Psychological stress occurs when one’s perceived demands exceed one’s perceived capacity to cope [1,2]. Excessive and prolonged stress can have multiple negative health effects, including heightened risk of obesity [3], impaired working memory [4,5], and cardiac arrest [6]. Adverse stress effects on mental health are also common, including an increased risk of depression and anxiety [7], insomnia [8], and psychosis [9]. Excessive stress is widely recognized as a major health burden [10]. Young adults and late adolescents are at particular risk of experiencing heightened stress levels owing to life transitions from childhood to adulthood and taking on new responsibilities [11,12] or uncertainties relating to career prospects [13]. Equipping young adults with effective stress-coping skills is a viable attempt to lower the public burden of excessive stress [9]. Digital tools such as mobile health (mHealth) apps are likely going to play an increasingly important role in public mental health [14].

Self-guided interventions (SGIs) on smartphones are easily accessible, affordable, flexible, and convenient to use [15]. SGIs can also promote well-being and prevent illness by equipping individuals with the skills to address minor stressors before they evolve into larger problems [16]. Various randomized controlled trials (RCTs) have examined the efficacy of SGIs on mental health outcomes [17], and meta-analyses have confirmed that SGIs can improve symptoms of anxiety and depression across age groups [18,19]. A recent meta-analysis by Linardon et al [20] specifically found that SGIs were efficacious in reducing stress levels. However, the effect sizes for most interventions on stress (Hedges g=0.35), anxiety, and depression were rather small (Hedges g=0.19-0.21) and inconsistent across studies [18,19]. Although many SGIs are freely available on smartphones, only a few are supported by theoretical and empirical data [21]. In a review of 62 stress management SGIs [22], a quarter of them did not use any evidence-based stress reduction strategies. These results correspond to a separate review of 60 stress-coping SGIs whereby a third of them either failed to provide any empirical strategies or delivered a different strategy than that stated in its description [23]. In both reviews, most of the remaining SGIs focused on momentary breathing and mindfulness exercises, which produced short-term modest effects on stress and anxiety levels. Only a minority of SGIs involved evidence-based interventions such as cognitive behavioral therapy (CBT) [21,23]. Indeed, the meta-analysis by Linardon et al [20] found substantially larger effect sizes for SGIs that involved CBT-based practices compared with waitlist control group. These CBT-based effects on stress, anxiety, and mood were also superior to alternative interventions and could be strengthened when combined with brief professional guidance within the mHealth app. The same authors also concluded that research on the overall efficacy and dissemination of evidence-based SGIs is in its infancy because of the small number of studies, especially those involving younger adults (ie, 5 studies). In addition, the meta-analysis by Linardon et al [20] highlighted methodological problems with RCTs involving stress-coping SGIs, such as high risk of bias and lack of follow-up data.

Furthermore, very little is known about who benefits the most from stress-coping SGIs [24]. Previous authors were unable to analyze any moderators of web- and computer-based stress interventions owing to the “lack and inconsistency of information provided by the [reviewed] studies” [25]. To our knowledge, only Coudray et al [26] evaluated moderators in a sample of 920 college students and found that those with lower perceived present control (ie, aspects of stressors perceived to be controllable in the present) experienced significantly greater reduction in stress after 1 to 3 weeks of web-based stress management interventions compared with those with higher perceived present control. The authors concluded their findings by encouraging similar research on moderators using diverse samples. A potential moderator that may influence the efficacy of CBT-based SGIs on mental health outcomes is coping self-efficacy (CSE). CSE refers to one’s confidence in coping strategies [27], particularly in times of hardship [28,29] and threat [30]. Slightly different from perceived control, which refers to one’s perception of the availability of any effective response, CSE describes one’s confidence in their ability to actually effect that response [31]. Some RCTs have found CSE to be a mediator of the relationship between CBT-based SGIs and improved stress, anxiety, and depression outcomes [32-34], but researchers are less clear about CSE as a moderator. For instance, individuals with higher perceived confidence in coping with stressors were expected to maximize their gains from CBT-based treatments [35], or similar to the findings on perceived present control, it is also possible for individuals who perceive that they need more help with coping have higher motivation to benefit from using a CBT-based SGI, relative to those with higher CSE. Another potential moderator of the relationship between CBT-based SGI and outcomes is psychological mindedness (PM). PM refers to one’s predisposition to be aware of, assess, and reflect on one’s mental states and behaviors cognitively and emotionally [36,37]. Self-reflective practices within CBT were involved in improving depression symptomatology in depressive disorders [38], anxiety disorders [39], and social phobia [40], as it is thought to enhance individuals’ abilities and frequency in self-monitoring and self-evaluating their cognitions, emotions, and behaviors. Despite PM being central to the successful practice of CBT [41], very little is known about the differences in an individual’s capacity for self-reflection and insight that could moderate the efficacy of the modality [36,42]. Wiles et al [43], for example, did not find PM to be a treatment moderator in CBT, although they acknowledged that their small sample size was inadequate for testing interaction effects. Some studies have recommended higher initial levels of self-reflection and insight that may predict faster responses to CBT as potential avenues for future research [36,42]. These authors speculated that metacognitive processes such as higher PM might predispose an individual’s capacity to appreciate and understand the cause-and-effect concepts of CBT-based techniques. As CBT principles are often adopted by evidence-based SGIs, it would be interesting to test how initial levels of PM can influence CBT-based therapy outcomes, as it would allow personalizing CBT-based SGI and optimizing treatment outcomes.

