Tip:
Highlight text to annotate it
X
I’ve talked a lot about how Zeo works and some of the features of PSG and how it’s
different, but I want to very clearly state why what we’re doing does not qualify as
PSG It does not qualify as a full polysomnography.
There are similarities. We’re measuring brainwaves and we’re defining it as though
we were recording by those standards that have been around for a while.
We’re actually able to stage sleep pretty reliably; we’re able to get REM, Deep, Light,
that sort of thing, but there are a whole lot of differences.
We lump some of the stages together; we don’t look for really brief awakenings like you’d
look for in PSG; we don’t measure a lot of the other whole body signals that I mentioned
before. We don’t take up breathing, we don’t look at blood pressure, we don’t look at
oxygen saturation, and that all leads to the technology.
It’s not our disclaimer of “We’re not a medical device”; it’s not just because
we don’t want to be a medical device, it’s “you can’t do these things without some
of these other measures.”
And, one of the advantages over PSG is that it’s a lot easier; there’s no cleaning,
glues, gels, wires, any of that. So we’ve tried to take the bulk of the advantage of
doing a PSG and put it into something that anybody can use.
I’m going to talk a little bit about the accuracy of Zeo. We’ve run this validation
in three separate studies and have come up with very consistent results. This is all
internally through the organization or through studies that we’ve sponsored. I am very
much looking forward to the day when someone independent of us takes up the mantle and
runs this validation stuff and publishes it, because I am confident that they will come
up with the same results.
When we look at that 75% as good as PSG, you have to remember that PSG is only good about
85% as good as itself, so we’re 90%; we’re B+/A range in terms of what you can do in
ambulatory, consumer product environment.
We validated the technology not only against PSG but against actigraphy. So this is for
sleep stages, the 75%: if you just wanted to know if someone was asleep or awake over
the course of the night, we are 90% plus accurate, with very good sensitivity, specificity, and
positive values.
And we’ve actually be working on a manuscript that we want to submit for peer review so
hopefully we’ll get to the real deal behind the validation.
All that said, the Zeo isn’t perfect. Some of the explanations that I’ve laid out up
front, like wake is happening a lot in the back of the brain, we’re going to have some
difficulties around wake, especially if you’re in REM or a very light sleep, where the rest
of your brain looks very much like you’re awake.
Honestly, sometimes I look at the raw data and I can’t tell if somebody is in Light
sleep or REM sleep. So the algorithm is actually much better than me being a trained expert
at looking at the differences between these stages, but it’s still going to have troubles.
Clearly, there’s room to improve; we’re 75% as good, the standard is 85%, we could
definitely bump it up. It’s just a matter of how much work and expertise we can muster
and put into it.
All that said, it is a pretty good technology. It’s fairly accurate but there’s definitely
a whole range. Some of the data we’ve looked at is, for some people, it works fantastically
well. The first night I slept in a sleep lab with PSG and Zeo, I got the report back that
it was 99% in agreement with the unit.
I’m kind of baffled by that--I don’t think that should have happened--but the low end
happens as well. You can get reports back where the agreement was 25-30%, but in terms
of the bell curve that’s both tails. For most people most of the time, it’s around
that 75%--very well established.