1. Introduction
Physical inactivity is one of the most important public health problems of the 21st century [
1,
2]. Regular physical activity (PA) contributes to the prevention of obesity and chronic diseases [
2,
3]. However, different reasons (e.g., lack of time, feeling too tired [
4], or bad weather [
5]) might impede people from engaging in a healthy amount of PA, as defined by the American College of Sports Medicine [
1]. In recent years, different prevention strategies have sought to overcome these barriers to physical activity. One of these strategies relates to the progressive technologization and affects current tendencies in the fields of sport and prevention. One popular tool to increase PA behavior are wearables, mobile intelligent devices that track one’s steps or heart rate with a wristband.
Wearable devices have gained tremendous momentum and have become part of people’s daily lives by tracking one’s level of activity. In Germany, 6.9 million people used a wearable in 2020 [
6]. Since 2016, wearable technology has been ranked in the top three fitness trends [
7], showing the growing use of activity trackers.
Health insurance companies recognize the underlying potential of compelling people to engage in PA by using activity trackers. However, related offers were not always successful, likely to be because they did not correspond to individual preferences or previous experiences. Our aim was to evaluate the usage of such devices in the first weeks after delivery and check possible affecting external parameters.
Previous systematic reviews and meta-analyses indicate that eHealth interventions that implement activity trackers might be an effective tool to promote PA among adults of various ages [
8,
9,
10,
11,
12]. The general assumption is that once people use a device, they will become concerned with their daily PA, although the positive effects and exact conditions of this usage over the long term are unclear [
13,
14]. For example, in overweight people, activity monitors were shown to increase the PA level in combination with lifestyle or activity interventions [
15]. A cross sectional online survey in a large cohort of mostly young to middle aged adults revealed that the majority of current (81.4%) and former (51.3%) users believed that they incorporated more PA into their days while wearing their activity trackers [
15]. Other researchers found increased moderate to vigorous PA and increased steps/day after 16 weeks of activity tracker usage [
16]. In this case, wearable technology helped to increase daily PA levels, although this increase seemed to be more effective in individuals with lower baseline PA [
17]. Studies on the use of wearables among patients showed different acceptance levels, with about one-third of participants dropping out and only half achieving their daily target goals (e.g., 10,000 steps per day) [
5,
18] for at least 50% of days [
19]. Only about 1/3 of subjects suffering from a myocardial infarct actually used an offered tracker [
20]. In the older population, wearable studies led to inconsistent results [
19,
21]. However, there is no evidence of a positive effect when such interventions are compared to alternative interventions [
14], regardless of target group. For example, studies stated no significant difference in PA behavior compared to a control group, at least after six months of wearing an activity tracker [
22].
Results on usage behavior over time are also heterogenous. The majority of participants show phases of very consistent activity tracker usage [
20,
21]. While some are very enthusiastic for months or even years, others lose their interest after some time or do not participate at all. Indeed, the use of activity trackers is often abandoned after a few weeks or months [
15,
22,
23], and only a little over 40% still use such devices after 24 months [
23]. Among undergraduate students, 65% stopped using the devices within the first two weeks [
24]. To date, little detailed data are available on short-term PA effects (on the first weeks of usage) of wearables.
Moreover, studies have mostly evaluated the summed steps per day [
18,
25,
26,
27,
28] instead of the steps per worn hour or minute (cadence), i.e., the intensity of PA while wearing the device. The few available studies focused mostly on the usefulness of the measure itself instead of the usage behavior [
29,
30,
31,
32]. Attaching a device, however, might not only encourage people to do more steps per day, but also to perform more steps per hour, thereby increasing daily physical intensity or walking speed [
19,
33]. More research is needed to evaluate the possible changes in steps per hour while wearing such a device.
To sum up, it remains unclear if the usage of wearables leads to an increase in PA since some analyzed cohorts are not representative of others. Furthermore, a possible increase in PA might be accompanied by increased PA over a longer time-period [
18,
34] and/or by increased physical intensity.
