Physical inactivity is highly prevalent in modern society, with an average global prevalence of 31% [1]. It is associated with noncommunicable diseases such as stroke, hypertension, and diabetes [2] and is now the fourth leading risk factor for mortality [3]. To promote physical activity (PA), behavior change techniques (BCTs) have been developed [4], which can be delivered either in-person (eg, by health care professionals) or digitally (eg, through smartphones and wearable activity trackers). These digital mobile health (mHealth) interventions are expected to play a pivotal role in promoting PA, particularly during and after the COVID-19 pandemic, when in-person contact is highly restricted. These digital PA tools typically offer measurement and monitoring (eg, activity logs), information and analysis (eg, progress and individual exercise data), and support and feedback (eg, advice on PA and goal setting) [5]. The most frequently implemented BCTs include self-monitoring, providing feedback on performance, and goal-setting [6]. Observational studies have found that fitness app users are more physically active than nonusers [7], with app users having approximately twice the odds of meeting aerobic PA guidelines compared with nonusers, even during the COVID-19 pandemic [8]. One important limitation in the literature is that most studies on PA apps (and health care apps in general) have used a cross-sectional design, meaning longitudinal evidence is still lacking. We aimed to fill this gap and investigate whether, and how, daily use of PA apps and wearables (not necessarily as part of a clinical intervention or treatment) is prospectively associated with increased levels of PA.

A large number of randomized and nonrandomized trials have been published on this topic, not limited to daily use of commercial PA apps and wearables. To our knowledge, 3 umbrella reviews have been conducted, focusing on digital interventions for improving PA. An early umbrella review [9] synthesized 11 systematic reviews and meta-analyses on eHealth or mHealth interventions targeting PA, sedentary behavior, and healthy eating for healthy individuals. The authors concluded that the majority of eHealth/mHealth interventions were reported as effective, although high heterogeneity was observed across multiple studies. Another umbrella review [10] focused on interventions using wearable activity trackers to improve PA. A synthesis of 39 systematic reviews and meta-analyses indicated a moderate effect size (standardized mean difference 0.3-0.6). A more recent umbrella review [11] identified 17 systematic reviews and meta-analyses on digital interventions specifically targeting PA and sedentary behavior to prevent or manage noncommunicable diseases. The results suggest that digital interventions have a small to moderate effect on increasing PA, although heterogeneity is documented across multiple reviews. For example, 3 systematic reviews concluded that mHealth interventions are effective, particularly those involving gamification [12], personalization [13], or delivery in workplace settings [14]. By contrast, a meta-analysis [15] found no significant effects of mobile interventions on total PA, moderate to vigorous PA, or walking. Similarly, a review of mHealth interventions equipped with social features found a nonsignificant effect on PA outcomes [16].

Researchers have also explored specific components or features of apps and wearables that are key to improving PA. Apps and digital interventions offering richer content and a larger number of BCTs are found to be more favored by users [17] and are associated with better health outcomes [18,19], although users also appreciate simplicity (eg, ease of use) [20]. Researchers have also found that certain app features and characteristics are more favored than others, such as data export, usability, and cost [20]; tracking (eg, steps, heart rate, and ovulation) [21,22]; and health information and medical reminders [23]. However, the umbrella reviews [9,11] concluded that the evidence for the effectiveness of specific BCTs or combinations of BCTs in digital PA interventions is largely mixed. In the search for successful digital implementations of BCTs, meta-regressions and systematic reviews highlighted the importance of behavioral goals and self-monitoring [24]; SMS text messaging, personalization, goal setting and planning, and graded tasks [25]; goal setting, prompts/cues, feedback on behavior, and action planning [26]; and personalized goal setting with motivational feedback [27]. By contrast, several meta-analyses reported no significant associations between intervention efficacy and the number or types of BCTs implemented [28,29].