Using a randomized controlled design, we examined the possible efficacy of a recently developed CBT-based stress-coping SGI in the mHealth app (Intellect) on a sample of Singaporean university students compared with an active control group. We first hypothesized that participants in the intervention group (stress SGI) would report lower perceived stress than the active control group after the intervention. Although the effects of self-guided CBT interventions for stress coping are typically small, we predicted possible gains for the intervention group to be maintained at 1-month follow-up, in line with related RCTs that also found small effects present at follow-up assessments [44,45].

Second, we predicted that participants in the intervention group would report lower anxiety and depressive symptoms than the active control group after the intervention. Third, we hypothesized that users with higher levels of PM and CSE would experience the greatest reduction in perceived stress levels. As studies have shown that individuals who benefited most from stress interventions typically also experienced greater improvement in anxiety and depressive symptoms [46], we also predicted that this moderation effect by CSE and PM also applies to secondary outcomes.

This study was an RCT with two groups: (1) an 8-day stress-coping self-guided program (intervention group) and (2) an 8-day cooperation self-guided program (active control group). A 2 × 3 mixed factorial experimental design was used with condition (intervention vs active control) as a between-group factor and time of assessment (baseline vs postintervention vs 1-month follow-up) as a within-group factor. The primary dependent variable measured was perceived stress levels. The secondary dependent variables were depression and anxiety levels. The CSE and Psychological Mindedness Scale (PMS) were also used for independent moderation analyses.

A total of 321 participants were recruited through the Psychology Department’s Research Participation Programme and the university’s research recruitment platform. Recruitment posters comprising the study procedures, inclusion criteria, and reimbursement for participation were distributed on the web and among the faculty. Of the 321 participants, 46 (14.3%) participants were excluded owing to withdrawal and failing data quality checks, whereas 3 (0.9%) participants were lost to follow-up (Figure 1). Of the 321 participants, the final sample of 264 (82.2%) participants was predominantly female (188/264, 71.2%), with a mean age of 22.45 (SD 5.41; range 18-59) years. All the participants were undergraduate students who were able to read and understand English. Elevated perceived stress was not a requirement for inclusion. Participants also received either course credits or monetary reimbursement for their participation.

Most studies that investigated app-based stress-coping programs found small to moderate effect sizes for the primary outcome measures of stress, mood, and anxiety [47]. Subjecting a small effect size of Cohen f=0.1, an α level of .05, and a power of 0.9 to G*Power revealed a minimum sample size of 214. Web-based studies are prone to attrition, and in line with similar studies, we expected 30% to 50% attrition rate [47,48] and considered that 10% of the questionnaire data may be inconsistent or invalid as commonly found in web-based survey studies [49]. Thus, we aimed to recruit 321 participants.

Participants signed up for the study on the university’s recruitment sites via a link that directed them to the web-based survey platform hosted on Qualtrics (SAP America Inc). Upon consenting to the web-based Participants Information Sheet, participants completed the primary outcome measure on perceived stress (Psychological Stress Measure [PSM]; PSM-9), secondary outcome measures on mental health (Patient Health Questionnaire [PHQ]; PHQ-9 and Generalized Anxiety Disorder [GAD]; GAD-7), CSE (Coping Self-Efficacy Scale [CSES]), PMS, and demographic information.

Next, participants were randomly assigned to either the intervention or active control condition using simple randomization procedures through the Qualtrics software. They were not informed about the conditions they were allocated to and the real nature of the study. Instead, they were informed that the study would examine the efficacy of SGIs in promoting well-being. They were then guided to download the mHealth app Intellect on their personal smartphones from the Apple App Store (Apple Inc) or Google Play Store (Google LLC). According to their assigned conditions, a number code was provided to them to unlock the app. Participants in the intervention condition participated in the 8-day stress-coping program, whereas those in the active control condition took part in the 8-day cooperation learning program. Both programs involved fulfilling a series of tasks aimed at improving well-being. This included content education and short daily activities, averaging 5 minutes. To promote adherence to the study, standardized daily reminders to complete the program were sent via SMS text messages by the researcher to participants. Participants were only allowed to proceed onto the next page of the app after they had completed the preceding exercises. It was expected that every participant would complete all the self-guided daily activities. All the participants were instructed to refrain from using any other SGIs that affect well-being other than the given SGI throughout the entire duration of the study, lasting from the beginning until the end of the 1-month follow-up. This minimized the potential confounding effects.

