This study reports primary findings from an app-based, 2-condition, randomized controlled trial (RCT) on healthy smartphone use among adults from the general population. The preregistration for the RCT can be accessed at the German Clinical Trials Register (DRKS00017606; date of registration: August 9, 2019; first participant enrolled: October 9, 2019; targeted sample size: 200). To provide a deeper focus on smartphone-related outcomes and intervention mechanisms, this paper reports findings on the primary outcome (problematic smartphone use), whereas findings on the secondary outcome of the RCT, psychological well-being, are not reported. The Ethics Committee of the Humboldt Universität zu Berlin granted ethics approval for this study (registration #2019-14R1).

Eligible participants were at least 18 years old, owned and used a smartphone with an iOS operating system (minimum Apple iPhone 5, iOS system 10+), and had sufficient visual ability and skills to understand and complete the English language study materials. Participants were recruited by using reactive strategies such as flyers, online postings, and email lists. As an incentive for study participation, participants took part in a lottery of 4 online shopping vouchers worth €25 (US $29) each and received course credits if needed. Data collection ranged between October 2019 and December 2019.

After downloading the study app and providing informed consent, participants responded to the baseline questionnaire. Subsequently, they were randomly assigned to either the intervention (intervention=1) or active control condition (control=0) using a simple (“flipping a coin”) randomization procedure via a web-based tool. No blinding procedures were used. Based on randomization and throughout the following 20 days (D; D1-D20), participants received daily app-based sessions on goal-directed smartphone use in 5 modules each spanning 4 days (intervention condition) or on defining daily time-outs (active control condition).

Throughout the 20-day intervention period, participants completed brief questionnaires on D4, D8, D12, D16, and D20, corresponding to the completion of the 4-day modules from the intervention condition. Moreover, participants responded to longer questionnaires at postintervention (D21) and at a 3-week follow-up (D42; Figure 1). Multimedia Appendix 1 (Figure S1) provides a Consolidated Standards of Reporting Trials (CONSORT) flow diagram.

To foster usability and acceptability, the development of the content in the intervention condition followed a user-centered design and person-based approach [23,24]. Initially, a large pool of brief daily exercises was developed based on elements used in cognitive behavior therapy, acceptance and commitment therapy¹⁶, and behavior change interventions [22]. The pool of exercises was tested with a total of 44 volunteering smartphone users. Moreover, guidance from 7 experts was received, which resulted in a final set of 20 exercises. The material of 5 out of 20 daily exercises (duration: 2 to 10 minutes per day) from the intervention app can be found in Multimedia Appendix 2 and 3.

The five 4-day modules were the following: Observe (D1-D4), Reflect (D5-D8), Vision (D9-D12), Plan (D13-D16), and Support (D17-D20). In the Observe module, participants observed their physical reactions (eg, posture, impulses of checking behavior) to the smartphone in exercises on impulse control and mindfulness. The Reflect module included educational elements regarding a better understanding of habitual and problematic forms of participants’ smartphone use [10]. In the Vision module, smartphone-related exercises on mindfulness, value-related committed action [19], and goal-setting were conducted [20]. The Plan module focused on strategies toward goal-directed smartphone use, such as developing if-then plans on when, where, and how to use the smartphone (action planning) or executing alternative behavioral responses in critical situations (coping planning) [25]. In the Support module, practical tips to support sustainable behavior change (eg, redesign of the home screen) were provided.

Similar to earlier digital detox intervention studies [12], participants in the active control condition received a daily time-out treatment. Across the intervention phase (D1-D20), they were asked to plan a smartphone time-out of at least 1 hour within the next 24 hours. Participants could freely choose whether and how long they executed their planned smartphone time-out; that is, nonaccess to the smartphone was not technically enforced.

Randomization to the 2 experimental arms was based on 232 enrolled individuals (205 women, 23 men, 1 diverse, 3 missing values) with a mean age of 29.62 years (SD 8.09, range 18-60 years). Further baseline sample characteristics are displayed in Multimedia Appendix 1 (Table S1).

After providing informed consent and responding to the baseline questionnaire, 114 participants were assigned to receive the intervention on goal-directed smartphone use, and 118 participants were assigned to receive the time-out treatment in the active control condition on D1. A randomization check (χ2 and t tests, followed by logistic regressions) using the experimental condition variable as the outcome revealed no unique between-condition differences in baseline variables, pointing to a successful randomization.

