What the in-home sensors actually detect in dementia
The clearest signal comes from passively collected sleep-activity data in people's own homes. In a combined dataset of 100 people and more than 9,500 person-nights, unsupervised clustering sorted sleep patterns into three groups. One group was 72% people living with dementia and had the highest average age, about 78. That group showed more light sleep and more wake after sleep onset—time spent awake after initially falling asleep—along with less rapid eye movement (REM) sleep and less non-REM sleep than the general-population comparison [1].
This is a group-level pattern, not a diagnostic test. The dementia-heavy cluster was defined by its sleep architecture, and the authors describe the result as suggesting detectable dementia sleep phenotypes and the feasibility of passive monitoring for disease tracking [1]. In other words, the sensor data reveal a recognizable sleep profile that is more common in dementia, which is different from saying any individual's sensor readings can confirm or rule out dementia.
How close are these home metrics to polysomnography?
Polysomnography (PSG) is the lab-based gold standard: it records brain waves, eye movements, and muscle activity to score sleep stages precisely, but it is expensive, invasive, and not a natural night's sleep. A radar-based home system tried to close that gap by training a neural network on large PSG datasets and then adapting it to radar signals from 47 older adults, 18 of whom had prodromal or mild Alzheimer's disease. When classifying wakefulness, REM, light sleep, and deep sleep, it reached 79.5% accuracy with a Kappa of 0.65—a score that reflects agreement beyond chance [2].
That result supports the idea that a contactless sensor can approximate PSG-scored sleep stages in a mixed older-adult cohort that includes people with early dementia [2]. But it measures stage-classification agreement, not the ability to sort dementia from non-dementia. The study's value for the comparison question is narrower: it shows the passive signal carries real sleep-stage information, which is the raw material any dementia-versus-population comparison would rely on.
Where self-report fits—and where the evidence runs out
Self-report is the other common yardstick, and it behaves differently from sensors. In a six-month study of 68 people referred for suspected obstructive sleep apnoea, participants completed app-based daily check-ins on about 52% of possible days, averaging 3.6 reports per week; compliance was highest early and declined over time, and people with more severe apnoea tended to sustain reporting longer [4]. That study is about a different condition and does not involve dementia, so it cannot say how self-report performs in dementia cohorts. It does illustrate a general point: self-report is intermittent and depends on engagement, whereas passive sensors collect every night without asking anything of the person.
A contactless under-mattress pressure sensor has also been used to follow sleep patterns longitudinally in the home, comparing a general-population cohort with a dementia cohort and describing sleep disturbances associated with dementia [3]. The available summary of that work does not report a direct comparison against self-report or PSG, so it adds longitudinal support for home-based dementia sleep differences rather than resolving how the methods stack up.
Put together, the evidence supports a practical conclusion with a clear boundary. Passive in-home monitoring distinguishes dementia from general-population cohorts at the group level through sleep-architecture differences [1], and radar-based systems can recover PSG-style sleep stages in older adults with reasonable accuracy [2]. What has not been shown in these studies is a three-way test—passive versus self-report versus PSG—for classifying dementia status, so claims that one method outperforms the others for that purpose go beyond what the data establish.
Sources used in this answer
Assessment of sleep patterns in dementia and general population cohorts using passive in-home monitoring technologies
Clustering of passive in-home sleep data from 100 people and over 9,500 person-nights identified a dementia-predominant cluster (72% people living with dementia) with more light sleep and wake after sleep onset and less REM and non-REM sleep than the general population.
Using Legacy Polysomnography Data to Train a Radar System to Quantify Sleep in Older Adults and People living with Dementia
A radar-based deep transfer learning system trained with polysomnography data classified wake, REM, light, and deep sleep in 47 older adults, including 18 with prodromal or mild Alzheimer's disease, with 79.5% accuracy and a Kappa of 0.65.
Contactless longitudinal monitoring of ageing and dementia-related sleep trajectories in the home
A contactless under-mattress pressure sensor was used to monitor sleep longitudinally in the home and compare a general-population cohort with a dementia cohort, describing dementia-associated sleep disturbances.
A mobile-based daily check-in approach to monitor patient reported outcomes in obstructive sleep apnoea: Results from a six-month home-based observation study
In 68 people referred for suspected obstructive sleep apnoea, app-based daily self-reports were completed on 52.1% of possible days over six months, with compliance declining over time and greater apnoea severity associated with slower decline in adequate reporting.
Evaluation of a Wireless Home Sleep Monitoring System Compared to Polysomnography
A wireless home sleep monitoring system was evaluated against polysomnography for sleep-stage classification, showing promising but preliminary agreement that the authors say needs further hospital testing.