Blog

An article published on January 29 in The Lancet Regional Health – Americas presents a critical response to a previous study by USP professors Daniel Tausk and Beny Spira, published in BMC Public Health, which suggested that mask use was associated with an increase in deaths during the Covid-19 pandemic. The authors of the new article argue that Spira and Tausk’s conclusion is incorrect due to significant flaws in the way the data were analyzed.

In addition to identifying biases and problems in data selection, when recalculating the analysis using a different data selection they consider more accurate (with different time intervals), the authors found no association between mask use and increased mortality.

Among the main problems, the authors of the critique, including researchers from USP, point out that the first study ignored the fact that human behavior changes according to the severity of the situation. The earlier study argued that in countries where mask use was higher, there were more deaths; therefore, masks could have caused the deaths. What actually occurred, however, was that people and governments increased mask use after deaths had already begun to rise.

According to Pedro Hallal, a professor at the University of Illinois at Urbana-Champaign (USA) and one of the authors of the critical commentary, the analyzed study fails to properly assess temporality. “This is the technical term we use to determine what came first,” he said. “Whenever the pandemic situation worsened, mask use increased. So, in reality, it was not mask use that was influencing Covid-19 [deaths], which was their mistaken interpretation; rather, it was the severity of Covid-19 that was influencing mask use”.

One reason for the mistaken interpretation stems from the fact that the initial analysis considered aggregated data over two consecutive years, without accounting for pandemic peaks, and from the fact that the strengthening of restrictive measures and adherence to mask use fluctuated during those two years, even within each country. By examining weekly data, the authors of the new analysis observed substantial fluctuations in mask use, as expected, and found that in several countries increases in deaths were followed by increased mask use.

The use of national average data, without considering that a country’s population is large and heterogeneous in its habits, also biased the conclusions, according to Bruno Gualano, a professor at USP’s Medical School (FM) and one of the authors of the critique in The Lancet RH – Americas. “We have no way of knowing whether the people who died were the same ones who wore masks, and whether those who survived were the ones who did not”, he said. This bias is known in science as the “ecological fallacy”.

“We cannot take the population average to infer what happens with the individual”, said Gualano. He cites a real example: in the United States in the early twentieth century, it was observed that states with more immigrants had lower illiteracy rates. This led to a hasty initial conclusion that immigrants were more literate.

When analyzed at the individual level — among immigrants themselves — the opposite was observed: they had lower literacy levels. “What happened is that these immigrants sought out more developed states, which also had higher overall literacy rates. This is the typical ecological bias we see occurring in a similar way in the mask study. The authors look at a general average and estimate individual mortality. Could it be that those who use masks more are also more exposed to risk, such as older adults or people with chronic illness? This is the type of question that ecological analysis cannot answer”, explained the professor. “Ecological studies rank among the lowest levels on the scale of strength of scientific evidence”, added Hallal.

“But even in our reanalysis, using data in a more detailed way, ecological bias still occurs”, acknowledged Felipe Nogueira Barbara, a researcher at the Rio de Janeiro State University (UERJ) and coauthor of the critical commentary — an observation also emphasized by Hallal. “Ecological bias is inherent to this type of country-level analysis”, said Barbara.

For this reason, he stresses that beyond ecological bias, the most important issue is to question whether what the authors of the original article (Tausk and Spira) used as data for “mask use” actually corresponds to mask use.

“This is the point that I think we need to highlight more. Even accepting that we are working within this ecological bias, we should eliminate that problem first, because if the variable they claim measures mask use does not accurately measure mask use, in my view this is a prior issue and even more important than ecological bias”, Barbara added.

Another aspect emphasized in the critique is the lack of plausibility for the claim that masks could be harmful to health. The idea that masks caused deaths makes no clinical or biological sense, the authors say. There is accumulated evidence that masks act as barriers against the virus. At worst, if not used correctly, they may fail to protect, but they never “cause” infections or deaths, said Gualano. He also recalls that during the pandemic, his group designed experiments to test whether masks could harm health or impair exercise performance, and found evidence that they did not (read more details here and here).

Hallal reiterates the central issue of temporality, that is, greater mask use followed increases in deaths, not the other way around.

Gualano believes that the article published in BMC Public Health should be retracted because it presents false conclusions and could negatively influence public policymaking. Hallal, however, believes this would not be necessary, noting that retractions are more commonly used in cases of fraud, data manipulation, or ethical issues, which was not the case here. In his view, what occurred was a misinterpretation by the authors and a demonstration of problems with the journal’s peer-review system, which failed to identify these issues when accepting and publishing the article.