A subsample of 110 (47% out of 232; n=55 in each condition) participants provided data at postintervention, and 88 (38% out of 232; n=44 in each condition) participants did so at the follow-up. Attrition rates within the range of this study are normal for online interventions because researchers do not have much control over the attrition of anonymous participants [36]. Participants from the longitudinal sample (n=88) showed a high response rate to questionnaires between D1 and D42 with a mean response rate of 93% (7.47 out of 8 assessments; SD 0.96, range 4-8).

To examine attrition bias, χ2 tests, t tests, and logistic regressions were performed across baseline variables as well as baseline variable x experimental condition interactions, with a dummy-coded attrition variable (0=dropped out; 1=remained in the study) as the outcome. A significant, unique difference emerged for age (dropped out: mean 31.11 years, SD 8; remaining in the study: mean 27.28 years, SD 7.7), baseline problematic smartphone use (dropped out: mean 3.74, SD 0.77; remaining in the study: mean 3.44, SD 0.75), and baseline action control (dropped out: mean 2.21, SD 1.01; remaining in the study: mean 2.55, SD 1.08). This indicates that participants in the longitudinal sample were younger and demonstrated lower problematic smartphone use and higher action control at baseline when compared to those who dropped out. Subsequent analyses therefore controlled for attrition variables of age, problematic smartphone use, and action control [31].

Regarding user engagement of the total sample (N=232), participants executed on average 11.34 (SD 7.87) daily time-out sessions in the active control condition compared to 12.84 (SD 7.41) daily exercise sessions in the intervention condition (Table S2, Multimedia Appendix 1). No between-conditions differences were found (F_1,230=2.24; P=.14; η²=0.01). A high user engagement across those who were retained for analyses on D21 (n=110) was found, with an average completion rate of 93% of the sessions (18.64 out of 20; SD 3.42) in the active control condition and 97% of the exercises (19.40 out of 20; SD 1.55) in the intervention condition. Between-condition differences in postintervention levels of perceived impact revealed that participants reported a higher impact of exercises in the intervention condition (mean 4.88, SD 0.80) as opposed to the active control condition (mean 3.76, SD 1.21; F_1,101=30.53; P<.001; η²=0.23).

In a first step, mean levels of problematic smartphone use (Figure 2) and additional study variables (Figure S3, Multimedia Appendix 1) over time and across both conditions were visualized.

In the 2-level model with problematic smartphone use as the outcome, a negative linear day prediction was found, indicating that problematic smartphone use decreased over time in the active control condition (Table 1; model A: b=–0.04; 95% CI –0.04 to –0.03; intraclass correlation [ICC]=0.21). Not in line with our hypothesis on the primary outcome, the nonsignificant linear day x condition prediction indicated that problematic smartphone use showed a similar decrease over time in the intervention (vs control) condition. Descriptive analyses (Table S2; Multimedia Appendix 1) showed that problematic smartphone use changed from mean 3.60 to mean 2.30 in the active control condition (ie, a decrease of 36%) and from mean 3.65 to mean 2.12 in the intervention condition (ie, a decrease of 42%) throughout the intervention period.

Moreover, daily smartphone unlocks did not change over time (model B: b=–0.13; 95% CI –0.29 to 0.03; ICC=0.65), whereas daily time of smartphone use decreased over time (model C: b=–0.77; 95% CI –1.12 to –0.43; ICC=0.67). In the testing of the equivalence hypothesis, alternative models with the intervention condition variable as an additional moderator did not yield a better fit to the data when compared to null models (model B: Δχ2_3.15, Δdf=4, P=.53; model C: Δχ2_2.14, Δdf=4, P=.71) [33]. In the alternative models, linear day trend x condition interactions were nonsignificant predictors of daily smartphone unlocks (b=–0.15; 95% CI –0.49 to 0.20) and daily time of smartphone use (b=0.36; 95% CI –0.32 to 1.05). This indicates that the intervention condition showed similar patterns of change over time in daily smartphone unlocks and daily time of smartphone use when compared to the active control condition.

