“All models are wrong, but some are useful.” This famous quote by George Box applies to many different forecasts, including models to predict COVID-19 hospital admissions and deaths.
What exactly is a ‘predictive model’ and how does it work for a pandemic virus? A century of studying epidemics has led to a mathematical understanding of different infectious diseases and how they spread in a population. The COVID-19 virus can spread easily from person-to-person, and the epidemic continues until a large fraction of people are immune due to infection or vaccination.
The spread of the virus can be described in different mathematical models. One approach uses computer software to simulate the spread of the virus in a large group of people who are not immune. The models require many inputs such as length of time from exposure to getting sick, average number of new infections caused by a single infected person and death rate. Sometimes these are known and sometimes the researchers have to make their best guess. This can lead to errors in the models.
“A model is a ‘black box’ that relies on an oversimplified model of the real world,” said Dr. Ed Belongia, an infectious disease epidemiologist at Marshfield Clinic Research Institute. “These models can be useful for planning even if the predictions are not accurate. They help define the spectrum of possibilities and support planning for hospital admissions, intensive care units and public health response.”
Many different models are used that try to predict COVID-19 hospitalizations and deaths. No one model is clearly better than all the others. CDC summarizes a variety of state and national models on their website.
Modeling for future outbreaks
A group of Harvard researchers has recently published a model that predicts wintertime outbreaks of COVID-19 will continue in the coming years if (as expected) immunity declines over time after people are infected.
The model predicts that we may face a cycle of repeated ‘lockdowns’ and gradual easing of restrictions to avoid overwhelming hospitals and intensive care units with critically ill patients. Perhaps less extreme measures also will work. A modelling team from the United Kingdom has concluded that wearing a facemask, along with social distancing, may provide an effective way to control COVID-19 without the social and economic disruption of lockdowns.
Mathematical models provide one piece of the COVID-19 puzzle. It is a large puzzle and we still have a long way to go before it is solved.