Proactive vs. reactive COVID-19 screening in schools: Lessons from experimental protocols in France during the Delta and Omicron waves

Abstract

The COVID-19 pandemic prompted diverse policies to manage safety in schools, balancing transmission control with educational continuity. We evaluated an experimental weekly screening protocol through salivary PCR tests against nationally implemented reactive strategies (class closure or class screening following case detection) in 25 primary schools in the Auvergne-Rhône-Alpes region of France during the Delta (November–December 2021) and Omicron (January–February 2022) waves. We used an agent-based model of severe acute respiratory syndrome coronavirus 2 transmission in schools, parameterized with empirical contact data and fitted to observed prevalence in 18 schools selected for the analysis under variant-specific epidemiological conditions. We simulated the weekly screening following the experimental protocol, with 7-day isolation of positive cases and class closure after three detected cases. We quantified reductions in within-school transmission and student-days lost and combined efficacy. The experimental weekly screening protocol reduced within-school transmission by 40% (interquartile range [IQR] 18-53%) during the Delta wave and 39% (IQR 31-46%) during the Omicron wave, relative to reactive strategies implemented nationally. Across both waves, weekly screening without class closure achieved the greatest overall efficiency, balancing transmission reduction and educational continuity. By mitigating asymptomatic spread through a structured and predictable testing schedule, weekly screening offered operational advantages over reactive strategies. These findings explicitly quantify key trade-offs between infection control and educational continuity and inform the design of proactive school-based interventions in future pandemics.

Significance Statement.

This study evaluates a real-world weekly COVID-19 screening protocol in 25 French primary schools during the Delta and Omicron waves. Using empirical data and transmission modeling, we show that school-based transmission accounted for up to two-thirds of infections among children and that weekly screening reduced in-school spread by 40% (interquartile range [IQR] 18–53%) and 39% (IQR 31–46%) compared with national reactive strategies. We further found that weekly screening led to broadly comparable student-days lost to national strategies, with differences largely driven by whether class closures were mandated. These findings demonstrate the importance of proactive testing in limiting silent transmission and limiting educational disruption. By combining modeling with field data, our results provide critical evidence supporting systematic screening with no class closure as a powerful, scalable approach to safeguard in-person learning and enhance pandemic response in educational settings.

Introduction

The COVID-19 pandemic has highlighted the diverse approaches adopted by governments worldwide to manage school safety while balancing the need for educational continuity (1). Strategies have ranged from prolonged school closures (2) and strict quarantines to the implementation of proactive testing and screening protocols. This variability reflects differences in public health priorities, resource availability (3), and the perceived role of schools in viral transmission. These heterogeneous policies have underscored the complexity of managing outbreaks in educational settings, particularly given the significant social and developmental consequences of school disruptions (4, 5). France offers a unique case study of evolving school policies during the pandemic, transitioning from reactive class closures to reactive screening strategies (6) and later exploring weekly screening of students (7) in a pilot initiative.

In September 2021, as schools reopened nationwide, France reinstated class closures in response to COVID-19 cases. This policy mandated quarantine for all students in a class upon the detection of a positive case, aiming to prevent further transmission among close contacts. Additional measures, such as mandatory mask-wearing and restricting interactions during meals or sport activities, were also implemented to limit exposure (8). However, the start of the Delta wave in Fall 2021 led to an escalation of class closures, with nearly 2% of all classes disrupted by the end of November 2021, and 73 out of 48,950 primary schools completely closed (9). To mitigate the growing educational disruption, French authorities introduced a nationwide reactive screening strategy starting in week 49 of 2021 (2021 December 6) as an alternative to class closure (6). This strategy enabled in-person attendance by screening the entire class upon identifying a positive case, isolating only those who tested positive while allowing others to continue attending school. In January 2022, as the Omicron wave surged, France implemented a strengthened version of the reactive screening protocol, requiring repeated testing of classmates of a detected case but no longer mandating systematic class closures.

Although reactive screening marked a shift toward minimizing educational disruptions, increasing evidence from modeling studies suggested that systematic screening on a weekly or semiweekly basis would be more effective in controlling viral transmission (10–12). Unlike reactive screening, which is triggered after a symptomatic case is detected, systematic screening proactively identifies asymptomatic infections, facilitating the timely isolation of positive cases and interruption of transmission chains. Reactive screening, on the other hand, risks missing undetected introductions and secondary infections, reducing its overall efficacy (13). Based on these findings, French authorities piloted a weekly screening protocol as an experimental initiative in selected primary schools in the Auvergne-Rhône-Alpes region from week 47 to week 50 of 2021 (7). The objective of this protocol was to assess its feasibility and contribution to preventing class closures. Several participating schools opted to extend the weekly screening into January 2022, during the Omicron wave, providing further data for analysis.

In this study, we used an agent-based transmission model parameterized with empirical contact data from primary schools and fitted to the weekly screening results of the experimental protocol. We evaluated the impact of school transmission on viral circulation among primary school-aged children and assessed the effectiveness of weekly screening in reducing transmission and school disruption during the second Delta wave (Fall 2021) and the subsequent Omicron wave. These results were compared with the nationally implemented reactive screening strategy. Our findings have important implications for informing strategies to maintain safe educational environments during future pandemics involving pathogens transmitted through the air. By providing evidence-based insights, this study contributes to the development of more effective and proactive public health policies for educational settings.

