Incidence and risk factors of COVID-19-like symptoms in the French general population during the lockdown period: a multi-cohort study

Following the identification of a novel coronavirus (SARS-CoV-2) in Wuhan, China in December 2019 and its worldwide spread [1], the first imported COVID-19 cases were initially reported in France on January 24, 2020 [2]. Less than 2 months later, the French government declared a nationwide epidemic (phase 3) and a generalized lockdown procedure was set-up on March 17, 2020 [3]. The lockdown included banning of any non-essential public gatherings, closure of educational and public/cultural institutions, ordering people to stay home apart from exercise and essential tasks. Children and their parents were required to stay at home as much as possible [4]. Public health reports have shown that lockdown had a marked impact on the dynamics of the pandemic with a clear downward trend in new hospitalizations from April 1, 2020, and a consecutive decrease in the number of deaths from April 7, 2020 [4, 5]. Thus, the French government eased these restrictions on May 11, 2020 [3]. Although lockdown appeared to successfully alleviate the burden of severe COVID-19 [6], estimates of its impact on mild-to-moderate COVID-19 are based on modelling studies [7], and are not yet supported by clinical evidence.

Our main goals were 1) to estimate the incidence of illnesses presumably caused by SARS-CoV-2 infection during the lockdown period; 2) to identify the associated risk factors. We also described associated symptoms, preventive behaviors and healthcare in relation to these illnesses.

A total of 116,903 of the 279,478 participants (42%) who were invited to participate in the survey completed the questionnaire. The participation rate was 69% in the CONSTANCES cohort, 51% in the E3NE4N cohorts and 26% in the NutriNet-Santé cohort (Fig. 1).

Table 1 presents the characteristics of included participants. Median age was 59 years old (Q1-Q3: 46 to 71 years), and 66% of the participants were women. Twenty-six percent were residents of the Ile-de-France or GrandEst regions – the two regions with the highest rate of SARS-CoV-2 in metropolitan France, while 23% lived in rural areas and 44% lived in cities of more than 100,000 inhabitants. At least one child or adolescent was present at home in 25%. Forty-three percent were retired and 50% were working adults, but only 8% worked outside the home during the lockdown period. Ten percent of the participants were obese and a chronic disease was reported in 34% of participants.

Participants who were living outside mainland France (n = 1588) or with missing information on their exact address location (n = 61) or who reported a positive SARS-CoV-2 test result (n = 342) and/or a CLS onset (n = 8289) before the at-risk period were excluded from the primary analysis. The primary analysis evaluated 106,848 participants: 3035 CLS were reported during 62,099 person-months of follow-up. The unweighted cumulative incidences of CLS were 6.2% (95% Confidence Interval (95%CI): 5.7%; 6.6%), 7.7% (95%CI 7.3%; 8.2%) and 8.8% (95%CI 8.3%; 9.2%) on days 15, 30 and 45 of lockdown, respectively. The weighted cumulative incidences were 7.2% (95%CI 6.7%; 7.8%), 9.0% (95%CI 8.4%; 9.5%) and 10.1% (95%CI 9.6%; 10.6%) on days 15, 30 and 45 of lockdown, respectively.

Extending CLS definition to illness that lasted less than 4 days, 5313 cases were identified during 59,768 person-months of follow-up with unweighted cumulative incidences of 9.7% (95%CI 9.2%; 10.3%), 12.6% (95%CI 12.0%; 13.1%) and 14.3% (95%CI 13.7%;14.9%) on days 15, 30 and 45 of lockdown, respectively. Sensitivity analysis of all cases of CLS onset after March 16, 2020 excluded 171 participants with a positive test result and/or 4084 with the onset of CLS before March 16, 2020 and identified 7240 cases in 110,207 person-months of follow-up with unweighted cumulative incidences of 4.7% (95%CI 4.5%; 4.8%), 6.6% (95%CI 6.5%; 6.8%) and 7.7% (95%CI 7.5%;7.9%) on days 15, 30 and 45 of lockdown, respectively.

