Google Cloud, Harvard release COVID-19 Public Forecasts
Google Cloud, in partnership with the Harvard Global Health Institute, released public models that can predict new COVID-19 cases, deaths and hospitalizations 14 days out.
Google Cloud, in partnership with the Harvard Global Health Institute, released COVID-19 Public Forecasts -- models that offer predictions of COVID-19's impact in the U.S. at a state and national level over the next 14 days.
The models, which can make predictions including deaths, new cases, and hospitalizations, could help healthcare organizations and researchers better predict disease spread and target care, said Cynthia Burghard, research director at IDC Health Insights.
Predictions for healthcare organizations
Public health organizations, for example, could use the COVID-19 Public Forecasts to focus educational or prevention programs in areas showing higher impacts, make screening tools from the Centers for Disease Control and Prevention available, ensure financially vulnerable populations have needed resources, set up testing sites, and assign contact tracers, Burghard said.
Health insurance payers may do similar education and prevention programs and outreach to vulnerable members. Care providers, meanwhile, could use the models in similar ways, or use them as additional data points to forecast demand and availability for medical equipment, such as beds and ventilators, and personal protective equipment, Burghard continued.
Google unveiled the tool Aug. 3.
Developed with a machine learning, time series forecasting approach, COVID-19 Public Forecasts are trained on public data from sources including Johns Hopkins University, Descartes Labs and the United States Census Bureau.
They are freely available on Google's Big Query cloud data warehouse platform, on the tech giant's Data Studio Dashboard, and as downloadable data files. Google Cloud and the Harvard Global Health Institute will continuously update the models to ensure accuracy.
COVID-19 Public Forecasts, created by Google Cloud and the Harvard Global Health Institute, can predict coronavirus cases, deaths and hospitalizations.
Potential problems
Still, in an accompanying user guide, Google Cloud notes some potential limitations with the models, including lags in how often training data sources update their data.
We won't know accuracy until the predictions have been tested.
Cynthia BurghardResearch director, IDC Health Insights
Not only that, but the forecasts can't account for outbreaks that are a result of misbehavior, Burghard said, such as large gatherings without social distancing or mask requirements.
"Increasingly, we are hearing more about the outbreaks from misbehavior or exposing people by opening up geographies too early," Burghard said.
These incidents, impossible to predict, can spark many new cases and throw off model predictions.
"We won't know accuracy until the predictions have been tested," Burghard said.