Results

A total of 28,643 salivary PCR tests were performed as part of the experimental weekly screening protocol in schools in the Auvergne-Rhône-Alpes region (Fig. 1). During period 1 (weeks 47–50, 2021 November 22–2021 December 19), characterized by the Delta wave, 14,630 tests were collected in 24 schools across four departments: 3,103 tests in five schools in Isère; 1,230 in two schools in Puy-de-Dôme; 7,434 in nine schools in Rhône; and 2,863 in eight schools in Savoie. In period 2 (weeks 1–6, 2022 January 3–2022 February 13), characterized by the Omicron wave, 14,013 tests were carried out in 14 schools: 550 tests in three schools in Isère; 13,143 in 10 schools in Rhône; and 320 in one school in Savoie. After applying the inclusion criteria, the total number of tests was reduced to 10,297 in period 1 (7,434 tests in nine schools in Rhône and 2,863 tests in eight schools in Savoie) and 13,143 in period 2, all of which were from the Rhône department, where 10 schools decided to continue adopting the experimental protocol beyond the initial study period (Table 1). For sensitivity, we conducted the analysis on all schools participating in the experimental protocol, resulting in additional five primary schools in the Isère department and two primary schools in Puy-de-Dôme and including the totality of the test in period 1 (Supplementary Material, pp. 4, 15).

Fig. 1.

Overview of school protocols and epidemiological context from week 47, 2021 to week 6, 2022 (2021 November 22–2022 February 13). A) Map of departments where the experimental weekly screening was implemented. Administrative boundaries were obtained from the France GeoJSON repository (gregoiredavid/france-geojson/departments, GitHub), used under its open-source license. B) Number of screened schools for each department (solid bars) between week 47, 2021 and week 6, 2022. The four selected departments are: Isère, Puy-de-Dôme, Rhône, and Savoie. The dashed line indicates the weekly incidence (cases per 100,000) from community surveillance in children aged 6–10 years old, corresponding to primary school students. The Christmas holiday period (weeks 51–52, 2021 December 20–2022 January 2) is shaded in light gray. C) Timeline of the school protocols.

Table 1.

Number of tests, students at school, observed adherence, and prevalence by weeks and departments in the primary schools participating in the experimental weekly screening and included in the study.

		Period 1—Delta wave (2021)				Period 2—Omicron wave (2022)
		W47	W48	W49	W50	W01	W02	W03	W04	W05	W06
Tests	Rhône	1,369	1,847	2,111	2,107	1,769	2,246	2,514	2,503	1,827	2,284
	Savoie	537	692	804	830	/	/	/	/	/	/
Students	Rhône	1,965	2,713	2,590	2,566	2,482	3,021	3,006	3,057	3,070	3,072
	Savoie	912	1,063	1,066	1,213	/	/	/	/	/	/
Adherence	Rhône	69.7	68.1	81.5	82.1	71.3	74.3	83.6	81.9	59.5	74.3
	Savoie	58.9	65.1	75.4	68.4	/	/	/	/	/	/
Prevalence	Rhône	1.53	2.33	1.33	0.95	5.03	8.06	8.15	5.07	5.09	3.20
	Savoie	1.30	2.17	2.24	0.96	/	/	/	/	/	/

Adherence rates varied across departments and periods. In period 1, the schools of Rhône showed the highest adherence, ranging from 69.7 to 82.1% with an increasing trend over time (Table 1). In Savoie, adherence increased from 58.9 to 75.4% before dropping to 68.4% by the end of the period. In period 2, adherence in Rhône rose in parallel with the Omicron surge, peaking during week 3 of 2022.

During the first 2 weeks of school following the holidays (weeks 45 and 46), participating schools adhered to the reactive class-closure protocol, which was applied nationwide, before the experimental weekly screening started in week 47. In period 1, prevalence rates in participating schools peaked during the second week of the protocol in Rhône (week 48) and the third week in Savoie (week 49; Fig. 2). In period 2, the prevalence peak in Rhône occurred during the third week after schools reopened. Table 1 summarizes the number of tests, number of students, adherence rates, and prevalence after applying the inclusion criteria. The corresponding information on the full dataset is reported in the Supplementary material (pp. 2–4).

Fig. 2.

Estimated and observed prevalence at school in period 1 and period 2 of experimental weekly screening in the Rhône and Savoie departments. A) The solid line represents the prevalence estimated by the model in period 1 of experimentation during the Delta wave in the Rhône department selected through the inclusion criteria. The shaded area surrounding the solid line represents the IQR. Dots represent the observed prevalence at school as obtained by experimental screening data. The error bars represent 95% CIs. The gray area corresponds to holiday periods. Dashed lines show the protocols in place at the national level. B) As in (A) for the Savoie department in period 1 during the Delta wave. C) As in (A) for the Rhône department in period 2 during the Omicron wave. In A and B, the simulated school protocols were reactive class closure from week 45 to week 46, 2021 and experimental weekly screening from week 47 to week 50 of 2021. In C, the simulated school protocol was the weekly screening from week 1 to week 6 of 2022. In all the panels, we considered salivary PCR tests.

The prevalence predicted by the fitted model closely matched the observed data for the departments included in the study: Rhône and Savoie during period 1 (Fig. 2A and B) and Rhône during period 2 (Fig. 2C). To evaluate the contribution of within-school transmission, we compared the fit estimates for the number of infections acquired at school and those introduced from the community (Supplementary material, p. 14). In period 1, within-school transmission accounted for 67% of all infections in both departments, with interquartile range (IQR) ranges of 53–78% for Rhône and 50–82% for Savoie (Fig. 3A). In period 2, within-school transmission represented 52% (IQR 47–57%) of total infections (Fig. 3B). Sensitivity analyses confirmed the stability of these estimates when varying model parameters, including extended durations in the postsymptomatic compartment $(R_{+})$ (10 days instead of 6), reduced test sensitivity for asymptomatic and presymptomatic phases, and variations in cohorting effects (Supplementary material, pp. 20–23). Estimates of the contribution of within-school transmission during the Delta and Omicron waves were similar to those obtained during the Alpha wave in the departments of the same region: 64% (IQR 20–81%) in the Ain department; 62% (IQR 14–80%) in the Loire department; and 59% (IQR 17–76%) in the Rhône department (Supplementary Material, p. 17).

