COVID-19 Forecasting

A Discussion with Ryan McCorvie

by Christina Schrier, June 24, 2020

During the past six years statistician Ryan McCorvie, has been researching random processes with cascades, crashes or contagion. McCorvie came to disease modeling in a rather circuitous route — before pursuing his Ph.D. in statistics his background was in finance. “It turns out that some of the same techniques used to calibrate forecasts for currency exchange rates can also be applied when you are trying to forecast ICU bed demand,” according to McCorvie. “There are general statistical principles which apply to many situations.”

One of the first goals in epidemiology is to gather the basic data about the disease, such as the number of cases, demographic data about those afflicted, and the amount of testing being done. Modeling takes the next step, integrating the raw data with mathematical models to gain additional insights. Models outputs have been featured prominently in policy discussions, such as Trump’s Coronavirus Task Force briefing which incorporated projections from the University of Washington’s IHME group.

According to McCorvie, there are three major areas where modeling can provide an advantage. First is in determining the current growth rate of the disease, something which is sometimes called a “nowcast” to distinguish it from forward-looking models. For infectious disease, the current growth rate is captured using a statistic called “R.” “This represents the speed at which the disease is spreading. ‘R’ is the number of people infected by an individual. In this way, an ‘R’ value above ‘1’ corresponds to exponential growth, whereas an ‘R’ below ‘1’ corresponds to a disease which is dissipating,” states McCorvie.

Secondly, forecasting can help in near-term planning. Models can project the utilization of limited resources, like hospital beds or personal protective equipment. These types of projections are precise and quantitative, since policy makers need to know what kind of planning and reallocation is needed. Finally, model results can be used to simulate different policy approaches and provide needed insights. Usually these planning models are longer term, and the results are considered holistically rather than as precise quantitative estimates. For example, one can simulate different approaches to contact tracing, or different schemes for vaccination, to evaluate their relative effectiveness.

“One of the hardest problems is how to accommodate and communicate uncertainty,” says Mr. McCorvie. Forecasting is a tricky business. To be accurate, forecast modeling depends on two primary criteria: accurate data and some historical behavioral perspective. Neither are completely available with COVID-19 modeling. While there is an abundance of data it is not gathered and stored in consistent formats. Collected data can vary country-by-country and even state-by-state. This is exacerbated by a lack of historical information. In short, we must rely on a very short history and many unknowns when it comes to how the virus behaves and is transmitted. Any model must try to compensate for these uncertainties and inconsistencies. Weather forecasters are long familiar with this problem, as most everyone is used to the cones of probability that accompany forecasts about hurricanes. We have all seen the spaghetti art diagrams that result from variations between hurricane path forecast models. Similar principles apply to forecasts of infectious diseases. Sources of uncertainty, including the fact that key model inputs are estimates instead of being precisely known, and uncertainty stemming from the intrinsic random nature of many models, must be included in any forecast.

By bringing together experts from various fields, and asking them to collaborate with similar data groups around the country, the hope is that predictive modeling can become more accurate. “That is why I wanted to get involved,” McCorvie says. “I know how important accurate forecasting will be to how we end the pandemic. I hope our efforts help people find their way through this and maybe even save lives.”

(copied from seekerstime)