Upon completion of the 8-day program through technical verification, participants received a survey link to complete the outcome measures and the App Engagement Scale (AES). A month after the completion of the SGI, participants were provided with a survey link to complete the outcome measures. Reimbursement was given upon completion of the postintervention measures with either 3 course credits or Singapore $10 (US $7.5) and after 1-month follow-up measures with an additional 1 course credit or Singapore $5 (US $3.75). University students intending to major in psychology were required to collect a minimum of 28 course credits over the course of an academic year. After the follow-up assessment, participants in the active control group were also given access to the stress-coping SGI.

Ethics approval to conduct the study was obtained from the Institutional Review Board of the National University of Singapore (NUS-IRB-2021-85). The study was preregistered with ClinicalTrials.gov (NCT04978896). All the participants provided electronic consent before participating in the study. Data collection took place in an entirely web-based setting, and all user data were deidentified before any analyses.

The descriptive statistics for both groups are presented in Table 3. Independent 2-tailed t tests and chi-square tests did not find any significant differences in age, sex, and baseline scores for PSM-9, PHQ-9, GAD-7, CSES, and PMS (all P>.05) between groups. Participants in both conditions did not show any significant differences in their degree of engagement with their SGIs (P>.05) as measured with the AES. The mean scores for the whole sample’s characteristics were as follows (PSM-9: mean 37.78, SD 9.58; PHQ-9: mean 7.07, SD 5.27; GAD-7: mean 6.60, SD 4.57; CSES: mean 142.9, SD 42.2; and PMS: mean 126.2, SD 10.5), indicating that the average student was mildly distressed, depressed, and anxious.

The mean values of the outcome measures are listed in Table 4. The intervention group reported significantly lower perceived stress levels at postintervention and follow-up periods compared with the active control group. The small effect sizes were between 0.015 and 0.019. This indicated that the stress-coping SGI was moderately more effective than the cooperation SGI (control) in reducing perceived stress over time. The intervention group also reported significantly lower anxiety levels than the active control group at postintervention period but not at follow-up. No significant differences were observed between the intervention and active control groups for depressive symptoms at the 2 time points.

For the ANCOVA results (controlling for all baseline measurements) at postintervention period, the time × group interaction exhibited significant differences for PSM-9 (F_1,261=7.16; P=.008) and GAD-7 (F_1,261=5.86; P=.02) but not for PHQ-9 (F_1,261=1.65; P=.20). Only the interaction effect for PSM-9 (F_1,261=6.08; P=.01) remained significant at follow-up (GAD-7: F_1,261=.796; P=.37; and PHQ-9: F_1,261=3.24; P=.07). Therefore, the main effects on psychological stress scores indicated a significant intervention effect over time.

At the postintervention period, the results revealed a significant effect in the overall moderation model with PSM-9 as the dependent variable and CSES and PMS as moderators (R²=0.39; F_9,253=18.3; P<.001). Both interaction terms were significant at postintervention period (CSES × condition: b=−0.0655, SE=0.0231; t₂₆₃=−2.83; P=.005; 95% Cl −0.111 to −0.020 and PMS × condition: b=0.181, SE=0.0917; t₂₆₃=1.97; P=.05; 95% Cl 0.0002-0.362), indicating that these specific conditions must be met for the intervention condition to predict lower psychological stress [66]. The stress-coping SGI significantly predicted lower scores on the PSM-9 at postintervention period in participants who (1) experienced lower CSE and (2) had moderate to high baseline PM (low CSE and moderate PM: b=4.56, SE=1.30; t₂₆₃=3.52; P<.001; 95% Cl 2.01-7.11 and low CSE and high PM: b=6.37, SE=1.72; t₂₆₃=3.71; P<.001; 95% Cl 2.98-9.75). Similarly, the intervention condition significantly predicted lower psychological stress at postintervention period for participants with (1) moderate CSE and (2) moderate to high baseline PM (moderate CSE and moderate PM: b=1.91, SE=0.920; t₂₆₃=2.07; P=.04; 95% Cl 0.0953-3.72 and moderate CSE and high PM: b=3.72, SE=1.27; t₂₆₃=2.92; P=.004; 95% Cl 1.21-6.22). The results are presented in Table 5. Finally, regardless of the level of PMS, participants who began the stress-coping SGI with high SCE did not experience significantly lower psychological stress at postintervention period. These moderation effects were not observed at follow-up, even as the stress-coping SGI group reported significantly lower perceived stress levels at postintervention period (CSE × condition: b=−0.0415, SE=0.0241; t₂₆₃=−1.72; P=.09; 95% Cl −0.0891 to 0.0060 and PMS × condition: b=−0.0071, SE=0.0958; t₂₆₃=−.0740; P=.94; 95% Cl −0.196 to 0.182). A visualization of these interactions is shown in Figure 3.

