8 April 2020 Blog Post: Modeling COVID-19
There is an adage in Epidemiology which is as follows: “all statistical models are wrong, but some are useful.”
Statistical models of infectious disease transmission come in a variety of forms, from highly complex models that need a range of experts to create and maintain them (the “jet aircraft” model) to a simple model that can be easily understood, adapted and maintained (the “toy” model).
Formulating any model is a trade-off between three important and often conflicting elements: accuracy, transparency, and flexibility (is this starting to sound like Politics or what?). Accuracy is vital. Transparency can be lost as we successively build upon a base model and attempt to assess the role of each component plus interactions between components. Transparency can be in direct opposition to accuracy. And flexibility is important to adapt to new situations.
Models have two distinct roles: prediction and understanding. But models also have limitations because it is impossible to build a fully accurate model. There is always some quirk of human behavior or disease transmission that is unknown or even unknowable.
So with that less than transparent introduction, we today have an ‘updated’ model from The Institute for Health Metrics and Evaluation (IHME) at the University of Washington School of Medicine. It projects 81,766 deaths in the US by August 4th, 2020. Peak mortality is expected to be on April 16th with 3,130 deaths (more than 250% higher than cardiovascular disease which is the usual leading cause of daily death in the US). We remain 36,654 hospital beds and 16,323 ICU beds short of our expected peak demand. See: https://covid19.healthdata.org/united-states-of-america?view=cumulative-deaths&tab=trend
Here’s the fine print. First. these projections assume that social distancing measures will continue until the end of May 2020. Yet we still have 8 states that do not have such orders and, nationally, distancing measures are only expected to remain in place until the end of April. The model does not account for the inevitable surge of cases in states not applying such measures, nor does it account for our current reality which is a relaxing of control measures on April 30th.
When looking at the model curve itself there is a distinct flattening of COVID-19 deaths beginning in mid-May. This runs counter to all coherent discussions thus far which have pointed to inevitable coronavirus ‘flare ups.’ These will need to be quickly identified, quarantined and extinguished. This model does not account for such events.
Lastly, the authors have a very wide range of their August 81,766 mortality estimate – ranging from 49,431 to 136,401. So up to 32,335 fewer but potentially 54,635 more.
So this model too while useful, is wrong. Its inherent limitations are best summed up by Governor Cuomo of New York (in his state IHME’s predictions have been spot on). He said there is a “danger” in being “over-confident.” Other entities have made that error, he said, “and we’re not going to make that mistake.”
𝗦𝗶𝗴𝗻 𝗨𝗽 𝗳𝗼𝗿 𝗢𝘂𝗿 𝗡𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿
by submitting this form you indicate you have
of Use. Please contact us to for us for more