Fig. 3.

Estimated school transmission contribution and experimental weekly screening impact under the Delta and Omicron waves by departments. A) Bars show the school transmission contribution predicted by the model in the selected departments of Rhône and Savoie in period 1 of the experimentation (i.e. weeks 47–50 of 2021) during the Delta wave. The error bars represent the IQR. B) As in A period 2, corresponding to the Omicron wave. C) Lines represent the prevalence estimated by the model in the Rhône department across various school protocols: reactive class closure from week 45 to week 48 (brown line), reactive screening on days 1 and 7 from week 49 to week 50 (dashed brown line), and experimental weekly screening from week 47 to week 50 of 2021 (green line). Areas represent the IQRs. The black dots represent the observed prevalence at school as obtained by experimental screening data. The error bars represent 95% CIs. D) As in C, but for period 2 (week 1 to week 6, 2022) during the Omicron wave, when strengthened national reactive screening on days 0, 2, and 4 was implemented at the national. E) As in C for the Savoie department. F) Percentage of averted school transmissions achieved by weekly screening (with and without class closure) compared with the reactive strategies in the two periods of experimentation (i.e. weeks 47–50 of 2021 in period 1 and weeks 1–6 of 2022 in period 2) when considering the same participation rates as observed during the experimental weekly screening. In all the panels, we considered salivary PCR tests.

To assess the relative effectiveness of the experimental weekly screening, we simulated school prevalence under the nationally implemented reactive strategies, using the number of introductions estimated under the weekly screening and the time-varying participation rates observed during the pilot. School prevalence remained consistently lower under weekly screening compared with the national reactive strategy, leading to a lower attack rate. During period 1, the nationally applied reactive screening strategy introduced in week 49 of 2021 was predicted to result in peak student prevalence rates of 3.1% (IQR 1.9–4.6%) in Rhône and 2.2% (IQR 1.1–3.4%) in Savoie—53 and 33% higher, respectively, than the peaks observed under weekly screening (Fig. 3C and E). In period 2, the strengthened reactive screening strategy applied between weeks 1 and 6 of 2022 was predicted to result in a school peak prevalence of 12.2% (IQR 10.0–14.5%) in Rhône, which was 42% higher than the peak observed under the experimental weekly screening (Fig. 3D). Weekly screening also led to an earlier decline in prevalence below key thresholds (2% during the Delta wave and 10% during the Omicron wave), further highlighting its greater effectiveness in controlling transmission. Overall, weekly screening was estimated to avert 40% (IQR 18–53%) of within-school transmissions during period 1 and 39% (IQR 31–46%) during period 2 compared with the reactive strategies (Fig. 3F). When the class-closure rule was removed, the proportion of averted transmissions remained similar in period 1 (34%, IQR 13–55%), but decreased in period 2, reaching 26% (16–38%) during the Omicron wave. These results indicate that class closures are not required for weekly screening to remain effective during periods of less sustained circulation, whereas in high-incidence contexts class closure provides additional epidemiological benefit by repeatedly interrupting within-school transmission chains. These estimates remained consistent across sensitivity analyses, even when varying key parameters (Supplementary material, pp. 26–28).

We evaluated school disruption associated with each strategy by estimating the number of student-days lost. In period 1, the number of student-days lost under weekly screening with class closure was comparable to that observed under the national reactive strategies (217 [IQR 93–294] vs. 183 [IQR 68–335] days lost in Rhône and 143 [IQR 57–231] vs. 158 [IQR 1–241] in Savoie, respectively; Fig. 4A). Under the scenario of weekly screening with no class closure, absences under weekly screening decreased by roughly half. In period 2, corresponding to the Omicron wave, student-days lost increased substantially under weekly screening with class closure, reaching 1,347 (IQR 1,147–1,537) days lost, compared with 215 (IQR 158–277) under the strengthened reactive screening strategy (Fig. 4B). Removing the closure rule markedly reduced absences, to 526 (IQR 468–585) student-days lost. Nevertheless, disruption under weekly screening without class closure remained higher, at about twice the level observed under strengthened reactive screening (215 [IQR 158–277] days lost).

Fig. 4.

Estimated school disruption and efficiency ratio. A) Bars show the number of student-days lost predicted in period 1 by the model in the departments of Rhône and Savoie during the Delta wave under the weekly screening (solid green), compared with that estimated for the nationally implemented protocols (brown). We also tested simulations with weekly screening without class closure after three cases (hatched green). Error bars represent the IQR. B) As in (A) for the Rhône department and period 2. C) Bars show the estimated efficiency index combining epidemiological benefit (reduction in school transmission) and educational cost (additional student-days lost) during the Delta and Omicron waves under: the reactive class closure followed by reactive screening in period 1 (brown), the strengthened reactive screening in period 2 (purple), and the weekly screening with class closure (green) and without class closure (hatched green) in both periods. Higher index values correspond to greater efficiency.

