Do sleep trackers help or hurt?

How accurate are they?

The range is the answer. Lee and colleagues tested 11 consumer sleep trackers against polysomnography, the laboratory standard, across 75 participants at two centres in Korea. The devices spanned wearables, bed sensors and smartphone apps.

Macro F1 scores ran from 0.6863 for the best performing device down to 0.2588 for the worst. That is the difference between broadly useful and close to guessing.

Sleep staging is the harder problem. Even the best device managed 0.5933 on detecting deep sleep specifically, which is the number most people care about and the one worth trusting least.

Why is staging so difficult?

Because the laboratory standard reads your brain and your wrist cannot. Polysomnography uses electrical activity from the scalp to distinguish light, deep and REM sleep. A consumer device infers those stages from movement and heart rate, which are downstream signals.

Detecting whether you are asleep at all is comparatively tractable. Deciding which stage you are in from a wrist is an inference, and the accuracy figures show how far that inference can drift.

The researchers also note they could not obtain transparency from manufacturers about the data used to train these algorithms, so how a given score is produced remains largely unexaminable.

What is orthosomnia?

Anxiety about sleep driven by tracking it. Jahrami and colleagues surveyed 523 adults and found 176 of them, 35.8 per cent, used sleep-tracking devices regularly.

Prevalence depends on where you draw the line. At a conservative threshold, 3.0 per cent qualified. At a moderate one, 8.6 per cent. At a lenient one, 14.0 per cent.

The consistent finding across all three thresholds is that people with orthosomnia had significantly higher insomnia severity scores. Which direction that runs remains unestablished, and this was cross-sectional.

Does the number make the sleep worse?

The design here cannot tell you, and the mechanism is plausible in both directions. Someone already sleeping badly is more likely to buy a tracker and more likely to be alarmed by it. Equally, a poor score in the morning gives you something to worry about that night.

What the accuracy data adds is that the score doing the alarming may be wrong. A device reporting 40 minutes of deep sleep is making a claim it is measurably unreliable at.

This sits awkwardly with the rest of the sleep evidence. Effort and vigilance are among the things that keep people awake, and a device that quantifies your failure every morning is well designed to produce both. The sleep hygiene evidence covers what actually moves the outcome.

What is worth doing?

Read the trend and ignore the stages. Night-to-night patterns over weeks are the part these devices handle best, and the deep sleep percentage is the part to disregard.

Stop looking in the morning if it changes your mood. There is no intervention available at 7am based on last night's score, so the information has no use and a measurable cost.

Take it off if you are already anxious about sleeping. People with orthosomnia scored higher on insomnia severity at every threshold tested, and a tracker gives that anxiety a daily number to attach itself to.