Annals of Internal Medicine today published an RCT comparing the effectiveness of N95s vs. medical masks to protect healthcare staff from COVID-19. It’s a great piece of work, conducted over many years, and whilst the study has some important limitations, suggests that N95s don’t offer significantly increased protection than medical masks for healthcare staff caring for patients with confirmed or suspected COVID-19.
The multi-centre and international study included 29 healthcare facilities from four countries, and randomised 1009 staff who provided direct care for patients with confirmed or suspected COVID-19 in a variety of healthcare settings to either medical masks or N95s. This included emergency departments, general wards (adult and paediatric), long-term care facilities, but not ICUs. The study was conducted between May 2020 and March 2022, so much changed during this period. This made it difficult to have a consistent set of exclusion criteria for staff, as testing and vaccination patterns changed. But the key exclusions for staff were: not having had a fit test for an N95, recent COVID-19 infection, vaccination, or clinical risk factors for COVID-19 (which may have made them more susceptible to SARS-CoV-2 infection). The study hospitals all had universal medical masking in place, and a policy to care for patients with suspected or confirmed COVID-19 using medical masks as baseline (and so staff randomised to N95 were going above and beyond what was recommended in that facility). All of the study hospitals required N95s for “AGPs”, and recommended additional PPE when caring for patients with confirmed or suspected COVID-19 including eye protection, gowns, and gloves. Active surveillance for symptoms of COVID-19 for the staff participating in the study was done by regular text messaging, prompting PCR testing in the event of symptoms developing.
The primary outcome in the intention-to-treat analysis, PCR–confirmed COVID-19, occurred in 52 of 497 (10.46%) in the medial mask group versus 47 of 507 (9.27%) in the N95 group (hazard ratio, 1.14 [95% CI: 0.77 to 1.69]). So, on the face of it, being randomised to wearing an N95 instead of a medical mask did not significantly reduce the risk of healthcare staff exposure to COVID-19. Although, since this was a non-inferiority trial, we can only really conclude that medical masks were not inferior to N95s given the parameters set for non-inferiority (which in this case was a doubling in hazard of staff becoming symptomatic with COVID-19). Or put another way, medical masks did not double the risk of COVID-19 compared with N95s! If you breakdown the data by country, there was a trend towards a significant reduction in the hazard ratio in Canada.
The study collected data to show that there was no significant differences in terms of minutes of exposure to patients with confirmed or suspected COVID-19, “AGPs”, and community exposures between the N95 and medical mask group. There was an attempt to measure adherence to the intervention by asking staff whether they actually used their assigned mask “always”, “sometimes”, or “never”. In addition, observational audits of mask use were undertaken periodically during the study. Both self-reported adherence to the intervention, and adherence measured by audit appeared to be similar in both groups, with self-reported adherence a bit lower in the N95 group.
This is a great study and a good read. But in some ways, frustratingly inconclusive. The study was done in such different settings (Canada vs. Pakistan vs. Egypt vs. Israel) that they are almost non-comparable. The set-up of healthcare systems and hence background COVID-19 risk is so very variable that you begin to question the logic of including these extra centres over and above the initial recruitment of study participants in Canada. Other limitations include that only symptomatic testing of staff was recommended, so there could be a whole different story about asymptomatic staff infection, which we won’t ever get to read (because the testing wasn’t done). During the course of the study, different variants came and went, each of which had substantially different transmission dynamics. There were also changes in lab testing methods and availability, changes in patient testing protocols and management, and constant rising and falling of prevalence. Also, and importantly, ventilation in the various settings that the study was undertaken was not taken into account.
This is a useful study and congrats to the team behind it for the huge amount of energy that went into this. The bottom line finding of this study is that N95 seem to offer little protection to staff over and above medical masks, which is consistent with previous studies around the risk of influenza acquisition. However, the study is far from definitive and, as ever, further studies are required!
Reflections on Infection Prevention and Control 
This is another excellent post on an excellent blog. I just wanted to take issue with the following remark: “Although, since this was a non-inferiority trial, we can only really conclude that medical masks were not inferior to N95s given the parameters set for non-inferiority (which in this case was a doubling in hazard of staff becoming symptomatic with COVID-19). Or put another way, medical masks did not double the risk of COVID-19 compared with N95s!”
Andrew Copas and I published an article on the false dichotomy between superiority and non-inferiority trials. (https://trialsjournal.biomedcentral.com/articles/10.1186/s13063-018-2885-z). Within this article we point out that the pre-specified non-inferiority margin is usually entirely arbitrary (with no real justification), and argued that inference should be based primarily on the observed point estimates and confidence intervals rather whether the non-inferiority criterion was met (i.e. a significance test).
Returning to trial in question, the primary outcome in the intention-to-treat analysis, PCR–confirmed COVID-19, occurred in 52 of 497 (10.46%) in the medial mask group versus 47 of 507 (9.27%) in the N95 group (hazard ratio, 1.14 [95% CI: 0.77 to 1.69]). Focussing on the upper confidence limit, we can conclude that if medical masks did increase risk then this increase was less than 1.69-fold. OK, not a million miles from doubling, but why invoke the arbitrary 2-fold non-inferiority margin at all?
David Dunn
Professor of Medical Statistics, UCL
LikeLike
Thanks for taking the time to clarify this Prof Dunn
LikeLike
I’ve been interested to read a few critiques of this study, including changes to the statistical analysis after the start of the study (with 3 unexplained), HCWs could have been infected outside of the hospital, differences in vaccination coverage in different countries. and the study also showing a general trend to N95s being superior to surgical masks at all sites except Egypt, where 50% of participants reported known exposure in their household/community. Masking was also only required when within 3 feet of a symptomatically infected patient and not at all if asymptomatic or pauci-symptomatic.
Also interesting to note a couple of conflicts of interest: 8 of the authors signed a witness statement that was apparently used to prevent nurses from obtaining N95 masks and one of the authors made a public statement against N95 masking because of harms, including acne.
Also worth mentioning that the Ontario Nurses’ Association told its members not to take part as they were concerned that it was unethical to place HCWs at risk for the sake of research.
Only just started looking into it, but I thought the above points were interesting.
LikeLike
The comments from Ros are untrue. All changes to the analyses were fully explained. Differences in vaccination coverage make no difference, in fact the attack rates were higher in Egypt and Pakistan. The policy was univeral masking which participants were expected to abide by. There were no actual conflicts of interest by the authors. There was absolutely nothing unethical about this trial. As investigators, it was very concerning to us that an organization would ask its members not to participate in an important study. [This comment has been edited from the original version at the request of the submitter, 05/01/23, Ed]
LikeLike