Let's talk about the "risk factors" for COVID-19 for a moment

1/n
We talk about risk factors all the time. Not just in the medical scientific literature: you will find risk factors being discussed in the popular media and on social media too

Exhibit A: https://www.nytimes.com/reuters/2020/04/08/world/europe/08reuters-health-coronavirus-france-confinement.html?searchResultPosition=2

2/n
That brings us to the first problem: an important "risk factor" for, say, a bad outcome (death, for instance) after COVID-19 diagnosis can make good TV AND be completely meaningless for doctors and patients

4/n
It depends on what the risk factor represents

Is it a factor that when changed immediately changes the risk of a bad outcome?

Is it a factor that we should keep in mind when making prognosis?

Or is it a factor that simply covaries with other "known risk factors"?

5/n
Sounds simple enough?

Well, take this example: https://academic.oup.com/cid/advance-article/doi/10.1093/cid/ciaa415/5818333

Conclusion: "Unfortunately, obesity in people <60 years is
a newly identified epidemiologic risk factor which may contribute to increased morbidity rates ..."

6/n
Should people with overweight start losing weight quickly to avoid COVID-19 related morbidity?

Or are we looking at a complex relationship between access to care, history of diseases, social economic status and a bad outcome?

And why below 60 you ask, and not, say, 53?

7/n
That brings us to the second, related issue: there are no rules in the find-a-risk-factor game

8/n
Take the example above, the researchers decided to take 60 years as the cut-off for a "young" versus "old" people analysis. Why? Who knows!

And why were people with a BMI of 35 and over compared only to people with a BMI in the 30-34 range?

There. Are. No. Rules

9/n
With COVID-19, finding risk factors and high risk groups clearly isn't just an academic exercise

If we aren't careful, COVID-19 patients will suffer from arbitrary risk grouping, statistical malpractices and/or hidden health(care) inequalities

10/n
This thread is not just to complain (okay, maybe a little)

We can do and should do better. To start, by defining what it is we are after when we study these dangerously ambiguous "risk factors"

11/n
Many will already be familiar with the often helpful distinction between prediction, explanation (or counter-factual prediction) and description

Those who are not, this is a great place to start: https://www.stat.berkeley.edu/~aldous/157/Papers/shmueli.pdf

12/n
And there is a place for each, IMHO:

It is helpful to identify strong predictive factors that help with diagnosis and prognosis

It is helpful to identify the true causes of bad outcomes after infection

It is helpful to describe which patients ended up in ICU

13/n
This was just to say: finding risk factors is a risky business if we are not specific our aims and methods

end rant.
Apparently this thread could have been summarized in just one tweet https://twitter.com/statsepi/status/1249680569463721984?s=21 https://twitter.com/statsepi/status/1249680569463721984
You can follow @MaartenvSmeden.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: