We recently changed a few of our model parameters. We wanted to share what changed and why.

The epidemiological model we use to track COVID is an SEIR model. An SEIR epidemiology model tracks the flow of a population between four states in relation to a disease (in this case, COVID). Those four states are susceptible (S), exposed (E), infected (I), and recovered (R). Mash those four together and, voila, “SEIR.”

Within the “infected” (I) category, our SEIR model projects how many patients will need to be hospitalized, how many will require ICU treatment, and how many will die. When we first created the model, we chose parameters based on the best available COVID information at the time.

(You might ask, what are “parameters”? Parameters are wonky modeling lingo for a model’s fundamental assumptions. Examples of parameters in an SEIR model: what percentage of people progress into each state of “SEIR”? How long does the average person spend in each of those four states?)

Our knowledge of COVID is always improving; as new data comes in, we update our parameters to fit what we know about the disease.

Given new available information and data, we are revising three different parameters in our SEIR model. The cumulative effect of these three parameter changes is that projected death rates remain about the same and that projected hospitalization rates have become more optimistic (i.e., fewer projected hospitalizations).

Here are the nitty-gritty specifics on all three parameter changes:

1. We have revised our estimate of the COVID hospitalization rate from 4% to 2%. The model’s initial assumed hospitalization rate (based on data from Italy and elsewhere) is approximately double the observed, empirical hospitalization rate in the United States. We have revised the assumed hospitalization rate to better reflect the newer, more accurate data.

2. While the model’s initial assumed hospitalization rate was too high, the model’s assumed death rate has accurately reflected available data. Therefore, as we lower the assumed rate that infected people require hospitalization, we commensurately increase the assumed death rates of patients in the ICU and on ventilators. (Our assumed rate of patients with disease severe enough to be hospitalized but *not* severe enough to go to the ICU has accurately reflected available data.)

3. We shortened the assumed average amount of time that people spend in the hospital, in the ICU, and on the ventilator — all to reflect new data. This is good news in the sense that COVID patients spending less time in the hospital reduces strain on hospitals. Unfortunately, though, one of the reasons COVID patients are spending less time in the hospital is that they are dying more quickly than initial data suggested.

So what is the take away? Taken together, these three parameter changes result in reduced projected COVID hospitalizations and no significant change in projected deaths. Our job as modelers is to continuously edit our model to reflect reality. We hope that a more accurate, up-to-date model will empower decision makers to make better decisions.