Weeks following the change in RPE are highlighted in grey. Community incidence (total cases per week) is shown for the East of England, UK, with raw data shown in Figure 1–source data 1 .
To further quantify the risk of infection for HCWs working on red and green wards, we generated a simple mathematical model. According to this model, the total risk of infection is divided into a risk from community-based exposure, and a risk from direct, ward-based exposure to patients. The risk from direct exposure on red wards was allowed to vary upon the introduction of FFP3 respirators, and was fitted to a maximum likelihood model. Inferred parameters and their confidence intervals are shown in Table 2 . Our model produced a qualitatively close fit to the observed numbers of cases (Figures 3A-B ).
The inferred risk of direct infection from working on a green ward was low throughout the study period, and consistently lower than the risk of community-based exposure, which increased in proportion to rising levels of community incidence (Figure 3C ). By contrast, the risk of direct infection from working on a red ward before the change in RPE was considerably higher than the risk of community-based exposure, and approximately 47-fold greater than the corresponding risk from working on a green ward (confidence interval [7.92, ∞]). Thus, whilst almost all cases on green wards were likely caused by infection in the community, cases on red wards at the beginning of the study period were attributed mainly to direct, ward-based exposure (Figures 3D-E ). Critically, our model further suggests that the introduction of FFP3 respirators provided 100% (confidence interval [31.3%, 100%]) protection against direct, ward-based COVID-19 infection (Table 2 , r2/r1).