Counts describe this library, not the strength of evidence.
Cómo interpretar la evidencia
Validation is specific to a metric, model, reference and setting. A useful heart-rate measurement does not validate calorie estimates or sleep stages. Even processing the same accelerometer data differently can change estimated activity–disease relationships.
Compared with 2017-2018, children aged 10-11 in 2021 had 7.7 fewer weekday minutes (95% CI 3.5 to 11.9) and 6.9 fewer weekend minutes (95% CI 0.9 to 12.9) of accelerometer-measured moderate-to-vigorous activity, and more sedentary time. The comparison is between two different groups of children, from 393 in 2021 versus 1296 earlier, so differences could reflect more than the lockdowns.
The international journal of behavioral nutrition and physical activity
A scoping review of articles from 2011 to June 25, 2021 found 344 records, of which 19 (5.5%) met criteria, covering autism, ADHD and anxiety-type disorders with chest straps or wrist-worn devices such as Fitbit watches. The studies varied widely and were mostly pilot or feasibility work, so none shows that wearable AI works for diagnosis.
In 100 patients in atrial fibrillation and 100 in sinus rhythm, Apple and Samsung watch algorithms diagnosed AF better than Withings (sensitivity/specificity 87%/86% and 88%/81% vs 78%/80%, p < 0.05), mainly because Withings gave more inconclusive ECGs. Expert reading ranked the watches differently, and the patients were recorded in a clinical setting.
Pooling 41 studies, programs using consumer activity trackers increased school-aged children's daily steps (d = 0.612, 95% CI 0.477-0.746) and moderate-to-vigorous activity (d = 0.220), but sedentary time did not improve (d = 0.172, unfavorable). The certainty of evidence was low to moderate, and many included studies were not randomized.
A review of 65 articles found Fitbit Charge models had a mean absolute percentage error (MAPE) under 25% for steps across 20 studies, and the Apple Watch under 10% for heart rate in 2 studies, while energy expenditure error exceeded 30% for every brand. Studies differed too much to pool, so these are descriptive patterns rather than a ranking of devices.
In 267 working adults wearing Fitbit Charge 2 trackers for 14 days, more variable nighttime heart rate and less regular step rhythms went with more depressive symptoms, but detection was limited in the whole sample. The 80% accuracy came only from contrasting depressed participants with those who had no or minimal symptoms, which is much easier than real-world screening.
Among 89,205 UK Biobank participants, those with a lifetime inpatient psychiatric diagnosis differed from others on a median of 8.5 of 10 wrist-accelerometer sleep measures, but effects were small (largest beta -0.11, 95% CI -0.13 to -0.10, for depression and sleep efficiency). This is cross-sectional, and diagnoses often came before the sleep recording.
Across 29 studies of community-dwelling adults over 60, all but one found an association between physical activity measured by wearables and frailty level, and every study reporting gait speed found it correlated with frailty. The studies used 13 different sensors in 8 body locations, so no single device or placement is yet established.
Of 1981 articles screened, 26 studies used mobile or wearable devices to monitor diabetes-related measures, mostly with accelerometers, and the authors judge the field to be at an early stage. Few studies focused on type 2 diabetes, none on complications, and clinical trials were scarce, so the review cannot say whether the devices improve care.
Synthesising 20 qualitative papers covering the views of 349 participants aged 51-94, the review found that motivation, ease of use and a useful purpose decide whether older adults fold a wearable into daily life. These are self-reported experiences from mostly activity-tracker studies, so they describe attitudes rather than measuring who actually keeps wearing a device.
In a sleep laboratory, 34 healthy young adults wore a subset of seven consumer sleep trackers for three nights. Against polysomnography every device detected sleep very well (sensitivity 0.93 or higher) but caught wake poorly (specificity 0.18–0.54), and five of the seven matched or beat research actigraphy (specificity 0.39). Sleep-stage estimates were inconsistent, and the participants were healthy adults aged 18 to 35 in a fixed 8-hour bed window, with each device tested in only 11 to 20 of them, so the results say little about people with sleep problems.
Using Fitbit, indoor air-quality and symptom data from 67 people with COPD, a model predicted whether a flare-up would occur within the next 7 days with 92.1% accuracy, 94% sensitivity and 90.4% specificity. Only 25 flare-ups were recorded at one hospital, so the model has not been shown to work in other people.
In 10 community-dwelling older adults, step counts fell by 55.1% and walking time by 52.2% after the pandemic declaration, while depression symptoms rose by 150%. With only 10 people and no comparison group, these are descriptive changes, not proof that reduced movement caused the depression.
Using 2812 questionnaire records from 368 participants at 3 European sites, 14 wristband-derived sleep features were significantly associated with depression symptom scores (PHQ-8), led by mean sleep offset time (z=6.19, P<.001) and awakening times (z=5.53, P<.001). These are associations between sleep and symptoms across people and time, and the pattern varied by site, so they do not show that sleep changes predict or cause depression.
A rapid review of 19 papers found three groups of factors that damage the quality of data from person-owned wearables: device and technical, user-related, and data governance. They can leave data incomplete, incorrect and heterogeneous. It is a narrative synthesis of a small literature and gives no measured error rates.
In health care workers wearing an Apple Watch, the circadian amplitude of an HRV measure (SDNN) differed between people with and without COVID-19 (P=.006) and in the 7 days around diagnosis versus uninfected periods (P=.01). The abstract gives no participant numbers or accuracy, so its value as an individual warning is unproven.
A narrative review of 82 papers (from 960 screened) found that medical wearables fall into 4 application areas: health and safety monitoring, chronic disease management, disease diagnosis and treatment, and rehabilitation. Because it is a narrative review that does not pool results, it maps what exists rather than showing how well any device works.
Over 7 nights in 45 healthy adults, the Oura ring's total sleep time tracked a medical-grade actigraphy device closely (r=0.86), while the Samsung Gear Sport watch tracked it less well (r=0.59). Participants were healthy and wore the devices at home, so the results say little about people with sleep disorders or about sleep staging.