There are enough models, we need accurate inputs!
For near term capacity planning we need to report hospitalizations correctly
Multiple data scientists, epidemiologists and computer scientists are building models to forecast how COVID-19 will spread and affect their communities.
Such models require knowing the disease burden and estimate that from confirmed cases, SARS-COV-2 test results, hospitalizations, or deaths. It is widely acknowledged that estimating prevalence from either of these inputs is imprecise.
While it is definitely important to obtain true prevalence, measure transmission dynamics, incubation time, case fatality rate, infection rate and various such parameters for understanding the pathogen we face, for near term planning we can do a lot better relying on hospitalizations. People with mild infections certainly need to be tracked and studied to understand the disease biology, to learn why some are affected severely and others get mild symptoms, and to assess the impact of policy interventions such as shelter-in-place.
However, for the purpose of planning a health system’s response, I argue that counting hospitalizations is the most useful. For effective planning to save lives, we need to know how many people will need a hospital bed (and perhaps a ventilator), tomorrow, in the next few days, and the next week or two. For such planning, a calculation via total cases is an indirect way to arrive at needed beds in a city, county or state.
Projecting needed beds in a region based on counts of hospitalized patients makes the least assumptions, has lower uncertainty, and can serve as the foundation for reliable planning. Stanford Medicine’s tool to project based on hospitalized patient counts is here and a manuscript by Ferstad et al has details.
How can we enable our hospitals to share admitted patient counts widely, and safely, for the public good?
