The Effect of Mobility Restrictions on the SARS-CoV-2 Diffusion During the First Wave: What Are the Impacts in Sweden, USA, France and Colombia

Abstract

Background: The control of human mobility has quickly been targeted as a major leverage to contain the spread of SARS-CoV-2 in a great majority of countries worldwide. Using the anonymised data collected by one of the major social media platforms combined with spatial and temporal Covid-19 data, the objective of this research is to understand how mobility patterns and SARS-CoV-2 diffusion during the first wave are connected in four different countries: the west coast of the USA, Colombia, Sweden and France.Method: To measure mobility patterns during the pandemic, we accessed Facebook mobile users’ data provided within the company’s Data for Good framework through a dedicated platform. Measured every eight hours, they provide the information on how more than the 10 million users moved from a given administrative unit (county in USA, département in France, municipios in Colombia, county in Sweden) to another between two time periods. The number of cases in the unit of origin per week was multiplied by the summed Facebook movements from unit of origin to unit of destination. This thus generated a potential incoming force of infection (FoI) for every unit of destination based on every unit of origin number of cases and incoming mobility. A Generalized linear model (GLM) with a logarithmic Quasi-Poisson distribution was then fitted to the number of cases per week per administrative unit as the response variable with the new number of cases registered in each unit during the previous week, the logFoI from the previous week and their interaction as explanatory variables.Findings: Despite a varying impact of lockdown on mobility reduction in these countries (83% in France and Colombia, 55% in USA, 10% in Sweden), our analyses reveal a high spatial concentration of user’s mobility in all the countries. The Gini index related to the number of incoming individuals per unit was high and stable before, during and after the lockdown (0·95-0·96 in Columbia, 0·74-0·79 in France, 0·72-0·77 in USA and 0·59-0·76 in Sweden). In three of the four countries, the logFoI was an important predictor of the number of cases or infections registered in a week throughout the period. For Colombia the exponential of the regression coefficients related to logFoI was constant, with an average of 1·30 over all the time period. In France, although more variable, the strength of the association was stronger, averaging at 1·47. For these two countries, there is a common temporal tendency with a decrease of the logFoI effect following the first lockdown. In the USA, the effect was very mild, with an average of 1·15 and peaking in June. By comparison the effect was only 1·18 on average for Sweden with confidence intervals often overlapping with 1. Finally, our analyses show that the integration of mobility data considerably improved the epidemiological model (as revealed by the QAIC).Interpretation: Although the impact of the lockdown on the diffusion of SARS-CoV-2 among geographical units was modest, the integration of mobility data considerably improved our understanding of the spatial spread of Covid. Such predictability of mobility patterns and their significant role in viral diffusion could enable optimisation of public health mitigation strategies at a local scale. Administrations can observe changes in mobility patterns, generate vulnerability maps and better locate where to implement disease control measures.Funding Statement: This study was funded by PICS Urb India (CNRS grant).Declaration of Interests: Authors declare no conflict of interests.