Vitamin D?

Covid-19 discussion, bring your own statistics
User avatar
sTeamTraen
Catbabel
Posts: 934
Joined: Mon Nov 11, 2019 4:24 pm

Re: Vitamin D?

Post by sTeamTraen » Tue Sep 29, 2020 7:10 pm

bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- Why do they take the unusual step of reporting mean 25(OH)D as a natural logarithm? i.e. as ln 25(OH)D? I don't think they used the natural log in any tests?
They say that the log-transformed this (and two other) variables. Whether that was necessary or not is another question. Having done that, it would be normal for them only to report the log values, and coefficients based on them. If they added "ln" to the labels inconsistently, that's a bit sloppy but not a huge problem, I'd have thought.
bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- Why don't they treat 25(OH)D as a continuous variable and test whether it predicts a binary outcome (e.g. survival vs death)? They have dichotomized vitamin D status. What's the betting they tried dichotomizing it in various ways before settling on this particular threshold.
They should have presented the results with continuous predictors as well, but if their aim was really to show that deficient levels of Vitamin D are a problem then the dichotomy doesn't seem to me to be totally worthless.
bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- In any case, they state (for Figure 1) that "the number of red dots (inpatient mortality) above the solid line is significantly less compared to the dots below the line", the solid line being the 30 ng/mL 25(OH)D line. Well, counting the dots I can't get this result with a chi-square.
I count 68 white and 7 red dots above the line, versus 125 and 28 below. Accordingly:

Code: Select all

> chisq.test(matrix(c(68,7,125,28), nrow=2))

	Pearson's Chi-squared test with Yates' continuity correction

data:  matrix(c(68, 7, 125, 28), nrow = 2)
X-squared = 2.4626, df = 1, p-value = 0.1166
Nope, not significant, even if they cheat (as Raoult, doubtless "accidentally", also did in the first HCQ paper) by not using Yates' correction:

Code: Select all

> chisq.test(matrix(c(68,7,125,28), nrow=2), correct=FALSE)

	Pearson's Chi-squared test

data:  matrix(c(68, 7, 125, 28), nrow = 2)
X-squared = 3.1145, df = 1, p-value = 0.0776
Something something hammer something something nail

User avatar
sTeamTraen
Catbabel
Posts: 934
Joined: Mon Nov 11, 2019 4:24 pm

Re: Vitamin D?

Post by sTeamTraen » Tue Sep 29, 2020 10:34 pm

More on this (thanks to Bob for making a much better job than I did of the data): https://twitter.com/sTeamTraen/status/1 ... 18017?s=20
Something something hammer something something nail

User avatar
sTeamTraen
Catbabel
Posts: 934
Joined: Mon Nov 11, 2019 4:24 pm

Re: Vitamin D?

Post by sTeamTraen » Wed Sep 30, 2020 12:18 am

Oh dear. Here is a story about the senior (and only non-Iranian; presumably added for a bit of Western credibility) author on the paper; he seems to be a shill for Big D (not the popular brand of peanuts sold in pubs on ingenious displays that revealed more of a scantily-clad woman with every purchase).

The article states that "The authors have declared that no competing interests exist." I have alerted the Editor in Chief of the journal.
Something something hammer something something nail

User avatar
jimbob
After Pie
Posts: 1623
Joined: Mon Nov 11, 2019 4:04 pm
Location: High Peak/Manchester

Re: Vitamin D?

Post by jimbob » Wed Sep 30, 2020 7:45 am

sTeamTraen wrote:
Tue Sep 29, 2020 10:34 pm
More on this (thanks to Bob for making a much better job than I did of the data): https://twitter.com/sTeamTraen/status/1 ... 18017?s=20
I have used https://apps.automeris.io/wpd/ (Webplotdigitizer - web based version before now to look at a badly drawn excel curve: it's pretty straightforward

https://twitter.com/ParkinJim/status/12 ... 5290027009
Have you considered stupidity as an explanation

geejaytee
Stargoon
Posts: 91
Joined: Mon Nov 11, 2019 1:32 pm
Location: Colchester

Re: Vitamin D?

Post by geejaytee » Wed Sep 30, 2020 12:18 pm

sTeamTraen wrote:
Tue Sep 29, 2020 7:10 pm
bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- Why do they take the unusual step of reporting mean 25(OH)D as a natural logarithm? i.e. as ln 25(OH)D? I don't think they used the natural log in any tests?
They say that the log-transformed this (and two other) variables. Whether that was necessary or not is another question. Having done that, it would be normal for them only to report the log values, and coefficients based on them. If they added "ln" to the labels inconsistently, that's a bit sloppy but not a huge problem, I'd have thought.
bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- Why don't they treat 25(OH)D as a continuous variable and test whether it predicts a binary outcome (e.g. survival vs death)? They have dichotomized vitamin D status. What's the betting they tried dichotomizing it in various ways before settling on this particular threshold.
They should have presented the results with continuous predictors as well, but if their aim was really to show that deficient levels of Vitamin D are a problem then the dichotomy doesn't seem to me to be totally worthless.
bob sterman wrote:
Sun Sep 27, 2020 5:58 pm
- In any case, they state (for Figure 1) that "the number of red dots (inpatient mortality) above the solid line is significantly less compared to the dots below the line", the solid line being the 30 ng/mL 25(OH)D line. Well, counting the dots I can't get this result with a chi-square.
I count 68 white and 7 red dots above the line, versus 125 and 28 below. Accordingly:

