Thread by @ProfDFrancis

"Why oh why is NIAID, in their new trial of baricitinib, giving remdesivir to both arms? We will never find out about interactions between remdesivir and baricitinib."

Wondering this?

Could it be:

If you voted for "Known not to interact" you are a dimbo.

Remdesivir was designed for Ebola and stuff.

Baricitinib was designed for inflammatory conditions like Rheumatoid Arthritis.

Met many people with both?
No?
Well that's why the Rheumebola-1 trial didn't recruit well, doh

Here is the structure of Remdesivir.

What is this bit called?

Options

And this bit...

what is it, roughly?

Peeps, if you are having difficulty, remember that Sucrose, the sugar we like eating (to make us fat and die) is made of two bits?

Glucose, which we give i.v. to people, plus fructose, which we don't?

Split up by sucrase in the GI tract?

So Sucrose (below) ain't it.

And this last bit:

Should immediately strike a chord with you as very similar to two of these things (the big ones).

So, working backwards, we have:

Base
Sugar
Phosphate

which is how you make what?

So at what point is remdesivir working?

OK so we know, just from looking at the picture of it, where remdesivir acts.

Now let's look at where baricitinib acts. It is the JAK-STAT pathway, in which I am an expert. (Well I had only vaguely heard about it 10 minutes ago, but thanks to Twitter friends I have Wiki Links)

But wait! Who are you people?

Don't tell me Aseem Malhotra is here, fighting the good fight by saying *everything* is carbohydrate and cholesterol?

Listen, that was the big discovery of Crick and Watson: the structure of DNA, Deoxy ribo nucleic acid

In the staircase, the things reaching out from one side to the other are the bases, Adenine, Thymine, Cytosine, Guanine.

The vertical bits in the staircase are sugar-phosphate-sugar-phosphate.

The sugar is ribose in RNA and deoxyribose in DNA (so it can be more coily).

We gotta know this stuff ... it's the basis of all life on earth ... big thing ... people got Nobel prizes and stuff (though Rosalind Franklin missed out).

How come people aren't fluent on this but seem to have heard of these people, whom I struggle to understand the utility of?

Serves me right for reading the daily mail.

My excuse was that someone sent me a link to it.

Anyway, back to the role of the JAK-STAT pathway in popular culture.

After memorising the following and reading all the cited papers in depth:

... I have concluded that this JAK-STAT pathway is

"something to do with sending signals from OUTSIDE THE ENTIRE CELL to INSIDE THE NUCLEUS",

i.e. getting across two membranes: the cell membrane and the nuclear membrane.

Step 1.

Cytokine thingy, outside the cell, comes to bind to the external part of the cell surface receptor.

JAK is attached to the internal part of the cell surface receptor, poised like a Crouching Buddha.

Step 2.

JAK leaps into action, phosphorylating some other intracellular bit of the receptor protein.

Step 3.

STAT, whatever that may be, shows up magically, and binds to that phosphorylated bit of the receptor.

JAK phosphorylates it too, for good measure!
"Take that! And don't come back"

This makes the STATs angry. "I didn't come here to be treated like that!"

Step 4.

The STATs now dimerise and flee to the nucleus, taking refuge from undignified manhandling by the JAK enzyme.

Inside the nucleus at last, the STAT does some sort of thing to transcription, in a manner which I can't quite discern. Lots of acronyms and even greek letters

But luckily I am saved from any more brain-melting reading of biochem.

Because Baracitinib, and drugs like it, work on JAK, very early in this pathway.

They inhibit JAK, so turn down the volume on the whole pathway.

Wait, is this a proper sciency thing to say?

Let's fix it. It takes 5 seconds.

Careful not to mess up the other stuff which looks fiddly, but find the bit:

Give a quick explanation of why you made the change.

And preview to make sure you haven't utterly buggered up the page for everyone.

Looks fine, so let's publish.

I do encourage anyone who uses Wikipedia to contribute to it by fixing small errors as you see them, making small improvements, or even big ones.

All changes can be tracked and can be undone if others don't like it.

I have made lots of contributions and very rarely been undone.

Anyway!

The good news is Baricitinib works at the JAK end of the JAK-STAT pathway of signal transduction, i.e. it interferes with the JAK protein. (It also apparently inhibits Endocytosis).

