Tag: AI Clinical Research

The generation of RWE at multisite level using Natural Language Processing on Electronic Medical Records.

Tip to find patients with a certain pathology.

A fellow doctor recently told me how he once was on a flight in business class with 30 oncologists and that, when asked “is there a doctor on the plane?, they all bowed their heads and none stood up.

I find this surprising because I know more doctors who find it natural to help than the opposite.

And furthermore, it can only be explained by laziness, not by legal risk, because when I was at the hospital and we were the reference site for the airport (many inflight strokes were brought to us), reviewing the subject for a clinical session, there was no precedent of complaints against doctors for assisting on an flight.

I once took care of a first seizure right next to me.

I did nothing more than reassuring, but isn’t it a big coincidence that you have a first seizure on a flight and the person next to you is a neurologist?

And again, and perhaps one of the most beautiful stories of my life, I decided in 60 seconds like in a movie (because that was the time they gave us at the New York airport before losing the slot to leave) that a flight attendant with a decreased level of consciousness had nothing serious and that we could take off.

The funny thing is that the flight attendant had a history of alcoholism and apparently she was gambling with the custody of her daughter, and if we had not taken off, with the millions of dollars in losses that this would have entailed, it is likely that her life would have been derailed.

I´m not making up the story.

This was told to me months later by the pilot, whom I befriended and remain friends with to this day.

Things connect, my friend, and it’s not like Steve Jobs on Linkedin talk; they really connect. All the time.

That’s why it’s better not to keep your head down too much.

Or at least to be as little of an asshole as possible.

And I’m an asshole everyday and sometimes I don’t recycle.

But the chained and collaborative stories lead me to the Pharmacy Service of a big hospital (Sant Pau in Barcelona), where they have always welcomed our AI technology with interest and have persisted in its use until, finally, we have been able to demonstrate that with Savana we have found 257 patients of a certain pathology that they were looking for (I will not say which one in case of privacy), when with their gold standard information system they found 206 patients.

And this is less glamorous than attending stewardesses at JFK with 300 people looking at you and then getting out of the plane like a bullfighter.

But I think the thing is much more serious and more important than the ego of a simple doctor.

It’s about something as basic as using technological tools that already exist in other fields to count patients.

Counting patients. That’s what we do at Savana sometimes.

How small.

And how big.

To implement a boring Artificial Intelligence system, with no colored lights, but which finds any patient with anything, at any point of care, in English, French, German, Portuguese or Spanish, this is where you have to ask.

2.5 million patients analyzed with AI.

When you watch an interview with a well-known actor, you realize that one thing they have in common is that they tend to be quite shy.

This may come as a shock at first, but what happens is that good actors are precisely shy people who became actors to hide their shyness behind a character.

It is as if, while they are acting, they rest from their social anxiety because the mask protects them.

These paradoxes or psychological compensations are interesting and they are everywhere…

and I think that’s what happened to me somehow, when I got into Artificial Intelligence.

I’m so impatient that it kept me from doing research.

And it bothered me because I think research is the most noble and interesting thing a person can do.

So I invented a world in which I could do research by being impatient.

Lots of data, lots of patients, but with a fraction of the effort.

Speed and immediacy, without spending a lot of time entering data.

Associations and correlations without having to spend 20 years in a practice to intuit them.

That’s why I often describe machine learning as “sensing machines”.

If we had not done all this with Savana, it would be impossible to have published the article we have just published called “LiverTAI, a retrospective analysis of EHRs through Natural Language Processing”.

And which has the peculiarity of having studied the patient journey of almost 2.5 million patients in terms of HCV testing.

2.5 million is a lot of patients.

So, if you want to study a population and you don’t have the time or the desire to do it by the traditional method, let me know, maybe we can do something.

Article about AI and Value Based Contracting.

We don’t know if you know that Russian soldiers were caught flirting with Ukrainian women on Tinder.

That was in February.

And since they were still doing it in September, Ukraine used hackers to impersonate these women and thus located Russian army positions.

This reflection on limitless human stupidity reminds us once again that digital data is and will be around.

And that its use can be beneficial or dangerous, depending.