The primary daily incidence rate peaked on day four of lockdown (March 19, 2020; unweighted estimate 5.57 per 1000 person-days (95%CI 4.45; 6.89) – Fig. 2) and showed a sharp and constant decrease to reach less than 1 per 1000 person-days after day 25 (April 9, 2020). Similar findings were observed in the weighted incidence rates and the sensitivity analysis considering a different at-risk period (supplementary Figs. 1&2). Daily incidence rates were higher but showed a similar temporal pattern when the case-definition of CLS included illness that lasted less than 4 days (supplementary Fig. 3).

Eighty out of 189 participants who experienced CLS and were tested reported a positive (RT-PCR) test result (supplementary Table 2). Headaches, rhinorrhea and fatigue were frequently reported in addition to the symptoms defining CLS. Eight hundred and forty-eight (28%) participants with CLS had a GP or a hospital visit, and a diagnosis of COVID-19 was considered to be very likely or likely by the physician in 62% cases. Paracetamol was taken by 62% and antibiotics by 6% of participants with CLS. Only 8 participants used chloroquine or hydroxychloroquine. Forty percent participants stayed strictly confined at home following symptoms onset.

Table 2 presents the unweighted incidence rates of CLS and the hazard ratios obtained from the univariable Cox models with stratification on source cohort. Almost all tested factors were found to be associated with CLS. A positive RT-PCR in another household member was strongly associated with CLS in the participant; this variable was not included in the multivariable analysis to avoid overfitting. On multivariable analysis (Table 3), the risk of COVID-19 was lower in older age groups and was higher in the Ile-de-France and GrandEst regions (compared to other French metropolitan regions), in those living in cities > 100,000 inhabitants (vs rural areas), when at least one child or adolescent was living in the same household, in overweight or obese participants, and in people with chronic respiratory diseases, anxiety or depression and chronic diseases other than diabetes, cancer, hypertension or other cardiovascular diseases. The observed associations were confirmed in the sensitivity analyses, except that male gender, living in a household of size 2 and being retired were negatively associated with the risk of CLS in addition to factors identified in the primary analysis (supplementary Tables 3 & 4).

Lockdown was associated with a strong decrease in the incidence of CLS in the French adult population that participated in this survey. This study shows that the cumulative incidence of CLS on day 45 of lockdown ranged from 7.7 to 10.2% depending on the estimation method, that more than 60 % of new cases occurred within the first 2 weeks, and that the daily incidence remained at a sustained low level 1 month after lockdown and thereafter. In addition, we identified several risk factors of CLS during this period, and have described the immediate consequences in terms of access to healthcare and treatment associated with these syndromes. To our knowledge this is the first study to report the signs and symptoms of COVID-19 on a nationwide scale and during lockdown.

Only 28% of the participants with CLS had a medical visit. This result is in line with estimates based on a digital participatory system in France during the same period, in which 31% of COVID-19 patients sought medical advice [13]. Forty percent of the participants with symptoms remained strictly confined without leaving their homes, following the government’s recommendations. This low proportion can be explained in part by the need for some participant to go out for basic necessities, for example if the participant lived alone.

Considering the estimated 5-day median incubation time of COVID-19 and the appearance of symptoms within twelve days after infection [14], a large proportion of participants who developed CLS in the first two weeks were probably infected before lockdown, most of them in the community or at the workplace. It is therefore not surprising to find the association of CLS in adults with decreasing age [15], living in urban versus rural environments [16], in highly prevalent French regions [17], all factors that were reported in other studies performed before lockdown.

