Study design and PHCWs recruitment
We used data from the COVID-SéroPRIM study described elsewhere [17, 18]. Briefly, this nationwide cross-sectional study was conducted between May 10, 2021 and August 31, 2021 among GPs, pediatricians, dental workers (dentists and assistants), and pharmacy workers (pharmacists and assistants) in primary care thorough metropolitan France. The survey was conducted after the third wave of COVID-19 in France. COVID-19 vaccination of HCWs was available for HCWs without limitations from early February 2021. The PHCWs were volunteers recruited from the following four primary care research and monitoring networks: the French Sentinelles Network (GPs), the French Association of Ambulatory Pediatrics (pediatricians), the ReCOL network (dental workers), and IQVIA (pharmacy workers). All PHWCs were eligible to participate in the study except those who had previously taken part in a clinical trial for chemoprophylaxis against SARS-CoV-2 infection.
Data collection and serological analysis
After providing online consent, PHCWs were invited to fill a self-administered electronic questionnaire and to perform a capillary blood sampling. The questionnaire collected data on socio-demographic characteristics, household size and composition, smoking status, clinical characteristics (chronic disease, history of SARS-CoV-2 RT-PCR/antigenic or ELISA testing, and of COVID-19 vaccination), history of unprotected COVID-19 case contact (defined as face-to-face contact with a confirmed COVID-19 case without the use of recommended PPE), and occupational activities during the first lockdown and the following period (place of work, care of COVID-19 patients, performance of COVID-19 tests, use of PPE).
PHCWs received a dried-blood collection card (DBS) kit to be returned to the centralized biobank (CEPH Biobank, Paris, France) after self-sampling of capillary blood. Samples were prepared and send for serological analyses (Unité des Virus Emergents, Marseille, France). More details on serological methods can be found in previous published work [17, 18].
All samples were tested for IgG antibodies against the Spike (S) and the Nucleocapsid (N) proteins as well as neutralizing activity against SARS-CoV-2. An ELISA test (Euroimmun®, Lübeck, Germany) was used to detect anti-SARS-CoV-2 IgG against the S1 domain of the S protein (ELISA-S). In accordance with the manufacturer’s instructions, a test was considered ELISA-S-positive if the sample density ratio ≥ 1.1 (sensitivity, 87%; specificity, 97.5%) [19]. An immunoassay on Luminex was used to detect IgG directed against the N-protein (CTD and NTD domains, N-immunoassay). The cut-off values and assay performance indicators were calculated by receiver operating characteristic curve (ROC) analysis [20]. The specificity/sensitivity values for the CTD and NTD domains in the duplex assay were 96.1%/97.8% and 87.8%/88.5%, respectively. An in-house microneutralization assay was used to detect neutralizing anti-SARS-CoV-2 antibodies [21]. The neutralization titer referred to the highest dilution of serum with a positive result. Specimens with a VNT titer ≥ 20 were considered positive.
Outcome
The main outcome was a history of SARS-CoV-2 infection, as defined by the following criteria. (1) A positive N-immunoassay from DBS samples. Indeed, anti-N antibodies are not elicited by COVID-19 vaccines that target the S protein, including all vaccines that had been used in France at the time of the survey, and are developed as a result of SARS-CoV-2 infection [22]. However, using anti-N antibodies alone as a marker for natural infection may be problematic as anti-N antibodies have been shown to wane quickly in the first months after infection [23]. Thus, to avoid the risk of misclassification among individuals with negative N-immunoassay despite previous infection, SARS-CoV-2 infection was also determined by the following information: (2) a positive S-ELISA or neutralizing assay from DBS samples in unvaccinated individuals; (3) a self-reporting of a positive SARS-CoV-2 confirmed by reverse transcription-quantitative polymerase chain reaction (RT-qPCR) or antigenic test; (4) a self-reporting of a positive ELISA test (before the first dose in vaccinated individuals) (See Supplementary Table S1 for details).
