JMIR Publications

ABSTRACT

Background:

A novel coronavirus disease (COVID-19) sparked significant individual and societal health concerns worldwide. As a result, policy makers and health care experts all over the world searched for ways to mitigate the spread of this new disease by designing and implementing population-level public health interventions such as stay-at-home recommendations. These interventions are economically costly and thus, should be only implemented in moments when disease activity may increase uncontrollably. New digital data sources such as geo-social media posts (posts with an explicit geo-reference), have shown great promise to anticipate these moments of potential healthcare crises. However, their ability to serve as epidemiological early warning signals has only been assessed during short time frames and frequently in coarse spatial resolutions. Furthermore, the extent to which local political beliefs might influence geo-social media data's early warning capabilities has not yet been extensively characterized.

Objective:

To assess how the epidemiological early warning capabilities of geo-social media posts for COVID-19 vary over time and across fine spatial regions with diverging political beliefs.

Methods:

We classified U.S. counties into three different political clusters, namely Democrat, Republican and swing counties, based on voting data from the last six federal election cycles. Across these three political clusters, we compared the epidemiological early warning capabilities of geo-social media data for six consecutive waves of COVID-19 from February 2020 to April 2022. More concretely, we examined the early warning capabilities in terms of the temporal lag, i.e., the number of days by which signals of geo-social media posts preceded COVID-19 cases and the respective correlation between their time series.

Results:

The results of this study illustrate that the number of days, by which geo-social media posts preceded COVID-19 cases, and the correlation between their time series differed across political clusters and individual waves of COVID-19. We observed signals in geo-social media data to precede increases in COVID-19 cases on average 6.4 days early for Republican (21 days) counties than for Democrat counties (14.6). In general, geo-social media posts exhibited early warning capabilities for COVID-19 cases in five out of six epidemiological waves across all political clusters. The results further elucidated challenges for early warning such as a decrease over time in the number of days that geo-social media posts preceded COVID-19 cases for Democrat and Republican counties. Additionally, a decrease in signal strength of geo-social media data over time and a susceptibility towards trending topics might pose obstacles for epidemiological early warning systems employing geo-social media data.

Conclusions:

The results of this study provide an in-depth understanding of strengths, weaknesses, and future paths for improving geo-social media based epidemiological early warning through accounting for differences across political beliefs. Clinical Trial: Spatio-temporal epidemiology; geo-social media data; digital disease surveillance; political polarization; epidemiological early warning; digital early warning