Want to know where the coronavirus pandemic is heading? Here’s another website to check: Experts at Harvard and Google have collaborated to come up with a 14-day look at what’s ahead.
The forecast, which is constructed with the help of artificial intelligence techniques, projects COVID-19 cases, deaths, and other metrics over the next 14 days for states and counties.
The forecasts are intended to “serve as an additional resource for first responders in healthcare, the public sector, and other impacted organizations preparing for what lies ahead,” according to a Google Cloud blog release.
The Massachusetts forecast for Aug. 4 though Aug. 17 predicts the state’s death tally will rise to 8,880 by Aug. 17, which is similar to the rough estimate of 8,862 created by the University of Massachusetts ensemble model.
Google Cloud said it was releasing the forecasts in partnership with the Harvard Global Health Institute.
The model combines artificial intelligence techniques “with a robust epidemiological foundation,” Google said. The model uses a variety of publicly available data.
“Data is one of the greatest assets of the modern era, including for healthcare. We aim to exploit the abundance of available data online to generate more accurate COVID-19 forecasts,” the researchers said in a white paper explaining their methods.
The model “integrates machine learning” into a well-known type of epidemiological model in which a population is separated into groups such as Susceptible, Exposed, Infected, and Recovered, and the progress of people transitioning from one group to the other is estimated, the white paper said.
Dr. Thomas Tsai, a surgeon and an assistant professor in health policy and management at Harvard Chan school of public health, who worked with Google researchers on the project, said the forecast “provides actionable insights at the local, county level,” which is crucial because the pandemic has “hyperlocal effects.”
“There aren’t a lot of other predictive models on the county level” and the Google Cloud model had “improved performance,” compared with the others, he said.
He said the model is expected to be updated daily. It only looks ahead 14 days, he said, because researchers were striking a balance between “usefulness and accuracy.”
He noted that if the model is really successful, it will end up being wrong because government officials will take steps, ordering “non-pharmaceutical interventions,” like shutdowns or masking, for example, if they see a grim future ahead.
“It’s a sort of self-defeating prophecy,” he said.
Tsai said the model could be used to “aid both public health officials in planning for testing and surveillance efforts, policymakers on the need for non-pharmaceutical interventions, and health systems and hospital leaders in planning for potential surges of cases.”
Martin Finucane can be reached at firstname.lastname@example.org.