Modelling safe protocols for reopening schools during the COVID-19 pandemic in France

European countries adopted heterogeneous exit strategies from the first COVID-19 lockdown, with particular debate around reopening schools. France announced a progressive reopening beginning May 11, 2020, prioritizing pre-schools and primary schools with limited class sizes and voluntary attendance, and deferring decisions for secondary schools based on local conditions. A key challenge for assessing reopening risk is the uncertain role of children in SARS-CoV-2 transmission: evidence suggests lower susceptibility and more asymptomatic or paucisymptomatic infections among children, with adolescents potentially playing a different role than younger children. This study aims to evaluate how various school reopening protocols (partial, progressive, full; by school level) could impact transmission and healthcare demand in Île-de-France under moderate social distancing combined with case finding, testing, and isolation. The work was conducted during lockdown as scenario analysis and is complemented with an ex-post assessment using data available after reopening.

The paper reviews evidence indicating children, especially those under 10 years, are less susceptible to infection and more likely to have asymptomatic or paucisymptomatic disease than adults. Household and contact tracing studies, as well as large-scale testing in Iceland and Vo', Italy, indicate lower incidence in younger children compared with adolescents and adults. Retrospective investigations of clusters in Oise, France, found substantial asymptomatic circulation in a high school, while primary school introductions did not lead to onward transmission. South Korean contact tracing showed the lowest household attack rates when index cases were under 10 years. Viral load appears similar across ages and between symptomatic and asymptomatic infections, but transmission risk varies with symptom presence and severity. These findings motivate modeling different transmissibility assumptions for younger children versus adolescents.

The study uses a stochastic discrete age-structured compartmental model for Île-de-France, with four age classes (0–11, 11–19, 19–65, 65+). Mixing patterns are based on 2012 French contact matrices by setting (household, school, workplace, transport, leisure, other) and contact type. Disease progression includes susceptible, exposed, infectious (prodromic), and infectious states stratified into asymptomatic, paucisymptomatic, mild symptomatic, and severe symptomatic, followed by hospitalization, ICU, recovery, or death. Asymptomatic/paucisymptomatic individuals transmit with reduced transmissibility (r = 0.55). Children (0–11 and 11–19) are modeled as half as susceptible as adults and predominantly asymptomatic/paucisymptomatic. Adolescents follow the same reduction in transmissibility when asymptomatic as adults. The relative transmissibility of younger children (0–11) compared to adolescents is varied in sensitivity analyses: 0.1, 0.25, 0.33, 0.55. Interventions are modeled by altering contact matrices: lockdown includes school closure, high telework, reduced senior contacts, and closure of non-essential activities; physical contacts outside households are removed. Model calibration is via maximum likelihood to hospital and ICU admissions data before and during lockdown (using data up to April 26, 2020), estimating transmission rates and the reproduction number. Mobility data inform the fraction staying at home during lockdown. The estimated reproduction number decreased from R0 ≈ 3.28 pre-lockdown to RLD ≈ 0.71 during lockdown. Post-lockdown scenarios assume moderate social distancing (50% telework, 50% closure of non-essential activities, 30% reduction in senior contacts) and efficient test-trace-isolate (50% of cases isolated with a 90% reduction in contacts), yielding an effective reproduction number around 1 with schools closed (R ≈ 1.02). School reopening scenarios are structured in three sets: (1) only pre-/primary schools reopen on May 11; (2) pre-/primary reopen May 11 plus middle/high schools reopen June 8; (3) all schools reopen May 11. Each set considers four protocols: Progressive to 100% (25% increments weekly to full), Progressive to 50% (25% then 50%), Prompt 50% (immediate 50% attendance), Prompt 100% (immediate full attendance). Attendance scales school contacts proportionally. Holidays are approximated using spring holiday contact data for July; adult holidays are not modeled. Each scenario is simulated with 500 stochastic runs, reporting medians and 95% ranges. Outcomes include daily clinical cases on July 5 and ICU occupancy on August 1. Sensitivities include lower case isolation (25%), ±10% variation of RLD, full reopening of younger children on May 11 in set (2), and a 50% reduction in adolescent contacts.

