Google Cloud AI and Harvard Global Health Institute Collaborate on new COVID-19 forecasting model
Product Manager, Google Cloud AI
The COVID-19 pandemic has had a tremendous impact on the world, from changing the way we live to driving extraordinary acts of human compassion. Nowhere have both the disruption and perseverance been more evident than among the front-line workers who continue to respond tirelessly.
In partnership with the Harvard Global Health Institute, Google Cloud is releasing the COVID-19 Public Forecasts to serve as an additional resource for first responders in healthcare, the public sector, and other impacted organizations preparing for what lies ahead. These forecasts are available for free and provide a projection of COVID-19 cases, deaths, and other metrics over the next 14 days for US counties and states.
The COVID-19 Public Forecasts are trained on public data such as those from Johns Hopkins University, Descartes Lab, and the United States Census Bureau and will continue to be updated with guidance from the Harvard Global Health Institute.
“The COVID-19 Public Forecasts model produces forecasts at the critical jurisdiction of public health action—the county. Coupled with the work of the Harvard Global Health Institute’s county-level COVID-19 Suppression Metrics, the COVID-19 Public Forecast Model will allow for targeted testing and public health interventions on a county-by-county basis. By providing accurate, timely predictions of cases, infections, hospitalizations, and deaths to both policy makers and the general public, it will enhance our ability to understand and respond to the rapidly evolving COVID-19 pandemic," said Dr. Thomas Tsai, surgeon and health policy researcher in the Department of Surgery at Brigham and Women's Hospital and in the Department of Health Policy and Management at Harvard T.H. Chan School of Public Health.
Alongside other data sources, the COVID-19 Public Forecasts can be a helpful resource for those at the front lines of responding to this pandemic who are seeking to better understand and prepare for the progression of COVID-19 in their region. For example, healthcare providers can incorporate the forecasted number of COVID-19 cases as one datapoint in resource planning for PPE, staffing, and scheduling. Similarly, state and county health departments can use the forecast of infections over the next two weeks to help inform their testing strategy and help identify areas at risk of new outbreaks.
“As healthcare providers, the ability to ever more accurately predict the evolution of this pandemic is vital to our ability to prepare for, and manage, the COVID-19 crisis,” said Dr. Edmund Jackson, Chief Data Officer at HCA Healthcare. “Having Google bring their unique compute and AI prowess to better answering this question is enormously helpful. We are excited to be part of this work.”
To generate the COVID-19 Public Forecasts, Google Cloud researchers developed a novel time series machine learning approach that combines AI with a robust epidemiological foundation. By design, this new model is trained on public data and leverages an architecture that allows researchers to dive into the different relationships that the model has learned to better interpret why it makes certain forecasts. We hope that these measures not only help the public understand how the model works, but can also enable further innovation in infectious disease modeling.
The COVID-19 Public Forecasts are free to query in BigQuery or to download as CSVs (state forecast CSV and county forecast CSV). Additionally, they are available through our Data Studio dashboard and as part of the National Response Portal. We are also publishing a full explanation of the novel methodology and the datasets used in our White Paper and User Guide. As with any forecasts, the COVID-19 Public Forecasts have limitations that should be carefully considered before being used to inform decisions. In order to download or use the forecasts, users must agree to the Google Terms of Service.
Google is committed to a core set of AI principles. In developing the COVID-19 Public Forecasts, we paid close attention to the disproportionate impact the disease has had and how that would impact our adherence to these principles, particularly principle #2: “Avoid creating or reinforcing unfair bias.” CDC research has shown that communities of color in the United States have been the hardest hit by COVID-19 with disproportionately high rates of cases and deaths. Our team has conducted a comprehensive fairness analysis to investigate how that disproportionate impact affects the accuracy of our forecasts and how they should be interpreted. We encourage all users who intend to make decisions in part based on the COVID-19 Public Forecasts to closely review the Fairness Analysis. Additionally, we call for an open dialog among public health officials and the AI community in how to address these inequities and measure how their impact may appear in various AI models.
We are excited to focus Google Cloud’s commitment to innovation in AI to help those on the front lines of the COVID-19 response. Learn more about the COVID-19 Public Forecasts from our User Guide and White Paper, or get started with the data now in BigQuery, Data Studio dashboard, the National Response Portal, or via the CSV data (state, county).