For our analysis, we chose wearables manufactured by leading companies in the market [10]. The 5 wrist-worn wearables were the Apple Watch Series 3 (Apple Inc), Fitbit Versa (Fitbit Inc), Garmin Vívoactive 3 (Garmin), Polar M600 (Polar Electro Oy), and Xiaomi Amazfit Stratos 2 (Huami Technology). Each wearable was installed as instructed by the respective manufacturer and was synchronized with the companion app for smartphones (Apple iPhones).

Wearables and companion apps were coded using the BCT Taxonomy Version 1 (BCTTv1), which was previously employed in similar studies [11-13]. The BCTTv1 is explained in detail by Michie et al [14]. Briefly, the BCTTv1 incorporated 93 nonredundant techniques, grouped into 16 hierarchical clusters in total, each coded using a dichotomous score of either 0 or 1, indicating nonpresence or presence, respectively [14].

Each wearable was worn by 2 researchers (MT and PD) for 1 week. The 2 researchers were well acquainted with the handling of the wearables, using the companion apps, and employing the BCTTv1. The researchers completed the training on the BCTTv1 website [15] before the analysis. Interrater reliability assessing each wearable incorporating a BCT was calculated using a kappa statistic in SPSS 22.0 (IBM Corp). Magnitude of agreement was interpreted as per the following criteria: 0.00=poor, 0.01-0.20=slight, 0.21-0.40=fair, 0.41-0.60=moderate, 0.61-0.80=substantial, and 0.81-1.00=almost perfect [16].

Coding disagreements were resolved by a discussion between the researchers. In case of disagreement, a third researcher's opinion (SK) was included to resolve the disagreement.

In line with the aim of this study, BCTs targeting PA were examined, while the feedback on other factors (eg, sleep or diet) was ignored. The researchers were instructed to include periods of physical inactivity as well as those of PA into the assessment week to verify the corresponding feedback by wearables.

To evaluate the potential effects of incorporated BCTs, we employed a previous list [9] created for the same purpose. The list is based on a meta-analysis [17,18], meta-regression [19], and systematic reviews [20-22] as well as recommendations from the US Preventive Services Taskforce [23]. These BCTs were marked with checkmarks in Table 1. As in the earlier study [9], we used this list to count the number of effective BCTs, which were incorporated in each wearable to promote PA.

Out of the 93 BCTs analyzed by the BCTTv1, 26 different BCTs were incorporated in the 5 wearables. On average, 19 BCTs (range 15-24) were incorporated in the wearables. Fitbit Versa and Garmin Vívoactive 3 incorporated the most BCTs (n=24), followed by Apple Watch Series 3 (n=18), Polar M600 (n=16), and Xiaomi Amazfit Stratos 2 (n=15). Due to technical issues with the Xiaomi Amazfit Stratos 2 device, we could not evaluate the BCT item of social support (emotional) (item 3.3); thereby, that item was not marked as incorporated with a checkmark in Table 1. Multimedia Appendix 1 provides detailed information about how often a BCT was incorporated in each wearable.

According to Lyons et al [9], 23 BCTs are effective in promoting PA. Out of these BCTs, Garmin Vívoactive 3 (n=16) and Fitbit Versa (n=14 + 3 paid BCTs) incorporate most BCTs, followed by Apple Watch Series 3 (n=12), Polar M600 (n=11), and Xiaomi Amazfit Stratos 2 (n=11). Only Fitbit Versa and Garmin Vívoactive 3 included the BCT items of instruction on how to perform behavior and behavioral practice/rehearsal. None of the wearables included the following BCT items: barrier identification/problem solving, self-reward, self-talk, commitment, information about social and environmental consequences, and information about emotional consequences.

On average, 11 (range 8-13) clusters were incorporated in the 5 wrist-worn wearables. Fitbit Versa and Garmin Vívoactive 3 incorporated most clusters (n=13), followed by Apple Watch Series 3 (n=10), Amazfit Stratos 2 (n=10), and Polar M600 (n=8). The 3 most common clusters were goals and planning (cluster 1) (n=4), feedback and monitoring (cluster 2) (n=3), and reward and threat (cluster 10) (n=3).

Although the taxonomy we applied to assess BCTs in wearables was employed previously [9], this taxonomy was not directly developed for machine to person (ie, wearable to consumer) interaction but for person to person (ie, psychologist to patient) interaction. This modification results in some difficulties in the application of this taxonomy. For example, different surrogate markers of PA (eg, number of steps or energy expenditure) can be employed to prompt the same BCT. To the best of our knowledge, it remains unclear which surrogate marker is most suitable as a basis for BCTs to improve PA.

Additionally, the BCTTv1 taxonomy does not allow certain functions, for example, evaluating how a BCT is visualized, evaluating how frequently BCTs are implemented (eg, daily report and hourly feedback), and determining whether a BCT is interactive (eg, a BCT disappears from the screen when an increase in PA is detected). However, we assume that these aspects affect the effectiveness of the BCTs and hence, that of the respective wearables in altering PA behavior among users.

Since the number of wearables promoting PA is rapidly increasing, we urge researchers to develop a taxonomy covering these aspects to assess the preferred BCTs employed by wearables.

Some BCTs displayed by the wearables addressed sedentary behavior as well as PA (ie, prompts/cues). Sedentary behavior is recognized as an independent behavior of PA [30,31]. In this study, we did not separate BCTs in terms of PA behavior and sedentary behavior. However, sedentary behavior and PA follow unique intervention logic [31,32], and research suggests that different factors influence each behavior. Future research should analyze the incorporated BCTs in wearables for sedentary behavior, separately of PA. Targeting both behaviors can have greater health benefits.

The shelf life of consumer-grade wearables, such as that of the devices we analyzed in this study, is short. Manufacturers frequently release new models, while older ones disappear from the market. Consequently, frequent assessment of new wearables, ideally before the market release (as recommended elsewhere [28]) is warranted.

Based on our analysis, we conclude that out of all the tested wearables, Fitbit Versa (n=14+3 paid) and Garmin Vívoactive 3 (n=16) included most of the BCTs with proved effectiveness, followed by Apple Watch 5 (n=12), Polar M600 (n=11), and Xiaomi Amazfit Stratos 2 (n=11). Out of all the tested wearables, Garmin Vívoactive 3 and Fitbit Versa might be the most promising wearables to promote PA from a psychological perspective. However, future studies need to evaluate the effectiveness of these devices in experimental studies along with the validity and reliability of variables obtained by these devices. Since no specific taxonomy is available to investigate BCTs incorporated in digital health tools, including wearables for promoting PA, we advise developing such a taxonomy given the global urgency to improve PA and the increasing popularity and availability of wearables. Since the effectiveness of BCTs is affected by individual characteristics, we recommend manufacturers to allow customization of BCTs.