Regarding significant predictions of covariates, daily smartphone unlocks were more likely when participants reported higher levels of problematic smartphone use at baseline or when they were younger or male; note that only a small group of men (n=23) participated in the RCT. Moreover, a longer duration of smartphone use was more likely when participants reported higher levels of action control and problematic smartphone use at baseline.

A simple mediation analysis involving experimental conditions, self-efficacy as the mediator at postintervention (D20), and problematic smartphone use as the outcome at follow-up (D42) was run (Figure 3). This mediation model including data from 231 participants fit well with the data (χ²₈=7.06, P=.53; CFI=1.00; RMSEA <0.01; SRMR=0.03). The intervention (vs active control) condition was positively related to changes in self-efficacy at postintervention (b=0.43; SE=0.21; P=.04; 95% CI_bc 0.01-0.85) which, in turn, were negatively linked to changes in follow-up problematic smartphone use (b=–0.21; SE=0.08; P=.01; 95% CI_bc –0.36 to –0.05). The mediation yielded a significant indirect effect of b=–0.09; 95% CI_bc –0.26 to –0.01). Thus, self-efficacy changes at the end of the intervention translated into substantial reductions in problematic smartphone use at 3 weeks following the intervention. Of the variance of self-efficacy and problematic smartphone use, 16% and 27% were accounted for by the joint set of predictors, respectively.

A second mediation model was specified with planning after the Plan module (D16 to D20) and smartphone unlocks at postintervention (D21) as sequential mediators (Figure 4) that fit well with the data (χ²₁₅=19.00; P=.21; CFI=0.98; RMSEA=0.03; SRMR=0.05). No between-condition differences for changes in planning were found (b=0.42; SE=0.25; P=.09). Changes in planning across both conditions were related to unlocks at postintervention (b=–6.22; SE=2.95; P=.04); that is, with each unit of higher planning toward goal-directed smartphone use, the frequency of smartphone unlocks decreased by approximately 6 units per day. Moreover, changes in daily smartphone unlocks across both conditions were significantly related to problematic smartphone use at follow-up (b=0.005; SE=0.002; P=.03). For the sequential mediation between intervention condition and problematic smartphone use via planning and daily smartphone unlocks, the 95% CI included 0.000 (95% CI_bc –0.050 to 0.000), whereas the 90% CI did not include 0 (b=–0.012; 90% CI_bc –0.043 to –0.002). When testing the simple mediation between planning and problematic smartphone use via daily smartphone unlocks, we found a significant indirect effect (b=–0.029; 95%CI_bc –0.078 to –0.003). This indicates that higher levels of planning in both conditions at the end of the intervention phase were associated with lower daily smartphone unlocks which, in turn, were connected with problematic smartphone use. The joint set of predictors explained 11%, 37%, and 27% of the variance in planning, daily smartphone unlocks, and problematic smartphone use, respectively.

An innovative, theory-guided intervention app was field-tested in an RCT by examining a sample of 232 participants up to a 3-week follow-up. The purpose of the app was to make people aware of their (problematic) smartphone use and to support them in establishing self-efficacious and goal-directed smartphone use in their daily life.

The results indicated that the intervention app was useful in lowering problematic smartphone use (primary outcome) as well as time spent with the smartphone. However, observed reductions in both outcomes were not superior to the active control condition (ie, digital detox treatment).

Findings on reductions of problematic smartphone use in both conditions are in line with previous evidence that interventions on psychological resources [18], along with digital detox interventions [37], can be beneficial for health-related outcomes. The findings revealed selective attrition related to higher levels of problematic smartphone use, and thus future interventions targeting problematic forms of smartphone use should add elements so that persons with severe smartphone-related issues receive support matched to their needs. Other strategies to prevent selective attrition in such online-based study designs include scheduling reminders and following up participants with lower study engagement using just-in-time messages or phone calls [38]. Regarding the time of smartphone use, our findings are consistent with evidence from interventions on goal-directed smartphone use [18] and smartphone time-outs [13], thereby testing both forms of interventions concurrently. Although planning a time-out of at least 1 hour, participants in the active control condition reduced their smartphone time by only 43 minutes per day (from 218 to 175 minutes per day; Table S2, Multimedia Appendix 1). Possibly, participants did not fully adhere to their time-out interval or compensated the abstinence from the smartphone by, for instance, catching up with new messages after their time-out [39]. Participants in the intervention condition, who learned and exercised on goal-directed smartphone use, reduced their smartphone time by 32 minutes per day (from 228 to 196 minutes per day; Table S2, Multimedia Appendix 1), an interesting finding given that less smartphone use was not primarily focused on by the intervention app.