Code: Select all

> chisq.test(matrix(c(68,7,125,28), nrow=2))

	Pearson's Chi-squared test with Yates' continuity correction

data:  matrix(c(68, 7, 125, 28), nrow = 2)
X-squared = 2.4626, df = 1, p-value = 0.1166
Nope, not significant, even if they cheat (as Raoult, doubtless "accidentally", also did in the first HCQ paper) by not using Yates' correction:

Code: Select all

> chisq.test(matrix(c(68,7,125,28), nrow=2), correct=FALSE)

	Pearson's Chi-squared test

data:  matrix(c(68, 7, 125, 28), nrow = 2)
X-squared = 3.1145, df = 1, p-value = 0.0776
As it's 2x2, there's an exact test you could run, rather than the chisq appoximation, and that gives:

Code: Select all

>  fisher.test(matrix(c(68,7,125,28), nrow=2))

        Fisher's Exact Test for Count Data

data:  matrix(c(68, 7, 125, 28), nrow = 2)
p-value = 0.08239
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
 0.8668037 6.1984502
sample estimates:
odds ratio 
  2.169318 
and the p-value is still not significant.

User avatar
bob sterman
Fuzzable
Posts: 275
Joined: Mon Nov 11, 2019 10:25 pm

Re: Vitamin D?

Post by bob sterman » Thu Oct 01, 2020 7:47 pm

It pains me to have to suggest that Matt Hancock might be taking some unnecessary flak - but the Daily Mail is trying to beat him with the unconvincing scatterplot....

https://www.dailymail.co.uk/news/articl ... id-19.html

raven
Fuzzable
Posts: 239
Joined: Mon Nov 11, 2019 8:58 pm

Re: Vitamin D?

Post by raven » Fri Oct 02, 2020 11:34 am

It occurs to me that even if there was some correlation, it's probably not causation. Low vitamin D could be a proxy for either underlying serious health issues, or working long shifts & thus poverty, as both can keep people indoors. And we know the first is already a risk factor independent of Vitamin D status.

User avatar
jdc
Hilda Ogden
Posts: 551
Joined: Wed Sep 25, 2019 4:31 pm

Re: Vitamin D?

Post by jdc » Sat Oct 03, 2020 1:28 am

raven wrote:
Fri Oct 02, 2020 11:34 am
It occurs to me that even if there was some correlation, it's probably not causation. Low vitamin D could be a proxy for either underlying serious health issues, or working long shifts & thus poverty, as both can keep people indoors. And we know the first is already a risk factor independent of Vitamin D status.
And looking at it the other way... maybe people who are health conscious dutifully meet their vitamin D requirements as advised but at the same time are also dutifully following all the other health advice they're given. A bit like that old 'bias of compliance' thing? https://www.nytimes.com/2007/09/16/maga ... ogy-t.html

Spoiler:

raven
Fuzzable
Posts: 239
Joined: Mon Nov 11, 2019 8:58 pm

Re: Vitamin D?

Post by raven » Sat Oct 03, 2020 12:49 pm

That research actually makes a lot of sense. Thanks, jdc.

User avatar
Woodchopper
After Pie
Posts: 1871
Joined: Sat Oct 12, 2019 9:05 am

Re: Vitamin D?

Post by Woodchopper » Tue Oct 27, 2020 4:28 pm

PLOS ONE has issued an expression of concern for a paper it published last month suggesting that vitamin D might protect against severe COVID-19.

Central to the concerns is that the authors seem to have been too far out over their skis in asserting a link between the vitamin and the response to the infection. But as the EoC reveals, many of the potential problems can fairly be attributed to porous peer review as much as over-ambitious authors.
https://retractionwatch.com/2020/10/26/ ... f-concern/

User avatar
sTeamTraen
Catbabel
Posts: 934
Joined: Mon Nov 11, 2019 4:24 pm

Re: Vitamin D?

Post by sTeamTraen » Tue Oct 27, 2020 5:36 pm

Woodchopper wrote:
Tue Oct 27, 2020 4:28 pm
PLOS ONE has issued an expression of concern for a paper it published last month suggesting that vitamin D might protect against severe COVID-19.

Central to the concerns is that the authors seem to have been too far out over their skis in asserting a link between the vitamin and the response to the infection. But as the EoC reveals, many of the potential problems can fairly be attributed to porous peer review as much as over-ambitious authors.
https://retractionwatch.com/2020/10/26/ ... f-concern/
I'm pleased to think that I may have been a modest part of the solution here. https://twitter.com/sTeamTraen/status/1 ... 51010?s=20
Something something hammer something something nail

Post Reply