ABSTRACT
A novel testing policy was implemented in May in France to systematically screen potential COVID-19 infections and suppress local outbreaks while lifting lockdown restrictions. 20,736 virologically-confirmed cases were reported in mainland France from May 13, 2020 (week 20, end of lockdown) to June 28 (week 26). Accounting for missing data and the delay from symptom onset to confirmation test, this corresponds to 7,258 [95% CI 7,160-7,336] cases with symptom onset during this period, a likely underestimation of the real number. Using age-stratified transmission models parameterized to behavioral data and calibrated to regional hospital admissions, we estimated that 69,115 [58,072-77,449] COVID-19 symptomatic cases occurred, suggesting that 9 out of 10 cases with symptoms were not ascertained. Median detection rate increased from 7% [6-9]% to 31% [28-35]% over time, with regional estimates varying from 11% (Grand Est) to 78% (Normandy) by the end of June. Healthcare-seeking behavior in COVID-19 suspect cases remained low (31%) throughout the period. Model projections for the incidence of symptomatic cases (4.5 [3.9-5.0] per 100,000) were compatible with estimates integrating participatory and virological surveillance data, assuming all suspect cases consulted. Encouraging healthcare-seeking behavior and awareness in suspect cases is critical to improve detection. Substantially more aggressive and efficient testing with easier access is required to act as a pandemic-fighting tool. These elements should be considered in light of the currently observed resurgence of cases in France and other European countries.
Introduction
As countries in Western Europe gradually relaxed lockdown restrictions, robust surveillance and detection systems became critical to monitor the epidemic situation and maintain activity at low levels1. The need is to rapidly identify and isolate cases to prevent onward transmission in the community and avoid substantial resurgence of cases. In France, the surveillance strategy implemented by authorities to exit lockdown on May 11, 2020 was multifold2,3 and based on an expanded case definition for COVID-19 suspect cases to guide clinical diagnosis4; recommendations to the general population to seek healthcare even in presence of mild symptoms; prescription of diagnostic tests to suspect cases by general practitioners for systematic and comprehensive testing; isolation of confirmed cases and tracing of their contacts.
The specific characteristics of COVID-19 epidemic, however, hinder the identification of cases5. Large proportions of asymptomatic infectious individuals6, and presence of mild or paucisymptomatic infections that easily go unobserved7,8 present serious challenges to detection and control8-10. This may potentially result in substantial underestimates of the real number of COVID-19 cases in the country. Here we estimated the rate of detection of COVID-19 symptomatic cases in France in May-June 2020 after lockdown, through the use of virological and participatory syndromic surveillance data coupled with mathematical transmission models calibrated to regional hospitalizations. The study focused on mainland France where the epidemic situation was comparable across regions, and excluded Corsica reporting a very limited epidemic activity and overseas territories characterized by increasing transmission11.
COVID-19 surveillance
COVID-19 epidemic management in France in the post-lockdown phase involved the creation of a centralized database collecting data on virological testing (SI-DEP, Information system for testing) to provide indicators to monitor the epidemic over time2,12. 20,736 virologically-confirmed cases were reported from May 13 (week 20) to June 28 (week 26) in mainland France. After imputation of missing data (see Methods), an estimated 9,326 [95%CI 9,234-9,403] cases with symptoms resulted in the study period (Figure 1). The average delay from symptom onset to testing decreased from 20.7 days in week 20 (w20) to 7.1 days in w26. Accounting for this delay (see Methods), we estimated that 7,258 [95% CI 7,160-7,336] confirmed symptomatic cases had onset in the study period, showing a decreasing trend over time (1,663 in w20, 892 in w26). The test positivity rate decreased in the first weeks and stabilized around 1.2% (average over w24-w26).
