First responders, such as police officers and firefighters, are regularly faced with highly demanding working conditions [1] such as threatening or emotionally demanding situations. These roles frequently involve irregular shifts and peak workloads, sometimes compounded by understaffing [2]. Such conditions can lead to increased stress levels [3], reduced mental well-being [1,4], sleep disturbances [5], and difficulties in maintaining a healthy lifestyle [6], thereby escalating the risks of long-term health issues [7] and absenteeism [8]. Although some of the inherent challenges of these professions are unavoidable, it could be beneficial for employees to gain insights into their own stress levels, recovery patterns, and lifestyle-related factors like sleep and physical activity [9]. Such personal monitoring could help them to improve their stress- and well-being–related awareness and self-efficacy to make behavioral adjustments in order to better manage demanding aspects of their professions and thus optimize their well-being.

The emergence of wearable sensor technology facilitates promising opportunities to monitor a broad set of health-related outcomes [10]. Besides the initial use cases of wearables in estimating daily physical activity levels [11] and sleep duration [12], the introduction of photoplethysmography sensors enabled the measurement of other metrics such as heart rate [13], heart rate variability (HRV) [14], oxygen saturation [15], and potentially blood pressure [16]. Wearable-based measurements such as sleep and resting HRV have already been linked to stress- and well-being–related outcomes in first responders [17,18], but are sometimes also used to estimate other relevant outcomes. Using measures such as HRV as inputs, wearable companies now also estimate mental stress levels [19,20] and physiological recovery [21] in an attempt to provide more easily interpretable feedback. Although conceptually useful, in consumer-available wearables this is almost exclusively done via black box algorithms and the created measures do not always correlate optimally with similar subjective outcomes [22]. Despite these limitations and need for more scientific evaluation of the performance of these metrics, the wearables that utilize them are already broadly commercially available and adopted and are being integrated into wearable-based interventions [23]. In these wearable-based interventions, individuals use real-time data from physiological measurements to improve their health and performance, also known as biofeedback [24]. Given the continuing adoption of these wearables, it is important to understand the impact of these devices on behavior change and stress- and well-being–related outcomes.

Wearable-based interventions can be supportive at increasing physical activity [25], reducing body weight [25-27], and improving sleep outcomes [28]. Regarding the application of wearables to reduce stress and support mental well-being, most research efforts are focused on the modeling of stress-related outcomes using wearables [29]. Although studies show that wearable-based stress detection is feasible, this is especially true in controlled (eg, laboratory) settings, whereas challenges remain for accurate measurement in real-life circumstances [30]. A few studies have been performed investigating the effectiveness of wearable-based interventions to reduce stress and improve well-being, and they found some promising results [31-33]. A recent review that surveyed the broader literature on this topic showed that these effect studies tend to assess acute stress responses in laboratory settings, often in young (student) populations [34]. The authors concluded that there is therefore a need for more studies that investigate the effects of longitudinal wearable-based interventions on stress- and well-being–related outcomes in a natural, real-life context.

To improve our understanding of the effects of wearable-based interventions, it is not only helpful to gain insight into their direct impact on outcomes such as stress and mental well-being, but also in the possible precursors to those outcomes. Effects on these precursors could be meaningful even in the absence of short-term effects on the primary outcomes and may substantiate why the found effects occur. With regard to these precursors, the Health Belief Model proposes that health-related awareness (eg, on the susceptibility for and severity of the potential health problem) and perceived self-efficacy to cope with the situation are necessary to come to action and change behavior, which in turn may improve health-related outcomes [35,36]. Metacognitive awareness (MCA) has been shown to be related to learning [37] in general and learning about coping with stress specifically [38]. MCA is defined as the ability to reflect upon, understand, and regulate one’s learning [37]. This concept distinguishes between insight in one’s responses and potential causes and ways to regulate these responses. MCA helps people to develop strategies to change their behavior in accordance with the situation.

Self-efficacy is considered a key mechanism through which behavioral change can be observed [39]. This has also been shown for the way people cope with stress [40]. People’s beliefs about their ability to change behavior is determined, among other factors, by mastery experiences (succeeding at changing behavior) and vicarious learning (seeing other succeed). Daily feedback through monitoring can provide valuable insight into how to change behavior and mastery of these behaviors in attainable steps.

