Oh
My
God
I _had_ given up annoying people, forever.
But this one is such a facepalm. And there is now online evidence of me completely failing to debug a numbers thing.
#MedEd #Foamed #cardiotwitter
All blame goes to @mshunshin for my relapse.
My
God
I _had_ given up annoying people, forever.
But this one is such a facepalm. And there is now online evidence of me completely failing to debug a numbers thing.
#MedEd #Foamed #cardiotwitter
All blame goes to @mshunshin for my relapse.
I told Matt I had read this paper and I was gonna get some of this eicosapentanoic acid stuff for my TGs.
Then he said, "don& #39;t bother, just look at Table 2"
Then he said, "don& #39;t bother, just look at Table 2"
So I looked up the EVAPORATE paper again, and thought the master was merely swatting the grasshopper aside.
"Good try, Matt, but it will be OK - they have an & #39;out& #39;"
"Good try, Matt, but it will be OK - they have an & #39;out& #39;"
So it doesn& #39;t matter that 3.5 to 3.6 is not +0.3 (even with rounding).
It could be something to do with log adjustment, whatever that may be.
It could be something to do with log adjustment, whatever that may be.
The bombshell.
So here is the problem.
You know how you do a trial, yeh?
You know how you do a trial, yeh?
And once randomized, you give real and placebo into the two arms, and then do the follow-up, yeh?
At BASELINE, what are the two groups?
At BASELINE, what are the two groups?
And if the damn thing works, at the end, the active group is what, in comparison to placebo?
So they go like this:
Just thought we would get that straight.
I know it is obvious.
But somehow I missed it. (Don& #39;t tell anyone.)
Here are the #EVAPORATE results, which I said to Matt was probably OK. (Me, the great skeptic.)
I know it is obvious.
But somehow I missed it. (Don& #39;t tell anyone.)
Here are the #EVAPORATE results, which I said to Matt was probably OK. (Me, the great skeptic.)
What is this showing?
Oooooooooooooo Kaaaaaaaaaaaaaay.
Maybe a typo?
Maybe the other ones are OK?
How many of these between-arm differences are this strange way round?
Maybe a typo?
Maybe the other ones are OK?
How many of these between-arm differences are this strange way round?
How many become more samey, versus become more differenty?
How to explain this?
Paper is here for all to review.
Go here and click the red "PDF" icon.
https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehaa652/5898836">https://academic.oup.com/eurheartj...
Go here and click the red "PDF" icon.
https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehaa652/5898836">https://academic.oup.com/eurheartj...
Seems straightforward enough.
No time travel backwards, or anything like that, to explain how they can end up the same after starting different.
No time travel backwards, or anything like that, to explain how they can end up the same after starting different.
Very good and kind suggestion @Biotech_beast
Next time I am had up for a beating by my university, I want you on the panel, as you will go lightly on me!
Next time I am had up for a beating by my university, I want you on the panel, as you will go lightly on me!
Quick sense check. How many SD& #39;s apart are the two arms at baseline?
Suppose each group is 25 patients.
(In reality they are each larger than this, so we are being generous.)
How much smaller is the SE than the SD,
TO AN ORDER OF MAGNITUDE
(Statisticians - I know this is the wrong SE, but it is a quick sense check only)
(In reality they are each larger than this, so we are being generous.)
How much smaller is the SE than the SD,
TO AN ORDER OF MAGNITUDE
(Statisticians - I know this is the wrong SE, but it is a quick sense check only)
So how many SE& #39;s are the means apart, at baseline?
0.5 SDs
SE is about 1/5th of an SD.
THINK and get it right way round.
Will it be more or less SE& #39;s?
0.5 SDs
SE is about 1/5th of an SD.
THINK and get it right way round.
Will it be more or less SE& #39;s?
So if you are that number of SE& #39;s apart, it is just a little more than 2 SE& #39;s
If you are 2 and a bit SE& #39;s away, ROUGHLY what is the p value?
If you are 2 and a bit SE& #39;s away, ROUGHLY what is the p value?
OK So now we know it is worth checking properly.
Find any online unpaired t test calculator.
For example, https://select-statistics.co.uk/calculators/two-sample-t-test-calculator/">https://select-statistics.co.uk/calculato...
Find any online unpaired t test calculator.
For example, https://select-statistics.co.uk/calculators/two-sample-t-test-calculator/">https://select-statistics.co.uk/calculato...
We don& #39;t have to do it for all of them. And this is why.
So, without doing any more thinking or typing or calculating, you know the P value for baseline comparison is what?
OK this is useful, isn& #39;t it?
The quicker you can do this in your head, the less effort it is, and the more willing you are to do it.
Most people don& #39;t realise they can do this.
The quicker you can do this in your head, the less effort it is, and the more willing you are to do it.
Most people don& #39;t realise they can do this.
How many out of 6, do you now KNOW instantly are different with p<0.05.
First person wrong.
Don& #39;t make blanket assumption! The top one has quite a wide SD, so is NOT at least 0.5 SDs apart.
All the others are.
Don& #39;t make blanket assumption! The top one has quite a wide SD, so is NOT at least 0.5 SDs apart.
All the others are.
How did this nonsense happen?
The authors told us!
The PATIENTS were blinded to which drug they were taking.
But the people making the CT measurements were not. Or at least they didn& #39;t say it was.
The authors told us!
The PATIENTS were blinded to which drug they were taking.
But the people making the CT measurements were not. Or at least they didn& #39;t say it was.
Heh heh https://twitter.com/egutibanes/status/1300006064130347008?s=20">https://twitter.com/egutibane...
Here is my reasoning.
Do you think the measurement was blinded?
Do you think the measurement was blinded?
