I. Introduction
Addressing the obesity problem worldwide is not only a focus of the pharmaceutical industry, but also the software and hardware technology industry. Nowadays, starting with simple pedometers, highly intelligent technology has been adopted [1]. The fitness-sensing market has been primarily dominated by body-worn sensors, which are often integrated or connected to discrete devices with global positioning system (GPS) receivers [1]. The activity-sensing market is expected to be worth circa $975 million by 2017 [1]. The use of wearable technologies, such as the Fitbit, is becoming increasingly common among the general public [1]. Seventeen million wearable fitness bands are expected to be sold in 2014, rising to 45 million by 2017 and 99 million annually by 2019 [2].
The use of smartphones and wireless smart trackers in healthcare systems depends on recording activity and monitoring vital signs, such as calorie consumption, fitness activity, pulse, weight, heart rate, oxygen level, and sleep pattern [34]. New trends of health and fitness trackers have been developed to track calorie intake and activity pattern along with calorie burning rate. Such trackers adopt MotionX technology using 3D accelerometers to identify movement and transform it to calories burnt. They also measure sleep pattern and pulse and translate all of these data into health information [5].
The use and implementation of various sensors, such as 3D accelerometers, pedometers, and heart rate monitors, in mobile and wearable devices has enabled the successful use of such devices in health applications [6]. A three-axis accelerometer measures change in X, Y, and Z coordinates to track activity [7]. An accelerometer also records sleep quality by watching movement during sleep [8].
Devices such as the Apple Watch, Samsung Galaxy Gear 2, and Samsung Galaxy S5 mobile phone include embedded heart rate monitors. They measure the heart rate by using light to track the blood [9]. Such devices illuminate the capillaries with a light-emitting diode (LED), a sensor that measures the frequency at which the blood pumps. Other trackers, such as Garmin Vivofit, are sold with a more accurate heart rate monitor to be used in conjugation with the tracker [1011]. Another trend is using embedded smartphone cameras to estimate heart beats accurately [12]. Poh et al. [12] developed an algorithm to detect any slight increase in blood volume via light absorption and reflection pattern from the user's face. This idea was commercialized in the Cardiio mobile application. It promises measurement accuracy within 3 beats/min of a clinical pulse oximeter when used at rest in a well-lit environment [12].
The use of smartphones, smart watches, wearable trackers, and new health applications has started a revolution in the healthcare system [1314]. These devices monitor physical activity and provide a convenient continuous feedback. Despite widespread sales of these devices, there has been little evaluation of their use, accuracy, or precision [15].
II. Case Description
1. Materials
The Apple Watch, Samsung Gear Fit, Samsung Gear 1, Samsung Gear 2, Samsung Gear S, iHealth Tracker (AM3), Pebble Steel, Pebble Watch, Qualcomm Toq, Motorola Moto 360, Garmin Vivofit, Mi Band, MisFit Shine, Jawbone Up, Nike+ Fuelband SE, Sony Smartwatch (SWR10), and FitBit Flex were purchased for the assessment of accuracy and precision (Supplementary Figure 1). All of these fitness wearable trackers were selected based on their popularity, availability, consumer surveys, price, and public sales figures during the period from late 2014 until mid-2015 [18]. Table 1 provides a detailed comparison of the various devices included in this study.
2. Methods
This prospective study recruited four healthy adults aged between 22 and 36 years through direct verbal communication. Participants gave verbal informed consent to walk 200, 500, and 1,000 steps. An observer counted steps using a tally counter throughout the period from March 2014 until June 2015. This study was approved by the home institution's ethical committee board.
On the wrist, each participant wore three accelerometers at a time. In one pants pocket, each carried either an Android or iOS device simultaneously running the three corresponding applications of each tracker. Each set was repeated 40 times. Data was recorded after each trial, and the mean step count, standard deviation, accuracy, and precision were estimated for each tracker.
Heart rate was measured by all trackers (if applicable), which support heart rate monitoring, and compared to a positive control, the Onyx Vantage 9590 professional clinical pulse oximeter, which has been well validated for research, measured at the same time on the same hand wearing the tracker. Thirty readings were recorded for each tracker simultaneously.
Finally, the consistency of the synchronization of these trackers with their corresponding mobile application was tested 20 successive times. The number of successful synchronization was counted for each tracker to its corresponding application.