Since most wearables are designed to improve their users’ awareness of their own health-related behavior and provide them with leads on what needs to change and how they are progressing, improvement in health outcomes may be preceded by increased awareness and self-efficacy. Some studies have confirmed that wearable-based interventions can indeed positively influence health-related awareness [41] and self-efficacy in patient populations [42,43], but it is unknown to what extent these findings translate to occupational settings. Health-related behavior change interventions generally require a duration of months to years in order to ingrain the new behaviors [44], but when studying awareness and other precursors of behavior change, a shorter duration (eg, weeks) is sufficient.

Therefore, the first research question that this study addresses is the following:

Research question 1: Does offering a 5-week wearable-based intervention improve awareness, self-efficacy, and stress- and well-being–related outcomes in police officers in comparison to a control period?

For policymakers considering the deployment of wearable-based interventions, it is also helpful to know to what extent it is sufficient to use an “off-the-shelf” solution, where consumer-available wearables are directly deployed to the target group, or whether it is necessary, for instance, to provide (1) more extensive feedback and (2) (peer) support for the intervention to be effective. Based on self-efficacy theory, the latter could be of added value as it provides potential for vicarious learning and supports awareness of ways to attain mastery experience, whereas the former may contribute to the development of increased awareness.

First, it is possible to additionally collect subjective data via brief daily questionnaires, also known as ecological momentary assessment (EMA), and utilize the data to provide feedback on outcomes that cannot be measured using wearables. These self-report data can also be used to provide contextual and subjective information to the objective wearable-based measures [45]. Providing this more enriched feedback could potentially have beneficial effects on awareness as it links personal experience to objective measurement and provides information on the behavioral context that affects stress and well-being. This in turn can inform users on strategies to attain behavior change.

Second, organizing peer support groups can be a way to facilitate collaborative reflection and group learning [46]. Literature on the effectiveness of peer consultation groups in a preventive occupational health context is scarce, but there are indications that these can positively contribute to recovery from mental health conditions [47-49]. Although different from peer consultation through the guidance of more experienced professionals, peer supervision has also been shown to have positive effects on self-awareness and self-efficacy in occupational settings [50,51]. It is therefore possible that utilizing peer support groups may benefit the effectiveness of wearable-based interventions that aim to achieve health-related behavior change and optimization of stress- and well-being–related outcomes.

To contrast the effects of a condition that is enriched by both enhanced feedback and added peer support to a more simple “off-the-shelf” intervention that just utilizes a consumer-available wearable, the second research question is as follows:

Research question 2: Is offering a 5-week wearable-based intervention that is extended by EMA, personalized weekly feedback, and peer support meetings (“Wearable+”) more effective at improving awareness, self-efficacy, and stress- and well-being–related outcomes in police officers than an “off-the-shelf” (“Wearable”) intervention?

The study protocol was approved (case 2022‐100) by the internal Research Ethics Committee of the Netherlands Organisation for Applied Scientific Research (TC-nWMO). Participation occurred voluntarily based on informed consent, and participants could opt out at any time without further consequences. Data were collected in a custom-built app that allowed for full control over the collected data without using consumer cloud services. During data collection, data were pseudonymously stored by using participant identifiers and anonymized after the data collection was completed by deleting the participant identifiers. Participants did not receive monetary compensation, but in each participating police team a lottery was held among the participants that adhered to the measurement protocol, with one wearable awarded per team.

A mixed research design was applied, combining elements of a within-subject design with those of a between-subject design (Figure 1). In this study, 2 teams from 2 separate units of the Dutch police force were recruited. Of the teams in each unit, one team enrolled into the Wearable condition, whereas the other team enrolled into the Wearable+ condition (the between-subject comparison). Before undergoing the respective 5-week interventions, the participants in each team would first partake in a 5-week baseline control period (the within-subject comparison). This approach ensured that we would have a well-comparable control condition. No a priori stratification methods were applied, but χ² tests to assess possible between-group differences in gender, age group, and wearable ownership before their participation were performed afterwards.

The recruitment of participants was initiated in 2 teams from 2 units of the Dutch police. In each unit, police officers of one team were recruited for the Wearable condition, whereas the police officers of the other team were recruited for the Wearable+ condition. In each team, up to 25 participants were recruited (up to 100 participants in total). This clustered recruitment method ensured geographical distribution of participants while preventing contamination of the different interventions within teams. Due to a lack of response in one team (the Wearable condition team in the rural unit), participants were also recruited from an additional team within that same unit.