A lower infection rate with increasing age was reported in several population-based serological studies [18] which is consistent with our findings, although the risk of severe illness or deaths exponentially increases with age among those infected [19]. As in other studies, univariable analysis identified the size of the household (including children), but only living with at least one child or adolescent remained associated with CLS on multivariable analysis, indicating that this age group could play an important role in household-related transmission [20]. We also identified other factors indicating potential secondary household-related transmission, such as living with another person with a positive diagnosis of SARS-CoV-2 [21]. However, it was impossible to determine a timeframe for this factor and identify whether the participant was the source of infection or was infected by a household member. Being a healthcare worker was associated with CLS in univariable analysis - as reported other studies [22, 23], but the association did not remain significant in the multivariable model, potentially due to a lack of statistical power. Obesity has been found to be linked with the risk of severe CLS in young patients [24], and also suspected to increase the susceptibility to infection [25]. Different theories suggest that asthma, COPD and other respiratory diseases may be negatively or positively associated with the susceptibility to SARS-CoV-2 infection due to up or down regulation of angiotensin-converting enzyme-2 expression. However, all of these respiratory diseases have been shown to be associated with the severity of illness in infected persons [26‐28]. Since 30 to 60% of SARS-CoV-2 infections are asymptomatic [29‐32] and were not included in our CLS cases, by definition, it is not surprising to find the presence of these conditions, which are known to be associated with more severe disease, in subjects with symptomatic SARS-CoV-2 infection. Finally, we found a strong association between CLS and anxiety or depression, which may be related either to a direct (i.e. causal) impact of these comorbidities on the risk of CLS, or to an over-reporting of CLS caused by increased anxiety or stress in this specific subgroup. Although psychiatric disorders have been reported during the acute phase of the infection [33], the risk of reverse causality explaining this association should however be limited as co-morbidities were collected prior to the survey.

Consistent results were obtained in the sensitivity analyses. An association with CLS was found with being retired compared with being working, but the strength of the association was of the same magnitude than estimated in the main analysis. This result can be the consequence of a higher power of the sensitivity analyses due to a higher number of events, and explained by a lower rate of social contacts in this category of persons compared with working people of the same age.

Our study has several limitations. The most important limitation is the lack of virological confirmation of CLS and the risk of misclassification of a SARS-CoV-2 infection and a disease from another etiology. During lockdown, French health authority recommendations limited SARS-CoV-2 testing with a RT-PCR test to patients with severe symptoms requiring hospitalization or to specific situations (e.g. healthcare workers with symptoms). Thus, testing was not available to most participants. Nevertheless, the influenza season peaked on week 6 and ended on week 10 to 12, just before lockdown, which limits the risk of acute respiratory infection caused by an influenza virus. In addition, 42% of the small group of participants who were tested for SARS-CoV-2 infection in our study reported a positive RT-PCR result. This positive rate was higher than that reported in the community (30% at its highest between March 23 and March 29, 2020) [34]. However, a 15 to 20% seroprevalence of SARS-CoV-2 was reported in Spain in individuals from the general population who presented symptoms compatible with COVID-19 [32]. It is therefore likely that the cause of illness was not SARS-CoV-2 infection in a significant proportion of our CLS cases and only studies using sensitive and specific virological methods can accurately quantify the extent of the SARS-CoV-2 epidemic. To avoid recall bias, which is another potential limitation of our study, we limited the questionnaire to the symptoms present in the past 14 days. To avoid a selection bias induced by different dates for filling in the questionnaires resulting in dates of ‘at-risk period’ that varied from one subject to another, we used a Cox model with delayed entry. Finally, although participation bias was accounted for with an appropriate weighting method, our findings should not be considered to be strictly representative of the general adult population in France. Nevertheless, the large number of subjects from all social categories allows us to draw robust conclusions on the factors associated with the occurrence of CLS in France.

In conclusion, to our knowledge this is the first study to quantify the incidence of CLS in the general population on a nationwide scale and during a lockdown, and it shows the significant impact of lockdown on the dynamics of the incidence of infection. A follow-up study is ongoing and will be combined with SARS-CoV-2 serological tests of all participants to estimate the seroprevalence and identify the associated factors.