The limits of human mobility traces to predict the spread of COVID-19

TL;DR

This study uses transfer entropy to evaluate how well mobile phone mobility data predict COVID-19 spread across European regions, revealing limited predictive power and regional differences.

Contribution

It introduces a model-free transfer entropy approach to assess the predictive value of mobility metrics for COVID-19, accounting for regional variations.

Findings

Mobility data often do not significantly predict COVID-19 cases or deaths.

Contact rate measures are more informative than movement data in some regions.

Geographic and demographic factors influence the predictive usefulness of mobility metrics.

Abstract

Mobile phone data have been widely used to model the spread of COVID-19, however, quantifying and comparing their predictive value across different settings is challenging. Their quality is affected by various factors and their relationship with epidemiological indicators varies over time. Here we adopt a model-free approach based on transfer entropy to quantify the relationship between mobile phone-derived mobility metrics and COVID-19 cases and deaths in more than 200 European subnational regions. We found that past knowledge of mobility does not provide statistically significant information on COVID-19 cases or deaths in most of the regions. In the remaining ones, measures of contact rates were often more informative than movements in predicting the spread of the disease, while the most predictive metrics between mid-range and short-range movements depended on the region considered. We finally identify geographic and demographic factors, such as users' coverage and commuting patterns, that can help determine the best metric for predicting disease incidence in a particular location. Our approach provides epidemiologists and public health officials with a general framework to evaluate the usefulness of human mobility data in responding to epidemics.