Apple Watch can detect diabetes with 85% accuracy

[DrMacintosh]

Quote

A new study from Cardiogram shows that by using heart rate monitors on wearables like the Apple Watch, neural networks can now detect whether the wearer shows early signs of diabetes with astonishing accuracy

Quote

Cardiogram’s findings came as part of its larger DeepHeart study with the University of California, San Francisco. Using data from 14,000 Apple Watch users, the team used its algorithms to detect that 462 people in the study had diabetes. That meant an 85 percent accuracy rating.

Quote

We’re not quite at that point yet, but the study’s recent finds are an encouraging sign to diabetics. More than 100 million adults in the United States live with diabetes or prediabetes. About one in four of those with diabetes don’t realize they have it.

Apple Watch is apparently almost at the point that it can accurately detect diabetes using neural networks and it’s existing sensors. 

 

This tech can can be added to Android Wear (and presumably Gear watches) as well but in my opinion their market presence makes support for those watches mostly being an after thought. 

 

This really does show that the Apple Watch was always geared towards fitness over being a “smart watch.” 

 

Source: https://www.cultofmac.com/527743/apple-watch-detects-diabetes-85-accuracy/

[Zodiark1593]

Yeah, now to hack the thing so I don't need an iPhone to use one.

 

[Guest]

And still apple watch cant have a always on display... android wear ftw

[GoodBytes]

The article says:

Quote

A new study from Cardiogram shows that by using heart rate monitors on wearables like the Apple Watch, neural networks can now detect whether the wearer shows early signs of diabetes with astonishing accuracy.

Keyword: LIKE the Apple Watch.

So, in other words: "Apple Watch" like products. So any smartwatch with a good heart rate sensor and not those found in cereal boxes.

 

The article that this news refers to, says:

Quote

Android Wear, Garmin, or Fitbits

Let an Apple fan site make it sound like it's only Apple Watch that can do the job, by excluding the other products mentioned in the study.

 

Also the study is from a company that collected and analysis user heart rate data from its own app, on Android and iOS:

https://play.google.com/store/apps/details?id=com.cardiogram.v1&hl=en

https://itunes.apple.com/us/app/cardiogram-for-apple-watch/id1000017994?mt=8

 

So again, highlighting Android wearable devices.

 

So the title of the news should be: Smartwatches can detect diabetes with 85% accuracy.

 

 

 

Another fault in the news you are quoting:

The study says:

Quote

14,011 users of Cardiogram for Apple Watch and Android Wear were recruited into UCSF’s Health eHeart Study

Yet, the news say "14,000 Apple Watch users". Clearly trying to boost Apple Watch user base. These numbers are not speerated in the study. In fact, you can't even find "14,000" anywhere.

[FakeNSA]

7 minutes ago, GoodBytes said:

 

 

So the title of the news should be: Smartwatches with decent HR monitors can detect diabetes with 85% accuracy.

FTFY, because some of them just suck!

[mr moose]

That's great, but the people most at risk probably don't use health/fitness apps. 

 

 

[TheRandomness]

1 minute ago, mr moose said:

That's great, but the people most at risk probably don't use health/fitness apps. 

 

 

Depends on if it's detecting type 1 or type 2.

[vanished]

So really this story is about how an AI can notice subtle patterns in a heartbeat that signals they may have diabetes (or are close to getting it), and how that heartbeat was detected is kind of totally irrelevant.

[mr moose]

7 minutes ago, TheRandomness said:

Depends on if it's detecting type 1 or type 2.

does it matter what it's detecting if they are using it?

[GoodBytes]

2 minutes ago, Ryan_Vickers said:

So really this story is about how an AI can notice subtle patterns in a heartbeat that signals they may have diabetes (or are close to getting it), and how that heartbeat was detected is kind of totally irrelevant.

Yup! And that can be done with any smartwatch with a good heart rate sensor, and not Apple Watch exclusive, such as Fitbit, Garmin, among other Androidwear devices, as the study clearly states.

[vanished]

5 minutes ago, mr moose said:

does it matter what it's detecting if they are using it?

I don't know what you mean, but the point is I strongly suspect that type 2 correlates with "unique" heartbeats more than type 1, so if they can also detect type 1 this way that would be interesting and impressive.  I also suspect that the "have diabetes, don't use fitness bands" idea only applies to people with type 2 as well.

[BuckGup]

Very cool. Getting my Apple Watch was one of best tech purchases I have made. Mainly because I just like data about my life and knowing what’s going on. So it’s cool to see things you would normally never been able to

[mr moose]

24 minutes ago, Ryan_Vickers said:

I don't know what you mean, but the point is I strongly suspect that type 2 correlates with "unique" heartbeats more than type 1, so if they can also detect type 1 this way that would be interesting and impressive.  I also suspect that the "have diabetes, don't use fitness bands" idea only applies to people with type 2 as well.

My previous post was meant to say that those who are at most risk of developing diabetes are not likely using fitness apps (overweight, busy and disinterested in such things),  so it doesn't really matter which type it detects as those most in need of earlier detection aren't using it.    Those who already have diabetes don't need a diagnosis. 