The accuracy and precision were calculated in each study. Accuracy refers to the closeness of measured values to the positive control in each study. The positive controls in case of step counting and heart rate measurements were the observer-counted steps using the tally counter and heart rate measurements obtained using the Onyx Vantage 9590 professional clinical pulse oximeter, respectively. Accuracy percentages were calculated based on the percent by which measurements deviated from the average. The coefficient of variability (CV%), between the repeated measurements for each tracker and user, represents the precision. All data and statistical analysis for each device was estimated using GraphPad Prism version 6.
3. Results
Across all devices, 200, 500, and 1,000 step count observations were recorded for four participants. The participants were all males and had a mean age of 26.5 years (standard deviation [SD] = 12.8 years).
Figure 1 shows the results for the 200, 500, and 1,000 steps trials counted by the tracker devices. Compared with direct observation, the accuracy and precision of the tested wearable devices ranged from 99.1% (MisFit Shine) to 79.8% (Samsung Gear 2) for accuracy and 4% (MisFit Shine and Qualcomm Toq) to 17.5% (Jawbone UP) for precision. Findings were generally consistent between the 200, 500, and 1,000 step trials.
The Apple Watch showed accuracy of 99.1% (SD = 16.6) for 200 step counts, rising to 99.5% (SD = 25.8) for 1,000 step counts. It showed the most precise results for 1,000 steps (CV = 2.6). MisFit Shine showed competitive accuracy of 98.3% (SD = 7.2) for 200 steps, rising to 99.7% (SD = 39.8) for 1,000 steps. However, Samsung Gear 1 showed 97% accuracy (SD = 8.5) for 200 steps, declining to 94% (SD= 103.9) for 1,000 steps. Qualcomm Toq showed about 97% accuracy (SD = 6.9) but it showed the most precise result with CV% of 3.6. Qualcomm Toq maintained very precise results for 500 steps (CV = 5.2) and 1,000 steps (CV = 3.4).
Figure 2 shows the heart rate measurements of the trackers against the Onyx Vantage 9590 professional clinical pulse oximeter as a positive control. Compared to the Onyx Vantage 9590 pulse oximeter, the accuracy and precision of the tested wearable devices ranged from 99.9% (Apple Watch) to 92.8% (Motorola Moto 360) for accuracy and from 5.9% (Apple Watch) to 20.6% (Samsung Gear S) for precision.
The accuracy percentages of heart rate measurements (Figure 2) obtained by the Apple Watch, Motorola Moto 360, Samsung Gear Fit, Samsung Gear 2, Samsung Gear S, Apple iPhone 6 (using Cardiio application), Apple iPhone 5S (using Cardiio application), Samsung Galaxy Note Edge, and Samsung Galaxy S6 Edge were 99.9% (SD = 5.7), 92.8% (SD = 14.1), 97.4% (SD = 28.8), 97.7% (SD = 16.5), 95.0% (SD = 20.9), 99.2% (SD = 6.3), 97.6% (SD = 12.4), 99.6% (SD = 14.4), and 98.8% (SD = 11.6), respectively.
Finally, there was no significant difference in the consistency of synchronization with the Apple iOS and Google Android devices. Within Android version 5 (Lollipop) showed the best consistency (Supplementary Figure 2).
III. Discussion
We found that several of the wearable fitness trackers and smart watches were relatively accurate for tracking step counts and heart rate. Generally, the data recorded were slightly different from observed step counts and heart rates, but they could deviate positively or negatively. Some devices reported step counts more than 15% than the observed count, but none exceeded 20% deviation.
Though the accuracy of a tracker is an important characteristic, it is not the only determinant of the quality of a tracker. The associated mobile application, compatibility with a variety of mobile operating systems, customization options, ease of use, efficacy of synchronization with the mobile devices, size, and external appearance affect the final appeal of a tracker to consumers. The mobile application has a major role as it is the interface for interpretation of data collected by a tracker.
In conclusion, consumers use these fitness trackers and smart watches to estimate physical activity, such as distance or calories burned based on step count and sleep monitoring. For such purposes, these trackers were found to be relatively accurate and beneficial. Increased physical activity facilitated by these devices could lead to clinical paybacks with low cost, as in case of the Xiaomi Mi Band priced at $14 [4]. Such devices, along with their tight integration with Apple Health and Google Fit platforms, can significantly help in improving the quality of life of consumers and help in integrating mobile technology into efforts to solve many health problems, such as obesity and heart disease.