It has also been discussed whether (beside other factors) the day of the week or weather conditions may play a role in human PA patterns. Generally, certain weather conditions might have greater influence on PA during the weekend than during workdays [
35]. As intervention studies run over several weeks and thus are likely to cross different seasons, it is important also to integrate weather-dependent activity changes. Diverse studies in different countries seem to show comparable reactions with a lower PA level during colder seasons [
36]. Extreme weather conditions with mean temperatures ≥ 29 °C, as well as higher amounts of rain or snow, were shown to lead to a decrease of PA in north America [
5,
36], especially under severe weather such as rain in summer or ice and snow in winter [
37,
38]. In Europe, where our investigation took place, lower PA was associated with lower temperatures, heavier rain, and shorter day lengths. Furthermore, lower PA corresponded to temperature cut-offs below 10 °C [
39,
40]. A decline in PA from November to March is, therefore, expected [
41]. Besides weather conditions, typical patterns for each week or day can be observed since usage days are not equally distributed each week [
20]. The distribution levels over the week are somewhat contradictory, with both less [
20] and more [
23] reported usage on Tuesdays and both higher [
20] and lower [
21] amounts of usage on weekends. Besides usage, more steps could be recorded on the weekend compared to weekdays [
20,
42]. Jeong et al. reported high average wear times of about 8 h per day at weekends versus 11 h per day on Mondays and Tuesdays [
21].
The Current Study
Thus, it remains unclear to what extent wearables are used and affect changes in the PA of new users. With diverse average usage times of just a few weeks up to about seven months in the adult population, more research is needed to understand the requirements for long-lasting use (and effect) of wearable devices such as activity trackers [
15,
20]. Existing studies focused on different populations (elderly, diseased individuals, or different age groups) and reported heterogenous usage behaviors. Therefore, it is difficult to determine the usage behavior of a recreationally active sample of young to middle-aged adults and what influences usage.
In this study, we aimed to evaluate how an activity tracker was integrated into daily life and whether there was a change in usage time across a 9-week period since former research indicated a decline in PA during this first phase after the device was introduced [
22,
23]. The usage time per day and changes after a certain amount of use might be an important indicator for a decline in usage. We hypothesized that interest in a newly implemented wearable device decreases in the first weeks but is highly diverse between individuals. Investigating how fast this process occurs and how it differs between individuals is one aim of this study.
Since about 70% of wearable users reported an initial increase of PA after starting tracking, and 10% of current and 27% of former users reported that their PA levels subsequently decreased to baseline levels within the first weeks of wearing the device [
15], a reduction in the amount of PA seems to be expectable. We were further interested in determining the intensity of PA behavior measured as steps per hour when using an activity tracker and if the steps per worn hour change across time. An altered rate of activity per hour might suggest an accompanying change in usage behavior. That is, some people might use such devices only when they are physically active.
Further, affecting parameters of usage behavior need to be better understood. Therefore, we analyzed usage behaviors over the whole week and on different days of the week. We hypothesized that there would be differences in usage behavior and activity level throughout the week. Further, it is not clear how and to what extent environmental factors such as weather conditions influence usage behavior. A natural response might involve reduced PA under “bad” weather situations such as extreme temperatures or a low amount of sunshine per day. We hypothesized reduced PA on extremely hot or cold days and a positive influence of sunshine.
The random transfer of such devices to, e.g., policyholders might increase their usage and, therefore, lead to a conscious and increased use of such devices in everyday activities. Therefore, we ran this investigation on a “normally” active (defined as being physically active at least two hours per week) population that was recruited by chance without specific interest in wearable technology. Besides individual interests and differences, this investigation intended to show what external parameters like weekdays or different weather conditions can affect usage. Existing studies often analyze these factors via questionnaires or retrospective surveys [
15,
43] among participants that acquired the devices by themselves. In our study, we used the quantitative and physiological data of the tracked subjects. These results will help to assess the acceptance level of wearables and the handler’s usage behavior to identify the influencing factors within the first nine weeks after delivery.
