The Mehra et al Lancet paper is clearly the most important recent "treatment update". WHO have stopped their HCQ arm in SOLIDARITY.
In order to trust these results, we need to KNOW about where the data come from. A thread about data digging... (1/n)
@TheLancet @richardhorton1
In order to trust these results, we need to KNOW about where the data come from. A thread about data digging... (1/n)
@TheLancet @richardhorton1
First, I have acknowledge the crowd sourcing that has gone into this @StatModeling blog
https://statmodeling.stat.columbia.edu/2020/05/25/hydroxychloroquine-update/
Thanks to Andrew, 1000s of people have scrutinised the paper. His blog is well-versed in bad/fake science.
So, the main issues as I see them: (2/n)
https://statmodeling.stat.columbia.edu/2020/05/25/hydroxychloroquine-update/
Thanks to Andrew, 1000s of people have scrutinised the paper. His blog is well-versed in bad/fake science.
So, the main issues as I see them: (2/n)
1/ No detail on which countries or hospitals contributed data (again only 4 authors and no acknowledgements)
We can work out approximately which hospitals by comparing with Table S1 in their previous NEJM paper, slightly smaller numbers (3/n)
https://www.nejm.org/doi/full/10.1056/NEJMoa2007621
We can work out approximately which hospitals by comparing with Table S1 in their previous NEJM paper, slightly smaller numbers (3/n)
https://www.nejm.org/doi/full/10.1056/NEJMoa2007621
2/ Imputation of missing data values by the default "absence" value: "if the patient's electronic health record did not include information on a clinical characteristic, it was assumed that the characteristic was not present."
I have never seen that assumption before!
(4/n)
I have never seen that assumption before!
(4/n)
Now to the really juicy stuff... Australia!
Australia is considered both continent & country. So we have a breakdown of numbers to compare with reported data.
They have 5 hospitals, 609 patients and 73 deaths (deaths up until 21st April) (5/n)
Australia is considered both continent & country. So we have a breakdown of numbers to compare with reported data.
They have 5 hospitals, 609 patients and 73 deaths (deaths up until 21st April) (5/n)
From the very reliable source, wikipedia, I find that on the 22nd of April there were 74 deaths reported in 7 different states
https://en.wikipedia.org/wiki/COVID-19_pandemic_in_Australia#April_2020
I imagine 1 hospital=1 state?
How to square these data? (6/n)
https://en.wikipedia.org/wiki/COVID-19_pandemic_in_Australia#April_2020
I imagine 1 hospital=1 state?
How to square these data? (6/n)
Asia: the argument is a bit more involved - see pic. Bottom line is: thousands of patients in a single hospital in Japan over 1 month. Seems unlikely! (7/n)
Last major point: dosing of hydroxychloroquine. Most patients are from USA, and the FDA recommends 800mg (salt) on day 1, followed by 400mg for up to 4-7 days. This is not weight-based.
The mean daily doses reported in the paper are about 100mg higher than expected. (8/n)
The mean daily doses reported in the paper are about 100mg higher than expected. (8/n)
Conclusion:
1/ Unknown methodology for producing the data
2/ Possibly biased imputation
3/ Numbers of patients that don't match
4/ High dosing of HCQ
Given the enormous importance of this paper, the authors need to address these issues ASAP
1/ Unknown methodology for producing the data
2/ Possibly biased imputation
3/ Numbers of patients that don't match
4/ High dosing of HCQ
Given the enormous importance of this paper, the authors need to address these issues ASAP