To jointly assess epidemiological benefit and educational impact, we defined an efficiency index (see Materials and methods), which quantifies the number of averted school-acquired infections per additional student-day lost relative to a minimal school intervention consisting of the isolation of symptomatic and test-positive cases only. This efficiency index provides a unified scale to compare the performance of all protocols across periods on both dimensions. Higher values therefore indicate more efficient protocols, achieving larger reductions in school transmission for a given educational cost. In period 1, weekly screening without class closure was the most efficient protocol, followed by weekly screening with class closure as implemented during the experimentation, while the national strategies of reactive class closure and reactive screening showed the lowest efficacy. In period 2, weekly screening without class closure remained the most efficient protocol, with the experimental protocol (weekly screening with class closure) showing efficiency comparable to that of the national strategy of strengthened reactive screening (Fig. 4C). Across all protocols, the estimated efficiency index exhibited substantial stochastic variability across simulations, as reflected by relatively wide IQRs, particularly for strengthened reactive screening.

Discussion

We analyzed virological test results from an experimental weekly screening protocol conducted in 17 primary schools in the Rhône and Savoie departments in France during the Delta wave (2021 November 22–December 17), and in 10 primary schools in the Rhône department during the Omicron wave (2022 January 3–February 13). We estimated that between half and two-thirds of student infections originated from school contacts, with systematic weekly screening reducing school-related transmissions by ∼40% compared with the national reactive strategies across both periods. Weekly screening also led to a number of student-days lost broadly comparable to those observed under the national strategies, although differences were strongly influenced by whether class closures were mandated. When closures were not applied, weekly screening achieved similar levels of attendance while ensuring the detection of infections, underscoring its value in limiting silent transmission without increasing educational disruption.

A key advantage of weekly screening was its ability to identify and isolate cases earlier than reactive protocols, effectively reducing student prevalence. Notably, prevalence began to decline 1 week earlier under the weekly screening protocol compared with predictions for reactive screening under the same epidemiological conditions in the community. This reflects the capacity of systematic screening to promptly detect and isolate presymptomatic and asymptomatic cases, which account for a substantial portion of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission, particularly in children (14–16). Empirical evidence supports this mechanism: the COVID-19 Schools Infection Survey in England (2020–2021) found that most cases identified through screening were asymptomatic (17), while a screening campaign in Piedmont, Italy, during early 2021 similarly showed that asymptomatic cases accounted for a substantial portion of infections during periods of high community transmission (18). Prior modeling studies (11, 12, 19, 20) further underscored the value of regular screening in curbing hidden spread. Our conclusions are likely to be broadly applicable to other regions of metropolitan France, since the participating schools reflect typical primary school structures and contact patterns, and the model incorporated surveillance and vaccination data. More broadly, they are consistent with studies across settings and variants showing that weekly screening outperforms symptomatic testing, reactive class closure, and reactive screening in limiting school-related infections (10, 13). Nonetheless, local variations in sociodemographic composition, school structures, resource availability, adherence to protocols, and the concurrent implementation of other public health measures may affect quantitative estimates.

Another important dimension of screening strategies is their impact on school attendance. In our study, weekly screening was penalized by the rule mandating a 7-day class closure after three detected cases. Under the national strategies, this rule applied only during period 1 and was then lifted under the strengthened reactive screening in period 2, whereas in the experimental weekly screening it remained in force throughout both periods. During period 1, the number of student-days lost under weekly screening with class closure was comparable to that observed under the national strategies; removing the class-closure rule, however, reduced absences by roughly half while maintaining similar epidemiological effectiveness. In period 2, weekly screening with class closure generated substantially greater disruption, largely because its systematic detection of infections in high-incidence conditions more frequently triggered class closures. By lifting the class closure in a counterfactual scenario, however, simulations show that the number of days lost under weekly screening would decrease markedly, but would still remain approximately twice as high as under the strengthened reactive screening.

To move beyond absences alone and capture the overall balance between transmission control and school continuity, we also examined an efficiency index that integrates both infections occurring in school and days of missed attendance. This index provides a combined view of efficiency by weighing the public health benefit of preventing cases against the educational cost of isolation. In both periods, weekly screening without class emerged as the most efficient strategy, achieving the highest efficiency index among all protocols. In period 1, this resulted from substantially fewer student-days lost compared with all other protocols, while achieving a number of school transmissions that was lower than that of the national strategies and comparable to that of weekly screening with class closure. In period 2, weekly screening without class closure again achieved the highest combined efficiency, despite being slightly less epidemiologically effective than weekly screening with class closure under high-incidence conditions. This difference likely reflects the reduced marginal benefit of systematic screening when repeated closures already interrupt transmission. However, the markedly lower educational disruption associated with the absence of class closure more than compensated for this reduction in transmission control, resulting in the highest overall efficiency. Importantly, the two strategies that were actually implemented during the study period—weekly screening with class closure in the experimental schools and strengthened reactive screening nationwide—achieved comparable efficiency levels in period 2, albeit through fundamentally different mechanisms. Weekly screening with closure provided better control of in-school transmission, while strengthened reactive screening resulted in moderately fewer absences. This was not because reactive screening preserved attendance more effectively; rather, it failed to detect a large proportion of infections—particularly asymptomatic cases—that weekly screening was able to identify. These findings highlight that the apparent lower disruption under reactive screening reflects limited case detection rather than a more balanced strategy and that the impact of weekly screening on absences critically depends on whether class-closure rules are mandated. This interpretation aligns with our previous modeling work, which showed that reactive class closures are highly costly in terms of absences, while regular testing (with no class closure) can reduce peak absences by up to 30%, depending on the incidence levels (13). Thus, while these strategies were implemented under evolving and uncertain conditions, our results clarify the trade-offs they entail between minimizing educational disruption and maximizing case detection and transmission control, offering guidance on the prioritization of objectives for future pandemic preparedness. Finally, despite these different pathways to similar efficiency levels, weekly screening without closure offers the added advantage of a predictable and manageable testing schedule, in contrast to the unpredictability and logistical strain created by strengthened reactive screening. Indeed, during the study period, the continuous occurrence of COVID-19 cases triggered surges in test demand (13), overburdening testing centers and straining pharmacies and laboratories. The lack of predictability associated with reactive screening, which depends on the detection of symptomatic cases, further complicated resource allocation and planning. This unpredictability causes additional stress for educators and parents, particularly during the Omicron wave. Teachers in France expressed concerns about the continuity of education (21), while families struggled to access timely testing, often resulting in children staying home unnecessarily and parents missing work.