(a) Number of virologically-confirmed symptomatic cases in mainland France by week of testing and week of onset, along with the test positivity rate. Week of symptom onset for symptomatic confirmed cases was estimated based on patients’ declarations (see panel b) through a Gamma distribution fitted to the data with a maximum likelihood approach. Missing data about presence/absence of symptoms and declaration of onset were imputed by region and by week, by sampling from a multinomial distribution according to the observed breakdown among cases with complete information (see Methods). Test positivity rate was computed on cases with complete information. Data for weeks 20-26 were consolidated in w27. (b) Breakdown of virologically-confirmed symptomatic cases by week of testing according to declared onset of symptoms, along with estimated mean time from onset to testing. (c) Incidence of COVID-19 suspect cases (estimates by week and 3-week moving average (thick line)), along with percentage of those seeking healthcare, estimated from participatory surveillance system COVIDnet.fr. (d) Number of COVID-19 suspect cases of the participatory cohort seeking healthcare, and among them those receiving a prescription, and performing a virological test given the prescription. COVIDnet.fr estimates were adjusted on age and sex of participants. (e) Estimated change in individuals attending their workplace locations over time and by region based on Google location history data15. Region acronyms are listed in Table 1. (f) Percentage of individuals avoiding physical contacts estimated from a large-scale survey conducted by Sante publique France16.
A digital participatory system was additionally considered for COVID-19 syndromic surveillance in the general population11, including those who do not consult a doctor. Called COVIDnet.fr, it was adapted from the platform GrippeNet.fr (dedicated to influenza-like-illness surveillance since 201113,14) to respond to the COVID-19 health crisis in early 2020. It is based on a set of volunteers who weekly self-declare their symptoms, along with socio-demographic information. Based on symptoms declared by approximately 7,500 participants each week, the estimated incidence of COVID-19 suspect cases decreased from about 1% to 0.8% over time (Figure 1), according to the expanded suspect case definition recommended by the High Council of Public Health for testing4 (Methods). 162 out of 524 suspect cases (31%) consulted a doctor in the study period. Among them, 89 (55%) received a prescription for a test, resulting in screening for 50 individuals (56% of those given the prescription).
Projections of COVID-19 epidemic trajectories and estimated detection rates
We used stochastic discrete age-stratified epidemic models17,18 based on demography, age profile19, and social contact data20 of the 12 regions of mainland France, to account for age-specific contact activity and role in COVID-19 transmission. Disease progression is specific to COVID-1917,18 and parameterized with current knowledge to include presymptomatic transmission21, asymptomatic6 and symptomatic infections with different degrees of severity (paucisymptomatic, with mild symptoms, with severe symptoms requiring hospitalization)8,22. The model was shown to capture the transmission dynamics of the epidemic in Île-de-France and was used to assess the impact of lockdown and exit strategies17,18. Full details are reported in the Methods section.
Intervention measures were modeled as modifications of the contact matrices, accounting for a reduction of the number of contacts engaged in specific settings, and were informed from empirical data. Lockdown data came from Refs.17,18. The exit phase was modeled considering region-specific attendance at school based on Ministry of Education’s data23, partial maintenance of telework according to estimated presence in workplaces from mobile phones location history data15 (Figure 1), reduction in adoption of physical distancing over time based on survey data16 (Figure 1), partial reopening of activities, senior protection17. A sensitivity analysis was performed on the reopening of activities and senior protection, as data were missing for an accurate parameterization. Testing and isolation of detected cases were implemented by considering a 90% reduction of contacts for the number of virologically-confirmed COVID-19 cases17,18. Region-specific models were calibrated to regional hospital admission data (Figure 2) through a maximum likelihood approach in the phase before lockdown, during lockdown, and in the exit phase. Further details are reported in the Methods section.
(a-c) Hospital admissions over time, data (points) and simulations (median and 95% probability range), for Île-de-France (a), Pays de la Loire (b), Normandie (c). Hospital admission data up to w27 (consolidated in w28) were used to calibrate the models. (d-f) Projected number of new symptomatic cases over time (median and 95% probability range) and estimated number of virologically-confirmed symptomatic cases by week of onset (points), for the same regions above. The estimated detection probability of symptomatic cases (%) is also shown (red points, median and 95% probability ranges, right y axis).
Projected number of cases decreased over time in all regions, in agreement with the decreasing tendency reported in hospital admissions in the study period (Figure 2). Overall, 69,115 [58,072-77,449, 95% probability range] new infections were predicted in mainland France in weeks 20-26 (from 22,882 [18,942-26,221] in w20 to 2,922 [2,530-3,248] in w26). Île-de-France was the region with the largest predicted number of cases (8,126 [4,848-10,305] to 944 [712-1,088] from w20 to w26), followed by Grand Est and Hauts-de-France (Table 1).
Number of virologically-confirmed symptomatic cases, number of projected symptomatic cases, estimated detection rate, estimated trend in detection rate, population per region. Regions are ranked by decreasing number of confirmed cases in w20. The trend is estimated comparing the average of the estimated detection rate in the weeks of June (w23-26) with the average in the weeks of May (w20-w22).
Projections were substantially higher than virologically-confirmed cases (Figures 2 and 3). The estimated detection rate for symptomatic infections in mainland France in the period w20-w26 was 11% [9-13%], suggesting that 9 out of 10 new cases with symptoms were not identified by the surveillance system. Estimated detection rate increased over time (7% [6-9]% in w20, 31% [28-35]% in w26). By the end of June, 9 regions had a median detection above or equal 25% (Figure 3), and 2 regions detected a number of cases within the probability ranges of model projections (Table 1). All regions except Pays de la Loire displayed increasing trends in the estimated detection rate.
We compared the projected incidence of COVID-19 symptomatic cases in w26 (4.5 [3.9-5.0] per 100,000) with the value obtained from confirmed cases (1.38 per 100,000) and two estimates based on COVIDnet.fr data (Figure 3). The first estimate applies the measured test positivity rate to the number of self-reported COVID-19 suspect cases (estimate #1, yielding 8.6 [6.2-11.5] per 100,000); the second additionally assumes that only 55% would be confirmed as suspect case by a physician and prescribed a test (according to COVIDnet.fr data on consulting participants, estimate #2, yielding 4.7 [3.4-6.3] per 100,000). Our projections are in line with plausible estimates from COVIDnet.fr.
(a) Projected number of new symptomatic cases over time (median and 95% probability range) and estimated number of virologically-confirmed symptomatic cases by week of onset (points) in mainland France. (b) Estimated detection rate of symptomatic cases (%) over time for mainland France (red dots) and for all regions (grey lines). (c) Map of the estimated detection rate (%) by region in w26 (June 22-28, 2020). (d) Estimated detection rate per region compared to the national estimate. Regions are ranked by increasing median detection rate. (e) Estimated incidence of symptomatic cases in mainland France in w26 from different sources: virological surveillance data (SI-DEP), participatory surveillance data (COVIDnet.fr, with two estimates), model projections. (f) Projected incidence per region compared to the national estimate. Regions are ranked as in panel d.
Sensitivity analysis showed that findings were robust against elements of the contact matrices that could not be informed by empirical data, and against current epidemiological uncertainties. Including in the analysis also asymptomatic cases led to higher detection rates, 43% [39-50]% in w26 compared to 31% [28-35]% for symptomatic cases only. This however assumes that asymptomatic cases were detected by the virological surveillance system in the week of infection, as no additional information was available to adjust for the possible delay.
Discussion
Despite a test positivity rate in mainland France well below WHO recommendations (5%)24, a substantial proportion of symptomatic cases (9 out of 10) remained undetected in the first 7 weeks following the end of lockdown. More than 60,000 symptomatic infections were not ascertained by the surveillance system from May 11 to June 28, 2020, according to our estimates.
Surveillance improved substantially over time. Detection rate was estimated to be 7% [6-9]% at the national level in mid-May, in line with estimates for the same period from a seroprevalence study in Geneva, Switzerland25. By the end of June, it increased to 31% [28-35]%, leaving about 2/3 of cases with symptoms undetected. Two regions (Occitanie, Normandy) reported cases compatible with model projections. These figures suggest that the new surveillance framework was progressively strengthened with increasing resources and capacity over time. Detection became also faster, with a 66% reduction of the delay from symptom onset to testing. At the same time, increasing performance benefited from a concurrent decrease of the epidemic activity in all regions.
