Tag: Machine Learning in healthcare

Machine Learning in healthcare is an AI capability. In Savana we use machine learning to generate predictive models in clinical research.

How to never become a world expert in AI.

A few days ago I was invited by Josep María Mascaró to give a talk at the Academy of Medicine of Catalonia.

Dr. Mascaró is ninety-something years old and he is still skiing. With that I tell you everything.

If you have not met him, I feel sorry for you.

Because he is one of those luminous people of whom you come across no more than 3 or 4 in life, who radiates kindness, positivity and lucidity. To the maximum.

What I won’t be at his age, no way.

So I accepted because it was him.

Because I hardly ever accept talks in which I don’t have a direct contact with the researchers or companies I have to reach.

If you think it’s bad, you tell my investors.

Better yet, tell it to the children of Savana’s one hundred and some employees.

Well, I will continue.

Dr. Mascaró, who looked at me with the same excited face for having gone there that my 1-year-old son looks at me with when I show him a cookie, introduced me as “one of the world’s leading experts in AI in Medicine”.

Wait, wait, wait.

One thing is that I do not hide the fact that I was one of the first doctors who saw AI coming. That I still say that is so.

But as a world expert, not at all.

I have just been sent the T-shirt of my town’s festival and on the front it says “what a fart we are carrying” and on the back “fuck, fuck, the world is coming to an end”.

No world expert in anything would wear that T-shirt.

And I have bought two in case I stain one of them.

Mascaró made a full-throated defense of machine learning when he introduced me.

It was great to hear.

Most doctors are not like him and have not yet realized the power of AI.

This is a bummer for patients, but, in truth, it is a unique and unrepeatable opportunity for those who have already seen it.

Like the internet or bitcoin back in the day.

In the talk I dedicated myself to debunking myths, which is what I do now when I am invited.

People like this, because they feel everything closer to them.

Look. And I am finishing.

Like Mascaró, I have a clock.

It reminds me that I have less and less time left to leave something of impact. in the world

Something my son will be proud of someday.

And sometimes the ticking is so strong, I can’t even sleep.

If the same thing happens to you, give me a call. Maybe there’s still time to do something.

The reason why Francisco Franco could not do AI.

One thing is clear.

In the end you can look outward or inward.

You can be in your territory, in your language, in your grandmother’s tortilla. As was Franco, the Spanish dictator, and many others.

Or you can be in the world.

And don’t tell me it’s compatible because it’s not.

At least not for people with the attention span of a moth, like me.

This subject of “being in the world” interests you for one thing.

Pay attention, a good lesson is coming today.

Science needs geographical representativeness.

It demands it.

If you are going to do an AI model in China, you have to know if you are going to be able to apply it in Canada.

We can call this transportability or transfer learning, although the fewer “new” words the better.

Many times we are bought for projects because of the scientific capabilities combined with AI capabilities.

But I kid you not.

We often get picked up because we have access to a lot of European data.

We move pretty well in GDPR, where American companies feel like a Belgian in a flamenco tablao.

That’s where we have a good part of the almost 200 hospitals in our network, although we also have in the Americas, the northern and the southern.

That is why we have read 3 billion clinical documents, more than anyone else I know of.

Because privacy is not a pain in the ass that has to be bullied.

Privacy, when you don’t talibanize it, is a very respectable value.

Because in the end your intentions will be transparent to everyone.

And when they see that you do not come to take any data on the sly to resell it to anyone, hospitals are very reasonable and are fine with working in the cloud.

In a cloud operated by them, of course.

In other words, the data is controlled by the hospital. But in a federated way it is uploaded to a cloud to be able to do research.

And that way you can make the studies international.

And AI becomes universal.

And everyone lives in equality, liberty and fraternity.

And that’s all for today.

Multi-continent AI projects, here.

The specific way to use data layers to predict.

They’ve opened a new restaurant right across the street.

Great news.

At first. Then less so.

Because the problem is that this week I went there and my illusions were instantly trampled to the ground.

As soon as I saw that the menu had 328 dishes.

Thanks to our spidey sense, we all know that the more dishes, the less quality.

Not everything can be done well.

Not even geniuses like politicians can.

You can’t with restaurants and you can’t with health data.

That is why the monster companies that offer you all the data end up providing poor things.

I am referring to those that do the same for a survey of doctors, or a clinical trial, or give you data from the pharmacy.

In other words, they give it to you, but of poor quality and a little bit without knowing what to do with it afterwards.

Quite a bit.

A project that on day 1 has genomics, proteomics, clinical variables, imaging, etc., is not reliable either.

Wait…

It is true that the more data layers, the greater the capacity to cluster patients.