Note that these analyses typically targeted clinical trials focused on specific populations, such as patients, older adults, and individuals with low socioeconomic status. Only a few studies have investigated how the spontaneous use of commercial apps and wearables can help improve PA in a community sample. Investigating app use in uncontrolled settings is particularly important to assess the potential efficacy of PA apps and wearables on the market. Studies have highlighted substantial differences in user behavior between controlled clinical contexts and real-world settings. For example, the average retention rate of mHealth interventions in published randomized controlled trials is about 91% [25], which is surprisingly high compared with the user engagement observed with commercial health care apps (eg, 4%, the median percentage of daily active users of mental health apps) [30]. An exceptional longitudinal study [31] investigated user engagement with a commercial app that rewards users with digital incentives for walking. The results showed that 60% of participants engaged with the app for at least 6 months. Interestingly, users who actively engaged with the app experienced larger increases in daily step count compared with less frequent users. These findings highlight the importance of continuous, long-term use of PA apps or wearables for users to fully benefit from these digital tools.

The general purpose of this study was to identify app-use patterns prospectively associated with increased levels of PA. To achieve this, we conducted a 2-wave longitudinal survey with a 6-month interval. At each survey point, participants reported their current levels of PA and whether they were using any PA apps or wearables. Each participant was classified into 1 of 4 categories: continued users (those who used apps or wearables at both baseline and follow-up), new users (those who started using apps before the follow-up), discontinued users (those who used apps at baseline but not at follow-up), and continued nonusers (those who had never used apps). We were particularly interested in (1) whether continued and new users would increase their PA levels or maintain high-level PA at the 6-month follow-up, and (2) whether any features of the PA apps would predict increases in PA.

Our recent cross-sectional study found that users typically engaged with a limited number of functions within an app (median 2 functions, IQR 1-4 functions). Physically active users tended to use functions such as sensor information (eg, step count and heart rate), goal setting (eg, setting a daily step goal), energy analysis (eg, estimating calories burned), journaling (eg, manually recording daily exercise), and global positioning system (GPS)/maps. We did not have a specific hypothesis regarding the prospective effects of individual app functions, so the overall analyses were conducted in an exploratory manner. However, we expected that sensor information (closely related to self-monitoring/tracking in the BCT taxonomy) and other functions implementing regulatory techniques [32] would be associated with increases in PA. Additionally, we expected that the number of functions might be linked to PA increases, as some studies have suggested that the amount of app content or the number of implemented BCTs is associated with the efficacy of mobile interventions [18,19].

In this study, we had 2 objectives: the primary objective was to investigate whether new and continued users would increase PA or maintain high levels of PA over 6 months. The secondary objective was to identify the functions and features of PA apps that would predict increases in and maintenance of PA.

Participants were recruited from the respondents who completed the first (baseline) survey, the results of which have been published elsewhere [33,34]. The baseline survey was conducted in 2023. Invitations to participate were sent to potential participants (residents of Japan who were registered in a database for online surveys) [33,34]. At baseline, 20,573 online respondents completed questionnaires regarding general health and health-related behaviors, including PA levels and use of mHealth apps. The only eligibility criterion was age (>18 years), with no additional criteria. Proficiency in Japanese was assumed, as the survey was written in Japanese. All participants were invited to the second (follow-up) survey, which took place approximately 6 months after the baseline survey. The follow-up survey, which included the same questions on PA and mHealth use, was completed by 16,286 of the 20,573 participants. Data from those who completed the follow-up were used for statistical analyses. Because of the online nature of the surveys, the reasons for dropout could not be tracked. Specifically, we had no means of reminding participants about the follow-up survey other than via email. Tracking dropouts was technically impossible, as they no longer responded to our emails. Participants received a small compensation for each survey (an online shopping voucher valued at approximately US $0.31).

The study protocol was approved by the Ethics Committee of the National Institute of Advanced Industrial Science and Technology (approval ID 2022-1279).