Participation rates were critical to the success of the protocol. In period 1, adherence reached over 80% in Rhône and over 67% in Savoie by the end of the period, contributing significantly to the flattening of the prevalence curve. Similarly, in period 2, adherence in Rhône exceeded 80% during the Omicron peak (weeks 3 and 4 of 2022), coinciding with a reduction in student prevalence. Modeling studies have consistently confirmed that such high adherence—at least three-quarters of students—is required to achieve meaningful reductions in transmission. In contrast, pilot screenings conducted in the same region during the Alpha wave in the Spring 2021 reached only ∼50% adherence in primary schools and as little as 10% in middle and high schools (10), highlighting the difficulty of sustaining engagement. Yet, examples from elsewhere show that high adherence can be maintained over extended periods. In Baselland, Switzerland, voluntary weekly salivary PCR testing achieved participation rates between 70 and 85% for several months in 2021, prior to becoming mandatory the following year (13). Sustaining such high adherence requires well-coordinated, effective communication strategies that address community concerns and encourage participation (22, 23). Despite substantial upfront investments, modeling studies indicate that regular screening strategies can ultimately be cost saving, by reducing the expenses associated with outbreak response (i.e. contact tracing) and school closures (i.e. online lessons) (24). However, implementing such programs remains challenging, particularly in low- and middle-income countries. According to the second UNESCO–UNICEF–World Bank survey on COVID-19 school responses (May–August 2020), only 19% of countries reported plans for school-based testing, and only 50% of low- and middle-income countries reported sufficient resources to implement even basic health protocols (25). Bridging these gaps requires strategic investments in education and health infrastructure, ensuring that proactive interventions are accessible and tailored to the unique needs of local contexts (26). Such efforts are critical for achieving equitable responses to future health crises.

The contribution of schools to SARS-CoV-2 transmission varied across periods and contexts, reflecting differences in variant transmissibility, age-specific attack rates, vaccination coverage, and mitigation measures. In this study, we estimated that 50–70% of infections detected through weekly screening originated in schools, with a lower proportion during the Omicron wave, likely due to increased community transmission during the Christmas break when schools were closed. These estimates are consistent with findings from the Alpha wave in the same region, where prevalence data from 71 primary schools in Ain, Loire, and Rhône (10) indicated that schools contribute ∼60% of overall infections. Comparable results have been reported in other European contexts. In Belgium, reconstruction of primary school outbreaks integrating epidemiological and genomic data found that 66% of transmissions originated in schools, with most transmission occurring between children or from children to adults (27). Another study confirmed that children aged 0–12 years were key drivers of transmission during the Alpha wave and again in autumn 2021, coinciding with school reopenings and delayed vaccination in younger age groups (28). Similarly, an analysis of 87 school outbreaks in Italy during March–April 2021 attributed nearly half of confirmed cases among school attendees to school contacts under reactive class-closure protocols (29). Cross-national differences in school contribution highlight the importance of local epidemiological and policy contexts. In Germany, schools accounted for ∼20% of cases during the Omicron wave due to strict masking and reduced class sizes (30). In the United States, modeling of primary schools participating in weekly testing programs from spring 2021 to March 2022 suggested smaller transmissibility within schools than in the broader community (31). However, this analysis relied on a compartmental framework rather than explicit contact network data within schools.

Beyond epidemiological factors, structural and societal determinants—such as intergenerational contact patterns and household crowding—likely modulate school contributions. Data from the United States (32, 33), Sweden (34), and France (35) indicated that crowded households with children or teachers attending in person were associated with higher infection risks and, in the United States, with increased severity of COVID-19. These findings underscore that while schools may not act as primary amplification hubs for SARS-CoV-2, as they do for influenza (36), they nonetheless play a substantial role in sustaining transmission when mitigation measures are insufficient. Secondary attack rates in educational settings remain low where nonpharmaceutical interventions (NPIs) such as masking and ventilation are applied effectively, as shown in studies from Switzerland (37) and Italy (38). Future research should further investigate environmental factors such as ventilation and the contribution of household spillover from school infections to community transmission (39, 40).

Our study has a set of limitations. First, we did not explicitly assess aerosol transmission or the effectiveness of ventilation measures during the experimental period. Applying natural ventilation was reported in 93.1% of schools more than twice daily, alongside compulsory mask use during period 1. However, our estimation of the transmission rate inherently accounts for the implemented preventive and control measures. Second, our analysis was limited to school-based transmission and did not account for infections introduced into households. Prior research suggests that active school surveillance can reduce such spillover effects (40). Third, the school contact network we used to parameterize the model was collected in 2009. However, the organization of French primary schools has remained structurally stable between 2009 and 2021, with only moderate increases in student numbers and class sizes that suggest expansion in scale rather than reorganization (41). Collecting high-resolution contact data in schools with wearable radio-frequency identification (RFID) sensors based on radio signals when individuals are in close proximity is logistically and ethically challenging, and very few examples exist. Those available cannot be directly applied to our context, either because they concern different school types and used different sensor technologies (42) or because of their limited sample size and short observation period (43). Synthetic networks have also been generated to approximate school settings where empirical data were missing (12, 20) or to extend available empirical data across time and school size (44). While relying on older contact data represents a limitation, the stability of French school organization and the lack of more recent high-resolution alternatives support our modeling choice. Fourth, our findings are limited to primary schools and may not generalize to secondary schools, where student density, mixing patterns, and activities—as well as student epidemiological and immunological characteristics—differ. Further research in real-world conditions is needed to assess the effectiveness of weekly screening in these contexts. Our previous modeling work showed that weekly screening remains the most effective strategy in controlling viral spread in high schools compared with reactive protocols (10), highlighting its broader potential across educational levels. Additionally, while systematic screening is resource intensive, its testing demand under high-incidence conditions is comparable to reactive strategies requiring three tests over 4 days (13). Even with reduced diagnostic sensitivity, such as with antigen tests, systematic screening was shown to consistently outperform reactive approaches in controlling disease spread (10, 13). This underscores its robustness as an effective strategy for managing school-based transmission across varied contexts.