2. Methods
2.1. Participants
Originally, 81 recreationally active participants between 21 and 47 years of age from two cities in Germany (Münster: 310,000 inhabitants; Cologne: 1.06 million inhabitants) were recruited by notice boards, personal networking, and email distribution through cooperative partners like a health insurance company and different industrial companies. Participants were not specifically informed that they would receive a wearable for this study to avoid impacting the recruitment by chance.
Subjects were screened by a short telephone interview. Inclusion criteria were an age between 18 and 48 years and being physically active for at least two hours per week. PA was defined as planned physical activity in a non-professional way. Participants were excluded if they owned a wearable, had used one before, or if they reported any history of disease that limited their ability to perform recreational sports or PA in their free time.
Half of the cohort (n = 39) was assigned to the wearable group (25 female, 14 male; age: 32.8 ± 6.9 years, 21–47 years; height: 174.9 ± 10.1 cm; weight: 72.3 ± 14.3 kg) by simple randomization (pulling a match) and received a Fitbit Charge 2 (Fitbit Inc., San Francisco, CA, USA) for a nine-week period. In this paper, we focused only on that cohort. All participants were well educated, i.e., had at least a high school diploma and professional training or course of studies. Recruited participants performed a wide variety of sports (e.g., strength training, team-ball sports, dancing, horseback riding, climbing, rowing, and endurance training like running or cycling). All participants were informed about the design and the procedure of the study and gave their written informed consent. The participants also signed an informed consent to allow us to access to their recorded wearable data. There were no dropouts over time because the usage of the wearable was tracked across the whole study period, and non-usage (without a specific reason, e.g., illness) was also treated as a result. This study was approved by the local Ethics Committee of the Department of Psychology and Sports Sciences, University of Muenster, Germany (2019-33-AH) and was conducted according to the Declaration of Helsinki.
2.2. Measures
Physical activity data and the usage of an activity tracker were evaluated via a conventional device (Fitbit Charge 2; Fitbit Inc., San Francisco, CA, USA). Participants were asked to wear the device on their preferred wrist and to secure data transmission by adjusting the device to a proper position. This wearable device is a wireless, triaxial accelerometer with a heart rate sensor. The raw acceleration signals are converted into sums to estimate a person’s steps. Further, one’s heart rate, activity level and energy expenditure per minute were extracted [
44,
45]. To secure their personal data, participants received anonymous email addresses to create an account, and help was offered to set up the account if a participant had any problems in doing so. The participants then synchronized the device with their smartphone app for better data visualization and to assure synchronization without connecting the device to a personal computer. Fitbit data synchronization and data securing were performed by the subjects and the scientific staff at least once a week. If a participant did not synchronize in time or did not connect to the webpage at all, he or she received an email or a phone call with help offered. Data were then downloaded anonymously as JSON files by the study staff and processed by the R 4.0.2 base package [
46]. Data analysis used the heart rate and steps recorded by the wearable.
The usage time was determined by the HR readings per minute. The heart rate was recorded constantly, and the data output generated small clusters of reading intervals produced by Fitbit©. Furthermore, Fitbit provides a confidence indicator for measuring parameters to indicate the level of confidence for data correctness. In this study, we only used values that had a confidence > 0. A valid use per minute was assumed if at least one HR value was available in that minute. From these valid uses minutes, the total usage time was calculated for each participant each day. The average usage time was determined as an indicator for acceptance. If there were no HR readings on one day, the usage time is set to 0. The usage hours were analyzed by the hours per day.
