Defining high-value information for COVID-19 decision-making

Joshua A Salomon

The first generation of models generated forecasts that focused public attention on worst-case scenarios in the absence of decisive policy action. Despite substantial uncertainties surrounding the biology of the disease and intervention costs and effects, most models arrived at the same qualitative finding: left unchecked, the spread of COVID-19 would overwhelm hospital capacity to care for acute and critical cases, resulting in intolerable potential morbidity and mortality. These projections underscored the imperative for bold and immediate measures to slow the spread of infection. [6] [7] [8] Policymakers will continue enlisting models to forecast the trajectory of COVID-19 in the coming weeks. 6, 9, 10 As many communities have implemented shelter-in-place orders and other non-pharmaceutical interventions (NPIs), vital questions now include how to plan for short-term and long-term health system needs and when and how policymakers might begin to modify or relax the imposition of NPIs. Answers to these questions will be much more sensitive to underlying uncertainties. They will require a different kind of modeling, focused less on conventional forecasting and more on program evaluation, resource allocation, careful characterization of uncertainty, and optimal inference from limited information.