Two New CDC Studies on Masking in School

Do they prove causality?

Two studies in MMWR claim to provide evidence that policies mandating masking children in school lowers rates of sars-cov-2 spread among children due to school.  Both suffer from limitations that preclude causal claims.  I will review both here.

CDC’s Maricopa and Pima County Study

https://www.cdc.gov/mmwr/volumes/70/wr/mm7039e1.htm

The US CDC has recommended masking for kids starting at age 2, while the World Health Organization has advised this only selectively for kids <12.  Peer nations, such as the UK did not mask kids <12, during the pandemic.  To date, no cluster RCTs have been conducted in children.

A new MMWR paper claims that schools with no mask requirement had were 3.5 times more likely to experience an outbreak than schools that mandated masks from day 1.  Here mask mandates are defined as a requirement for all teachers, staff and students. (all must be include).  

The paper has been used to claim a causal relationship—mask mandates led to this difference. Several reasons to doubt that interpretation.

First, schools with mask mandates were fundamentally different. It should not be surprising given political differences in mask acceptance that districts with masking are different than those without in ways apart from masking policies.  Specifically, schools with mandatory masking policies were more likely to be in Pima county; Maricopa county mostly did not have mask policies.  As illustrative of political differences, Pima county voted more strongly in favor of Joe Biden than Maricopa county.

Schools with masking requirements included younger kids (greater % elementary, fewer middle/ high), and had fewer students enrolled—also suggestive we may be comparing schools for younger kids vs. those for older kids. 

The endpoint of the study were not cases attributable to school spread, but rather the number of schools with 2 or more cases.  This weighs equally a school with 4 cases vs. 400 cases.

Second, there are many important variables the paper did not adjust for. Paper does not report the rate of testing/ including asymptomatic testing; It does not examine the rate of teacher’s being vaccinated. Does not report the rate of students >12 being vaccinated.  Does not report the type of masks being used, and compliance.

Strangely, the paper does not report the raw numbers of kids, teachers, and staff with covid19 infection thought attributable to school spread.

Third, the reported OR of 3.5 in adjusted analysis seems large, given that the cluster RCT of Bangladesh showed no effect for cloth masks in primary analysis, and a modest effect size for surgical masks.  Likely reflects residual confounding.

In short, the paper is ill-suited for causal claims.  Schools with mask requirements likely also vary in other important ways from those without them. Cluster RCTs needed.

The paired paper is

Pediatric cases in counties with or without mask requirements

https://www.cdc.gov/mmwr/volumes/70/wr/mm7039e3.htm?s_cid=mm7039e3_w#contribAff

This analysis compares US counties where all schools had mask requirements vs. all schools do not require masks. Notably this is only 16.5% of all counties.  I.e. most places (83.5%) in America excluded due to variable policies.

The Figure shows counties with consistent mask requirements were different at baseline from those with consistent no mask requirements. They consistently had a lower county level mean change in pediatric covid19 cases.

After school began, authors note counties with consistent no mask requirements “experienced larger increases in pediatric COVID-19 case rates after the start of school compared with counties that had school mask requirements (p<0.001)”.

Study did not control for pediatric or adult vaccination rates in the county. 

Study included covid cases in children who did not attend school, or were too young for school. 

Study does not report raw number of cases, but percent change in cases.

Overall, this is a very crude, high level analysis with severe missing data, and countless unmeasured confounders.  I conclude it is ill-suited for causal claims. 

Ultimately, cluster randomization needed, particularly in the USA where mask policies are a surrogate for political valence and other differences. Without these analyses, to date, this observational analysis from Spain is most compelling:

Rob Hughes @R_Hughes1

Dear #epitwitter and kids #MaskMandates Twitter, Please look at this 👇graph of covid transmission risk amongst 15,000 kids within bubbles, and tell me: *What age group of children were mandated to wear masks at school/pre-school😷?*

The answer is age 6, but the fact you cannot tell visually is the point.