Counts describe this library, not the strength of evidence.
How to read this evidence
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.
In a 10-week trial of 30 older adults with mild cognitive impairment, machine learning on wrist-sensor features, mainly HRV, predicted executive function scores with r = 0.69 and within-person changes with r = 0.61. Only 17 individuals and 96 data intervals were analyzed, so the model is a promising proof of concept, not a validated tool.
An umbrella review of 51 systematic reviews found that using wearable devices may raise physical activity by a median of 1,312.23 steps per day and 57.8 minutes per week of moderate-to-vigorous activity. Certainty was low to moderate and 72.5% of the reviews were rated critically low quality, so the true effect could differ.
Using smartphone and wearable data from 479 people with major depressive disorder, at least 8 days (range 2-12) in a 14-day window were needed to calculate most of 21 features reliably. Some features, such as sleep onset time, tracked depression scores better across people than within a person, while others, such as wakefulness after sleep onset, did the reverse. The abstract reports no prediction accuracy.
In a nationally representative survey of 9303 US adults, an estimated 18% of those with cardiovascular disease (CVD) and 26% of those at risk used wearable devices, versus 29% of all adults. Among people at risk, older age, lower education and lower income were independently linked with lower use, but the data are self-reported and cross-sectional.
Among 530 17-year-olds in a UK birth cohort, more accelerometer-measured sedentary time was associated with higher left ventricular mass (standardized β = 0.16, p = 0.01 in females), and the increase linked to sitting was about threefold that linked to moderate-to-vigorous activity. Heart structure was measured once at age 17 and the effects are small associations, so the findings cannot show that cutting sitting changes the heart.
Scandinavian journal of medicine & science in sports
Pooling wrist accelerometer data from 1250 English children aged 5 to 15 from 10 studies, the authors produced age- and sex-specific percentile curves; activity peaked between ages 6.5-10.5 years, boys were more active than girls (p < .001), and activity was lowest in school Years 8 and 9. The sample is a pooled convenience set, not nationally representative, so the curves are not national norms.
The international journal of behavioral nutrition and physical activity
Machine learning models using 21 days of wearable data from children in the Adolescent Brain Cognitive Development (ABCD) study separated children with ADHD from controls with an AUC of 0.798, and those with sleep problems with an AUC of 0.737. Positive predictive values were low (0.159 for ADHD, 0.036 for sleep problems), so most children flagged would not have the condition.
Of 10 414 children aged 11-13 in the Adolescent Brain Cognitive Development (ABCD) study, 7424 (71.3%) shared wearable data, but Black children were underrepresented (11.4% vs 19.3% of non-sharers; P < .001) and stayed in the substudy for fewer days (16 days vs 21 for White children). Wearable data from this study therefore do not represent all children, and the abstract does not explain why the differences arose.
Among 81 717 UK Biobank adults followed a median 6.8 years, each 1-SD higher accelerometer-measured activity was linked to lower hospitalization risk for 9 conditions (from gallbladder disease, HR 0.74, to colon polyps, HR 0.96) but higher risk for 3. These are associations from one week of wear in mostly White, middle-aged and older volunteers, so the 'preventable' estimates assume causation.
In a randomized trial of 162 adults wearing an Apple Watch for 5 weeks, showing steps deflated by 40% led to more negative affect, lower self-esteem and mental health, and higher blood pressure and heart rate than accurate steps, while actual step count did not change. The final sample was small after attrition and exclusion of 45 participants, and many outcomes were examined.
Of 1203 studies identified, 69 (5.74%) used AI-equipped wearables for anxiety or depression, about two-thirds for depression, mostly to diagnose and none to treat. Because this scoping review did not pool results, it shows what has been studied, not how well wearable AI works.
After wearing a waist-worn sensor for a week, 106 participants rated it comfortable (median Comfort Rating Scale 0.0 on a 0-20 scale where lower is better) and acceptable (median 5.0 on a 1-5 scale where higher is better). This reports comfort and opinion, not measurement accuracy, and the abstract does not describe who the participants were.
Using 107 144 Fitbit nights from 1041 working adults, 3 nights gave a "good" and 5 a "very good" weekly average of total sleep time, while monthly averages needed 5 and 10 nights. Sleep variability needed far more nights (11 and 18 for monthly), and the participants were healthy working adults aged 21-40 on one device brand.
Sleep advances : a journal of the Sleep Research Society
An umbrella review of 39 systematic reviews and meta-analyses, covering 163,992 participants, found that activity trackers were associated with more physical activity (SMD 0.3–0.6), equating to about 1,800 extra steps and 40 minutes more walking per day, and with a bodyweight reduction of about 1 kg. Effects on blood pressure, cholesterol, glycosylated haemoglobin, quality of life and pain were typically small and often non-significant. The abstract does not say what else came with the trackers in these programs or how much the reviews overlap, so the gains cannot be credited to the device alone.
In 122 people with a history of COVID-19 and 588 controls, wristband data showed less light sleep (244 vs 258, P=.003) and less deep sleep (123 vs 128, P=.02) in the COVID-19 group. The differences are small, the abstract does not give units or say how matching was done, and a wearable's sleep stages are estimates.
In 37 adolescents with depression symptoms whose phone and Fitbit data were analyzed, personalized models predicted weekly depression scores with a root mean squared error of 2.83 and weekly change with 3.21. This is a very small exploratory study, and the abstract gives no comparison against simple baselines, so it does not show a usable screening tool.
Among 175 older adults asked to wear a Fitbit Flex 2 for 30 days, participants wore it on an average of 95% of days, and 85% adhered to daily syncing. Wearing was higher in women and syncing was better with better memory, but all participants came from one brain-aging study, so adherence may differ elsewhere.
Among 7607 US adults aged 45 or older followed for a mean of 7.4 years, the highest versus lowest tertile of accelerometer-measured moderate-to-vigorous activity had a lower stroke hazard (HR 0.57, 95% CI 0.38-0.84), while longer average sedentary bouts had a higher one (HR 1.53). This is observational, with only 286 strokes and one week of wear, so it shows association, not that moving more prevents stroke.