IBM: Our AI correctly predicts onset of Alzheimer’s 71% of the time, better than standard clinical tests

Machine-learning algorithms analyzing human communication can predict whether someone will develop Alzheimer’s disease more accurately than standard biomedical screening, say IBM and Pfizer. IBM claims its software was capable of correctly predicting the onset of Alzheimer’s 71 per cent of the time, compared to 59 per cent for …

[Screwed]

Just what will, indeed could, be done with this information?

Assuming they can correctly predict Alzheimer's, so far as I know there is currently no effective treatment. In which case, about the only advantages I can see are a) allows time to make decisions and tidy things up; b) avoids treatment for other possibilities. And, just how long before any real deterioration would anyone wish to know this diagnosis?

But what if they wrongly predict Alzheimer's and the issue is actually a treatable condition? I have seen someone with a thyroid conditions more or less diagnosed with dementia and, possibly, Parkinson's. I am aware of people with Pernicious Anaemia who have been diagnosed with brain and mental conditions. Both of which can improve massively with treatment. (Though there are still major arguments about the exact choices of treatment.)

[BebopWeBop]

I take your point on incorrect diagnoses. and a 29% failure rate is probably not one that would cause me do all the paperwork and quietly top myself, but if it is an improvement on current diagnoses and can be used to direct other tests then it is better than what we can do at the moment.

On the sentence construction how would Hemmingway have done:-? Maybe I have misunderstood, but this would appear an appropriate way of monitoring changes in language and therefore part of a continuous monitoring system?

[Harry Kiri]

1. The supposed increase in performance uses a different data set. You cannot claim an increase in performance for your technique if you use different data.

2. You also cannot claim an increase in performance if the false positives aren't stated - just saying everyone is of class X, correctly identifies everyone may be in class X. They may not have Alzheimers or they may have something else as Screwed rightly points out.

[The Mole]

My (very limitted) understanding is that whilst there is no cure, there are treatments and therapies that can help slow down the progression of the disease.

If you catch the disease very early and can halt or dramatically slow down the progression then there is a chance the person can lead a nearly normal life. It may even be medicines can be developed to reserve small amounts of damage that aren't effective on later diagnosed patients with much more severe symptoms.

Part of the problem with slow progression diseases is that research into possible preventative medicines (like asprin for heart disease/stroke) takes a very very long time to detect meaningful results. A more sensitive means to track deterioration may help speed up those investigations even if it isn't 100% accurate.

[Jim84]

Leucadia Therapeutics are pursuing the hypothesis that ossification of the cribiform plate with aging is a cause of Alzheimers:

https://www.fightaging.org/archives/2018/08/undoing-aging-doug-ethells-presentation-on-the-leucadia-therapeutics-approach-to-treating-alzheimers-disease/

Finding people at risk of Alzheimer’s and treating them with Leuicadias solution and then seeing how many go on to develop full Alzhiemers May work better than trying it on already mentally stuffed individuals who may be too far gone.

[Danny 2]

@TheMole

You're completely correct about early diagnosis. A family friend who just happened to one of the first specialist dementia nurses in Scotland correctly diagnosed my father early on many years ago. He was prescribed medicine that slows the decline, and while he is not in great shape (he is currently on his hands and knees looking for a pill he dropped) he is in much better shape than my mates mum who was diagnosed years later. She is in a care home and in such a bad shape that her family have stopped visiting her.

I could expand in depth but for now I have to look for a missing pill.

[Lord Elpuss]

What The Mole says. There's currently no cure, but there are things you can do to slow onset; diet is believed to be a contributing factor, as is (mental + physical) exercise, and there are medicines that can also slow the onset, but can't reverse damage. Knowing it's on the cards might be the trigger some people need to eat better/get out more - I know this would probably work for me, plus you can start taking medicines to prevent damage before it occurs.

It's similar to people who are clinically obese but not (yet) suffering medical symptoms; if a doctor could tell you that there's a 71% chance you'll be dead in 5 years, that's a far more powerful message for behavioural change than "Don't be fat because fat people get ill more."

[RM Myers]

Screwed makes a good point about treatable conditions which can mimic Alzheimer's and other neurodegenerative disorders. There are a large number of autoimmune diseases and infections which sometimes get diagnosed as Alzheimer's, and most can be treated and, if not cured, at least be controlled. Since many of these are very difficult to diagnose, the most important step is to eliminate the neurodegenerative diseases. For Alzheimer's, the best way is a amyloid pet scan, which can eliminate Alzheimer's as a diagnosis even though it can't directly confirm Alzheimer's (you can have amyloid plaques without having Alzheimer's, but you can't have Alzheimer's without amyloid plaques). And yes, thyroid autoimmune disease (Hashimoto's) is an example of an autoimmune disease which is sometimes diagnosed as Alzheimer's, and can be treated.

[Anonymous Coward]

"Just what will, indeed could, be done with this information?"

Looking at their senior management and financial performance over the last 10+ years, I think they may use it for C-level recruitment.