Covariables
PHCWs were categorized in 6-level group categories according to occupation. Non-occupational factors potentially associated with SARS-CoV-2 infection comprised: age (<40/40-49/50-59/>60), sex, household factors (number of adults: 1/2/≥3; of children: 0/≥1; of rooms: <3/3/≥4), comorbidities (obesity, hypertension, diabetes, others chronic diseases), smoking status, unprotected contact with a COVID-19 case and region of workplace (a five-category variable defined according to the telephone area code and consistent with the various degrees of pandemic intensity across the regions : Île-de-France/North-West/North-East/South-East/South-West). Occupational factors included place of work (primary care only/other place), number of days worked per week (<3/3-4/>4), care of COVID-19 patients, performance of COVID-19 test, occupational activities during the first lockdown and access to PPE (FFP2 or surgical mask, gloves and coat, glasses and coverall).
Statistical analysis
Categorical variables were described by numbers and percentages, with comparison using Chi-square test or Fisher’s exact test when appropriate.
Region- and age-weighted prevalences were estimated for GPs, pediatricians, dentists and pharmacists. Our weights were the age-region proportion in the population (from the French 2021 census for each population) divided by the age-region proportion in our sample, for each age-region combination (Supplementary Table S2). Since national data were not available for dental and pharmacist assistants, we could not estimate weighted prevalence for these groups. 95% confidence intervals (CI) for estimates were Wald-type intervals computed on the log-odds scale, as implemented in the R “survey” package.
We compared the estimates among PHCWs with the proportion of the general population that has been infected by the SARS-CoV-2, at the national and regional levels, obtained with a mathematical model fitted to the virus spread in France [24]. For the national estimate, we used a stochastic age-stratified transmission model, integrating data on demography, age profile, and social contacts for the French population [24]. Four age classes were considered: [0–11), [11–19), [19–65) and 65+ years old. Transmission dynamics follows a compartmental scheme where individuals are divided into susceptible, exposed, infectious, hospitalized and recovered. For the regional estimates, we used a stochastic metapopulation transmission model, with individuals divided in the 12 regions of mainland France (excluding Corsica). Regions are interconnected by coupling probabilities, inferred from mobility data. Both models were parameterized with estimates from the literature on the infection-hospitalization ratio, and were fitted to hospital admission data, to reproduce the observed epidemic and estimate the total number of infections. Model predictions were validated against serological estimates [25]. We extracted the predicted proportion of SARS-CoV-2 infection as of June 1, 2021. Median and 95% probability ranges of the estimated proportions were computed from 100 independent stochastic runs for each model. We compared the estimates for adults (age bracket [19-65)) at national level with the prevalences among PHCW (non-overlapped 95%CI and probability ranges indicating a statistical difference between the two estimates). We analyzed the correlation of the estimates among PHCWs and the general population by using Pearson correlation test.
Poisson regression models with robust standard errors were used on unweighted data to identify the factors associated with the SARS-CoV-2 infection. A backward elimination procedure was used. The initial multivariable model included all factors with a p-value <0.20 in the univariable models. Elimination of covariates was based on the significance of the Wald chi-square test for parameter estimates (p-value <0.05). Improvement of model fit was determined through the Akaike Information Criterion (lowest value was preferable). To account for possible interactions, we compared model fit before and after addition of an interaction term between occupational group and occupational exposure factors using likelihood ratio tests, with interactions included where p-value <0.05 for the likelihood ratio test. Missing value were excluded from analysis. Statistically significant was considered if p value ≤ 0.05. All the analyses were computed with R software, version 4.0.3 (4.0.3, R Core Team, 2021, R Foundation for Statistical Computing, Vienna, Austria).
Ethical statements
The COVID-SéroPRIM study was approved by the “CPP Île de France V” ethics committee (ID RCB: 2020-A03298-31). Electronic informed consents were obtained from each participant before enrolment.