[vanished]

Just now, mr moose said:

My previous post was meant to say that those who are at most risk of developing diabetes are not likely using fitness apps (overweight, busy and disinterested in such things),  so it doesn't really matter which type it detects as those most in need of earlier detection aren't using it.    Those who already have diabetes don't need a diagnosis. 

and I'm saying that overweight stereotype doesn't apply to both types

[mr moose]

3 minutes ago, Ryan_Vickers said:

and I'm saying that overweight stereotype doesn't apply to both types

it wasn't a stereotype, it was just an example.  people most likely to use a fitness app/health tracker are already looking after their health and are by their very nature in a lower risk category for both type 1 and type 2.   those who do not use such an app are not going to be detected and given the latest statistics on weight in the US (and Australia for that matter) there is a higher chance they are overweight adding to the risk.

[Energycore]

Reminds me of a video from Veritasium on why you have to double check stuff like this: 85% accuracy on the Apple Watch test is a lot less than you'd think.

 

So just because the watch says you have it, doesn't mean you have it. Go see a doctor if it does.

 

UDPATE: Here's the actual probability that you have diabetes if you're diagnosed by Apple Watch, and live in the US

P(D): Probability that you have Diabetes (living in the US) = 9.4%

P(D*): Prob. that you don't have Diabetes = 90.6%

P(W|D): Prob. that Apple Watch says you have Diabetes, if you do have it = 85%

P(W|D*): Prob. that Apple Watch says you have Diabetes, but you don't = 15%

 

The computation then goes

 

P(D|W)=[P(W|D)*P(D)]/[P(W|D)*P(D)+P(W|D*)*P(D*)]

 

Which computes to

 

P(D|W)=[0.85*0.094]/[0.85*0.094+0.15*0.906]=0.3702...

 

So, assuming these calculations are correct, and Siri on your Apple Watch goes and says "hey, be careful there's a chance that you have diabetes", then you'll actually have diabetes 37.02% the time.

 

On the other hard, if you go see a doctor and they conduct tests that end up positive, then you're most likely to have it.

 

So please folks, double check every time. It's worth it for your health.

 

EDIT: Had some numbers wrong

[Master Disaster]

23 minutes ago, Energycore said:

Reminds me of a video from Veritasium on why you have to double check stuff like this: 85% accuracy on the Apple Watch test is a lot less than you'd think.

 

So just because the watch says you have it, doesn't mean you have it. Go see a doctor if it does.

That's just common sense though. 

 

I always find it incredible how many people are willing to blindly follow the advice of a machine without consulting any kind of expert first.

[Master Disaster]

1 hour ago, GoodBytes said:

The article says:

Keyword: LIKE the Apple Watch.

So, in other words: "Apple Watch" like products. So any smartwatch with a good heart rate sensor and not those found in cereal boxes.

 

The article that this news refers to, says:

Let an Apple fan site make it sound like it's only Apple Watch that can do the job, by excluding the other products mentioned in the study.

 

Also the study is from a company that collected and analysis user heart rate data from its own app, on Android and iOS:

https://play.google.com/store/apps/details?id=com.cardiogram.v1&hl=en

https://itunes.apple.com/us/app/cardiogram-for-apple-watch/id1000017994?mt=8

 

So again, highlighting Android wearable devices.

 

So the title of the news should be: Smartwatches can detect diabetes with 85% accuracy.

 

 

 

Another fault in the news you are quoting:

The study says:

Yet, the news say "14,000 Apple Watch users". Clearly trying to boost Apple Watch user base. These numbers are not speerated in the study. In fact, you can't even find "14,000" anywhere.

You're wasting your effort I'm afraid, I doubt the Doc would even have posted this at all had it contained anything about Android, FitBit etc in the article.

[Energycore]

Just now, Master Disaster said:

That's just common sense though. 

 

I always find it incredible how many people are willing to blindly follow the advice of a machine without consulting any kind of expert first.

I think you're misguided. If an expert ran a test on you with 95% accuracy and you got out positive, your probability as a US citizen to actually have it is still only 68%. It's a higher number, sure, but a bit less than a third of the time, you don't have diabetes.

 

The equation approaches 1 as the accuracy of the test approaches 1, but it's (almost) always a good idea to get a second opinion.

[seon123]

42 minutes ago, Energycore said:

P(D): Probability that you have Diabetes (living in the US) = 9.4%

P(D*): Prob. that you don't have Diabetes = 80.6%

What about the remaining 10%?

[Energycore]

Just now, seon123 said:

What about the remaining 10%?

OH SHIT I FUCKED UP

[seon123]

1 minute ago, Energycore said:

OH SHIT I FUCKED UP

37% in that case. That's probably worth checking with a doctor for

[Energycore]

Just now, seon123 said:

37% in that case. That's probably worth checking with a doctor for

I would say it's worth it; if you don't, 1 time out of 3 roughly, you will have diabetes and you will not have started treating it on time. I don't want to play with that risk personally.

[79wjd]

41 minutes ago, Master Disaster said:

I always find it incredible how many people are willing to blindly follow the advice of a machine without consulting any kind of expert first.

I'll defer you to Yahoo Answers -- the only medical advice needed. 

[vanished]

8 minutes ago, djdwosk97 said:

I'll defer you to Yahoo Answers -- the only medical advice needed. 

And don't forget WebMD.  It detects AIDS and Cancer with 100% accuracy by assuming everyone has it, thus missing an astounding 0 cases!