Despite this positive trend, our findings highlight a critical need for improvement. Some regions remained with limited diagnostic exhaustiveness by the end of June. This is particularly concerning in those regions predicted to have a large number of weekly infections, such as Île-de-France where approximately only 3 out of 10 cases with symptoms were detected by the end of June, and Grand Est (1 out of 10). Novel recommendations since the end of lockdown require that all patients with symptoms suggestive of COVID-19 (as well as contacts of a confirmed case) be screened for SARS-CoV-22. Almost all cases (92% since May 25) clinically diagnosed by sentinel general practitioners as possible COVID-19 cases were prescribed a test11. However, only 31% of individuals with COVID-19-like symptoms consulted a doctor in the study period according to participatory surveillance data. Overall, these figures suggest that a large number of symptomatic COVID-19 cases were not screened because they did not seek medical care despite recommendations. A similar evidence emerged from a large-scale serological study in Spain where only between 16% and 20% of symptomatic participants with antibodies against SARS-CoV-2 reported a previous virological screening26. By combining estimates from virological and participatory surveillance, together with measured rates for test recommendations by general practitioners (e.g., due to misclassification of self-reported symptoms), we extrapolated an incidence of symptomatic cases from crowdsourced data that is compatible with model projections. This finding further supports consultation for all COVID-19 suspect cases. Large-scale communication campaigns should reinforce recommendations to raise awareness in the population and strongly encourage healthcare-seeking behavior especially in patients with mild symptoms. At the same time, investigations to identify reasons for not consulting could be quickly performed through the participatory surveillance system.
Red tape might have contributed to low testing rates. Prescription for a test was deemed compulsory in the new testing policy to prevent misuse of diagnostic resources2, however the path involving consultation, prescription, and lab appointment may have discouraged mildly affected individuals not requiring medical assistance. To facilitate access, some local initiatives emerged recently that increase the number of drive-through testing facilities, mail test vouchers to promote massive screening in certain regions (e.g. in Île-de-France27), offer temporary mobile testing facilities (buses, pavilions) to increase proximity with the population28. These initiatives are particularly relevant for counteracting socio-economic inequalities in access to information and care in populations vulnerable to COVID-1929 and may be established in the longterm. Given the non-uniform detection rate estimated within the country, learning from specific regional realities may aid to improve detection. The recent change in screening policy no longer requiring a prescription for testing30 could further improve access.
Screening rates remained overall well below the objective fixed by authorities for the post-lockdown phase (average weekly number of tests in May-June was 250,000 vs. target of 700,000), and the delay from onset to screening was still very long (7 days) by the end of June, despite substantial reduction over time. The large demand for testing currently observed in certain regions, mainly as a result of imminent travels and protocols imposed by certain countries and air companies, is reportedly causing long waiting lists at overwhelmed testing sites31. Given pre-symptomatic transmission, notification to contacts should be almost immediate to allow the effective interruption of transmission chains21. For testing to be an actionable tool for surveillance and, most importantly, for control of COVID-19 transmission, screening rates should be radically increased and delays suppressed. The risk would otherwise be a rapid and uncontrolled resurgence of cases with potential transmission in the community10, as currently reported in some French areas (e.g. Mayenne district in Pays de la Loire region)32 and countries in Europe33. Such risk is expected to increase if the reported relaxation in preventive behaviors persists, due to adhesion fatigue16.
Aggressive and efficient testing will become increasingly more important in the Fall months, as other respiratory viruses, such as influenza, RSV, rhinoviruses, will start to circulate and influenza-like-illness incidence levels will become comparable with those of COVID-19. Reviewing the testing strategy over Summer, while at low COVID-19 epidemic activity, is an important opportunity to strengthen French response system for next season.