The COVID-19 pandemic pushed authorities worldwide to repeatedly close schools, exposing students to significant learning losses, increased anxiety, missed social opportunities, and exacerbated inequalities (45, 46). These consequences demonstrate that prolonged school closures are unsustainable, even during public health crises (47). Flexible strategies that adapt to evolving transmission risks should be considered to optimize the balance between public health measures and educational needs. Systematic screening offers a viable and proactive strategy to mitigate presymptomatic and asymptomatic transmission, ensuring safer access to education while highly transmissible variants circulate and limiting school disruption. Our findings provide evidence to improve the design of NPIs in schools and contribute to future pandemic preparedness plans to make schools more resilient in the event of a future pandemic.

Materials and methods

Data collection under the experimental weekly screening

The experimental screening campaign targeted students aged 6–10 years old from 25 primary schools randomly selected in the departments of Isère, Puy-de-Dôme, Rhône, and Savoie within the Auvergne-Rhône-Alpes region. The campaign was conducted over two separate periods, divided by the Christmas holidays. Period 1 spanned weeks 47–50 of 2021 (2021 November 22 to December 17) during the Delta wave, while period 2 occurred during weeks 1–6, 2022 (2022 January 3 to February 13) amid the Omicron wave. In period 1, over 99% of detected cases were identified as the Delta variant (week 43, 2021), with the peak incidence reaching 1,457 cases per 100,000 children aged 6–10 years (week 49, 2021). In contrast, period 2 coincided with the emergence of the Omicron variant (96% BA.1 Omicron in week 2, 2022), which saw a peak incidence of 7,652 cases per 100,000 children in the same age group (week 3, 2022; Fig. 1) (48). The primary objective of the experimental weekly screening was to assess its feasibility and contribution to preventing class closures. Building on the collected prevalence data, we further evaluated the effectiveness of weekly screening in preventing new cases.

Over the experimental period (week 47, 2021–week 6, 2022), weekly PCR tests on saliva samples were offered every Monday to students in participating primary schools. Participation was voluntary, requiring informed written consent from parents. Samples were collected at home, and the testing was conducted in collaboration between schools and medical laboratories. School principals provided class and participant lists to assigned laboratories, which managed the scheduling of tests and communicated results. Students who tested positive were required to isolate for 7 days, with their status reported to local health authorities for follow-up. If more than three positive cases were detected within a single class, the entire class was closed for 7 days.

For this study, we accessed anonymized, aggregated data for each participating school. These data included the number of students present on screening days, the number of tests conducted, and the number of positive cases. Only departments with at least five schools participating in weekly screenings (and at least 500 tested students per week) were included in the main analysis, in line with previous studies (10). Sensitivity analyses were also performed, relaxing this criterion to include all schools participating in the campaign. Adherence to screening was computed as the ratio of tested students to the total number of students present on the screening day. Observed prevalence was determined as the ratio of positive tests to the total number of tests conducted.

Nationwide school protocols

During the study period (week 45 of 2021 to week 6 of 2022), two changes in nationwide school protocols occurred in France. Initially, the school protocol relied on reactive class closures upon the detection of a case. Under this policy, the positive case and their classmates were required to isolate for 7 days.

As case numbers began to rise during the Delta wave, authorities introduced a reactive screening protocol starting in week 49 of 2021 (6). Under this updated strategy, all students in the same class of a detected case were required to perform an RT-PCR or lateral-flow device (LFD) test on day 1 and day 7 following the detection of the index case. Only students with a negative test result were allowed to continue attending school in person, while positive cases were required to isolate for 10 days. Additionally, the detection of three positive cases within a single class triggered a 7-day classroom closure.

When schools reopened in January 2022 following the Christmas holidays, a more stringent version of the reactive screening protocol (strengthened reactive screening) was introduced in response to the emergence of the Omicron variant. Starting in week 1 of 2022, students in the same class as a detected case were required to complete an LFD or RT-PCR test on day 0 following the case detection, with two self-tests on days 2 and 4. Students with positive test results were required to isolate for 7 days. Families were responsible for administering the tests, which could be performed at pharmacies, laboratories, or similar facilities outside the school. Beginning 2022 January 14, self-tests on day 0 were also accepted as an alternative to LFD or RT-PCR tests conducted at testing centers (49). Unlike the previous protocol, no classroom closures were mandated under this strengthened strategy. However, local authorities retained the discretion to close classes in the event of a high number of cases.

