Today we are seeing tons of hype around the corticosteroid dexamethasone, prompted by this press release.

It does indeed look like good news, but it is important to recognize the limitations of what was found.
https://www.recoverytrial.net/files/recovery_dexamethasone_statement_160620_v2final.pdf

The key thing to recognize is that this is not a general treatment for COVID-19. What it does is to improve the odds of survival in severely ill patients suffering respiratory distress.

Presumably it works by suppressing the overactive immune response in these patients.

The trial found no benefits in hospitalized patients not receiving respiratory support, and there is no reason to expect that the drug would improve the course of disease for patients without severe disease. As an immunosuppressant, it could even worsen it for those patients.

Note this is not a drug that one would take for COVID19 the way one takes tamiflu for influenza. It's specifically for dealing with severe respiratory complications.

Comments such as this one, from one of the scientists involved with the study, have muddied the waters there.

The other thing that we need to keep in mind is how likely it is to help. I find the concept of "Number needed to treat (NNT)" to be the most useful here, if somewhat nuanced.

The NNT tells you how many people you need to treat to have a beneficial outcome in 1 person.

In this case, the beneficial outcome we are looking for is "preventing death".

For patients on ventilators, you need to treat 8 to prevent one death.

For patients on oxygen, 25 to prevent one death.

(figure from https://www.bbc.com/news/health-53061281)

You can think about this in terms of your personal odds. Suppose you're in the hospital on oxygen for COVID. You are offered dexamethasone. All else equal, there is about a 4% chance it will make a difference in your outcome.

Would I take it? Sure.

It is a miracle drug? No.

So overall this result is certainly good news. It was considered good enough to halt that arms of the trial, for example.

It's great that we are starting to learn how to better care for patients suffering respiratory difficulty from COVID.

That should save many lives.

At the same time, it's key to recognize that taking it won't *prevent* you from getting COVID.

Nor, if you have COVID, will it prevent you from being hospitalized.

Nor, if you are hospitalized, is it guaranteed to save your life.

It's an incremental step, not a panacea.

Followup: @EricTopol points out that I fell into the trap of using the term "severe" disease to mean "very bad". Severe has a technical meaning for COVID (akin to ARDS?)

Study: Dexamethasone is useful for patients with non-severe, very bad disease who need supplemental O2.

Also, @knickerbacker makes a nice point about how my preferred metric NNT can be deceiving. If only 1 in 100 are going to die without treatment, but you don't know which one that is and have to treat everyone, NNT is at least 100. And yet the drug is in a sense 100% effective.

Finally, in response to comments I want to stress that NNT=8 is impressive. Even NNT=25 isn't so bad. But thinking about NNT is useful context in light of "miracle drug" headlines, and worth considering given that this is not necessarily the safest drug to be taking.

An additional note: some are unhappy that I have described this as science by press release. But @hardsci has an example of a vital question that seems to require access to a more detailed methodological description than is in the research protocol. https://twitter.com/hardsci/status/1273148676190416902