While Remdesivir is a nucleotide analogue, i.e. works INSIDE the nucleus.

So they are different-shaped drugs, working in different places. If you had to take a bet on whether they would interfere with each other, I would bet that they didn't.

NIAID are having to bet, and they chose like I would that they don't interfere.

If anyone believes Remdesivir is now the standard of care, *NIAID* must believe it! It was their trial that sealed the deal: they stopped it early for that reason.

So it is rational for them to have remdesivir as background therapy, and Baricitinib as the trialled drug.

The alternative would be to have a 2x2 factorial design.

Randomize to Baricitinib versus placebo
and separately but simultaneously
Randomize to Remdesivir versus placebo

Why is this 2x2 design worse?

In what situations would you be *glad* of doing the 2x2 factorial, rather than the actual NIAID design?

I'll come back later and see how people are getting on

ORBITA HQ'ers are chasing me for their weekend research meetings

Can't lie and say I am out playing football or rock climbing.

Also I have a nice tweetorial lined up for tomorrow suggested by Charlotte Manisty & James Moon

OK How have we got on?

This is, I have to say, disappointing.

A factorial trial is a fancy way to say people are randomized in several trials simultaneously. Why?

Let's say I want to test whether Vitamin A is any good. Or Vitamin B. Or Vitamin C. Or Vitamin D, etc. to Vitamin Z.

I could do a 27-arm trial.

26 arms each having a vitamin, and the 27th having placebo.

Trouble is, if I started with 1000 patients enrolling into the trial, how many will get Vitamin Z?

So when I do my "Vitamin Z versus placebo" comparison eagerly (since Francis Industries is the sole supplier of Vitamin Z), how many patients are in that comparison?

Yeah.

And when I count deaths from Covid, suppose I enrol ANY SYMPTOMATIC PERSON (who tests positive). Mortality, say, ~2%.

How many deaths do I expect in EACH of the two arms?

Yeh, this is gonna be the worlds stupidest trial.

(On par with those trials with only one arm, claiming efficacy.)

So let's try the "Factorial" design.

I line up the patients at one end of a big aircraft hangar, like the Excel centre or whatever. They sign up, and then flow forward on a conveyor belt.

Angharad, meets the first, and randomizes them to Vitamin A versus Placebo.

They slide on to meet Brian, who randomizes them to Vitamin B versus Placebo.

Then to Charles, who randomizes them between Vit C and placebo.

and so on.

When they fall off the end of the conveyor belt, Zeke has just randomized them between Vitamin Z and placebo.

Off they go to seek their fortune, with their 27 little canisters, one labelled "A or placebo", another labelled "B or placebo", etc.

With *one* lot of followup, we can capture all the outcomes, and then, if we wanted, write 26 papers.

Angharad would write the Vitamin A Versus Placebo paper.

How many people in her two arms: Vitamin A and Placebo, respectively?

And how many in each arm of Brian's "Vitamin B Versus Placebo" paper?

Sweet!

All 26 papers are nicely big in numbers!
One lot of followup was enough to cover all of it!

To be even more efficient in practical terms, we would arrange it not as 26 simultaneous RCTs, but 1 RCT with 26 branch points.

Technically there are 2^26 possible combinations, but it doesn't matter. We are not trying to evaluate each *combination*.

Rather, we are trying to evaluate each *drug*, on a background where the variability between arms in the other treatments is predictable.

(For short, we may say they are "balanced", but we mean variability is calculable so we can test whether outcomes differ more than expected)

That, in principle, is a factorial randomized controlled trial.

You bring the patient in to a production line and randomize them to multiple questions.

The most dramatically useful example of this in my mind is ISIS-2, done by Peter Sleight et al in the days that Myocardial Infarction was managed by:

Bed rest (which gave you DVT), and

Oxygen (which kills you).

In this unpromising environment they randomized people to:

Aspirin versus placebo

and also separately randomized them to:

Streptokinase versus placebo

Trial described here:
cardiologytrials.org/detail/22/

Full paper seems to be behind a paywall though for some reason. Nevertheless, due to my high-level connections (with Pubmed.org) I can reveal the highlight:

17000 patients randomized

Essentially all were in the aspirin - placebo tablet comparison

And

Essentially all were in the streptokinase - placebo infusion comparison

So this approach of Factorial Randomization, in this case 2 way for aspirin and 2 way for streptokinase, and so "2 x 2", gives you very good power for the "aspirin versus placebo" comparison.

