Is there a better way to define groups for COVID-19 impact?

post by jmh · 2020-03-04T13:24:51.221Z · LW · GW · 5 comments

This is a question post.

Contents

  Answers
    3 Bucky
None
5 comments

I think everyone who has posted a table of stats for COVID-19 infection or deaths seems to do a 10 year grouping. For example here [LW · GW]. (Used only because I was looking at that post when it occurred to me.)

However, my understanding is that physiological changes in the human body are not linear over time but tend to be more like state changes. Now, it is true these changes are not on any annual schedule either but we do have some average ages for when changes in the human body seem to occur.

Could using the 10 year grouping actually hide important implications for those trying to make personal decisions based on that data presentation?

Answers

answer by Bucky · 2020-03-04T21:01:35.308Z · LW(p) · GW(p)

My main worry originally was that 10 year groupings could hide effects on very young children (<1 year). As the parent of a 7 month old this was a concern to me.

However we see 0 deaths in 400+ cases in <10 year olds so this suggests a fairly low upper bound for fatality rates in the very young (compared to the rates in the very old).

Interestingly <20 year olds only accounted for 2% of cases so either there is a reporting issue or so far children are less likely to get the disease (or more likely to be asympomatic).

Flu death rates are low even when grouping 0-4 year olds together (.01%, the same as 5-17 yr olds according to the CDC) so this provides some supporting evidence that the very young are not at heightened risk.

5 comments

Comments sorted by top scores.

comment by gjm · 2020-03-04T15:04:49.348Z · LW(p) · GW(p)

A related thing I wonder about: as well as the variation of risk with age, we hear about increased risk if you have various other conditions -- diabetes, hypertension, etc. Many of these of course are also things that tend to appear and/or worsen with age, and it's not clear to me how the various numbers should be interpreted if you want to estimate the risk to someone with known age and known other conditions (or absence thereof).

Replies from: Bucky
↑ comment by Bucky · 2020-03-04T21:13:51.157Z · LW(p) · GW(p)

Not quite what you were looking for but this report gives an idea of risk of various conditions:

CFR was elevated among those with preexisting comorbid conditions—10.5% for cardiovascular disease, 7.3% for diabetes, 6.3% for chronic respiratory disease, 6.0% for hypertension, and 5.6% for cancer.

I agree it would be nicer to have the combined effects of age and condition.

Replies from: gjm
↑ comment by gjm · 2020-03-04T23:05:44.648Z · LW(p) · GW(p)

That's exactly what I meant by "we hear about increased risk if ...". Those figures don't do much to distinguish between e.g. "these specific conditions make it more likely to be bad, and if you're old but don't have them then you're fine" and "age makes it more likely to be bad, and if you're young but have those conditions then you're fine".

Do they do anything to distinguish those possibilities? Probably. How much depends on how strongly those various conditions correlate with age. But my feeling is that e.g. cardiovascular disease is a better indication of being old than hypertension or diabetes, which I think are more likely to crop up in middle age, so the percentages feel fairly compatible with the it's-just-age hypothesis. If it were 6% for CVD and 10% for hypertension instead then I'd be more confident that there's something specifically bad about hypertension that makes COVID-19 worse.

(If I have to guess, I guess that the answer is somewhere in the middle: almost any specific health issue makes something like COVID-19 more likely to kill you, including the specific ones listed there but others too, and being older is bad because of all the ways you're likely to be less healthy if older. But who knows?)

Replies from: Bucky
↑ comment by Bucky · 2020-03-04T23:35:17.273Z · LW(p) · GW(p)

Yes, you are stuck with predicting based on the two different models (or, as you say, something in between). For a given individual, if you use your credence in the models you can get your best guess as to the CFR via weighted average.

This has larger uncertainty than a proper analysis but until we have one it’s probably the best we can do.