Today, Covid Act Now’s U.S. Interventions Model launched “inference projections” at the state level and for select counties with sufficient data. You can see inference projections on state graphs at Covid Act Now by clicking on the blue label “projected based on current trends.”

This is a big deal!

Previously, the U.S. Interventions Model projected two static COVID hospitalization scenarios for each state — a best-case and a worst-case scenario — based on limited existing data.

Now, **we are using actual numbers of cases**, hospitalizations, and deaths in each state to infer (hence “inference projections”) the current rate of virus spread for each state.

These actual data are encompassed in the new, wonky-sounding metric: “R(t).” (R(t) is a variation on the all-important epidemiology metric, R_{0}, aka a disease’s “basic reproduction number.”) R(t) quantifies the rate at which a virus — in this case, COVID — spreads, based on **actual COVID case, hospitalization, and mortality data since interventions have been implemented**. R(t) is different for each state, because the virus is spreading more quickly in some states and more slowly in others.

Above is an internally-generated — and very complicated looking! — example plot of cases per day (represented by blue dots and blue line), total current hospitalizations* (represented by green), and deaths per day (represented by red) from New Jersey.

Up until April 22, the date this chart was generated, we plotted reported COVID data (those blue, green, and red dots), along with confidence intervals (the vertical lines extending above and below each dot).

We then fit three curves (again, in blue, green, and red) which are generated by our model to each of the three types of plotted data.

**R _{0} metric: **The rate of COVID growth

*before*public health interventions were implemented is represented by the metric R

_{0}(bolded to the right hand side of the graph, and pronounced “R-naught”), which is, in effect, the virus’s “true” rate of infection, galloping along relatively uninhibited and without public health interventions to slow its spread.

**R(t) metric: **The rate of COVID

growth *after* protective measures were implemented is represented by the variable R(t) (also referred to in the right hand side of the graph as Reff), which is the virus’s effective rate of infection after New Jersey pulled out the stops to slow its spread (e.g. stay at home orders, closing schools, and so on).

Unsurprisingly — and fortunately — **R(t) is a much smaller value than R _{0}**.

The shaded blue region in the chart represents the two-week intermediate period after public health interventions had been implemented to account for a transition period that is observed empirically as more stringent interventions were implemented across the U.S.

So back to inference projections. Now that we’ve explained what R(t) is and how it’s calculated, we use the R(t) value as the forward-looking growth rate to project the curve into the future — in this event, past the date of April 22 (when this chart was generated) — and predict what future COVID infection, hospitalization, and death data will be using our SEIR model.

Inference projections are a massive step forward. Among their many benefits is that projections now implicitly take into account local factors such as population density or compliance with anti-COVID interventions, because inference projections are calibrated by R(t), which reflects local realities.

More good news: In many states, inference projections closely match our previous “strict stay at home” projections (which were based on pure epidemiological theory and the known disease dynamics of COVID). In other words, that means our model’s theoretical “strict stay at home” projections closely matched reality.

Check it out yourself: you can compare New Jersey’s inference projections (the blue line on the graph) with the “strict stay at home” projections (the green line).

Launching inference projections is the latest example of our ongoing commitment to improve our models and provides critical disease intelligence to help us better understand how and when we can reopen safely.

* It appears that the number of cases per day and the number of current hospitalizations are similar. Read literally, one might think that everybody who tests positive for COVID ends up in the hospital. (That would be scary!) Fortunately, that’s not the case. The data for hospitalizations counts the total number of current hospitalizations, while the data for cases counts only new cases per day. People with COVID tend to spend multiple days (or weeks) in the hospital, so each hospitalization is reflected in the data numerous times. The number of *new* hospitalizations is actually much lower than the number of new cases, as one would expect.