Nationwide community protocols during the same period included vaccination campaigns (booster doses extended to those aged ≥12 years from late November 2021 (50), vaccination opened to children aged 5–11 years in December 2021 (51), and the vaccination pass replacing the health pass in January 2022 (52)), as well as reinforced workplace health measures, mandatory remote working from January 2022, and the progressive tightening of protective rules in response to the Omicron variant (Supplementary Material, p. 4). While we did not explicitly model these community policies or their direct effects on transmission, their impact is implicitly captured in the estimated number of introductions into schools, which reflect the epidemiological situation in the community.

Ethical statement

Contact studies were approved by the Commission Nationale de l'Informatique et des Libertés (the French national body responsible for ethics and privacy; 1719527 and 1427054) and school authorities. Informed consent was obtained from participants or their parents if they were minors (age <18 years). No personal information of participants was associated with the RFID identifier. No ethical committee was required for this study, as no individual data were collected in the study. All data retrieved during the screening campaign were anonymous and aggregated at the school level. Oral and written consent were obtained from parents ahead of the campaign.

Transmission model

We designed a stochastic agent-based model of SARS-CoV-2 transmission in schools, originally developed to study the Alpha wave in school settings (10), parameterized with empirical contact data. The model used data on face-to-face proximity interactions, captured through RFID sensors with a 20-s time resolution, in a French school comprising 232 primary students and 10 teachers divided into 10 classes (53). These sensor data were used to generate temporal contact networks, where nodes represent individuals (classified by class and student or teacher) and links represent empirically measured proximity contacts occurring at specific times. The contact networks revealed a strong community structure centered around classes, with students spending more time interacting within their own class than with students from other classes (10). Additionally, students were observed to have longer interactions than teachers.

Transmission in the model occurs with a specific transmissibility rate (β/min) per contact per unit time between an infectious individual and a susceptible one who are in contact. The transmission rate β was fitted to reproduce the observed student prevalence estimated from the experimental weekly protocol data (see the “Inference framework” section). Infection progression includes prodromic transmission, followed by clinical or subclinical disease stages, informed from empirical distributions (54–56). We also allowed individuals to enter the transient phase $(R_{+})$ following the clinical or subclinical phase, where they are no longer infectious but can still test positive through a PCR test. The model was parameterized with age-specific estimates of susceptibility, transmissibility, probability of developing symptoms, and probability to detect a case based on symptoms (14, 57–61). It also accounts for the epidemiological and immunological characteristics of the Delta and Omicron variants (62). To capture immune evasion during Omicron circulation in period 2, we considered an additional compartment (S*), representing individuals previously infected with pre-Omicron variants who had partial protection against reinfection (Supplementary material, pp. 9–11). Infection-induced immunity against Omicron was set at 61% from available estimates (62, 63). The initial number of individuals in the S* compartment was informed from pre-Omicron infection estimates (64). We distinguished between unvaccinated individuals, those fully vaccinated with a primary vaccination course, and boosted individuals (65). Vaccination coverage data were dynamically implemented and sourced from national registries (65). Furthermore, we considered age-specific vaccine effectiveness (VE) against infection, transmission, and clinical symptoms given infection, depending on the number of doses, time since the last dose, and the variant of concern (66–69). Specifically, we incorporated temporal changes in VE among children to reflect the rapid waning of protection observed after full vaccination during period 2, when vaccination first became available for this age group. Vaccination was therefore not considered in period 1. In adults, VE was assumed constant over the 6-week study period, as waning was substantially slower than in children (69, 70) (Supplementary material, pp. 9–11). Full details on the model structure, parameter values, and estimates are reported in the Supplementary material (pp. 4–12).

Sensitivity analyses were performed on the values of children’s susceptibility and transmissibility and the residency time in the $R_{+}$ compartment (Supplementary material, pp. 15–17).

Simulations

We simulated the school protocols that were implemented in France in the study period (Table S6).

The model fit was performed on prevalence estimates derived from the experimental weekly screening protocol, starting in week 47, 2021. Observed prevalence was defined as the proportion of positive screening tests on each screening day. Model-estimated prevalence was defined as the ratio between the number of students in a detectable state (prodromic, asymptomatic/symptomatic, or recovered with residual viral load) adjusted by the state-specific test sensitivity and the number of students present at school on the screening day (excluding those in isolation or quarantine). This measure thus represents the infection prevalence at school among those eligible for testing and provides a consistent metric across all simulation weeks—including periods with no testing, such as during reactive class closures—while being largely unaffected by screening adherence. In the weeks with screening in place, this quantity matches the predicted proportion of positive tests.

Separate fits were conducted for the Delta and Omicron waves (periods 1 and 2, respectively) to account for the substantial epidemiological and immunological differences between the two variants, while maintaining a single-strain model for each study period (Supplementary Material, pp. 14 and 15). In period 1, simulations started in week 45, when schools reopened after a 2-week break. During weeks 45 and 46 of 2021, we simulated the reactive class-closure protocol implemented in France at the time, which required a 7-day class closure following the detection of a positive case. Starting in week 47, this was replaced by the weekly screening protocol, which mandated a 7-day isolation for students testing positive and a class closure if three or more cases were detected. In period 2, simulations spanned from weeks 3 to 6 of 2022, when schools reopened after the Christmas holidays. During period 2, we simulated the strengthened reactive screening protocol, which required the isolation of positive cases only, without class closure.

Age-specific seroprevalence estimates and vaccination coverage were used to initialize the model (65, 71). Weekly introductions were stochastically estimated using age-specific community surveillance data by department and adjusted to account for detection rate and estimated within-school transmission dynamics (10). By using community surveillance data, we implicitly accounted for changes in community-based interventions.

All simulated protocols assumed the use of salivary PCR tests with a 1-day turnaround time. Test sensitivity was modeled as age-dependent and time-varying, with a peak sensitivity of 96% (72) (Supplementary material, p. 12).