In each team, recruitment was done via a combination of posters, flyers, a mention of the study during a team meeting, and an email to all police officers. Interested police officers would use an online tool to check if they met the inclusion criteria, which meant that they were (1) aged 18 years or older, (2) executive police officers, and (3) willing to put aside their own wearable during the duration of the study if they owned one. The police officers then received an information brochure about the study and filled in an online informed consent form before they received their baseline questionnaire and started their participation in the 5-week control period. To stimulate participant adherence during the intervention condition, within each team, a lottery was held that included the participants that collected complete daily data for at least 80% of their participation period and filled in all 5-weekly outcome questionnaires, where the winning participant got to keep the wearable that was used during the study.

Data management and statistical analyses were performed in RStudio (version 2023.06.2) [58] using R (version 4.2.2; R Foundation for Statistical Computing) [59].

A total of 95 participants were recruited (48 Wearable, 47 Wearable+), of which 16 participants (7 Wearable, 9 Wearable+) were excluded from statistical analysis. Of the excluded participants, 12 (3 Wearable, 9 Wearable+) used their own private wearable during the control period, 2 (Wearable) dropped out, and 2 (Wearable) did not fill in the final (T2) questionnaire. The data of the remaining 79 participants—who were predominantly males (n=57, 72%) in the age category of 30-45 years (n=46, 58%) and did not own a wearable before participating in the study (n=47, 59%)—were analyzed. Descriptive statistics of the analyzed participants of both intervention groups and the total sample are described in Table 1. No statistically significant baseline differences between the Wearable and Wearable+ groups in gender, age category, and wearable ownership prior to participation were identified based on χ² testing.

LMMs were created for each of the 15 awareness-, self-efficacy–, and well-being–related outcomes. Statistically significant 2-way interaction effects were found in 8 outcomes, namely MCA stress awareness and cause (P=.04, mR²=.02, cR²=.59, Hedges g=0.30, 95% CI 0.01-0.58), MCA stress regulation (P=.002, mR²=.02, cR²=.66, Hedges g=0.41, 95% CI 0.16-0.66), MCA physical activity awareness and cause (P=.03, mR²=.07, cR²=.64, Hedges g=0.30, 95% CI 0.04-0.57), self-efficacy behavior change (P=.002, mR²=.06, cR²=.73, Hedges g=0.37, 95% CI 0.13-0.60), self-efficacy stress resilience (P=.04, mR²=.03, cR²=.71, Hedges g=0.25, 95% CI 0.01-0.50), and self-efficacy task (P=.03, mR²=.02, cR²=.68, Hedges g=0.29, 95% CI 0.04-0.54), as well as the MHC-SF (P<.001, mR²=.02, cR²=.74, Hedges g=0.40, 95% CI 0.17-0.63) and PSS-10 (P<.001, mR²=.02, cR²=.71, Hedges g=0.46, 95% CI 0.22-0.70). The detailed results for the 3-step LMM process for each of the 15 outcomes is available in Multimedia Appendix 3. The interaction plots in Figure 3 summarize these findings by visualizing the average scores before (“0” on x-axis) and after (“1” on x-axis) the control (gray) and intervention (blue) conditions, and labeling the statistically significant interaction effects.

For each of the 8 statistically significant outcomes, the wearable-based intervention had a favorable effect in comparison to the control condition. For 6 of the remaining 7 nonstatistically significant outcomes, the direction of the effect also favored the intervention, with the exception of recovery during work. Therefore, the observed effects consistently favored the wearable-based intervention over the control period, for which the statistically significant effects were estimated to be “small” (Hedges g=0.25-0.46) in size. In these outcomes, the fixed effects (the control or intervention period) explained between 2% and 7% of the variance, whereas the full model including the random effect (the participant identifier) explained between 59% and 74% of the variance in the outcomes.

In the LMMs that were created for analysis 2, statistically significant 3-way interaction effects were found in 3 of 15 outcomes, namely self-efficacy behavior change (P=.02, mR²=.10, cR²=.75, Hedges g=0.58, 95% CI 0.11-1.05), recovery after work (P<.05, mR²=.01, cR²=.83, Hedges g=0.41, 95% CI 0.04-0.77), and sleep issues (P=.03, mR²=.02, cR²=.77, Hedges g=0.45, 95% CI 0.02-0.88). The detailed results for the 3-step LMM process are available in Multimedia Appendix 4. The interaction plots in Figure 4 summarize these findings by visualizing the average scores before (“0” on x-axis) and after (“1” on x-axis) the control (gray) and intervention (blue) conditions for both the Wearable (striped line) and Wearable+ (solid line) groups and labeling the statistically significant interaction effects.

