Quantifying the impact of social activities on SARS-CoV-2 transmission using Google mobility reports

We developed a state-space model to investigate which social behaviours had biggest impact on the spread of SARS-CoV-2. The analyses were based on reported hospitalizations, together with information on vaccinations, weather data, virus strains and, most importantly, Google mobility reports on 4 different types of social activities. While our new approach is general, we studied Sweden and Norway on a regional level over 75 weeks, and the major regions of Berlin and Bavaria in Germany over 10 months. Most results are shared for all three countries: Activity in four social settings explain between 40-60% of all infections; Public transport appears as an important setting for infections in all countries; and the transmission potential drops by 40-50% during the summer as compared to the winter peak. However, the analyses for Germany differ in that Retail and recreation is the other setting dominating transmission whereas it is contacts at the Workplace in Norway and Sweden, showing how our model is able to adapt to specific cases. Transmissions not captured by the Google data may happen in other settings, in particular in households. The statistical model has a deterministic time and region specific transmission rate with an additive component for the four Google settings, and a multiplicative part taking seasonality and circulating virus strains into account. Inference is performed in a Bayesian setting using Stan.