Multimodality in itself is good.

The problem is putting it into practice.

That’s why it has to be treated with care. It doesn’t work either way.

You have to choose the right data layers, depending on the problem.

And not in any order, no. First some and then others.

All according to the objective to be achieved.

It is simple, but you have to do it well and be very methodical.

And if it is done, then yes.

Then you can do like that institution that upon seeing our database in a type of a cancer has convened the entire team in Europe and beyond so that each member can make their list of what they have called “disruptive questions”.

And for the record, I don’t like to disrupt. Neither disrupting nor arguing.

If you want to set up a multimodal big data project with feet and head, here.

What I learnt about forescasting from the US Government.

If I ask you what predictions you’ve gotten right in your life, almost certainly one or two will come to mind.

What you saw coming before it happened.

It’s also likely that the guess evokes in you a kind of regret that you didn’t do anything about it.

“I got it right, but it didn’t do me any good because I didn’t do anything with it.”

There are some very clever people out there who actually do something with forescasting. Putting money in or whatever.

They don’t just sit at the bar, like you and me.

If I had done something about it I would have been able to monetize things I’ve gotten right in my life, like:

– The rise of China

– Craft beer

– Bitcoin.

A few years ago I studied in California with some gentlemen who worked for the American government and who had some methodologies to get things right.

They structured and systematized uncertainty. Obviously they didn’t get everything right, but they got something more than average.

It is curious that there are a couple of techniques for this.

And it’s not the tarot or a crystal ball.

One is to look for strange or surprising things. If something doesn’t add up on the street or in the newspaper, watch out. There’s a tendency there.

The other fascinates me…

And that is that some things are so big that we don’t see them.

Like the particular day when more people moved to live in cities than in the countryside. It was around 2010 or so.

The day Humankind reached its peak oil usage. Also around that time.

These are things that are so big that you don’t see them,

And there are trends there.

This happens with data in health.

It’s so big that most people don’t see it.

You don’t see what you can get out of it.

And we’re not talking about complicated things like asking disruptive predictive questions.

We’re talking about everyday things, like understanding the effectiveness of an intervention.

Or pulling out epidemiology where you only had indirect and biased studies.

Or knowing which biologics are being most cost-effective by patient phenotype.

Look, I only got going with one of my accurate predictions.

And that was with AI in Medicine.

It’s still a little while for most to realize its power.

If you want to get ahead of the curve, it’s here.

The obvious reason why you shouldn't say "big data" anymore.

Hey, maybe we could stop using the expression “big data”.

In the worlds that are more advanced than ours in technology, almost nobody says it anymore.

And I guess the reason is that it is redundant. In other words, it is already understood that in 2022 almost all “data” is “big”.

In any case, we could talk about “small data” when the opposite happens.

But, well, this is just me being a little bit of a fad.

What interests you is to realize that having brought massive data analysis techniques to healthcare has allowed us to see where we could not see before.

So please pay attention to this. It´s important,

It is possible that you have ever had your hands on a database.

And you thought you had a lot of patients.

But gradually there were fewer and fewer of them as you were left with:
– women

– of age between X and Y

– with breast cancer

– but not just any breast cancer but this breast cancer

– but not just this breast cancer and that’s it

– but it also has to have this marker
– and this staging

– and this mutation

– and this treatment

And yeah…

It turns out that in the end there are very few.

And what was big, a little while after the excitement, is already small.

Ooooops.

If you were trying to describe the epidemiology, you still don’t care….

But if you were trying to look at the effect of a drug…

…welcome to the “effect size dilution.”

Our inseparable companion.

Against which, however, we have recently found a solution.

It has to do with not taking a few data here and there, but taking them all.

But all means all.

Of course, this cannot be done by human beings, because it would be unmanageable and unaffordable.

A machine can do it.

In other words, if I want all the music, I don’t go to a store and get all the records, I subscribe to Spotify. You know what I mean.

The machine has to be able to read any format, any file, any system from any hospital. In multiple languages.

Otherwise, no party.

We call our machine to do that Savana Data Gatherer.

But well, the name is the least important thing.

What matters is that it works.

And yes, it is extremely respectful of privacy and data ownership.

You say how big you want it, from 1 hospital to 3 continents.

So if you want to avoid effect size dilution, it’s here.

Tag: Machine Learning in healthcare

How to never become a world expert in AI.

The reason why Francisco Franco could not do AI.

The specific way to use data layers to predict.

What I learnt about forescasting from the US Government.

The obvious reason why you shouldn't say "big data" anymore.

The AI + RWE Company

Want to use it?:

Start with your proposed AI + RWE use case:

This is the first step for AI + RWE: