How well does SARS-CoV-2 spread in hospitals?

Covid-19 poses significant risk of nosocomial transmission, and preventing this requires good estimates of the basic reproduction number R0 in hospitals and care facilities, but these are currently lacking. Such estimates are challenging due to small population sizes in these facilities and inconsistent testing practices over time. We estimate the patient-to-patient R0 and daily transmission rate of SARS-CoV-2 using data from a closely monitored hospital outbreak in Paris, March-April 2020. We use a realistic epidemic model which accounts for progressive stages of infection, stochastic effects and a large proportion of asymptomatic infections. Innovatively, we explicitly include changes in testing capacity over time, as well as the evolving sensitivity of PCR testing at different stages of infection. We conduct rigorous statistical inference using iterative particle filtering to fit the model to the observed patient data and validate this methodology using simulation. We provide estimates for R0 across the entire hospital (2.6) and in individual wards (from 3 to 15), possibly reflecting heterogeneity in contact patterns or control measures. An obligatory mask-wearing policy introduced during the outbreak is likely to have changed the R0, and we estimate values before (8.7) and after (1.3) its introduction, corresponding to a policy efficacy of 85%.