Thread by @COVID19Tracking

A methodological note. As we've been saying, we're continuing to revise and pin down the way we code hospitalizations data. The states are reporting two different numbers: current hospitalizations and cumulative hospitalizations. These are unlike numbers.

Cumulative hospitalizations tells you something (though far from a complete story) about the severity of the disease. Current hospitalizations measures something else, though: it quantifies the strain on the hospital system at a given time.

There's a little more complexity: some states that report current hospitalizations lump together laboratory-confirmed positive and suspected cases, and others (like California, seen below) break them out.
public.tableau.com/views/COVID-19…

Our default logic has been to use the "positive" cases only, rather than positive and suspected. But on-the-ground reporting shows that both buckets of patients create strain on the hospital system. Until someone has tested negative, PPE and caution must be used.

So, yesterday, we made the tough decision to begin to standardize the way we capture state data to sum testing-confirmed cases of COVID-19 and suspected cases in our "currently hospitalized" numbers.

The previous method also made sense. But if the primary use of this statistic is to understand the strain on the hospital system, then the total strain needs to be captured.

This will also bring the different state numbers closer to the same standard.

This is not an across-the-board methodological change. For cases, like Wyoming, where the state reports confirmed and suspected ~cases~ of COVID-19, we will report only the "confirmed" in our case count. This is just about current hospitalizations.

In an overarching sense: people would like all these numbers to be some sort of irreducible and inhuman truth. But data is full of human decisions. Good judgment requires understanding these caveats and enriching one's apprehension with other evidence.

thenewatlantis.com/publications/w…