At baseline, participants provided a binary response indicating whether they used any apps or wearables to support their PA and exercise. Those who answered affirmatively were then asked for details on how they used the apps and wearables. The questions included the duration of app use, with the following response options: <1 week, <1 month, <3 months, <6 months, <1 year, and ≥1 year. Participants were also asked about the functions and features of the apps they were using. They were presented with a list of 41 app functions (eg, sensor information, goal setting, and energy analysis) [33] and indicated any that applied to their usage [20,29]. However, as most of the listed functions were rarely used [33], we focused exclusively on the most frequently used functions for the current analysis: sensor information (eg, step count and heart rate), goal setting and progress tracking (eg, steps achieved), energy analysis (eg, estimated daily energy expenditure), weight recording, journaling (eg, manually entered diary or notes), GPS/maps, sleep information, reward points, and blood pressure recording.

At the 6-month follow-up, participants completed a similar questionnaire asking whether they were using apps and wearables. Unlike at baseline, participants were given 3 response options: (1) have been using apps and wearables for the past 6 months, (2) used them previously but no longer, and (3) have never used any app or wearable. Participants were also asked about the duration and frequency of app use (ie, how long and how often they had used/been using the app). However, questions regarding individual app functions and features were omitted due to limited space in the follow-up survey.

Responses from baseline and follow-up were interpreted as a 2×2 factorial matrix (user vs nonuser; baseline vs follow-up), classifying each participant into 4 categories (Figure 1): new users (those who began using apps before the follow-up), continued users (those who used apps or wearables at both baseline and follow-up), discontinued users (those who used apps at baseline but not at follow-up), and continued nonusers (those who never used apps). Participants who indicated at follow-up that they had used apps or wearables but were no longer using them (ie, those who selected option 2) were classified as discontinued users if they were identified as app users at baseline. If they were identified as nonusers at baseline, they were excluded from the statistical analyses for ease of interpretation. These participants were considered temporary app users—they may have used apps for a short period between baseline and follow-up but did not show significant changes in PA levels (P=.29; see Figure S1 in Multimedia Appendix 1 for details). We did not consider their usage comparable to that of the other participant groups.

First, we explored demographic and descriptive differences between the 4 types of users based on gender, age, education level, household income, PA level, and readiness. Second, logistic regression analyses were conducted to examine how continued and discontinued app usage are associated with changes in PA levels. Two binary dependent variables were used to represent the following contrasts: (1) individuals who maintained underrecommended levels of PA (<23 or 10 METs) at the 6-month follow-up versus those who increased their PA to recommended or higher levels over time, and (2) individuals who maintained the recommended PA levels versus those who showed decreases and no longer met the recommended levels at follow-up. We also calculated simple change scores for PA (follow-up minus baseline) to clarify the magnitude of change experienced by each type of user over time. Finally, 2 independent logistic regression analyses were conducted to determine which app features were associated with changes in PA levels. The logistic regression models predicted the 2 binary dependent variables, specifically changes in (and maintenance of) adherence to the recommended PA levels, based on the individual 10 app features and functions used at baseline. All analyses were conducted using R (version 4.2.2; R Foundation) with the following specific packages: chisq.posthoc.test [41], finalfit [42], ggpubr [43], and tidyverse [44].

Table 1 shows the demographic characteristics for each type of app users (N=16,286). We identified 1462 (8.98%) new users, 2150 (13.20%) continued users, 1899 (11.66%) discontinued users, and 10,775 (66.16%) continued nonusers in the data set. A 1-way analysis of variance indicated significant age differences between the user types (partial η²=.009, P<.001), implying that new users were older than continued users and continued nonusers, and discontinued users were the youngest among the 4 types of users (P<.001; adjusted by the Tukey method). The descriptives per age group showed that, among older participants (eg, ≥60 years) who were identified as app users at baseline, 899 continued app use to follow-up, whereas 675 discontinued their use. Younger app users at baseline (eg, <30 years) were, however, more likely to discontinue than continue app use (n=250 vs 203). Chi-square tests revealed overall significant gender differences (P<.001). Residual analyses detected significant gender differences (P<.001) within continued users, men (1295/2150, 60.23%) versus women (855/2150, 39.77%). However, within continued nonusers, women (5770/10,775, 53.55%) were more dominant than men (5005/10,775, 46.45%; P<.001). Furthermore, continued users were the most prevalent among individuals with the highest household income (≥10 million JPY; 1 JPY=US $0.0065) and education level (university or above). At baseline, most continued users (1609/2150, 74.84%) reported that they had been using an app for longer than 6 months (1128/1899, 59.39%, for discontinued users). Similarly, most continued users (1654/2150, 76.93%) reported using a PA app once or more each day (1153/1899, 60.71%, for discontinued users).

Table 2 illustrates changes in PA levels for each user type over 6 months. Continued nonusers generally maintained underrecommended levels of PA over time (5259/10,775, 48.81%). Among the 4 types of users, new users had the largest proportion of individuals who increased their PA to recommended levels (178/1462, 12.18%), although achieving these levels was uncommon in the current sample. Continued users typically maintained the recommended PA levels over time (1327/2150, 61.72%), while discontinued users had the largest proportion of individuals who failed to adhere to the recommended levels at follow-up (330/1899, 17.38%).

Logistic regression analyses were performed to examine how the 4 types of users were associated with adherence to the recommended PA levels (23 or 10 MET hours/week) over 6 months (primary objective). Discontinued users were used as the reference group, as they showed the largest proportion of individuals who decreased their PA to underrecommended levels at follow-up. Results (Table 3) showed that, compared with discontinued users, continued users (odds ratio [OR] 2.171, 95% CI 1.71-2.76, P<.001) and new users (OR 1.93, 95% CI 1.50-2.48, P<.001) were more likely to increase PA to the recommended levels at follow-up. Continued nonusers, compared with discontinued users, were more likely to maintain underrecommended PA levels at follow-up (OR 0.71, 95% CI 0.59-0.86, P<.001). Another logistic regression analysis showed that continued users (OR 2.56, 95% CI 2.11-3.10, P<.001) and new users (OR 2.29, 95% CI 1.84-2.85, P<.001) were more likely to maintain the recommended PA levels compared with discontinued users. Continued nonusers did not significantly differ from discontinued users (OR 1.12, 95% CI 0.97-1.29, P=.11).

We then calculated the simple change scores for PA levels (ie, follow-up minus baseline in MET hours/week) to estimate the changes experienced by each user type over 6 months (Figure 2). New users were the only group to show increases in PA levels (mean 1.71, SD 57.76), which was significantly larger than the changes (decreases) observed in continued nonusers (mean –2.95, SD 50.76; t_1,780.5= 2.94, P=.003). Continued users showed a decrease in PA on average (mean –3.85, SD 58.53), while discontinued users exhibited even larger decreases (mean –7.95, SD 60.52; t_3,949.1=2.19, P=.03).

At baseline, app users most frequently reported using sensor information, followed by goal setting, goal progress, energy analysis, and weight recording (see Table S1 in Multimedia Appendix 1 for details). To explore which app functions are associated with increases in or maintenance of PA levels (secondary objective), we estimated 2 logistic regression models. These analyses targeted continued users exclusively (ie, individuals who reported using apps at both baseline and follow-up), with nonusers excluded. The results (Table 4) showed that energy analysis and journaling were significantly associated with increases in PA to the recommended levels (OR 1.67, 95% CI 1.05-2.64, P=.03 and OR 1.76, 95% CI 1.12-2.76, P=.01, respectively). Maintenance of the recommended levels (vs a decrease to underrecommended levels) was predicted by goal setting (OR 1.73, 95% CI 1.21-2.48, P=.003), sleep information (OR 1.66, 95% CI 1.03-2.68, P=.04), and blood pressure recording (OR 2.05, 95% CI 1.10-3.83, P=.02). We also tested the association between the number of app functions in use (at baseline) and changes in PA level (simple change score, ie, follow-up minus baseline) among continued users, which did not reach statistical significance (r=–0.02, P=.48).