[RM Myers]

Oh come on, just because you are ignorant doesn't mean you're senile. Some of us Some people were just born with limited intelligence.

[David 132]

Some people were just born with limited intelligence.

And some of us had to work hard to acquire it. Weeks of watching TOWIE. So many hours spent reading particularly stupid tabloid gossip columns.

BUT IT WAS WORTH IT.

[b0llchit]

Perfection paradox

And when the AI system has been perfected to recognize the onset of the disease, then it will predict its own demise by forgetting to tell the user about its own condition. Perfection can only mean that it will share the human condition and ultimately degenerate until no viable matter remains.

Now, where is that AI that can ingest Beer?

[Anonymous Coward]

Re: Perfection paradox

But as the AI can recognise the disease earlier than any human without aid of the AI, it cannot be measured by the 'perfection' standard implied in your message - i.e. the assumption that being human is achieving a state of 'perfection', therefore it cannot be compared to a human, cannot be assumed to develop the disease, and cannot degenerate because of it being itself affected by the disease...

[JohnMurray]

I'm guessing...

...that there is no chance either of the US presidential candidates will be taking the test anytime soon.....although it looks as if Biden would trump Trump...

[Ken 16]

Re: I'm guessing...

I was just thinking they should play the presidential debates into it.

[IGotOut]

Re: I'm guessing...

It's odd, which one has repeatedly denied he has said something, despite it being recorded.

[IGotOut]

Meh.

I rank that along with 84% of women said their skin felt better after miracle cream (sample size 97 people).

BTW I bet the relatives of those people got WAY higher results.

[Magani]

Testing, testing...

"Speaking in short sentences with less complex grammatical structures, and a repetition of the same words, are often signals that a patient is at a higher risk of developing Alzheimer’s disease,..."

Now which well-known politician could that description fit?

[nxnwest]

Re: Testing, testing...

American or European?

[JohnMurray]

Re: Testing, testing...

Or British! #Brexit

[Lord Elpuss]

Re: Testing, testing...

"American or European?"

I know you're gunning for Boris here, but he has the opposite problem to the orange f*cktrumpet; he suffers from incurable verbal diarrhoea. He uses ten words where one will do, talks like he's swallowed a thesaurus, is verbose to the point of tedium, and seems incapable of condensing an argument down to concise points.

Which makes me wonder about the AI: how would it recognise the difference between somebody who uses few words and speaks in short sentences because they are inherently incapable of doing otherwise, and somebody who is highly educated and trained in concise communication, and can convey meaning with simplicity and clarity?

[CAPS LOCK]

I think I'm there already...

...what time is the next train to Switzerland?

[Anonymous Coward]

Specificity / Sensitivity?

In all of the AI hype in Medicine , (e.g. Cancer detection)

The numbers being hyped don't pan out.

Not knowing sensitivity/specificity makes any report meaningless.

[Snowy]

Yes...

but it is still not Ai!

[Val Halla]

Perhaps IBM could also solve the conundrum of Covid.

[Doctor Syntax]

IBM has yet to solve the conundrum of IBM.

[Anonymous Coward]

AI vs AD - IBM is Practicing Double Standards

It’s interesting that a company who has been actively targeting older loyal employees for repeated rounds of “Resource Actions” (RA’s) through the manipulation of redundancy scoring matrix and then forcing legal minimum redundancy payments by geography, seeks to develop AI diagnostic tools for degenerative deceases like Alzheimer’s.

Perhaps it’s AI capabilities should be used to review employees that have been made redundant under IBM HR / Executive driven head count refresh programmes like project “Baccarat”. As they say good practice on age discrimination (AD) vs AI should start at home ?

[Jay Lenovo]

Re: AI vs AD - IBM is Practicing Double Standards

Selective AI is good for you, not for me.

I'm IBM, I do what I wannt!

[Schultz]

This research is hard to assess ...

The claim is that "the software was able to successfully predict the onset of Alzheimer’s before an official diagnosis correctly seven in ten times." But there is no information about the error rate and the false positive rate of the diagnosis. Simple example: I can predict that all study participants have early stage Alzheimer's and I'll have a 100% success rate in predicting the onset of Alzheimer's disease (with a few false positives). And, of course, things only get more complicated from there, as the traditional diagnosis will have its own false positive and false negative rate (you have to wait for someone to die before you can diagnose the physical changes inflicted upon upon your brain by Alzheimer's disease).

That's the reason why it takes quite some time for a scientific break-through to make it into clinical practice and why most "break-throughs" slowly move towards the waste-bin of no relevance. Medicine is a complicated, messy business and I have my doubts whether the PR-driven stories from this week's or last actually matter. But then, if you can combine two or more hot topics into one story it clearly deserves attention!

[cd]

Can they use it to see when a company is senile? They should turn it on themselves.

[Doctor Syntax]

There seems to be an assumption that speech is in regular sentences whether long or short. What would it make of continuous rambling with no such structure? Just diagnose politician"? And no, I'm not necessarily thinking of BoJo.