The USA researchers, in a preliminary study published before validation and presented in NYtimes, demonstrated that the proportion of smokers among patients infected with SARS-CoV-2 (the virus responsible for the Covid-19 disease) was low, lower than that of smokers in the general population.
Other observational studies point in the same direction. As Makoto Miraya indicates, caution is required in drawing definitive conclusions about the “protective” role of tobacco against Covid-19. In our view, this caution must be all the more important since smoking patients seem to progress to more serious forms of the disease than non-smokers. How can we understand and explain the low proportion of smoking patients with Covid-19 and the higher risk of developing a serious form? In other words, how can we explain the role of tobacco, both “immunizing” and “aggravating”?
Despite these important questions, the conclusion is drawn from this difference in proportion that daily smokers have a much lower probability than the rest of the general population of developing an asymptomatic or severe form of SARS-CoV-2 infection. In addition, the hypothesis of a “protective” role for smoking is indicated as strongly suggested by these results.
A weakness of the biomedical sciences is the great malleability of the methods and their vulnerability to various types of bias. It was, however, precisely in the context of a long controversy over the role of tobacco in lung cancer that the methodology of epidemiology was strengthened and sharpened its tools for causal inference. The causal interpretation of a statistical association is indeed an exercise that requires great rigor not to fall into what the epidemiologist Peter Skrabanek denounces under the expression of “risk factorology”, that is to say, this tendency to confuse association and causality. It is therefore important to have a good knowledge of the methodology to avoid the risk of too hasty conclusions on a protective role of tobacco against Covid-19.
Table of Contents
The validity of the association
First of all, before being able to claim a causal inference from the demonstration of a positive statistical association, it is still advisable that this statistical association is very valid, that is to say, that, as l established epidemiologist Mark Elwood, it is necessary that at least the following three criteria are met:
- the association is statistically significant,
- the study carried out to highlight them does not introduce bias
- the association itself is not linked to a confounding factor.
For the study in question, it seems that the first point is respected. The fact remains that this criterion of significance is not without raising a certain number of difficulties, in particular in the use which is too often made of it as a guarantee of a firm conclusion.
Many questions arise about biases. The study protocol suffers from a significant number of biases. First of all, this cross-sectional study only takes into account cases of affected patients. A follow-up of smokers and their infection rate would have made it possible to respond more precisely to the protective hypothesis of tobacco. Ex-smokers were not taken into account (they are however around 60%), knowing that some of them could use e-cigarettes containing nicotine. In addition, the proportion of caregivers (supposed to smoke less as the study indicates) in the group of ambulatory patients is high. Finally, very serious patients hospitalized in intensive care were excluded from the study.
For the third point, the answer to this question remains largely insufficient. The number of risk factors for the severity of the infection currently recognized such as male sex, age over 65, tobacco, hypertension, diabetes, cardiovascular and respiratory diseases makes it difficult to identify a particular factor d ‘as far as they are often linked.
Association or Cause?
In an article published in 1965, Austin Bradford Hill, an English epidemiologist famous for his studies on tobacco and lung cancer, poses the question which will be central and structuring of modern epidemiology: under what conditions is it justified to go from the finding of a valid statistical association between variables with a causal judgment?
Hill offers a list of nine “aspects” of the association that the researcher must examine before “shouting causality”: the strength of the association, its reproducibility, its specificity, its temporality, the biological gradient, its plausibility, its consistency with biological and medical knowledge, the possibility of experimental confirmation, the analogy with other known data.
Let us, therefore, examine the available data concerning the association of tobacco and Covid-19 in the light of these “aspects” of Hill.
In terms of reproducibility: a meta-analysis of 12 studies shows that the prevalence of Covid-19 smoking patients is lower than the prevalence in the corresponding populations. For the 10 Chinese studies, for example, the rate of smokers varies from 3.8 to 14.6% against a reference prevalence rate of 27.7%. However, the rate of smokers increases in the most serious patients (around 26%), approaching the prevalence observed in the general population.
Regarding the biological gradient, no prospective study has been conducted to test the link between the amount of tobacco and the risk or not of contracting the disease.
For consistency: tobacco (nicotine and other molecules) is often associated with cardiovascular and respiratory pathologies and would decrease the action of the immune system, increasing the risk of infections and serious complications. The protective role of nicotine, if it exists, must be weighed against the observed risk of tobacco on the evolution of Covid-19. Coherence is very problematic here.
At the end of this analysis, it seems difficult to seriously base the hypothesis of a protective role of tobacco against Covid-19. Further studies will be required. In other words, association studies here are far from leading to the formation of a causal judgment.
Pluralism of evidence and causal models
For a few years a reflection on causality and causal inference in epidemiology, enriched by the contribution of the philosophy of biomedical sciences, has led to defend the need to adopt a pluralist approach to evidence. Different types of evidence would be necessary and complementary. Some consider it important to obtain evidence that the potential causal factor “makes a difference” in effect, on the one hand, and to identify a mechanism to explain this relationship, on the other. presumed causal.
Each of the aspects of Hill’s list falls under one of these two main types of evidence. In other words, it is both a question of being able to show that the relationship exists, that is to say, that the consumption of tobacco makes a difference (the epidemiological proof here) as to the risk of contracting the disease on the one hand, and to be able to explain how it is carried out, that is to say to account for the causal mechanism, on the other hand. Consequently, it would not be necessary that each aspect of Hill be respected, to conclude that it was causal, but if we were able to identify that our statistical association is valid, that the factor does indeed make a difference in effect, it would then also be necessary to be able to show the existence of a mechanism linking smoking and the protective effect.
Furthermore, and in a complementary way, another approach for causal inference is also used in epidemiology which consists of modeling the relationship between the factor and the effect by taking into account all the other factors and the context. The goal of these models, sometimes very sophisticated being, to integrate different types of evidence to form the best possible causal inference judgment. They allow us to test different causal hypotheses. To illustrate this by returning to our potential protective role of tobacco, we propose to imagine a graphic model where the different elements at our disposal would be tested.
In this model, the protective role of tobacco (dotted arrow between T (N) and D with the negation) would be tested using probabilistic calculations, as well as its aggravating role (rounded arrow between T (N ) and D * with positivity), taking into account the other variables of interest (pathologies linked to tobacco Z (T), unknown variables U).
Information of all kinds around the Covid-19 is accumulating, some continue to be controversial, others appear that open new controversies and still others move us forward to treat patients better. It will undoubtedly need to be armed with great patience and a lot of rigor to save the best hypotheses and reject the worst. Those dedicated to the role of tobacco – whose harmfulness is no longer in question – will still have to prove their interest. Above all, we can undoubtedly speed up research efforts under the pressure of the crisis, but let us not forget that the time of science is a time that can, in reality, be difficult to compress without risking giving up being scientific and leading to results, which then denied, will not fail to arouse skepticism then legitimate on science. This example and our present situation also testify to the difficulties and ambiguities of science in times of crisis.