Wrist activity trackers improved over two decades on what they measure, heart rate (correlations with the reference from as low as 0.50 in exercise before 2010 to typically above 0.80 now) and steps (errors up to 25% before 2017, now commonly under 5% in steady walking), but energy expenditure, which they estimate, still erred by 10-40% against indirect calorimetry: a systematic review of 47 validation studies in healthy or general-population samples, read from its abstract

Hogue and colleagues reviewed how the accuracy of wrist-worn activity trackers has changed as devices moved from simple accelerometers to multi-sensor watches with optical heart rate and machine-learning algorithms.

Searching four databases under a registered protocol, they included 47 studies in healthy or general-population samples, tested in laboratory or simulated free-living conditions against reference methods.

Heart-rate correlations, as low as 0.50 during exercise in devices before 2010, typically exceed 0.80 during exercise in modern devices; the abstract names Apple Watch and Garmin vivosmart as examples. Step-count errors of up to 25% before 2017 now commonly fall below 5% in steady walking. Misclassification of activity intensity fell from more than 20% to 10-15%.

Energy expenditure improved least: recent devices still erred by 10-40% against indirect calorimetry, which the authors attribute to estimating metabolic rate without measuring oxygen and carbon dioxide. Acceptability, judged by wearing time and satisfaction, was generally high. Variation by brand, activity and population persisted.