Moderation analyses for the secondary measures were not significant. For postintervention and follow-up periods, the interaction terms for both CSE and PM were not significant for both anxiety and depressive outcomes. This means that even as participants in the stress-coping intervention group experienced significantly lower anxiety levels than active control participants at postintervention period, no subgroup of individuals benefited more than the others. Similarly, as there were no significant main effects on anxiety and depressive symptoms at follow-up; there was no moderation effect.

The randomized controlled design allowed for causal conclusions [95]. The active waitlist control group also controlled for the effects of attention and other nonspecific factors from the intervention effects, thus strengthening the validity of the results [96]. Finally, significant effects on perceived stress levels at the 1-month follow-up period showed that the stress-coping SGI successfully improved university students’ well-being over a short period. These directly addressed the lack of follow-up measurements as stated in the most recent meta-analyses of mental health SGIs [20,25].

This study had several limitations. First, the actual amount of effort put in to complete the SGI was not controlled for. Data on the duration that participants spent thinking about or practicing the skills learned from the SGI beyond the intervention were not collected. It is plausible that the positive effects could also be explained by other variables that were not assessed during the intervention, such as better campus life or a less stressful curriculum. Second, the actual duration spent on the Intellect app was not collected or controlled for. This is a potentially serious concern, as low adherence and inconsistent engagement with SGIs were previously found to limit their positive effects [97]. However, efforts were made to ensure that the participants completed the app activities properly. Initially, the participants were instructed to complete every daily activity present in their respective SGIs during the intervention period. Intellect SGIs also did not allow the participants to proceed to the next page until they completed the preceding activities. We encouraged maximum adherence through daily text reminders. Importantly, every final participant was technically verified as having completed the activities in their SGI. We acknowledged that it is possible that the participants in the active control group completed the activities in less time and with less care than those in the intervention group, which would threaten the internal validity of our findings. However, both groups did not differ on the subjective AES, which gives some confidence that groups did not differ in their motivation to engage with the SGIs and the time they spent on them [98]. Nonetheless, future studies should control for the time spent on the SGI and evaluate its relationship with SGI outcomes. Third, subjective self-reports are prone to retrospective recall biases [99]. This limits the accuracy of our findings on perceived stress, anxiety, and depressive symptoms given that they vary from time to time. Therefore, our results should be interpreted with caution. Given that the PSM-9 asked for self-reported psychological stress over the last 4 to 5 days, we are less able to infer whether the stress-coping SGI could ecologically reduce perceived stress among university students without the implementation of momentary assessments. To further strengthen the reliability of these findings, future studies can use daily diary methods to evaluate the real-time efficacy of mobile interventions on university students’ well-being [100]. In place of self-reports, recent studies have recommended behavioral or physiological measures to objectively assess stress symptoms [101]. This method would also reduce the risk of recall bias and enhance the validity of our findings. Fourth, the study design only compared the intervention group with an active control group. The absence of a waitlist control condition prevented the exclusion of the possibility of third variables contributing to the findings, which can be addressed in future studies by including a waitlist control alongside an active control. Fifth, the absence of long-term follow-up assessments (eg, 3 or 6 months) makes it difficult to deduce whether improvements can be sustained over time. Therefore, our study could not evaluate the possible long-term benefits on symptoms of chronic stress. Finally, this study has limited external validity for the student population at large, as our sample contained mainly female students from Singapore with slightly elevated stress, anxiety, and depression levels. Therefore, our positive findings on perceived stress and anxiety symptoms may be generalized to the Western student population or clinical samples. Future researchers may consider replicating our findings with a more diverse college sample while administering even longer-term follow-up assessments.

In conclusion, this RCT found evidence for an 8-day stress-coping SGI in improving perceived stress and anxiety levels among Asian university students. The effects on perceived stress levels were sustained at 1-month follow-up, but not for anxiety, thus giving some confidence that a brief, time-limited, and CBT-based SGI can maintain its gains on perceived stress. However, depressive symptoms did not decrease. Students with lower CSE and higher PM experienced reduced stress faster than other students in the CBT-based SGI intervention group. The identification of these moderators can optimize the outcome and treatment delivery of such stress-coping mobile apps. Our findings are useful, given the potential for scaling up such easily accessible, and brief interventions. Several limitations were noted, including the lack of ecological momentary assessments to capture perceived stress more accurately, longer-term follow-up measures to evaluate sustainability of gains, and more diverse samples to evaluate transferability of the findings.