Counterfactual scenarios and effectiveness evaluation

To evaluate the effectiveness of the weekly screening protocol, we constructed counterfactual scenarios reproducing the national strategies applied in schools of the same departments during the study period. Effectiveness was assessed by comparing the number of school-acquired infections and the number of student-days lost under each strategy.

Counterfactuals were generated by parameterizing the model with the maximum likelihood estimate (MLE) of transmissibility (${\beta }_{\mathrm{MLE}}$) and with weekly introductions inferred under the weekly screening protocol. This allowed us to effectively reproduce the same community transmission levels. Using these parameters, we simulated the sequence of national strategies. In period 1, this included the reactive class-closure protocol applied in weeks 45–48 of 2021, followed by the reactive screening protocol introduced in week 49, which required tests on days 1 and 7 for classmates of a positive case, 10-day isolation for positives, and class closure if three cases occurred within a week. In period 2, we simulated the strengthened reactive screening protocol, which required tests on days 0, 2, and 4 for classmates of a positive case, with a 7-day isolation period but no class closure. For all simulations, weekly participation rates within each class were assumed equal to those observed during the experimental weekly screening.

We also simulated a counterfactual weekly screening scenario without class closure after three detected cases to isolate the effect of the closure rule itself.

We then compared weekly screening (with or without class closure) with the national strategies along two dimensions in each period. First, we computed the percentage of averted school transmission, defined as the relative reduction in the number of school-acquired infections between the weekly screening (with or without class closure) and the national protocol in that period, considering the national protocol as the reference. Second, we assessed school disruption by calculating the number of student-days lost, measured as the cumulative number of days students spent in isolation or quarantine during each period, for each protocol (weekly screening with or without class closure and the national reactive strategies).

To jointly assess epidemiological benefit and educational impact across all protocols, we defined an efficiency index (η) as the number of averted school transmissions per additional student-days lost. To move beyond the direct pairwise comparisons between weekly screening and period-specific national strategies, we used a common reference protocol consisting of symptomatic testing and isolation of positive individuals only, corresponding to the minimal intervention implemented during the COVID-19 pandemic. In each period and for each protocol, the number of averted school transmissions ($\mathrm{\Delta }T$) was computed as the difference in the number of school-acquired infections relative to this reference protocol. The additional educational cost ($\mathrm{\Delta }D$) was defined as the difference in student-days lost relative to the same reference protocol. The efficiency index was therefore defined as $\eta =\mathrm{\Delta }T/\mathrm{\Delta }D$. Higher values of η indicate more efficient protocols, achieving larger reductions in school transmission per additional student-day lost. This definition allows direct comparison of all protocols on a common efficiency scale, independently of the period-specific national strategies in place.

Inference framework

We used an MLE approach to fit the model to the student prevalence observed in primary schools implementing the experimental weekly screening. We estimated the transmissibility per contact per unit time (β/min) at school for the Delta and Omicron variants, considering time-varying adherence rates based on the observed participation data from each department.

In period 1, detection rates were explored in a grid and selected using the likelihood ratio test (Supplementary Material, p. 14). In period 2, the detection rate was estimated by comparing age-specific community surveillance data for children aged 5–11 years with the prevalence reported in schools (Supplementary Material, p. 15).

Transmissibility estimates for both periods were obtained from 2,000 simulated stochastic outbreaks per parameter set over a 6-week timeframe (42 days). Sensitivity analyses were conducted to examine the effects of cohorting and the impact of reduced test sensitivity for asymptomatic individuals during the presymptomatic and postsymptomatic phases.

Comparison with prior estimates from the Alpha wave

We extended the analyses we performed in a previous work to study SARS-CoV-2 transmission in primary schools during the Alpha wave (10) to estimate the contribution of within-school transmission in that wave and compare it with the results of this study from the Delta and Omicron waves. Details are available in the Supplementary material, p. 17.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This study was partly funded by the French Research Agency (Agence Nationale de la Recherche, ANR) through the project DATAREDUX (ANR-19-CE46-0008-03 to V.C.), European Union's Horizon 2020 under grant MOOD (H2020-874850 to V.C.; paper cataloged as MOOD 125), Horizon Europe grants VERDI (101045989 to V.C.), and Horizon Europe grants ESCAPE (101095619 to V.C.). The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission.

Author Contributions

Elisabetta Colosi (Data curation, Formal analysis, Methodology, Software, Visualization, Writing—original draft, Writing—review & editing), Bruno Lina (Data curation, Investigation, Writing—review & editing), Christelle Elias (Data curation, Investigation, Writing—review & editing), Philippe Vanhems (Data curation, Investigation, Writing—review & editing), and Vittoria Colizza (Conceptualization, Funding acquisition, Methodology, Supervision, Writing—original draft, Writing—review & editing)

Preprints

This manuscript was posted on a preprint at https://www.medrxiv.org/content/10.1101/2025.01.17.25320676v1.

Data Availability

De-identified individual data on contacts of the primary school under study are publicly available at the SocioPatterns project website (http://www.sociopatterns.org/datasets/). De-identified aggregated COVID-19 community surveillance data by age class are publicly available at Santé publique France data observatory platform (https://odisse.santepubliquefrance.fr/explore/dataset/covid-19-synthese-des-indicateurs-de-suivi-de-la-pandemie-dep/table/?sort=-tx_incid). De-identified aggregated COVID-19 prevalence data from experimental weekly screenings during the Delta and Omicron waves used in this study are available in the tables reported in the manuscript and Supplementary material. The code used to run the transmission model is publicly available on GitHub at https://github.com/EPIcx-lab/COVID-19/tree/master/School_transmission.