"In September 2020, the Institute for Health Metrics and Evaluation (IHME) at the University of Washington in Seattle, a prominent modeling group, released a COVID-19 forecast for the fall and winter. Their base case prediction was 410,000 deaths by January 1, 2021, a winter surge that would more than double the death toll. The estimate sparked a flurry of news articles—some nervous, others skeptical. Modeling experts commented that it was nearly impossible to make specific reliable predictions so far into the future; independent models aggregated by the Centers for Disease Control and Prevention (CDC) only include four-week death projections. A modeler with high-forecast accuracy during the summer of 2020 argued against both the IHME’s approach and its findings, proffering an “educated guess” of around 250,000 deaths by January with a 95 percent confidence interval from 210,000 to 300,000.
"It turned out that the IHME’s estimates were high, but the prediction of a considerable winter wave was correct. On January 1, 2021, the death count was 339,394; it passed 410,000 23 days later and currently stands above 600,000. At the time, however, it was unclear how policy makers were to interpret highly divergent predictions. Now, amid a fourth wave despite highly effective vaccines, we again face questions of how to best anticipate and address further COVID-19 resurgences and prepare for future pandemics. With expansions in data and modeling capacity promised by the recent launch of the Center for Forecasting and Outbreak Analytics, we believe that models can inform this process but will do so more effectively if policy recommendations are linked to observable, leading indicators. Here, we describe how this approach would build on current thinking as well as the necessary conditions that would make such an enterprise most useful for policy makers."