A research team from Florida International University compared four smartwatches of different brands to verify their accuracy in calculating calories: three out of four overestimated (by a lot) the calories burned, and one returned irregular and unpredictable results.
The aspect that most surprised the authors was that the margin of error increased as the user’s body fat increased: in short, those who are overweight or obese and want to lose weight would do better not to rely on these devices to keep track of the calories consumed through physical activity. Details of the study have been published on PLOS One.
The research involved 58 Hispanic adults with an average age of 23, and tested four different types of smartwatches – the Apple Watch Series 8, the Garmin Forerunner 955, the Samsung Galaxy Watch 5 and the Fitbit Sense 2. The participants wore the smartwatches and a COSMED K5, a wearable device used in sports and clinical settings to analyze the user’s breathing and calculate various parameters (including calorie consumption during exercise), and they pedaled a recumbent bike for 10 minutes, with a five-minute break before and after the workout.
Photoplethysmography
Smartwatches estimate calories burned indirectly, by combining different signals about what the body is doing and how hard it is working. Using a technique called photoplethysmography (the same as the oximeters we used during Covid) an optical sensor detects changes in blood flow under the skin to estimate the heart rate; at the same time, an integrated accelerometer records the speed and intensity of the wrist movement.
This data is then processed by an algorithm together with personal information such as age, gender, height and weight, to produce an estimate of energy expenditure.
Overestimated calories
On average the smartwatches deviated from the COSMED K5 calculations by around 15-25%, but some particularly large errors caused the averages to rise significantly: the Apple Watch overestimated energy consumption by around 22 calories on average, the Garmin by around 69 and the Samsung by around 57. The Fitbit reported moderate errors on average, but in 13% of the tests it produced extreme values, and in one session it did not return any estimates.
Fat increases inaccuracy
The most interesting (and unexpected) finding is that the inaccuracy is greater as the user’s body fat increases, across all four brands analyzed.
Why? The authors hypothesize that the quality of the optical signal worsens when it has to pass through fatty tissue, that the movement of soft tissue around the wrist interferes with the detections, and that the algorithms themselves were developed on populations that likely underrepresent people with a higher percentage of body fat.
