Skip to main content

Identifying airborne transmission as the dominant route for the spread of COVID-19

Our take —

While masks most likely prevent community spread of COVID-19, this highly flawed paper provides no evidence on mask effectiveness at the population level. The study also provides no information to demonstrate that airborne transmission — let alone “long-range airborne transmission” — is the dominant form of COVID-19 transmission. The claims made in this paper are not supported, and the journal editors should strongly consider retraction.

Study design

Other

Study population and setting

The objective of the study was to assess the impact of masks and other non-pharmaceutical interventions on COVID-19 transmission, as well to determine the predominant routes of viral spread (e.g., airborne versus droplet). The authors used linear regressions fit to case data from New York City and Italy to assess trends in cases occurring before and after implementation of various non-pharmaceutical interventions, particularly mask mandates in northern Italy on April 6, 2020 and in New York City on April 17, 2020. Qualitative comparisons were made to trends in Wuhan, China. Inference about the effectiveness of mask-wearing, and subsequently about the dominant route of SARS-CoV-2 transmission, was made on the basis of departures from linearity after the date of mask mandates.

Summary of Main Findings

This paper had two primary conclusions, neither of which were supported by the evidence presented. First, the authors concluded that mask mandates were the only factor that led to departures from the linear trend of case counts in Italy and New York City, and by implication, that no other non-pharmaceutical interventions (e.g., social distancing policies) were effective. Second, the authors concluded that airborne transmission is the major driver of COVID-19 spread. There were no measures of uncertainty reported.

Study Strengths

None

Limitations

The authors made two false statements, which were key assumptions underpinning their primary conclusions. The first statement, which was used to justify the conclusion that masks were the sole effective non-pharmaceutical intervention leading to reduction in the growth of new cases, was that, “after April 3, the only difference in regulatory measures between NYC and the United States lies in face coverings on April 17 in NYC”. This assertion is verifiably false based on publicly available resources (e.g., HIT-COVID) and represents a gross mischaracterization of the primary exposure variable (non-pharmaceutical interventions). Second, in justifying the conclusion that airborne transmission is the dominant mode of COVID-19 spread, they state that, “with social distancing, quarantine, and isolation in place worldwide and in the United States since the beginning of April, airborne transmission represents the only viable route for spreading the disease.” This statement is also verifiably false; many countries and jurisdictions were not in lockdown and did not have isolation and quarantine programs in place at the beginning of April (e.g., Sweden, parts of the US).

There were also serious methodological flaws, which may be broadly categorized as follows:

Exposure misclassification: the construct of interest is mask-wearing, but dates of mask mandates are poor proxies for behavior. No attempt was made to measure or describe actual mask-wearing behavior during the time periods and regions considered.

Confounding: the authors considered no other factors determining COVID-19 case counts that may have differed between time periods and regions. The time period under consideration in Italy and the United States saw sweeping, complex changes in a broad array of non-pharmaceutical interventions, personal behaviors, testing availability, and epidemic dynamics. None of these are included in the models. Similarly, the authors’ qualitative comparisons with Wuhan, China ignore substantive differences not only in the underlying populations, but in the intensity of non-pharmaceutical interventions such as case detection, contact tracing, and isolation.

Model misspecification: Linear models were fit to cumulative and daily COVID-19 case-count data, which do not follow linear growth patterns, except in specific circumstances for limited times. Infectious disease dynamics were entirely ignored by this simplistic approach. The analysis also did not account for the necessary lag between reduction in transmission (via mask-wearing, for example) and reduction in reported cases.

Measures of uncertainty: The authors reported no measures of uncertainty. The results include no p-values, confidence intervals, or any formal inference of any sort. This represents a departure from scientific orthodoxy and a gross oversight, particularly considering the strong conclusions the authors assert from their results.

Value added

None

This review was posted on: 17 June 2020