A team of researchers has developed an algorithm that analyzes data such as tweets and Google searches to forecast coronavirus outbreaks.

The team, led by Harvard scientists Mauricio Santillana and Nicole Kogan, presented the algorithm in a paper submitted last Thursday to the arXiv scientific paper repository. The New York Times, which interviewed Santillana about the project, reported that the model could potentially forecast outbreaks about two weeks before they occur.

The researchers’ goal is to give health officials the ability to detect upticks in coronavirus cases earlier so they can take more effective action to curb the pathogen’s spread. Their approach to forecasting, which has not yet been peer-reviewed, could be a step toward enabling that kind of visibility. “Such a combined indicator may provide timely information, like a ‘thermostat’  in a heating or cooling system, to guide intermittent activation, intensification, or relaxation of public health interventions,” Santillana, Kogan and their colleagues wrote in the arXiv paper.

The algorithm makes forecasts by analyzing five different kinds of data. These include Google Trends statistics on what keywords people are looking up, doctors’ search queries on the UpToDate medical knowledge platform, coronavirus-related tweets and anonymized smartphone mobility data. The algorithm also uses readings from San Francisco-based Kinsa Inc.’s smart thermometers.

The researchers determined that those data sources, when evaluated in aggregate, potentially can provide early insight into coronavirus outbreaks. They tested their model by analyzing logs from the period leading up to the surge in coronavirus cases that New York experienced mid-March.

They discovered a “sharp uptrend” in coronavirus-related tweets more than a week before cases began to spike, the Times reported. They also found a similar increase in relevant Google searches and Kinsa measures several days ahead of time. 

But the researchers cautioned that the project is still in an early stage. “These efforts represent an initial exploratory framework, and both continued study of the predictive power of digital indicators as well as further development of the statistical approach are needed,” they wrote in the paper.

Multiple academic institutions and private companies are working to develop better ways of tracking coronavirus outbreaks, in some cases by collaborating with one another. One of the most notable initiatives in this area is the collaboration between Facebook Inc. and Carnegie Mellon University. In April, Facebook launched a freely accessible COVID-19 symptom map as part of the initiative. 

Image: Pixabay