Survey Launched at Pandemic’s Onset Proves It Can Function as Early Warning System for Changes in Outbreaks

Compared to local governments’ attempts at predicting COVID-19 case count numbers, a global Facebook survey launched by University of Maryland researchers last spring to identify and track disease symptoms gave more accurate results in 77% of the 114 countries and territories included in the survey.

The team from the Joint Program in Survey Methodology (JPSM), College of Information Studies and Department of Geographical Sciences successfully trained a machine learning algorithm to identify which COVID-19 symptoms were most often associated with a positive result—critical knowledge during the early days of the pandemic when telltale symptoms were still being identified. They used that information, plus additional responses, to make close-to-reality predictions about case counts at that time.

The findings from the University of Maryland Global COVID Trends and Impact Survey (UMD-CTIS) were presented in a paper published today in the Proceedings of the National Academy of Sciences.

“As intended, this is an early warning system for changes in outbreaks distributed geographically over time,” said Professor Frauke Kreuter, JPSM director, whose team worked with Facebook, the World Health Organization and researchers at other universities to develop the questionnaire. “It was reassuring to see we don't have to wait for someone to show up in a hospital, that there are other ways of collecting data very quickly and globally to get a read on what is going on.”

The researchers also learned that “a pretty strong predictor of COVID-19 cases isn't just about individual symptoms, but symptoms [respondents] are aware of in their local community,” she said, with answers to a question about whether survey-takers personally knew anyone with symptoms most closely matching benchmark case data.

With new variants—each potentially with unique symptoms—cropping up regularly, and unequal vaccine availability from country to country, accurate, real-time insights like those from UMD-CTIS are essential for many policymakers, the researchers say.

Survey findings are easily accessible and actionable via a user-friendly dashboard created by Kathleen Stewart, a geographical sciences professor, and her team.

“As the pandemic has evolved so has the survey—from variables looked at to questions asked,” said Stewart. “Very quickly after an individual completes the survey, we go through a set of processes to make that data available as quickly as possible. We offer data at both the regional and country level, and because it is a daily survey with machine learning models, local experts are able to take advantage of those granularities to support their decisions.”

With the rise of the omicron variant, Kreuter said it’s unclear if the questionnaire will need to morph as well. “There are a lot of methodological questions yet to come.”

This article by Rachael Grahame originally appeared in Maryland Today. The illustration is courtesy of iStock.