If you voted for the lamb, remember where lambs are taken. 8-)
If you think the measurement was blinded, you will believe anything, and I can& #39;t help you.
If you think the measurement was blinded, you will believe anything, and I can& #39;t help you.
I agree with the majority that measurement appears to have been not blinded.
Do you take the effect sizes as unbiased estimates of the effect of the drug, therefore?
Do you take the effect sizes as unbiased estimates of the effect of the drug, therefore?
Ah, Andre Plumb - I slightly disagree. It doesn& #39;t say it WAS blinded.
It says they were PLANNING to do it blinded.
Well, I am planning to be the Queen of England.
It says they were PLANNING to do it blinded.
Well, I am planning to be the Queen of England.
Maybe I am too suspicious, but maybe you are too trusting, Andrew Althouse.
I am gonna put you down as reviewer for everything now.
8-)
I am gonna put you down as reviewer for everything now.
8-)
Quick summary for those still puzzled.
Statisticians telling me there is no problem. I believe, howerever, that they are describing the statistical calculation process.
In contrast, I am using the statistics to test a different hypothesis.
Statisticians telling me there is no problem. I believe, howerever, that they are describing the statistical calculation process.
In contrast, I am using the statistics to test a different hypothesis.
Patients are KNOWN to be randomized into two groups.
Question is, were the 6 endpoint variables measured blinded or unblinded?
H0: Blinded
H1: Unblinded
So here goes my hypothesis test.
Question is, were the 6 endpoint variables measured blinded or unblinded?
H0: Blinded
H1: Unblinded
So here goes my hypothesis test.
"Under H0, the two-tailed P value for obtaining this level of difference, or larger, for each variable in turn is as follows."
What would you do?
You should never compare your own trial& #39;s two arms& #39; baseline characteristics, because:
1. You KNOW you randomized
2. About 1 in 20 rows will be significant, through chance (unless you did fancy stuff to prevent it)
So it is a waste of time and shows unintelligence.
1. You KNOW you randomized
2. About 1 in 20 rows will be significant, through chance (unless you did fancy stuff to prevent it)
So it is a waste of time and shows unintelligence.
However if you KNOW you randomized,
and you are concerned that your MEASUREMENT PROCESS was unintentionally biased,
that is a perfectly good comparison to make.
The Hypothesis 1 you are testing is that "I accidentally biased myself during measurement."
and you are concerned that your MEASUREMENT PROCESS was unintentionally biased,
that is a perfectly good comparison to make.
The Hypothesis 1 you are testing is that "I accidentally biased myself during measurement."
Here, it is as significant as Mr Significant, on a day he is feeling particularly significant.
In fact, he might just have been appointed to Oxford University as Professor of Significance.
In fact, he might just have been appointed to Oxford University as Professor of Significance.
Now this puts a different complexion on things.
People had been whatsapping me with messages like "Mineral Oil is not neutral".
I answered "What do I care if it is acid, base, or neutral? Or democrat, republican, or independent?"
People had been whatsapping me with messages like "Mineral Oil is not neutral".
I answered "What do I care if it is acid, base, or neutral? Or democrat, republican, or independent?"
Here I have photoshopped the data (in a good way! before you make any wisecracks!) to reduce the columns.
For each arm, this shows the median change and the p-value for the change, versus baseline.
For each arm, this shows the median change and the p-value for the change, versus baseline.
In the non-HDL-C, which arm had bigger changes?
Which arm had bigger changes in CRP?
Now I understand why people were talking about the placebo not being "neutral".
What they meant was:
It raises your nonHDL, by more than the active reduces it
It raises your CRP, by more than the active reduces it
What they meant was:
It raises your nonHDL, by more than the active reduces it
It raises your CRP, by more than the active reduces it
I wouldn& #39;t call this a placebo.
Placebo means "I will please you".
No, it would definitely not please me to have that stuff done to me!
It would annoy the hell out of me.
So let& #39;s call it what it is.
Placebo means "I will please you".
No, it would definitely not please me to have that stuff done to me!
It would annoy the hell out of me.
So let& #39;s call it what it is.
"VEXABO"
noun. (Clinical Trials)
A thing you give the other arm,
that makes & #39;em worse.
Etymology: Latin - "I will irritate".
noun. (Clinical Trials)
A thing you give the other arm,
that makes & #39;em worse.
Etymology: Latin - "I will irritate".
So now I don& #39;t have faith in last year& #39;s endpoint trial either.
Shame. Was looking for a way to make myself healthier without having to go on a diet or do exercise etc.
Never mind.
Definitely cancelling my plans to smuggle this stuff in from USA.
Shame. Was looking for a way to make myself healthier without having to go on a diet or do exercise etc.
Never mind.
Definitely cancelling my plans to smuggle this stuff in from USA.
And no, you can& #39;t interest me in any HDL reduction chit-chat.
Not after this miserable attempt to find anything good about HDL reduction, from Dan Keene.
https://www.bmj.com/content/349/bmj.g4379.full.pdf">https://www.bmj.com/content/3...
Not after this miserable attempt to find anything good about HDL reduction, from Dan Keene.
https://www.bmj.com/content/349/bmj.g4379.full.pdf">https://www.bmj.com/content/3...
Hmmm... Maybe we could go for the "Cohen Plus" option?
Mixed up time points AND mixed up arms.
That would work!
Now they would start samey, and become differenty, and the new thing would be better!
DONE!
Can Francis Industries bill Amarin Inc, or will Cohen Corp be doing it?
Mixed up time points AND mixed up arms.
That would work!
Now they would start samey, and become differenty, and the new thing would be better!
DONE!
Can Francis Industries bill Amarin Inc, or will Cohen Corp be doing it?