Models were region-based and did not consider a possible coupling between regional epidemics caused by mobility. This choice was supported by stringent movement restrictions during lockdown34, and by the limited mobility increase in May-June35, before important inter-regional displacements took place at the start of summer holidays in July. Foreign importations were neglected9,36 as France reopened its borders with EU Member States on June 15, and the Schengen area remained closed till July. COVIDnet.fr cohort is not representative of the general population14, however the agreement found with sentinel incidence trends for influenza-like-illness suggests that these limitations have little effect once results are adjusted for lack of representativeness13. Underdetection may also proceed from the imperfect characteristics of RT-PCR (reverse transcription-polymerase chain reaction) tests used to identify infected cases37. Some cases tested for SARS-CoV-2 could have been falsely negative. This would affect the analysis presented in the manuscript and would be in line with our conclusion that a large part of cases may have been undetected. Asymptomatic infections were not considered in the main analysis, as we lacked information on the likely time of infection. The median duration from first to last positive nasopharyngeal swab was estimated to be 19 days in asymptomatic patients in China, with the longest duration at 45 days38. No such analysis has been performed in France yet. Assuming that asymptomatic infections were rapidly identified through contact tracing yielded higher detection rates than estimated for symptomatic cases only. This is due to a proportionally higher fraction of asymptomatic cases among the confirmed ones. Though limited by the underlying assumption, this result further strengthens the main conclusion that detection of symptomatic index cases is the key aspect that requires fundamental improvement.
Our findings identify critical needs of improvement to increase case ascertainment in France and the performance of the response system to monitor and control COVID-19 epidemic. Substantially more aggressive and efficient testing needs to be achieved to act as a pandemic-fighting tool. These elements should be considered in light of the resurgence of cases currently observed in some regions in France and in other countries with similar response systems.
Methods
Virological surveillance data and analysis
The centralized database SI-DEP for virological surveillance12 collects detailed information on patients tested in France, including (i) date of test, (ii) result of test (positive or negative), (iii) location (region), (iv) absence or presence of symptoms, (v) self-declared delay between onset to test in presence of symptoms. The delay is provided with the following breakdown: onset date occurring 0-1 day before date of test, 2-4 days before, 5-7 days before, 8-15 days before, or >15 days before. The SI-DEP database provided complete information for 13,887 (62%) out of 23,053 laboratory-confirmed COVID-19 cases tested between week 20 (May 11-May 17) and week 27 (June 29-July 5), with an increasing trend of complete information over time. The study referred to the period from w20 to w26; data of w27 were used to account for the delays. Data were consolidated in w27.
For cases with complete information, we estimated the number of symptomatic laboratory-confirmed COVID-19 cases by date of onset, using the information on the date of test and the time-interval of onset-to-test delay declared by the patient. We fitted a Gamma distribution to these data with a maximum likelihood approach, obtaining a shape parameter equal to 0.64, and expected value of delay equal to 8 days. Given a symptomatic confirmed case tested on a specific date, we assigned the onset date by sampling the onset-to-testing delay from the fitted distribution, conditional to the fact that the delay lies in the corresponding time-interval declared by the patient. To account for the changes in the distribution of self-declared delays over time, we also fitted the distribution to three periods of time, obtaining no significant difference.
Cases with missing data were imputed by sampling from a multinomial distribution with probabilities equal to the rate of occurrence of the labels reported for cases with complete information. Imputation was performed by region and by week. Onset date was then estimated for the imputed symptomatic cases.
Participatory surveillance data and analysis
COVIDnet.fr is a participatory online system for the surveillance of COVID-19, available at www.covidnet.fr. It was adapted from GrippeNet.fr to respond to the COVID-19 health crisis in March 2020. GrippeNet.fr is a participatory system for the surveillance of influenza-like-illness available in France since 2011 through a collaboration between Inserm, Sorbonne Université and Santé publique France13,14,39, supplementing sentinel surveillance. The system is based on a dedicated website to conduct syndromic surveillance through self-reported symptoms volunteered by participants resident in France. Data are collected on a weekly basis; participants also provide detailed profile information at enrollment. In addition to tracking influenza-like-illness incidence13,39, GrippeNet.fr was used to estimate vaccine coverage in specific subgroups40 individual perceptions toward vaccination41 and health-seeking behavior42. It was also used to assess behaviors and perceptions related to other diseases beyond influenza43.
Participants are on average older and include a larger proportion of women compared to the general population14,44. Participating population is however representative in terms of health indicators such as diabetes and asthma conditions. Despite these discrepancies, trends of estimated influenza-like-illness incidence from GrippeNet.fr reports compared well with those of the national sentinel system13,39. All analyses were adjusted by age and sex of participants.