And that to walk on the Internet mindless is like letting people drive without a license.

Or like crossing the street without looking at the traffic lights.

You can’t do that.

AI is not good or bad, it depends.

What does it depend on?

On how you look at it.

At what?

Well, sometimes you’ll be using it to see things you can’t see with normal human eyes.

Or, to do things that you can do, but it takes too much time and effort, and it is possible to automate.

We have written an article.

We wrote it together with José Luis Poveda, Head of Pharmacy at La Fe, in Valencia.

We talk about the use of AI to measure the results of drugs in all patients and thus be able to make risk-sharing agreements.

In other words, you only pay for drugs that work.

But watch out, you charge well for those.

No self-respecting modern company at this point is interested in having a drug that doesn’t work.

Some call it providing value.

We call it not paying for the car if the car doesn’t work.

But it used to be very difficult, because you had to pick them up by hand.

Now a machine does it.

We’re going to try to make a first case for a value based contract using AI.

Are you in?

How to write a headline involving AI.

When you have your AI model done, validated and published, it’s time to communicate it to the world.

And the important thing to realize is that you can go from less sensationalist to more sensationalist as you see fit.

Level 1: scientific purist.

The biomarker NG17 may have implications for the development of hypophosphatemia.

Only later in the body of the text will you mention that Random Forest (a form of machine learning) was used for its development.

Pros: disease experts will respect you.

Cons: you reach 0.001% of the audience.

Level 2: you still retain dignity, but you are already starting to have fun.

Researchers develop a predictive model for the development of hypophosphatemia.

Again, it is in the body of the message when you mention that machine learning has been used. Moreover, you still stress that it is researchers, not machines, who do things.

Pros: no one spits in your face at the congress.

Cons: you are still only reaching people with an interest in this pathology.

Level 3: it’s all about impact, not accuracy.

A machine learning model is able to predict the development of hypophosphatemia.

Bye bye humans. Here we have already crossed the abyss, by putting the machine learning concept in the headline.

Pros: journalistic arrival level is high and there will be many likes on Linkedin.

Cons: your fellow residents hate you .

Level 4: what the hell, we are here to play.

An Artificial Intelligence predicts who will develop metabolic diseases.

There is no shame. AI is presented as an individual and autonomous entity that can destroy humanity at will. Moreover, you generalize, as if it can maneuver in any disease.

Pros: you might get it published in general media.

Cons: you can never set foot in a hospital or apply for a serious grant again in your life.

Look,making an AI algorithm has its own set-up.

Then you have to publish it, communicate it, use it with the agency or administration you are interested in… and even implement it in clinical practice.

If you’re going to do it, do it with the best. Here.

Evaluating a drug in a ridiculous amount of time.

To write a book, to take care of an orchid garden or to decorate a living room, amateurs are often better than professionals.

What happens is that amateurs usually have a passion that you don’t find in those who do things as a way of life.

Because what you do for money and routine becomes mechanical, lacking creativity, without soul.

That’s why for many things I don’t believe in professionals, because it seems to me that sometimes professionals are like prostitutes to love; I hope f I’m making myself clear.

I see this with medical publications.

The “professionals” of the publications, that is, those who only dedicate themselves to doing studies, do not usually publish things that change clinical practice.

I’m not saying that what they research isn’t interesting, but often, it has little impact.

While clinicians, in addition to seeing patients, publish about what they see and, of course, they do have an impact.

At Savana we had a department dedicated to talking to clinicians.

And we closed it.

We removed it because departments are not necessary when something comes out on its own.

We are good at talking to doctors because we speak their real language.

That of having little time and a lot of desire to do things.

That is why our obsession, our obstinate obsession, is on how to shorten the time of the studies more and more.

How to have the data clean, refined,… as ready as possible so that when someone thinks of launching a question, we only have to execute it.

This is what happened with the article “Use of N-Acetylcysteine at high doses as an oral treatment for patients hospitalized with COVID-19”, done by pulmonologists in ridiculously short times, thanks to the fact that they had the data in Savana ready for the time of their question.

So, if you want us to use our AI to set you up with an automated registry on any cohort of patients, against which to throw unlimited questions, we can talk about it here.