P < .05) in the community (only one RCT). The sensitivity analysis showed results remained unchanged after excluding each trial.