[Side note before I get trashed by the statisticians:

I have excluded the Bonferroni-like consideration for simplicity, but of course that does reduce power and would be vital if we had 26 randomizations!]

Factorial Randomization also allows us to detect "Interaction".

Here are some examples of interaction.

"Resect left lung for bronchiectasis" and "Resect right for bronchiectasis" show an interaction.

Doing one can be beneficial, or doing the other can be beneficial, but doing both together is very, very bad.

Good + Good = Bad.

That is an interaction.

Puffer fish.

"Cook carefully"

"Eat"

Doing either one, on its own, will do nothing for your survival, if you are starving and it is the only food available.

But doing both, is a delicacy.

That is another example of an interaction.

Factorial trials let you see the interactions easily.

One person starved to death (double control)
One person poisoned to death (eat, but not cook)
Another person starved to death (cook, but not eat)
And one happy customer (cook + eat)

The person with double-active did not get what you would expect from adding the effects of the two separate interventions in isolation.

That is the definition of interaction (I think!)

Factorial randomization would be fine for power, and absolutely essential to detect interactions.

So why did they not do factorial randomization?

It's because half the patients would not get remdesivir, which they have already satisfied themselves (and me) is beneficial.

And indeed they will never find out about interactions.

As pointed out by @curious30525151:

Yes, most interactions between drugs are through their metabolism.

But that does not mean that *most drugs* significantly interact with each other through metabolism.

Of the thousands of drugs we have on formulary, there are millions of combinations of drugs, only a few tens of thousands of which are listed as significant interactions, and only a minority of which would be reason not to give one or the other, in a fatal disase such as COVID.

If you pick two drugs at random from the formulary, what are the chances that there is a clinically significant interaction between them (i.e. that you would actually HAVE TO CHANGE your dosing or not give one)

There are plenty of pairs of people in the UK who have a personal dislike of each other. Most often it is because of something one person has said or done.

However, is it true to (therefore) conclude that "Most pairs of people, chosen at random from the UK, have a dislike of each other because of something one person has said or done"?

Of course it is false.

Otherwise there could never be a bus, or a train, or an aeroplane, because passengers would be throttling each other.

From this reasoning, I am happy to work on the assumption that remdesivir and the new thing (whose name I have forgotten just now) will not interact in a significantly harmful way.

And therefore, NIAID/NIH's trial design looks brilliant to me.

I see the trainee pharmacists are out in force with their green pens.

8-)

Sorry to tease, we love you all really!

Thanks Larry. My colleague, @mshunshin Matt Shun-Shin whom you kindly invited to @CVCTForum last year, often brings up ISIS-2, not to praise antiplatelets or fibrinolytics, but to mock requests for endless subgroup analysis, in the hope of finding "something interesting".

I think he tells me the story that "The Lancet forced Peto and Sleight to have subgroup analysis, which they did not want to do. So they did a deal with the Lancet: We will give you your desired subgroup analysis, but only if we can put in an astrological one."

Annoyingly, I can't access the paper itself.

Does anyone have a bootleg copy of ISIS-2?

Dunno about this one.

1. I don't think remdesivir is an antiretroviral, is it?

2. I would be delighted to sweat over the anti-arrhythmics (just slow them down with anything, and leave them in AF) and anticoag (i.v. heparin should do it), if the drug kills the coronavirus!

Erratum!

Coronaviruses are not retroviruses, so remdesivir is doing its thing in the cytoplasm, not the nucleus.

I still am happy to assume it won't have a meaningful adverse interaction with Baricitinib.

p.s. Confidentially, Graham Cole @drgrahamcole has a motto.

"What's the difference between the Mielewczik and a Rottweiler?"

"The Rottweiler eventually lets go."

Michael managed to almost single handedly debunk the world's biggest cardiac stem cell scamster, Dr Bodo Strauer of Dusseldorf.

Didn't quite manage to get him into jail, despite forcing his institution to report him to the police.

But an "instant retirement" to avoid justice.