Diabetes: Why @IBMWatson Sucks The Big One

You might remember IBM’s Deep Blue computer, that managed to beat reigning Chess World Champion Gary Kasparov. In essence, rather than simply playing chess, DeepBlue looked up a huge list of games from other chess champions, and replayed them against Kasparov. In the end, an usual move by DeepBlue, actually the result of a programming bug only admitted 15 years later, sent Kasparov into a tail-spin.

So fast-forward to 2016, and DeepBlue is now remarketed as IBMWatson, in a fairly thin reference to Sherlock Holmes’ assistant (who was a doctor).

For the last 2 years, every digital health meetup, or medtech meeting I have been to, both in Australia and in the USA, has a speaker from IBM, who simply shows a concept video of how IBM Watson is going to revolutionize health care.

But for 2 years, I have not seen a product demo. And I have not spoken to anyone who has actually used it.

But I keep seeing the same marketing video – over and over again.

And when we were looking at organizing online courses for nurses to learn about PredictBGL, we found that IBM had paid to put up its own promotional videos! Imagine, an entire generation of nurses getting CPD (continuing professional development) points for watching IBM’s promo video! And yet, not one of these nurses will be using anything like IBMWatson for 15 years.

Vaporware, IBM?

And now IBM has overextend itself again, with a move to predict blood sugars for diabetics.

With diabetes – it’s very difficult to generalize about two patients. Two people with the same height, weight and level of fitness can have completely difference insulin doses. This is why diabetes requires self-management, because ultimately, you have to know yourself better than your doctor. Population models do not work.

But according to Medtronic’s cautious partnership release, IBMWatson will use “…data combined with numerous other sources of data such as electronic medical records, health insurance claims and population health data to uncover patterns and predict health risks using advanced analytics models.”

This ‘SugarWise’ looks like the DeepBlue model over again – find someone else from the population who ‘looks like’ person A, and then assume that their life, eating patterns, sleep/wake routines, medications, insulin doses, biometrics and metabolism are the same – and hey presto! A blood sugar prediction.

Except it will be crap.

Because there is no way that even my life today resembles yesterday, and there is no way that you can assume that someone else’s life can be replayed to predict mine.

Statistical population models cannot predict your blood sugars right now.

For example, the risk of you having bowel cancer aged 50+ might be 0.2%. But what if you actually have bowel cancer, and the doctor refuses to test for it because of a population-based cost-benefit model? Your personal statistics are different to population statistics.

The same with the risk of a Hypo (low blood sugar). The population model might say that few people have a Hypo at 8:15am. But how does that help you if you DO have a hypo at 8:15am?

If you want to see the revolution in diabetes self management, and REAL prediction of future blood sugars – checkout PredictBGL.com.

No Watson involved. No Hype. No Vaporware. No invasive devices required.

And we’ve been doing it for 2.5 years.

In the App Store and Google Play right now.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s