This study investigated whether continued use of commercial PA apps and wearables over 6 months contributes to increasing or maintaining PA levels. Overall, the results emphasize the importance of continued app use in maintaining PA levels. Most continued users (1538/2150, 71.53%) either maintained or improved their PA to the recommended levels over 6 months, whereas 51% (968/1899) of discontinued users failed to meet the recommended levels at follow-up (compared with 797/1899, 41.96%, who were nonadherent at baseline). New users were found to experience the largest increase in PA levels, while discontinued users showed the largest reduction among the 4 types of users. These results suggest that individuals who recently started using apps and wearables saw the greatest improvement in their PA levels. By contrast, continued app use helped maintain PA levels (despite slight decreases), whereas discontinuation led to a substantial reduction in PA levels, equivalent to a decrease of more than 1 hour of vigorous PA per week [45].

More than half (2150/4049, 53.09%) of the app users identified at baseline reported continuing to use the app at the 6-month follow-up. While this retention rate may seem high compared with the reported daily engagement rates for health care apps [30], it is comparable to the 60% active user engagement rate observed in a longitudinal study of a commercial PA app over 6 months [31]. Our analysis of the demographic characteristics of continued (vs discontinued) users indicated that continued users were older, more likely to be men, and had higher education levels and incomes. Previous cross-sectional studies have found that mHealth/eHealth users tend to be younger, more educated, and have higher (digital) health literacy than nonusers [46-48]. Educational attainment is thought to reflect both literacy and skills (including confidence with digital and smart devices) as well as social norms related to the perceived value of health [47]. In general, women are the dominant users of health care apps (for diet, nutrition, and self-care), while fitness apps tend to be more popular among men [46]. Older adults typically avoid new technologies and mHealth services [49]. However, our results showed that continued users were more prevalent than discontinued users among older participants, suggesting that older users were more likely to continue using apps. We do not have data to readily explain this unexpected finding. However, given that most of the continued users in our data had already been using PA apps for a long time (>6 months) at baseline, even older users may have developed high self-efficacy and perceived ease of use, which could reduce technology anxiety. Studies have identified various facilitators and barriers to technology adoption among older adults (eg, personal experiences and subjective norms) [50], which could serve as a basis for future research to explore how older users successfully adapt and integrate mHealth tools into their daily routines.

We found that energy analysis and journaling were predictive of increases in PA to the recommended levels over 6 months. Additionally, goal setting, sleep information, and blood pressure recording were commonly used by individuals who maintained the recommended PA levels at follow-up. In a previous cross-sectional analysis, we reported associations between PA levels and individual app functions [33], which indicated that individuals with health-enhancing PA levels typically used functions such as sensor information (eg, step count and heart rate), goal setting, goal progress, energy analysis, journaling, and GPS/maps. The current prospective analyses emphasize the particular importance of energy analysis and journaling in improving PA over time. While sensor information was commonly used by PA app users (Table S1 in Multimedia Appendix 1), the findings suggest that, in addition to the automatically recorded PA data (eg, step count), users may benefit from additional analyses of physiological data (eg, energy expenditure calculations) and more deliberate engagement with the app, such as journaling and manually logging daily exercise and PA.