Based on the interaction plots of the outcomes with statistically significant 3-way interaction effects (Figure 4), the enhanced Wearable+ intervention was more effective than the “off-the-shelf” Wearable intervention for improving recovery after work, but less effective for improving self-efficacy in behavior change and perceived sleep issues. However, the plots of these 3 outcomes all showed a noticeable change during the control period, all of which benefited the intervention group, which demonstrated statistically significant improvements in the respective analyses. Therefore, these changes during the control period may have influenced the results for these outcomes. Furthermore, the broad 95% CIs of the estimated effect sizes, compared to those in analysis 1, indicate decreased certainty in the magnitude of the reported effects. The results of analysis 2 should therefore be interpreted more cautiously than the results of analysis 1.

This research contributes to the literature by assessing the effectiveness of wearable-based interventions for reducing stress and improving well-being in a real-world context. By utilizing consumer-available wearables and investigating the impact of the interventions over several weeks in a real-life setting, where the participants are dealing with true life demands, the current findings can be generalized toward occupational settings, especially highly demanding work settings. Additionally, this study used a nested design that optimized the comparability between the control and intervention periods (ie, for analysis 1). Due to the clustered recruitment strategy and geographical distribution across teams, the potential for spillover effects between the 2 investigated wearable-based interventions was considered to be low (ie, for analysis 2).

A limitation of the study that could have caused a potential underestimation of the true effect sizes via ceiling effects [64] is the inclusion of a relatively healthy population. To at least partially account for this limitation, an LMM approach that was optimized for repeated measures (and therefore accounted for potential baseline differences between participants) was used in favor of using change scores [65]. Another limitation was that the Wearable+ intervention simulated the application of enriched feedback by providing custom-made feedback reports. It is possible that more integrated and continuously available feedback could be more user-friendly and potentially more effective.

The results of analysis 1, which showed that offering a wearable-based intervention improved stress- and well-being–related outcomes, are in line with those of prior studies that investigated the effectiveness of wearables in real-life settings. For instance, Ponzo et al [31] reported positive effects on anxiety, depression, and general well-being following a 4-week intervention that used a wearable device (BioBeam) that measured sleep, physical activity, and heart rate, as well as an app that utilized EMA and relaxation exercises based on a randomized controlled trial in a student population. Another randomized controlled trial, by Smith et al [32], showed favorable effects from a 4-week intervention on stress and anxiety, but used a wearable (Spire Stone) that primarily focuses on optimizing breathing patterns. Other studies on the effectiveness of wearable-based interventions were often performed in laboratory settings, often in young (student) populations [34]. The results of this study therefore expand upon this prior work by exploring the effectiveness of wearable-based interventions in a working population with high job demands and assessing actual changes in lifestyle outcomes and potential precursors of those in a real-life context.

We are not aware of similar prior work comparing the effectiveness of enriched and basic wearable-based interventions as in analysis 2. The difference between our enriched and basic wearable-based intervention can be broken down into the addition of (1) EMA questionnaires, (2) complementary and personalized feedback, and (3) peer support groups. The potential of combining wearable data with EMA questionnaires and complementary feedback is an emerging research topic in this field [66-68], but information about the benefits of EMA and expanded feedback next to wearable data is scarce. Similarly, there are indications that peer support groups can have a positive effect on the recovery of patients with mental health conditions [47-49] and may benefit self-care in patients with medically unexplained symptoms [46], but the effects in a preventive occupational setting are unknown. The lack of prior research on these specific topics highlights the relevance of this study, though based on our results, no clear benefits of these 3 extensions to the basic wearable-based intervention were identified.

Given the evidence obtained from our analyses, offering a 5-week wearable-based intervention to police officers has small but consistent positive effects on their stress, well-being, and related MCA and self-efficacy. The results also showed that there are large interindividual differences in the outcomes that were analyzed, which may be a sign that there is potential for further assessment of possible differences in effectiveness in subgroups, as well as for improvements in the overall effectiveness via more personalized approaches. Complementing the basic “off-the-shelf” wearable-based intervention with additional EMA questionnaires, weekly personalized feedback reports, and peer support groups did not appear to improve the effectiveness of the intervention in its current form. Future research is needed to better understand how the properties of the enriched condition can be improved to increase the intervention’s effectiveness and which participants may or may not benefit from such extended feedback and support.