To explain the observed changes in problematic smartphone use, intervention mechanisms were systematically examined. According to theories such as the HAPA model [21], self-efficacy toward goal-directed smartphone use should play a role in the process of behavior change. Similar to previous evidence on resource-oriented interventions [18], the present intervention was successfully fostering self-efficacy beliefs which, in turn, are a relevant resource for reductions in problematic smartphone use. The finding that fostering self-efficacious smartphone use is important to improving problematic smartphone use highlights how resource-oriented interventions can even outperform common digital detox interventions. Future interventions targeting problematic forms of smartphone use should follow up on present evidence by enabling persons to control their smartphone use by themselves.

Moreover, precise planning on when, where, and how to use one’s smartphone should also make a difference in behavior change because habitual checking behavior might be reduced [4]. Participants made plans on their smartphone use during the fourth training module (after D13), which was related to postintervention smartphone unlocks; that is, 1 unit of higher planning was linked to a lower daily unlock frequency of 6 units. Planning of smartphone use might result in the planning of smartphone sessions, in which persons take their time to use their phone for current smartphone-related tasks or leisure time activities. This, in turn, might reduce urges towards smartphone-checking behavior and thereby reduce the amount of unlocking of one’s phone. Moreover, the results indicate that daily unlocks at postintervention were linked with problematic smartphone use at follow-up. Next to addressing the time of smartphone use, future research should additionally focus on daily smartphone unlocks—as an indicator of checking behavior—and link smartphone unlocks to clinical-, health-, and work-related outcomes [4,40].

Overall, in terms of practical implications, the findings suggest following up with the currently existing version of the app as a means to change problematic smartphone use by scaling the app.

This study has several strengths. A comprehensive 20-day intervention app was evaluated by contrasting it with a treatment-as-usual control condition with various smartphone-related outcomes and testing intervention mechanisms being examined. However, some limitations need to be considered. First, the substantial reductions in problematic smartphone use were confirmed at a 3-week follow-up, but one cannot be sure whether there would be long-term maintenance of the improved behavior. Second, the primary outcome was a self-report assessment because objective data could not be obtained for this criterion. Third, further smartphone-related outcomes and mechanisms should be examined, such as smartphone-related impulsivity or habitual smartphone use [4]. Fourth, although the intervention condition included comprehensive theory-based content, this condition was not found to be superior in the final evaluation compared to the active control condition. Active control conditions benefit from the attention they received from the researchers or the software, and one can assume that volunteering participants come along with a high level of curiosity and motivation to succeed. However, a passive control condition was intentionally missing in this research design. It should be noted that the effects of a digital detox intervention, as opposed to a passive control condition, were examined by prior research [13]. Fifth, regarding sample characteristics, reactive recruitment procedures (eg, online postings and email lists) resulted in high participation rates of women and younger persons. Thus, associations of study variables with gender and age should be interpreted with caution. Moreover, the distribution of gender and age does not allow for inferences regarding the general population to be made. Further studies need to find representative samples for defined populations, for instance, by using proactive recruitment strategies. Finally, high attrition rates were observed, which is a general issue in online-based research [36]. Future studies could add just-in-time procedures matched to smartphone users’ needs to maintain user engagement [38].

An innovative, theory-based intervention app was successfully evaluated as being capable of changing problematic smartphone use. The app was found to be useful for lowering problematic smartphone use and daily time spent with the device. However, observed reductions in both outcomes were not superior to those in the active control condition (ie, digital detox treatment). As an intervention mechanism, the intervention condition developed increased self-efficacy toward goal-directed smartphone use, which was linked to a reduction in problematic smartphone use. Further, planning of smartphone use at the end of the intervention phase was connected with a lower frequency of daily smartphone unlocks, which, in turn, was related to less problematic smartphone use. Further research could build on theories of behavior change and identify more psychological intervention content to improve these types of intervention apps. The app in its current form appears suitable for practical use as an alternative to common digital detox apps.