• Calibration and projections to May 11, 2020: R0 = 3.28 [3.20, 3.39] pre-lockdown; RLD = 0.71 [0.69, 0.74] during lockdown. Projected on May 11: ~945 [802, 1076] new clinical cases/day (≈2391 [2025, 2722] infections/day), ~18 [11, 29] ICU admissions/day, ICU occupancy at 47% [37, 57]% of strengthened capacity.
• Under moderate social distancing + 50% case isolation with schools closed: effective R ≈ 1.02 [0.99, 1.06] (stable activity).
• Reopening only pre-/primary schools (May 11): daily clinical cases by July 5 increase to 2.0–2.4× compared to school closure, depending on younger child transmissibility and protocol; ICU demand peaks at 62% [54, 68]% of a 1500-bed capacity by August 1. Differences between progressive vs prompt protocols are minimal when younger children are less transmissible than adolescents.
• Adding middle/high schools on June 8: with 50% attendance or progressive reopening to full, epidemic levels are similar to the most active scenario of reopening only younger levels; ICU demand ~64% [56, 70]% by August 1. If middle/high schools reopen at full attendance on June 8 (Prompt 100%), daily clinical cases reach 2.6–2.9× the school closure scenario by early July; ICU occupancy rises to 54% [46, 60]%–76% [67, 84]% by mid-summer.
• Reopening all schools May 11 at full attendance: ICU capacity is saturated before end of July (Prompt 100%). Progressive to full attendance approaches ICU saturation by mid-summer: 98% [85, 107]%. Limiting attendance to 50% avoids saturation, keeping ICU demand between 45% [40, 51]% and 61% [55, 68]% (August 1).
• Control depends on efficient case isolation: with only 25% case isolation, all scenarios overwhelm ICU by mid-summer. A 10% increase in RLD does not change main conclusions if full attendance of adolescents is avoided. Reducing adolescent contacts by 50% lowers the risk when reopening secondary schools.
• Operational implications: scenarios avoiding ICU saturation imply isolating ~5,000 infected individuals per day at peak, requiring ≥100,000 tests/day in Île-de-France to maintain ≤5% positivity (WHO benchmark), exceeding expected regional capacity at the time.
• Ex-post assessment (May–July 2020): actual school attendance was low (14.5% in pre-/primary; secondary schools remained closed), detection ~10%; the epidemic declined with REXIT = 0.83 [0.81, 0.85]. Fitting indicates ~90% of the population avoided physical contacts, bringing R below 1; altering only physical-contact adoption could reproduce a range from decreasing to increasing trajectories.

The study shows that school reopening increases transmission relative to keeping schools closed, but appropriate protocols can keep the epidemic under control and prevent ICU saturation. Prioritizing younger children (pre-/primary) for in-person instruction is feasible under moderate social distancing and efficient test-trace-isolate, with limited impact on ICU occupancy. Full attendance of adolescents (middle/high schools) is not recommended when community transmission is stable or rising; limiting attendance to 50%, progressive reopening, and reducing adolescent contacts are effective mitigations. Progressive vs prompt reopening among younger children has little differential effect on epidemic risk, suggesting full attendance for this group could be allowed given the short remaining school term. Robust testing, tracing, and isolation are crucial; inadequate case isolation (e.g., 25%) leads to overwhelming ICU demand. The ex-post analysis underscores the critical role of widespread adherence to preventive measures (e.g., avoiding physical contacts), which can reduce R below 1 even with some school reopening, explaining the observed decline after lockdown lifting. These findings inform decisions conditioned on the epidemic trajectory (R, community incidence) and emphasize targeted strategies for secondary education settings to reduce transmission (reduced attendance, cohorting, sanitary measures, staggered schedules, and potential screening).

Model-based scenario analysis for Île-de-France indicates that safe reopening of schools during COVID-19 is possible with priority to pre-/primary levels, limited or progressive attendance in secondary schools, and strong test-trace-isolate combined with moderate social distancing. Full, immediate reopening of all school levels risks exceeding ICU capacity, whereas capping attendance at 50% avoids saturation. Ex-post assessment revealed that high adherence to preventive measures substantially mitigated transmission during reopening. Future research should refine age-stratified transmission parameters (especially distinguishing middle vs high school ages), evaluate reactive and targeted closure strategies, quantify the impact of masks and other school-level interventions, incorporate updated contact patterns and seasonality, and assess implications under emerging variants and varying testing capacities.

• Children’s role in transmission is uncertain; younger children’s transmissibility was explored via scenarios but not empirically determined.
• Mask usage was not explicitly modeled in forward scenarios (only implicitly in ex-post calibration).
• Summer period parameterization is uncertain; child contacts use spring holiday proxies and adult holidays are not modeled.
• No seasonality in transmission was included.
• Contact matrices are from 2012; potential changes in mixing patterns, especially among adolescents, may alter results (a sensitivity with 50% fewer adolescent contacts was explored).
• Reactive school closures and spatially targeted reopening within Île-de-France were not modeled; focus was on reopening conditions for the last two months of the school calendar.
• ICU capacity was assumed at 1500 beds post-first wave; actual capacity and surge capabilities could differ.
• Analysis predates emergence of new variants; variant-specific transmissibility and severity were not considered.
• Calibration assumes stable R during lockdown; although sensitivity to ±10% was tested, other dynamics or reporting artifacts (e.g., ICU transfers) may affect projections.
• Testing and isolation effectiveness assumptions (50% or 25% case isolation with 90% contact reduction) may not reflect operational realities; detection was low (~10%) in ex-post period.