Thread by @ProfDFrancis

If anyone wants to know how incorrect causal inference arises in cardiology, there's no need to do a PhD on it.

It's encapsulated in this thread. Unlike most of my threads it has a happy ending though !

First, a whole load of unsuspecting patients have PCI.

Then a bunch of cardiologists who are normally ultracompetitive decide to do something constructive for a change, instead of just doing each other down in cross-London acrimony.

Let's get together and ...

google.co.uk/url?sa=t&sourc…

They do what we always do when we cardiologists get together. Tell stories.

"I had a guy with 3 vessel thrombosis in cardiogenic shock, going into asystole as he was put on the table, but we still saved him!"

"That's nothing. Last night I had a lady present with 48 hours of chest pain, 7 vessel disease and a blood pressure of 15 mmHg, and we sorted her out in no time! She was dancing round the ward this morning making the other patients tea."

When we run out of tall tales (which takes quite a few hours) it's almost midnight and we haven't done anything useful.

So we resolve to report out combined outcomes.

"Don't make me look bad though?"
"Nor me!"
"Wouldn't dream of it, chaps. Plenty of other things to talk about."

Time passes and people return to their usual sniping across London.

Eventually somebody actually does some work and collates the data.

"I've found something very highly significant. P is less than almost zero."

"Awesome! Let's publish it! Remember to put my middle initial."

Paper is circulated.

"It's fantastic. We were thinking of the following title: 'GOZUP'.

Here is the result."

"oh I get it! It gozup!
Awesome.
Really easy to understand. Go for it!"

A question comes back amonst the emails.

"By the way, which are we saying goes up, the white things compared to the black things, or are we saying they are all going up from left to right?"

"Great, that's what I thought, but I was just checking."

"And that's not all, both of them are statistically significant. p value is almost less than zero!!!"

Some time later, an unusually curious interventionist asks, "What are the white and black things? It's nothing to do with race or anything like that, is it? Cos I don't want to get tangled up in any hate speech type of stuff."

What do you think it is? white and black:

A. Cumulative number of men and women trained in intervention, year by year.

B. Survival (white) and event free survival (black) at 10 years, stratified by FFR (x axis).

C. Number of things we put in coronary arteries.

The commonest answer is the correct one.

We have diagnosed that we are putting more and more things into people's coronary arteries.

The white and black are two types of thing.

Problem is, it's not particularly exciting in this era of recognition that putting more things into more people isn't automatically something to boast about.

There's another interesting observation.

Patients having OCT had the most stent length.
Patients having IVUS were in the middle.
Patients having neither had the least stent length.

Does this mean:

A. seeing stuff up close with IVUS and especially OCT, makes you want to put more metal in?

B. people with more badder disease get more stents and more pondering with IVUS/OCT?

C. Some docs just like IVUS, OCT and lengthy stents?

Or of course, it all looks a bit moderate so you stick in some imaging and Pow! Atherosclerosis Everywhere!

There is good news for these people where you had to IVUS or OCT to find a reason to put extra much metal in:

Yup. If you had to IVUS to see it, and especially if you had to go totally up close and personal with OCT to see it, the patient is gonna be OK.

I have to admit a small conflict of interests. I can hide it no longer.

Yes I admit it.

You thought I mocked IVUS, OCT, FFR, iFR, AFR, BFR, CFR, DFR and EFR just for the sheer fun of needling the hell out of pompous pronouncements?

Nope.

Francis Industries has just been awarded a patent on a test.

Just DOING this test on cath lqb patients reduces long term cardiovascular mortality by five fold.

(Apols to Simon Redwood.

I didn't think the Lancet would publish our rude reply to your amusing ORBITA letter, so I had to have something else as revenge. I therefore decided to edit out the data from the London dataset that would have allowed you to patent this genius test.)

Francis Industries will be sponsoring a large number of CME sessions at each forthcoming conference describing the world conquering reliability of the

francis Fecundity Radix

fFR

Quicker than IVUS
Cheaper than OCT
Less wheezy than FFR
Less invasive the iFR

A wonder of the 21st century.

f F R

Come and get it.
80% less MACE.
if you don't use it you are backward.

Any ideas how my test works?
How it lowers events so dramatically?

Clue:

The Francis Fecundity Radix, fFR, is a pregnancy test.

"We have convincingly shown, and published under expert peer review in the New England Journal of Mistakes," thunders the shiny handout from our reps.

"... that patients who had fFR measured had more than 5 times lower mortality.

This is the new gold standard.

blah blah blah."

Compelling stuff, eh?

I can tell the secret story now. I noticed that whenever we do a pregnancy test on a patient coming into the Cath lab, THAT PATIENT DOESN'T DIE!

Sorry Simon I didn't let you have that for the Pan London paper, but money is money after all...

Anyway Francis Industries has tested this across thousands of patients (Thanks for the data Simon) and it works everywhere, always.

