COVID-19 disease severity correlates with smoking Status

The novel COVID-19 disease is a contagious acute respiratory infectious disease whose causative agent has been demonstrated to be a new virus of the coronavirus family, SARS-CoV-2. Multiple studies have already reported that risk factors for severe disease include older age and the presence of at least one of several underlying health conditions. However, a recent physiopathological report and the French COVID-19 scientific council have postulated a protective effect of tobacco smoking. We have been able to demonstrate the statistical significance in this regard of recent series from both China and in the US, reporting smoking status as well as disease severity (p-values of 2.27*10 and 11.7*10, respectively). Subsequently and using a Bayesian approach we have established that disease severity is positively associated with smoking status. Finally, we refute claims linking general population smoking status (N=O(10) or O(10)) to much smaller disease course series (N=O(10)). The latter point in particular is presented to stimulate academic discussion, and must be further investigated by welldesigned studies.


INTRODUCTION
The novel coronavirus pneumonia (COVID-19) is a contagious acute respiratory infectious disease whose causative agent has been demonstrated to be a new virus of the coronavirus family, SARS-CoV-2. This illness was first evinced in December 2019 in the Seafood Market of Wuhan, Hubei Province, in southern China 1,2 . Patients with the coronavirus pneumonia have a fever, and the temperature above 38 degrees Celsius with symptoms such as dry cough, fatigue, dyspnea, difficulty breathing, and diarrhea [1][2][3][4][5] . There is a high transmission for this disease, which, as result of this and other factors such as international travel and trade, has now turned into a pandemic, with hundred thousands of individuals confirmed to be infected worldwideand most likely millions of unreported cases 5 .
Recent preliminary estimates of the prevalence of selected underlying health conditions among patients with COVID-19, in China as well as in the United States, show that individuals with underlying health conditions appear to be at higher risk for more severe COVID-19 3,6 . A recent Greek epidemiological review suggests that smoking is most likely associated with the negative progression and adverse outcomes of COVID-19 7 . Meanwhile, a recent physio-pathology study has formulated the hypothesis that the effectiveness of ACE inhibitors/angiotensin II receptor blockers and nicotine in patients infected with COVID-19 8 .
The smoking prevalence in China and in the U.S. have been previously and broadly documented: these are, respectively, of 26.6%, and 19.7% 9,10 . Furthermore, the president of the COVID-19 Scientific Council of France recently postulated that nicotine may have a protective role against COVID-19 infection 11 . Thanks to these multiple studies, we can observe that the prevalence of past/current smokers is higher in COVID-19 severe infected patients. In this context, we would like to know whether the past/current smoking status entails higher probability of a severe course of COVID-19 illness, as compared to having never smoked.

Study and Participants
We conducted a descriptive observational cross-sectional study in order to define a hypothetical relationship between smoker status and SARS-CoV-2 severity infection. The source and targeted populations are the whole humanity in view of the ongoing COVID-19 pandemic. The eligible population is constituted by all Chinese and American already included in previous papers.
The study was conducted by a consortium of two data analysts, a MD specialized in radiology, and a medical student in clinical years. NexGen Analytics had no role in designing the study, nor in making the decision to submit manuscript to the publication, nor did it receive any fee or compensation in the context of this work. The first author vouches for the data and analyses, as well as for the fidelity of this report to the study protocol.

Outcome Measures
Our null hypothesis (H0) was the orthogonality between two variables: the smoking status (X) and the COVID-19 Symptoms Severity (Y ). Each variable can take two values: for the smoking status (X), it was a never smoker (X1) or a current/past smoker (X2); for the COVID-19 Symptoms Severity (Y ), it was a mild (non-hospitalised) (Y1) or a severe (hospitalised, ICU or not) (Y2) symptoms severity.

Statistical Analysis
Each sample was analysed separately. First, we had to calculate for each sample different conditional probabilities: the probability to develop severe symptoms for non-smokers (i.e., P (Y2|X1)) and the probability to develop severe symptoms for former and current smokers combined (i.e., P (Y2|X2)).
We then calculated, for each observed sample, the expected cardinalities assuming H0 expected sample: in each case, we multiplied the two corresponding sums and subsequently divided by the actual total. In this way, we were able to apply a χ 2 test to compare the observed table to the expected table, for each sample. The p-value was subsequently computed in order to evince a statistically significant difference between the observed and the expected tables.

Populations
After a short treatment of the initial data from both Chinese and U.S. series (Table 1), in particular to eliminate incomplete data, we have obtained the observed data for each sample with their respective conditional probabilities (Tables 2, 3). For the Chinese sample, the probability to develop severe COVID-19 symptoms for non-smokers is of 14.5% and the probability to develop severe COVID-19 symptoms for former/current smokers is of 24.1% For the American sample, the probability to develop severe COVID-19 symptoms for non-smokers is of 21.7% and the probability to develop severe COVID-19 symptoms for former/current smokers is of 42.7%.

Outcomes
We estimated the expected data on each case by multiplying the two corresponding sums and then by dividing by the total effective, for each sample (Tables 4,5). In these conditions, we were able to estimate the χ 2 parameter, for Chinese and American samples, and the p-values where the difference between observed and expected data are significant ( Table 6). Indeed, the difference observed between them is significant for China and US with respectively a p-value of 2.27*10 -3 and 11.7*10 -15 .

DISCUSSION
Subsequently, by computing the empirical probability distributions conditioned on smoking status, we have established that disease severity is positively associated with smoking status, in those samples.
Nevertheless, our study has some limitations: 1. The difference between smoking status as observed across the entire Chinese and American populations 9,10 , and the same observed in our small clinical datasets ( Table 1) is significant. As a result, the previously claimed relationship between former/current smoking status and a COVID-19 disease severity may not be confidently estimated from a global population.
2. Only adults were considered to compute smoking prevalence 10 .
However, regarding the first limitation above, we contend that claims linking general population smoking status, computed across populations with cardinalities N in the order of O(10 8 ) or even O (10 9 ) with broadly varying smoking habits (e.g. already rural vs urban divide in this regard), to much smaller disease course series (N = O(10 3 )), are at best imprudent and at worst marred by selection bias, especially as the latter regard a fraction of patients from the same urban area. We therefore claim that specific smoking status data, relevant to the concerned geographic areas, is required to evince a statistically significant under-representation of smokers among COVID-19 patients. An alternative argument could be made from a physio-pathological standpoint as well: indeed, previous studies and assertions describe a hypothetical protector role of nicotine against COVID-19 infection 8 , but these are largely speculative at this point and require laboratory confirmation. We therefore conclude that the only significant findings that can be confidently asserted from the series discussed in this article, is that COVID-19 disease severity is positively associated with smoking status.