Public Mobility Data Enables COVID-19 Forecasting and Management at Local and Global Scales

Ilin, Cornelia; Annan-Phan, Sébastien; Tai, Xiao Hui; Mehra, Shikhar; Hsiang, Solomon M; Blumenstock, Joshua E

Policymakers everywhere are working to determine the set of restrictions that will effectively contain the spread of COVID-19 without excessively stifling economic activity. We show that publicly available data on human mobility -- collected by Google, Facebook, and other providers -- can be used to evaluate the effectiveness of non-pharmaceutical interventions and forecast the spread of COVID-19. This approach relies on simple and transparent statistical models, and involves minimal assumptions about disease dynamics. We demonstrate the effectiveness of this approach using local and regional data from China, France, Italy, South Korea, and the United States, as well as national data from 80 countries around the world. SummaryO_ST_ABSBackgroundC_ST_ABSPolicymakers everywhere are working to determine the set of restrictions that will effectively contain the spread of COVID-19 without excessively stifling economic activity. In some contexts, decision-makers have access to sophisticated epidemiological models and detailed case data. However, a large number of decisions, particularly in low-income and vulnerable communities, are being made with limited or no modeling support. We examine how public human mobility data can be combined with simple statistical models to provide near real-time feedback on non-pharmaceutical policy interventions. Our objective is to provide a simple framework that can be easily implemented and adapted by local decision-makers. MethodsWe develop simple statistical models to measure the effectiveness of non-pharmaceutical interventions (NPIs) and forecast the spread of COVID-19 at local, state, and national levels. The method integrates concepts from econometrics and machine learning, and relies only upon publicly available data on human mobility. The approach does not require explicit epidemiological modeling, and involves minimal assumptions about disease dynamics. We evaluate this approach using local and regional data from China, France, Italy, South Korea, and the United States, as well as national data from 80 countries around the world. FindingsWe find that NPIs are associated with significant reductions in human mobility, and that changes in mobility can be used to forecast COVID-19 infections. The first set of results show the impact of NPIs on human mobility at all geographic scales. While different policies have different effects on different populations, we observed total reductions in mobility between 40 and 84 percent. The second set of results indicate that -- even in the absence of other epidemiological information -- mobility data substantially improves 10-day case rates forecasts at the county (20.75% error, US), state (21.82 % error, US), and global (15.24% error) level. Finally, for example, country-level results suggest that a shelter-in-place policy targeting a 10% increase in the amount of time spent at home would decrease the propagation of new cases by 32% by the end of a 10 day period. InterpretationIn rapidly evolving disease outbreaks, decision-makers do not always have immediate access to sophisticated epidemiological models. In such cases, valuable insight can still be derived from simple statistic models and readily-available public data. These models can be quickly fit with a populations own data and updated over time, thereby capturing social and epidemiological dynamics that are unique to a specific locality or time period. Our results suggest that this approach can effectively support decision-making from local (e.g., city) to national scales.