In the past few weeks, we have become all too familiar with the phrase "flatten the curve."
The idea is that by staying home and staying apart, we'll slow the spread of the coronavirus (COVID-19) and leave enough room in our hospitals to care for the severely ill. If we do it right, the "curve" — the new cases of the virus — will resemble a broad hill more than a steep mountain.
So who is drawing those hills and mountains?
The estimated cases are predicted by scientists who build models based on the latest information we have about COVID-19 and its spread. To learn more about the predictive modeling developed at Ohio State and around the world, we spoke to Michael Oglesbee, director of Ohio State's Infectious Diseases Institute.
Predictive modeling serves two very important purposes.
- Predictions allow health care facilities to anticipate needs related to managing patients suffering from severe COVID-19 health outcomes. These needs include hospital beds, intensive care facilities and ventilators, and personal protective equipment for hospital staff.
- Predictions are used to confirm that our strategy of distancing is indeed having a significant impact on reducing the number of COVID-19-symptomatic individuals and the length for which such measures must be sustained.
The curve we are talking about is the daily count of new cases when a virus is introduced into a population with no protective immunity (i.e., they are immunologically naive). If there are no measures to suppress spread, that peak can come early and be of very high magnitude, a scenario that can readily overwhelm our health care system.
The goal of our distancing measure is to suppress that initial surge in cases (i.e., flatten the curve) so that we can better manage those with COVID-19. An outcome of this is that the peak in cases will be delayed, but it is currently unknown as to when that will occur. The answer lies in real-time monitoring of the surveillance data.
Data comes from the Ohio Department of Health and is based on COVID case counts and other characteristics reported throughout the state. Test results are generated from the Ohio Department of Health and approved testing centers, several of which have recently come online in the state of Ohio.
Different models make different assumptions and take different approaches. Ours is a contact network model that assumes differences in potential virus exposure (and thus the potential to contract COVID-19) within and between communities, and these differences can reflect locations within the state.
A model used by the Cleveland Clinic works on the idea of a larger susceptible population, in which geographical location does not play a significant role in predicting the likelihood of contacting an infected individual. This model predicts a large number of infected symptomatic individuals during the initial COVID-19 surge in Ohio.
A model from the University of Washington works on the number of severe COVID-19 outcomes in hospitalized patients, and then works backward (statistically) to predict what a surge in Ohio COVID-19 cases might look like. These results are more closely aligned with results of our contact network predictive model.
The biggest challenge to predictions has been the lack of surveillance data — the ability to test for the presence of virus in individuals exhibiting symptoms of COVID-19. Due to limitations in capacity, we only have been able to test those most at risk for severe outcomes of COVID-19 or the need to focus on those health care workers in the front line. Testing capacity is increasing as new testing centers come online, but there has been a backlog in reporting results of tests.
As a result, the flow of information has been uneven, and this, in turn, makes the predictions fluctuate. Those fluctuations impact predictions on the timing of the peak as well as the magnitude. As these data pipelines become more even, and we have greater confidence that the data received reflects the reality for the population at risk, then we can place greater confidence in the model predictions. Fortunately, our confidence increases daily.
But the point to emphasize is that the modeling suggests that distancing measures are making a difference. Results predict (and data show) a significantly reduced incidence of infected symptomatic patients compared to predictions where no distancing measures are implemented. The question is one of magnitude, and it is increasingly looking like our distancing measures are having a positive impact beyond expectations.
This is a bit like watching the stock market. The trend is what you need to focus upon. In our predictive model, we have been examining just that — the trend that emerges from daily fluctuations in reported cases. Our network model is a learning model, comparing yesterday’s data to today’s reality, and making appropriate adjustments in predictions.