A mixture of mobility and meteorological data provides a high correlation with COVID-19 growth in an infection-naive population: a study for Spanish provinces

• Published in: Frontiers in public health Vol. 12; p. 1288531
• Main Authors: Conesa, David, López de Rioja, Víctor, Gullón, Tania, Tauste Campo, Adriá, Prats, Clara, Alvarez-Lacalle, Enrique, Echebarria, Blas
• Format: Journal Article
• Language: English
• Published: Switzerland Frontiers Media S.A 07-03-2024
• Subjects: COVID-19; COVID-19 - epidemiology; epidemiology; Humans; meteorological data; mobility; Multivariate Analysis; Public Health; statistical correlations; Temperature; COVID-19; epidemiology; meteorological data; mobility; statistical correlations
• ISSN: 2296-2565; 2296-2565
• DOI: 10.3389/fpubh.2024.1288531

We use Spanish data from August 2020 to March 2021 as a natural experiment to analyze how a standardized measure of COVID-19 growth correlates with asymmetric meteorological and mobility situations in 48 Spanish provinces. The period of time is selected prior to vaccination so that the level of susceptibility was high, and during geographically asymmetric implementation of non-pharmacological interventions.
We develop reliable aggregated mobility data from different public sources and also compute the average meteorological time series of temperature, dew point, and UV radiance in each Spanish province from satellite data. We perform a dimensionality reduction of the data using principal component analysis and investigate univariate and multivariate correlations of mobility and meteorological data with COVID-19 growth.
We find significant, but generally weak, univariate correlations for weekday aggregated mobility in some, but not all, provinces. On the other hand, principal component analysis shows that the different mobility time series can be properly reduced to three time series. A multivariate time-lagged canonical correlation analysis of the COVID-19 growth rate with these three time series reveals a highly significant correlation, with a median R-squared of 0.65. The univariate correlation between meteorological data and COVID-19 growth is generally not significant, but adding its two main principal components to the mobility multivariate analysis increases correlations significantly, reaching correlation coefficients between 0.6 and 0.98 in all provinces with a median R-squared of 0.85. This result is robust to different approaches in the reduction of dimensionality of the data series.
Our results suggest an important effect of mobility on COVID-19 cases growth rate. This effect is generally not observed for meteorological variables, although in some Spanish provinces it can become relevant. The correlation between mobility and growth rate is maximal at a time delay of 2-3 weeks, which agrees well with the expected 5?10 day delays between infection, development of symptoms, and the detection/report of the case.