Sleep Monitor Analysis

Note: this analysis was scored using the criteria described here in late 2011/early 2012. Since then some device capabilities have changed, some devices are no longer sold.

Measuring how active someone is with an accelerometer is relatively easy to implement and for the user to understand, but measuring sleep quality is quite different. The simplest to measure and easiest to understand proxy for sleep quality is the number of minutes spent in bed and the consistency of the time to bed each night. However, this is only a reasonable proxy for people who fall asleep relatively quickly and have a normal pattern of REM and non-REM (deep) sleep (Iber & Medicine, 2007). The medical gold standard for measuring sleep quality is lab-based polysomnography (PSG), which involves a variety of sensors including accelerometers and EEG sensor pads with very skilled technicians needed to interpret the data. One of the less invasive measures used by sleep clinicians is actigraphy (Ancoli-Israel et al., 2003) where the amount of movement of the non-dominant arm indicates either deep sleep (no movement) or light sleep/REM sleep/awake. This was seized upon by app and device developers to give people an indication of their sleep quality. One of the first and most popular apps was Sleep Cycle for iOS. It works by placing the phone on the mattress near the pillow and using the lack of movement as a proxy for depth of sleep (a kind of proxy for actigraphy). Many of the manufacturers of activity logging devices also offer actigraphy based sleep logging by strapping the sensor to the wrist and pushing a button upon attempting to go to sleep and again upon waking up (to differentiate normal activity logging from sleep logging). The only consumer level sleep logger measuring some brain activity is the Zeo which used a dry headband sensor. Since the Zeo is the only consumer device with research results correlating it to PSG (Shambroom, FáBregas, & Johnstone, 2012) we considered it as the closest to ground truth.

Table 4 shows the results for the analysis of the sleep measuring devices and apps. As the table shows, each of the three methods  (smartphone on mattress, activity monitor on wrist, dry headband) requires increasing levels of effort and invasiveness but also result but these also seem to result in increased accuracy (Tonetti et al., 2012). My studies showed a relatively low accuracy for the mattress-based apps and roughly concur with Tonetti et al. (2012) showing at least 75% accuracy for wrist-based devices correctly detecting sleep/wake. It should be noted that even gold standard sleep measurements do not have an agreed accuracy:  PSG measurements are scored manually and inter-scorer reliability may be only 85% (Kelly, Strecker, & Bianchi, 2012). My measurements also showed a high variability between individuals using mattress and wrist-based measurements. Accurate measurement of sleep quality seems to have very high individual difference variability (unlike activity life-logging) and it appears that individuals need to make a judgement as to which technology combines the minimum invasiveness and maximum subjective accuracy based on their comparison of sleep score with how they feel in the morning. Although my results and the literature shows the Zeo headband produces the most accurate results, its cost and the reports from participants of the invasiveness of the elastic headband led me to conclude that it was not appropriate for an initial study given the high variability between individuals. The mattress-based smartphone apps were an interesting alternative as they were least expensive and least invasive but the manual data export made them less appropriate for a study with non-specialist participants. Given these criteria and the fact that the FitBit was also the most appropriate activity monitor, I chose it as the sleep measurement device so that participants had the reduced overhead of having to look after only one device.

Name/URL Charge/ Sync Invasive -ness Accuracy Open Data
WakeMatewww.wakemate.com

 

discontinued

2

-daily cable charge, Bluetooth manual sync with smartphone app

3

-worn as soft bracelet, small switch to activate

3

-2 minute granularity motion detection of wrist infers sleep depth (actigraphy)

1

-detailed graphical analytics on website, not exportable

Sleep Cycle iOS appwww.sleepcycle.com

or

Sleep As An Droid

https://sites.google.com/
site/sleepasandroid/

 

(less than £5)

5

4

-non invasive, placed on mattress, need to remember to switch on/off

1

– movement of mattress is inferred movement of sleeper

3

-daily start, stop and sleep quality manually exportable

Zeo 

£200

www.myzeo.com

4

-daily charging on stand when not in use

2

-headband worn when sleeping, automatic activation

5

– the only commericial device to measure (some) brain activity

3

-login to proprietary website, manual export, some partner data sharing

FitBit£65
www.fitbit.com

(original model evaluated, new models released in 2012)

5

-once/week 1 hour charge on stand, wireless sync

2

-inserted into wristband when going to sleep, requires button push on sleep/wake

3

-2 minute granularity motion detection of wrist infers sleep depth (actigraphy)

4

– API allows access to most data

BodyMediawww.kiperformance.co.uk

 

2

-2-3 day capacity, cable charge, cable sync with custom software

1

-strapped to upper arm, uncomfortable, automatic detection of sleep position

3

-uses actigraphy at 2 minute intervals to measure sleep quality

1

-custom software to access data, subscription required to access data, no export available

Table 4: Sleep Life-logging Device Evaluation

Experiments with Personal Informatics Devices (Lifelogging) for self-hacking, persuasion, influence, nudge, and coercion (PINC)