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.
Among 90,237 UK Biobank adults free of cardiovascular disease at baseline, followed until December 2022 for 1,298 strokes and 2,031 myocardial infarctions, the shape of the link between moderate-vigorous physical activity (MVPA) and each outcome depended on how the accelerometer data were processed. For stroke, hazard fell in a straight line only with the machine-learning metric; for myocardial infarction, two metrics gave a curved decline and activity counts a straight one. The abstract gives no hazard ratios, so it shows that methods disagree, not which one is right, and the findings are observational associations.
A systematic review and meta-analysis of studies on smartwatches from four major brands, framed around digital forensics, found that heart rate was both the most studied and the most accurate measure. Step count, distance, energy expenditure, activity classification and sleep were less consistently validated, and results varied by brand, protocol and reporting. The abstract gives no pooled numbers, and most studies were run in healthy people in laboratories, so it cannot say how a particular watch performs for a particular person in daily life.
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 a 3-month pilot, 51 of 62 pediatric clinicians (82%) wore a Garmin Venu 3 smartwatch for at least 70% of the study, and a machine learning model built on 41 of them classified burnout risk with what the authors call strong sensitivity and specificity. The abstract gives no accuracy figures, and the sample is tiny and from one health system.
A machine-learning model using 15 nights of Fitbit sleep heart rate plus prior thyroid blood tests classified thyroid status with 74.6% accuracy (sensitivity 82%, specificity 73.5%) in the harder euthyroid-versus-hypothyroid subset (744 paired observations). The 413 participants were 76% women and largely in a hormone-withdrawal setting, so it is unproven in typical primary hypothyroidism.
Among 29 young adults aged 19-39 using the ANZEILAX-Green app with a smartwatch for 8 weeks, GAD-7 anxiety scores fell by a mean 4.79 points (95% CI -6.56 to -3.02; P<.001). There was no control group, so the drop could reflect expectations, time or other factors rather than the app.
Across about 3.8 million nights from 192,500 female Galaxy Smartwatch users aged 20-65, time awake after sleep onset (WASO) was +11 minutes (+32%) higher at 60-64 years than at 20-24, with smaller differences in total sleep time (17 minutes or less). Users were watch owners tracked for one month, so the results may not represent all women.
In 30 women (15 pregnant, 15 non-pregnant) doing high-intensity interval running, a wearable sensor differed from a rectal probe by more than the 0.3°C limit in 47% of pregnant and 43% of non-pregnant measurements. It overestimated core temperature in pregnancy and underestimated it in non-pregnant women, from a small sample in one protocol.
In 51 Japanese outpatients with heart failure or ischemic heart disease (mean age 78.4), a Fitbit Inspire 3 overestimated sedentary time by 62.0 minutes (95% CI 51.5-72.5) in a 12-hour daytime window compared with a waist-worn research accelerometer. The two devices tracked each other well in a group (r = 0.905), but individual limits of agreement were wide (-11.3 to 135.2 minutes).
In a single-arm pilot of 21 older adults (15 dementia caregivers, 6 people with dementia), everyone wore an Apple Watch long enough to chart a rest-activity rhythm, and all 15 caregivers still used the app at week 6. With no control group and only 11 caregivers completing mood and sleep measures, the symptom improvements cannot be credited to the monitoring.
In 355 older adults in Japan, a 6-month Fitbit-based programme raised MoCA by 0.58 points versus control (95% CI 0.03-1.13; p = 0.039), but not after Holm adjustment (p = 0.12), and TMT and muscle mass showed no difference. The trial was open-label with unmasked assessors, and the gain is of uncertain clinical value.
In a randomized trial of 46 sedentary postmenopausal women, 12 weeks of Baguazhang gave systolic blood pressure 7.2 mmHg lower than habitual lifestyle (adjusted difference -7.2; 95% CI -10.5 to -3.9; p < 0.001), with sessions averaging 118.5 beats/min. The trial is small and the secondary outcomes were not adjusted for multiple testing.
In 95,559 UK Biobank participants wearing wrist accelerometers, sleep patterns were associated with 156 incident diseases over a median 8.9 years; sleeping under 5 hours showed 37 of 41 significant adverse associations versus 6-8 hours. The design is observational with algorithm-estimated sleep stages, so the links are not causal.
Among 147,414 adults aged 40-74 with wrist-device activity data, 150 minutes a week of moderate activity was linked to lower mortality than no activity (HR 0.70, 95% CI 0.67-0.72) over a mean 7.9 years. These are associations from observational cohorts, so time-swap estimates do not prove that changing activity would change survival.
The international journal of behavioral nutrition and physical activity
Across 16 randomized trials (N=910), wearable digital interventions in adults with insomnia were linked to a 4.52-minute shorter objective sleep-onset latency (95% CI -8.38 to -0.67) and 2.00% higher subjective sleep efficiency, but no improvement in the insomnia severity index (ISI). The prediction intervals for some outcomes include no effect, so the authors call the findings preliminary.
Simulating three smartwatch algorithms on continuous implant data from 346 ablation patients, sensitivity for atrial fibrillation (AF) recurrence was 64.4% to 82.2%, and AF burden correlated with the implant above r = 0.97. These are simulated watches applied to recorded data, not people wearing watches in daily life.
Europace : European pacing, arrhythmias, and cardiac electrophysiology : journal of the working groups on cardiac pacing, arrhythmias, and cardiac cellular electrophysiology of the European Society of Cardiology
Among 90,237 UK Biobank adults free of cardiovascular disease at baseline, followed until December 2022 for 1,298 strokes and 2,031 myocardial infarctions, the shape of the link between moderate-vigorous physical activity (MVPA) and each outcome depended on how the accelerometer data were processed. For stroke, hazard fell in a straight line only with the machine-learning metric; for myocardial infarction, two metrics gave a curved decline and activity counts a straight one. The abstract gives no hazard ratios, so it shows that methods disagree, not which one is right, and the findings are observational associations.
A systematic review and meta-analysis of studies on smartwatches from four major brands, framed around digital forensics, found that heart rate was both the most studied and the most accurate measure. Step count, distance, energy expenditure, activity classification and sleep were less consistently validated, and results varied by brand, protocol and reporting. The abstract gives no pooled numbers, and most studies were run in healthy people in laboratories, so it cannot say how a particular watch performs for a particular person in daily life.
In a single-arm 12-week trial with 200 Taiwanese university students, fruit and vegetable intake rose by 1.1 servings a day and dietary self-regulation scores rose from 64.59 to 79.50 (p < 0.001) after using a diet-tracking app. With no control group and self-reported outcomes, the gains cannot be credited to the app alone.
During controlled low-oxygen conditions, both the Apple Watch Series 7 and a Masimo MightySat Rx finger oximeter overestimated oxygen saturation, and below 88% the Apple Watch error (Arms 5.82%, 95% CI 5.32%-6.31%) was larger than the Masimo's (3.52%, 3.18%-3.86%). The sample was small and not diverse in skin tone or age, so the error sizes may differ in other people.
Across 61 studies of wearables for remote chronic disease monitoring, 39 (64%) used wearable bands, accelerometers were the most common sensor (n=39, 64%), and the wrist was the most common site. Most of the work targeted neurological (34%) and cardiovascular (25%) disease, and this mapping review does not test whether the devices improve health.