For the physical activity analysis, a maximum of 16 h was analyzed to exclude possible sleep time. The physical activity behavior was measured by the number of steps recorded with the triaxial accelerometer of the Fitbit each day. If at least one step per hour was detected, the hour was considered. In 303 cases (out of 2379 cases, n = 39, 61 days), there were no steps recorded although the wearable was used. These cases were set as “missing”. Steps were analyzed as steps per day.
Additionally, the number of steps per use hour were assessed as a measure of intensity. Since participants were not forced to wear the device constantly, we could only verify the parameters of the hours when the wearable was attached. We excluded days where no step was recorded but used the HR to avoid measurement errors. If the wearable was not used at all during that day, the steps per hour were determined as 0. Further indicators of usage behavior were the number of days since the wearable was received (61 days in total), the day of the week, the number of hours of sunshine in the day (M = 6.0, SD = 4.5, 5 categories: 0 h (13.4%), 1–4 h, (30.2%), 5–8 h (22.3%), 9–12 h (23.8%), > 12 h (10.3%)), and the maximum temperature that day (M = 19.5, SD = 6.3, 5 categories: ≤ 10 °C (5.7%), 11–15 °C (25.1%), 16–20 °C (27.3%), 21–25 °C (21.7%), > 25 °C (20.2%)). Weather information was obtained from a weather data platform for that specific region and was freely available with an hourly resolution [
47]. The hours of sunshine and temperature were divided in consistent segments of five degree or four hours, respectively. Weather conditions were calculated individually for the nine-week time span for each participant. Since our subjects lived in an area close to the city center of their hometown and made no long journeys or holidays during the study, weather conditions from the city were used for the analysis. For each participant, weather data of their specific hometown (the two relevant cities lying 150 km apart) are used.
2.3. Procedure and Intervention
The study ran between April and November 2018 to exclude the cold weather phase. During their first visit to the investigation center, the subjects filled out questionnaires regarding their usual weekly PA followed by an assessment of anthropometric data. Then, a Fitbit Charge 2 was handed out with the request to start usage the upcoming weekend. Participants were asked to wear the Fitbit during the following nine weeks as much as they desired. The device must be charged about every five days for a maximum of two hours. This time was not documented since we assumed that only a few subjects wore the device all the time. The participants received no specific information about the purpose of the study but were informed that we were interested in how they would use the activity tracker and how active they are. The default settings of the device aimed for 10,000 steps/day [
48]. If this goal was reached, fireworks combined with vibrations appeared on the wearable’s display. The subjects were encouraged (not forced) to adjust this goal or to test the tracking mode of the device’s sleeping states. They were further told that they were free to handle the device but had to synchronize the device at least once a week. After the nine-week period, participants returned the wearable.
2.4. Statistical Analysis
Pre-processing of the data and statistical analysis were performed using the R 4.0.2 base package [
49]. The additional modules “lmerTest” [
49] and “lme4” [
50] were used for statistical modelling. Statistical analyses were constructed in the same way for both dependent variables (usage time (h) and steps/h). Because the subjects were not forced to wear the device, PA was detected only when the wearable was applied. Therefore, only steps per hour instead of steps per day (PA) were evaluated. In the first step, we assessed the need to use multilevel analysis by comparing a baseline linear mixed effect model that includes only the intercept with the same model, allowing the intercepts to vary across participants. Due to the significant difference between these models, we concluded that intercepts varied significantly across participants and thus used random effects in subsequent analyses. To examine the main effects for the factors week of intervention, weekday, hours of sunshine, and temperature, a linear mixed effects model was calculated for both the outcome variables of usage time and steps per hour, using Satterthwaite’s degrees of freedom method to calculate the p-values [
49]. For the main effects, F-statistics were reported, and the omega squared (ω
2) was reported for the effect size. Upon reaching significance (
p < 0.05), the model contrasts were further inspected. In the second step, for each dependent variable (usage time and steps per hour), a full linear mixed model with all predictors (weeks, weekdays, temperatures, and hours of sunshine) and a random intercept for the participants was conducted.