Interestingly, maintenance of the recommended PA level was associated with the use of sleep information, blood pressure recording, and goal setting. This suggests that users who are already sufficiently active may value functions that support general health care, rather than those focused solely on fitness and exercise. It is also possible that some users were prompted to use PA apps due to specific health concerns. Published meta-analytic studies have identified key app components that enhance PA, including self-monitoring, goal setting and planning, prompts/cues, feedback on behavior, and action planning [24-27], most of which are specifically designed to support PA. It may be important for future research to broadly explore app functions and features (not limited to PA-related functions) to identify more effective combinations, especially when the focus is on maintaining rather than increasing PA levels. This could reinforce the usefulness of the stages of change model [39,51] (eg, to guide the best interventions for those in the action or maintenance stages) and highlight the importance of tailoring digital behavior interventions.

Another interesting finding from this analysis is that the number of app functions reported as being in use was not significantly associated with increases in PA over 6 months. While several studies have suggested that the amount of app content or the number of implemented BCTs is linked to the efficacy of mHealth interventions [18,19], this association has not always been replicated [29]. As our analyses utilized self-report data rather than actual logs of user behavior, we cannot exclude the possibility that participants may not have accurately or exhaustively reported all the functions they used. However, our findings suggest that (1) users may not be fully aware of every function available in an app (or, at least, they do not consciously use them all), and (2) they may not necessarily benefit from multifunctionality. Instead, a limited number of functions (eg, goal setting and journaling) may be more effective in improving PA. Indeed, it is known that users appreciate the simplicity of an app [20], and as Michie et al [32] found, interventions that combine self-monitoring with at least one regulatory technique (eg, goal setting) can form the most cost-effective, minimal set of interventions.

The results reported here should be interpreted with caution due to several important limitations. First, we targeted Japanese-speaking adults exclusively, which may limit the generalizability of the findings. The apps and products available on the Japanese eHealth/mHealth market may differ from those in other regions and countries. While there are similarities in user behavior between Japan and Western countries, exploring country- or culture-specific aspects would be an interesting direction for future research. Second, we cannot rule out the possibility of sampling bias. As reported elsewhere [33], the current sample exhibited higher PA levels than the general population in Japan, likely because the study was advertised as a survey on PA and health. Additionally, attrition could introduce bias, as some participants (4287/20,573, 20.83%) dropped out by the follow-up. Third, the follow-up survey did not include questions regarding how participants used specific app functions and features (assessed only at baseline). As this study was part of a larger project, there was a limit on the length of each survey. Future research should explore how changes in PA influence the use patterns of individual app functions, which could provide insights into how tailoring and personalization can be incorporated during app use adaptation. Fourth, we relied exclusively on self-reported PA, which may not always align with objective measures, such as accelerometers, due to self-reporting bias and other assessment artifacts. Similarly, user behavior and individual app function usage can be monitored automatically or made publicly available (eg, [52]). However, a downside of such an approach is that the analysis would be limited to a specific app or platform, potentially sacrificing the generalizability of the results. Finally, we cannot rule out the possibility of selection bias. As the surveys were administered online, participants were likely familiar with the internet and possibly mobile technology as well. This could explain the unexpected finding that older individuals were more likely to continue using the app.

This study demonstrated that continued use of apps and wearables contributes to both increasing and maintaining PA levels over 6 months. The results also revealed that app features associated with increases in PA differ from those linked to the maintenance of PA, highlighting the importance of tailoring apps to users’ PA levels and readiness. We believe these findings make a meaningful contribution to the literature by highlighting the continued use of apps and wearables as key factors in enhancing and maintaining high levels of PA. Additionally, it is noteworthy that more than half of the users continued using the apps through the 6-month follow-up, despite the poor retention rates and barriers commonly reported in the literature (eg, [53,54]). This may suggest that attitudes toward and the acceptability of mHealth apps are changing, with the COVID-19 pandemic potentially serving as an opportunity. Health care practitioners could increasingly rely on app-based approaches in their intervention repertoires, although integrating mHealth into routine practice remains a challenge [55]. Unfortunately, we did not assess the barriers preventing users from continuous engagement (eg, [56,57]), and these should be explored in future research to identify effective strategies for maintaining active user engagement.