Impact of non-pharmaceutical interventions, weather, vaccination, and variants on COVID-19 transmission across departments in France

Abstract

Background: Multiple factors shape the temporal dynamics of the COVID-19 pandemic. Quantifying their relative contributions is key to guide future control strategies. Our objective was to disentangle the individual effects of non-pharmaceutical interventions (NPIs), weather, vaccination, and variants of concern (VOC) on local SARS-CoV-2 transmission.

Methods: We developed a log-linear model for the weekly reproduction number (R) of hospital admissions in 92 French metropolitan departments. We leveraged (i) the homogeneity in data collection and NPI definitions across departments, (ii) the spatial heterogeneity in the timing of NPIs, and (iii) an extensive observation period (14 months) covering different weather conditions, VOC proportions, and vaccine coverage levels.

Findings: Three lockdowns reduced R by 72.7% (95% CI 71.3-74.1), 70.4% (69.2-71.6) and 60.7% (56.4-64.5), respectively. Curfews implemented at 6/7 pm and 8/9 pm reduced R by 34.3% (27.9-40.2) and 18.9% (12.04-25.3), respectively. School closures reduced R by only 4.9% (2.0-7.8). We estimated that vaccination of the entire population would have reduced R by 71.7% (56.4-81.6), whereas the emergence of VOC (mainly Alpha during the study period) increased transmission by 44.6% (36.1-53.6) compared with the historical variant. Winter weather conditions (lower temperature and absolute humidity) increased R by 42.2% (37.3-47.3) compared to summer weather conditions. Additionally, we explored counterfactual scenarios (absence of VOC or vaccination) to assess their impact on hospital admissions.

Interpretation: Our study demonstrates the strong effectiveness of NPIs and vaccination and quantifies the role of weather while adjusting for other confounders. It highlights the importance of retrospective evaluation of interventions to inform future decision-making.

Keywords: COVID-19; Non-pharmaceutical interventions; Regression model; Reproduction number; Vaccination; Weather.

Fig. 1

Dynamics of the COVID-19 epidemic and timing of non-pharmaceutical interventions across departments in metropolitan France, week 11–2020 to week 20–2021. A Map of departments, colored by region. The two departments marked by an asterisk were excluded, as well as Corsica (not shown), due to missing covariates. B Time series of new daily hospital admissions by department (logarithmic scale). C Temporal evolution of the reproduction number R by department and by week, overlaid with the timeline of non-pharmaceutical interventions. Boxplots feature the 2.5th, 25th, 50th, 75th, and 97.5th percentiles. Non-pharmaceutical interventions are shown taking the Rhône department as an example. D Timing of the third lockdown by department. E Timing of the 6 pm curfew by department. F Start dates of 2021 winter holidays by department

Fig. 2

Goodness-of-fit and estimated effects of covariates included in the multivariable model on the reproduction number R (in percentage of variation). A Trajectory of R estimated by the full model for an average department (red) compared to mean R (black). B Effects of linear and categorical covariates. Of note, the effect shown for first-dose vaccine coverage and the proportion of VOC corresponds to a covariate value of 100% (i.e., reflecting a fully vaccinated population and maximum prevalence of variants). C Non-linear effect of temperature. D Non-linear effect of absolute humidity. E Estimated seasonality of COVID-19 based on average temperature and absolute humidity observed over 1981–2010 in metropolitan France, after adjusting for other covariates. For C, D and E, fitted lines and their 95% confidence intervals show the estimated percentage of variation in R with respect to a baseline set to the value at the trough of the corresponding curve for weather variables. For C and D, the range of the x-axis is determined by the 2.5–97.5th percentile of the weather variable distribution

Fig. 3

Projected trajectory of R and hospital admissions from January 11, 2021, under different scenarios: the full model, a model without vaccination and a model without VOC. A Mean trajectory of R. B Number of national-level weekly hospital admissions (summed across all departments)