Sunday, February 20, 2011

A few things about sleep that I've been reading

(I think I'll try to post overviews like this every time I read a bunch of stuff. 3 purposes: help me organize my thoughts and cement them in my brain, help me remember what I was thinking about in the future, and let others know what I'm thinking about. Realize that these overviews will be out of date as soon as I post them. Consider this a snapshot of my Workflowy/mindmap/whatever. And incidentally, I apologize for linking to so many abstracts and paywalls! I wish these things were available freely. Journals and conference proceedings: what a mess.)

Sleep. A few of us at UW are going to work on sleep. We're in the Ubicomp lab, so it'll involve computers. Somehow. Probably.

Our goal is ultimately to help you sleep better. Why? Well, what can you do to be healthier? Eat better, exercise better, and sleep better. Eat and exercise have a lot of research, even in computer science; sleep, not so much.

Okay, but that's about as vague as "we want to cure headaches" or "look, we'll figure out math" or "let's, um, build a boat or something?" Where do we start?

I've started from this wonderful overview by Eun Kyoung Choe (a member of our group) et al, "Opportunities for Computing to Support Healthy Sleep Behavior." Another paper with the same title is due to appear in CHI 2011.

How could we help you sleep better? What does "sleep better" even mean? What's the problem? I think I'll wave my hands a little bit at this point and say "y'know, sleep better!"


One thing you could do is sleep sensing. How do you know if you have sleep apnea or something? It'd be nice to know exactly when you're asleep and awake during the night, and when you're asleep, what stage of sleep you're in.
- The gold standard is called Polysomnography, in which you sleep in a clinic overnight and get EEG (brain), EOG (eye), EMG (muscle), ECG (heart), and other readings. You have to get pretty wired up, you have to sleep in a foreign place, and it's expensive.
- An alternative is the Actigraph, a little wristband that measures your movement during the night; looks like the most official clinical one that's available is still order of $300. Nowadays devices like the WakeMate and FitBit are making this technology cheaper and easier (down to $60).
- There's also the Zeo, which reads... brain waves? They seem a little cagey about what exactly they read, but they've done some research to show that their device compares well to polysomnography.
- The downside with all of the above is that the user has to wear something. People forget, they run out of charge, they lose them, etc. Another approach, which uses load cells (big sensors you put under the bedposts), has been tried by Adami, Pavel, Hayes, and Singer at Oregon Health and Science University. They've met with good success telling when you're in bed, and even whether you're on your back, on your side, or sitting on the edge, but they haven't been focusing on detecting sleep states.
- Another group at UW, Peng, Ling, Sun, and Landis, has used a passive infrared sensor, a heart rate sensor, a video camera, and a microphone to determine sleep states less intrusively. Seems like it works about as well as actigraphy. This is pretty neat from a computer-science point of view, or maybe electrical engineering, because you're dealing with multiple sensors- how do you combine the data to say "you are or are not sleeping now"? That's cool. However, I guess I wonder "if you have to wear a heart rate sensor, why not just wear an actigraph?"
- You could also just ask people how they're sleeping. The Pittsburgh Sleep Quality Index (1988) does just that. Or you could have them fill out the Pittsburgh Sleep Diary. Perhaps the Sleep-Wake Activity Inventory? Naturally, these have the upside that they're easy and cheap to administer, and the downside that they're less accurate.

Ultimately, sleep sensing doesn't interest me as much, because actigraphy seems pretty good. What about, when we know how you're sleeping, how can we help you sleep better? Well, we could prescribe how you should sleep. Or we could let you prescribe it, and help you stick to your goals. The latter sounds better to me, because the former is a lot of work! Let's leave that to the doctors for now.

So it looks to me like there are two main challenges left for us. 1. Help you stick to your goals, and 2. Help you figure out if they're working. I guess solving #2 will help with #1. (If you can tell that your goals are helping you, you're more likely to stick with them. Probably. People are strange creatures.)

As far as #1, helping you stick to your sleep goals, Landry, Pierce, and Isbell have come up with a smart alarm clock that helps you make simple everyday decisions, like "when to wake up"; that could help. I guess that's one of many nifty new alarm clocks out there. I should look into this to see what other persuasive sleep technology exists. Luckily, there's a lot of persuasive technology work being done right at UW; for example, this theoretical paper by Sunny Consolvo, David McDonald, and James Landay. A quick google scholaring for "persuasive sleep technology" is not super fruitful.

And #2, helping you figure out if your sleep changes are working. This goes a step above "sleep sensing"; we have a vague idea of what your ideal sleep night might look like: a few cycles of REM, light, and deep sleep. But ultimately, we don't want someone whose actigraph reading looks nice; we want someone who is full of energy and healthy! How do we know if it's working?

I'm going to propose experience sampling: ask people every couple hours "how energetic do you feel, on a scale of 1-5?" If they constantly report high values, well, that's good enough, right?

This paper gives a little background on experience sampling; most of the downsides are not so relevant anymore, now that we live in the future. This paper too. Basically, it's a pretty well-established methodology, and suffers only from three things: self-report bias, reactivity (by making you think about something a lot, maybe we're changing your perception while we're asking you to report it), and participant attrition (these tend to be pretty long-term studies). Running an experience sampling study to evaluate effectiveness of sleep habit changes seems not completely crazy.

No comments:

Post a Comment