Of course some do die because of cancer, or because somebody threw a javelin through their chest etc, but it is rarely heart disease.

So now I have revealed all, what do people think of fFR?

A. We gotta get it in our lab

B. We already have it but should now do it in everyone.

C. Hey, this Francis Fecundity Radix is bullshit.*

(* if you vote for this I will sue you)

Why the skepticism?

Who gets pregnancy tests

Of the group or groups you picked, what do the patients who have pregnancy tests tend to be, compared with those who don't?

Choose one answer only

OK I am busted.

But why did this tweetorial class see though me so easily ?

I am assuming readers are cardiology fellows (the second funnest job in the world, other than being me).

For anyone else:

We do pregnancy tests before Cath lab in women rather than men, and in those of child bearing age rather than older women.

So it is blindingly obvious that almost all people who don't need a pregnancy test are either men or are older, two things that make LONG TERM CV outcomes after Cath lab procedures of all kinds, better.

Thus Francis Industries will be seen through by any intelligent observer.

(Of course if we pay doctors to do the test, we could probably keep the game going for a few years and make the odd billion...)

But what makes people do an IVUS/OCT?

Do they do it on barn door 99% lesions?

Hopefully not.

Hopefully they do it when they are not sure what's going on because there is nothing shockingly tight.

But actually medical decision making is complex and we rarely do it based solely on data in a database.

Propensity matching, I.e. adjusting for risk factor differences, is all well and good, if you have all the factors that matter.

But do we?

Ask yourself this.

When is the last time that you or an Interventionist you know, made a decision on what stent to put and where, based SOLELY on information in the database, I.e. without looking at the angiogram?

Why?

The reason for the answer is that the database is for summarising things that are NOT in the angiogram. The angiogram is key to deciding what to do because there is a lot of information there.

We don't know why people choose to do IVUS or OCT.

But we do know for 100% sure it was not based solely on the things that were typed in the database.

In fact if someone started reading out columns in a database when I was peering at an indistinct splodge on an angiogram, pondering a need for IVUS/OCT/etc, I would give them a piece of my mind.

So since the decision to IVUS comes overwhelmingly from NOT THE THINGS TYPED INTO THE DATABASE, what are the chances that statistical adjustment for these will restore the groups into magically retrospectively randomised status?

Well, you're right.

And Simon and his cheeky chappies know it. They're not cretins.

So they worded their Abstract Conclusion wisely.

So why am I bothering to harass poor Simon, other than for mere post-post-post-ORBITA hard feelings ?

Correct!

Much as I would like to give Simon and the the scribblers another drive-by beating, it's not their fault.

The abstract conclusions are correct.

(as is my Francis Fecundity Radix)

What went wrong was this:

.
.
sic transit gloria mundi
.
.

And I did promise there was a happy ending to this story.

It clearly wasn't my Pregancy Testing venture.

Nor did this give me a chance to berate Simon Redwood.

So what do you think is the saving grace in this cataclysm of cluelessness?

Hint

Why have I only been quoting from the Abstract? And not the Paper?

Hint 2

And that is the good news!

Our institution doesn't subscribe to this journal!

A rare time I have been happy with our poverty in UK.

Good day from the Primary PCI team at Hammersmith !

Thanks to Henry Seligman and Mark Sweeney for a fabulously excellent PPCI.

A complaint has flooded in.

Ironic that it is from a fellow.

This tweetorial, like most of my tweetorials, is for the benefit of fellows.

It does labour a very simple point, but one that people don't tell you when you are asking you to "analyse a big database we've got.

Biggest in the city/country/world"

The guy talking you into trawling this stuff isn't going to be sitting with you while you try to stitch the various bits of data together.

He isn't going to be sweating over the system crash when you finally get it all together and press "calculate P value".

He's not going to be spending the hundreds of hours with you that goes Into collating and analysing an observational database.

All he is going to be there for is to ensure you've spelt his middle name right.

So it's not a big investment for him.

It's a big investment for YOU.

So for him, it doesn't matter whether it is a good idea or not.

It only matters to you. It's your life. Think about it.

This tweetorial provides you with the questions to ask and the means to say no.

All this Propensity Matching business is about using the OTHER COLUMNS IN THE DATABASE, other than the treatment choice, to match people.

It ONLY works if the decision on treatment is almost always made exclusively using the information documented in those other columns.

If that is the case you are all clear.

Go ahead.

Propense away!
Match your little heart out!

But if people sometimes use non databased information, Just Say No.

"Sorry I've just become pregnant and I am scared of radiation from the computer."

"I'm a Zoroastrian and so I obviously can't help you here."

If they often use other information, don't say anything. Just edge away.

Later get a family member to email him to say you have unfortunately died.

If they ALWAYS use other information, don't do any of the above.

Call the police.