To monitor COVID-19 suspect cases in the general population, we used the expanded case definition recommended by the High Council of Public Health for systematic testing and described in their 20 April 2020 notice4:
(Sudden onset of symptoms OR sudden onset of fever) AND (fever OR chills) AND (cough OR shortness of breath OR (chest pain AND age > 5 years old))
OR (Sudden onset of symptoms OR (sudden onset of fever AND fever))
AND
(age > 5 years old AND (feeling tired or exhausted OR muscle/joint pain OR headache OR (loss of smell WITHOUT runny/blocked nose) OR loss of taste)
OR ((Age ≥ 80 years old OR Age < 18 years old) AND diarrhea)
OR (Age < 3 months old AND (fever WITHOUT other symptoms))).
Figure 3 reports two estimates obtained from COVIDnet.fr cohort data for the incidence of symptomatic cases in w26. They are computed as follows:
Estimate #1 = (COVIDnet.fr estimated incidence in w26) * (test positivity rate from SI-DEP in w26)
Estimate #2 = (COVIDnet.fr estimated incidence in w26) * (estimated proportion screened and confirmed as COVID-19 suspect case by a physician, and prescribed a test; estimates from COVIDnet.fr) * (test positivity rate from SI-DEP in w26)
Ethics statement
GrippeNet.fr/COVIDnet.fr was reviewed and approved by the French Advisory Committee for research on information treatment in the health sector (i.e. CCTIRS, authorization 11.565), and by the French National Commission on Informatics and Liberty (i.e. CNIL, authorization DR-2012-024) - the authorities ruling on all matters related to ethics, data, and privacy in the country.
Transmission models summary
We used a stochastic discrete age-structured epidemic model for each region based on demographic, contact20, and age profile data of French regions19. Four age classes were considered: [0-11), [11-19), [19-65), and 65+ years old. Transmission dynamics follows a compartmental scheme specific for COVID-19, where individuals were divided into susceptible, exposed, infectious, and hospitalized. We did not consider further progression from hospitalization (e.g. admission to ICU, recovery, death17) as it was not needed for the objective of the study. The infectious phase is divided into two steps: a prodromic phase (Ip) and a phase where individuals may remain either asymptomatic (Ia, with probability pa =40%6) or develop symptoms. In the latter case, we distinguished between different degrees of severity of symptoms, ranging from paucisymptomatic (Ips), to infectious individuals with mild (Ims) or severe (Iss) symptoms. Prodromic, asymptomatic and paucisymptomatic individuals have a reduced transmissibility rβ = 0.55, as estimated in Ref.7. A reduced susceptibility was considered for younger children and adolescents, along with a reduced relative transmissibility of younger children, following available evidence from household studies, contact tracing investigations, and modeling works45-50. A sensitivity analysis was performed on relative susceptibility and transmissibility of younger children, and on the proportion of asymptomatic infections.
Models calibration
Models were calibrated regionally to daily hospital admission data through a maximum likelihood approach. The likelihood function is of the form
where Θ indicates the set of parameters to be estimated, Hobs(t) is the observed number of hospital admissions on day t, Hpred (t) is the number of hospital admissions predicted by the model on day t, 
is the probability mass function of a Poisson distribution with mean Hpred(t), and [t1, tn] is the time window considered for the fit. Calibration involved three steps, each one orresponding to a different epidemic situation: pre-lockdown17, during lockdown17, post-lockdown.
Data Availability
All data are available at the references cited.
Acknowledgments
This study was partially supported by ANR project DATAREDUX (ANR-19-CE46-0008-03) and EVALCOVID-19 (ANR-20- 706 COVI-0007); EU H2020 grants MOOD (H2020-874850) and RECOVER (H2020-101003589); REACTing COVID-19 modeling and surveillance grants. We thank Pascal Crepey, Camille Pelat, Edouard Chatignoux, Juliette Paireau, Daniel Levy-Bruhl for useful discussions. We also thank all participants of COVIDnet.fr system.
References
