Hacker News Reader

Comment by pm90 | Hacker News Reader

The actual blog post is much more informative: https://blog.verily.com/2018/11/update-on-our-smart-lens-pro...

Sounds like they couldn’t, with today’s technology, measure glucose levels to the accuracy required for medical devices. But will continue work on measuring other things with those lenses.

I don’t know why the negative tone of the article. Sounds like they tried something and demonstrated their specific solution didn’t work. Most research experiments generally don’t.

View on HN

A lot of people in the scientific community had suggested for a while that glucose levels in eye fluid were not sufficiently correlated to blood glucose levels for glucose measured in tear fluid to be clinically useful. So even if you could detect glucose in eye fluid, it wouldn't help you manage disease. In the eyes of some this was an obvious limitation that the company brushed aside, and there was a fair amount of hype around it, so there were some ppl calling it vaporware

Although some of the core tech around engineering all the sensors, battery, Bluetooth etc into a contact lens is pretty cool. There are a lot of patents from the inventor, brian Otis (who wrote the blog post), if you want to take a look.

But based on the blog post it seems like they were having trouble measuring glucose in tear fluid, which is one step upstream from the aforementioned issue around lack of correlation btw eye and blood glucose. If you can't accurately measure the analyte it doesn't really matter if it's correlated w disease

"in the eyes of some..." - thought you could sneak that by HN? I see what you did.

You saw what they did there eh?

The technology is definately more than cool, hopefully some of that tech can be repurposed into something that can make it clinically.

Worth noting, tear glucose wouldn't need to be correlated in a linear fashion with BGL, as long as there was a way to derive BGL from tear glucose data.

I think the opposite is true. To achieve medical usage the correlation should be linear. If this is not the case, a small variation of glucose could lead to big jump in the reading, which would decrease the accuracy. In real life, most of the biological system or test exist in a buffered environment, which often show a range of linear relationship for substrates, and tests are often performed by adjusting into such linear range.

If a small variation in glucose led to a big jump in the reading, then the glucose calculation would end up with greater precision.

That is probably the catch. Tear glucose may not be all that well correlated with blood glucose.

The term "correlated" is often used as shorthand for having significant mutual information. The possibility of a nonlinear relationship is lost on very few.

Aside, spelling: it's definite, like finite/infinite. Think "the finite is more definite".

I had the idea of measuring the index of refraction in the aqueous humor using something like a multi-perspective speckle interferometry on the iris.

I think with all of these non-blood approaches is that only the blood matters, and the rate of change in the blood will always be much higher than anything else.

> Sounds like they couldn’t, with today’s technology, measure glucose levels to the accuracy required for medical devices.

There's a bit of spin there. Talking with friends working on it, glucose levels in tears simply can't be correlated with glucose levels in blood. A fundamental scientific issue, not an engineering one.

I was quite surprised when they tried that approach as such a correlation would be quite unexpected. Glucose measurement is an area of quite a lot of research, so their earlier announcements seemed premature at best.

I don't understand this -- we know there's pretty strong correlation between interstital fluid and blood sugar (which is how Dexcom and Lifestyle do their thing). If the correlation between blood sugar and tears isn't equivalent or better than that, what's the point?

Indeed there is no similar correlation. You are correct in "what's the point."

Indeed, most research projects generally don't.

In general, researchers don't report results of failures. How many academic papers reporting a negative would go through the review process...

So there's a bigger debate here. Negative results would also be useful for the broader community, but do not get published in most cases, because the processes may not scale to accomodate them (given most research experiments don't). What technical solutions can help here, and how would a solution become economically viable?

Pre-registration of experiments seems the best answer. Then you are motivated to add something to the registration, such as a negative results, rather than just looking lazy.

How do we know the failures aren’t due to incompetence. You can’t depend on a failure meaning impossible.

You can look at what they did , learn from it and draw your own conclusions.

That's true, but the same can be said for the reverse. How do you know that a research success is not really a failure. If anything, there is more incentive to forge successes than failures.

Indeed. If negative results can be published too and made useful for others, reduction in forged successes could be a side benefit.

Maybe because it was hyped up a lot, including on here.

The original announcement thread:

https://news.ycombinator.com/item?id=7073373

2016 update on Googlebet/Verily vaporware:

https://news.ycombinator.com/item?id=11845804

I didn't find the article's tone to be negative at all - it's dry and objective.

Reply on news.ycombinator.com