Between 25% and 50% of college students meet the criteria for at least one mental health disorder in a given year [1,2]. Data suggest that mental disorders account for about half the disease burden of young adults in developed countries [3] and are associated with a range of negative consequences, including lowered academic performance [4] and college attrition [5].

Despite the availability of effective treatment [6], only 1 in 5 students with mental disorders receives minimally adequate treatment [1]. Reasons for this treatment gap include attitudinal barriers such as stigma and a preference for self-help [7].

Internet- and mobile-based interventions [8] might help to increase the utilization of psychological interventions, as they can be easily accessed, allow for high scalability, and can be provided at a low cost [9,10]. Internet-based interventions may also be suitable for college student populations [11], with research indicating that preference for help-seeking through the internet is higher among younger and well-educated individuals [12]. There is meta-analytic evidence suggesting the efficacy of internet interventions for a range of conditions and populations [6,13-16], including college students [17], with effect sizes of technology-delivered interventions ranging from standardized mean difference (SMD) of 0.42 to 0.43 for depression, 0.30 to 0.56 for anxiety, and 0.73 to 0.82 for stress [17,18]. However, the few results for internet and mobile-based interventions targeting stress in students are conflicting in terms of their effectiveness [19,20], warranting further research.

A recent meta-analysis suggests that intervention effects are considerably higher in indicated compared with general student populations [18], stressing the importance of developing suitable intervention approaches for at-risk students. Internet-based interventions which are labeled to improve stress coping skills, as opposed to focusing on reducing symptoms of mental disorders, could represent a promising way to reach such burdened individuals. In an Australian investigation among severely distressed college students, 55.7% indicated that they were quite or very likely to use an internet-delivered program to seek help [21]. A significant association between heightened stress levels and positive attitudes toward internet intervention usage has also been found in a German general population sample [22].

If proven to be effective, internet- and mobile-based approaches could provide a feasible instrument to help avert the onset of more severe stress-related mental health concerns in at-risk college students [8]. More research is therefore required to corroborate results on the effectiveness of internet- and mobile-based stress interventions and assess the potential of such interventions to reach and be effective in burdened students who already show symptoms of mental illness such as depression. Facing the deleterious effects of poor mental health on academic functioning, it is also important to assess whether such interventions may have an impact on important college-related outcomes such as academic self-efficacy and impairment [17].

The aim of this study is thus to evaluate the effectiveness of an internet- and mobile-based intervention targeting university students with heightened stress levels. We hypothesized the internet intervention to be more effective in reducing symptoms of stress compared with a waitlist control group (WCG). It was furthermore assumed that more students participating in the intervention compared with the WCG would show a reliable change in perceived stress outcomes and attain close to symptom-free status. The second objective of this study was to investigate the hypothesized positive effect of the intervention on further mental health outcomes, modifiable risk and protective factors, and college-related outcomes compared with the WCG. Finally, our aim was to explore intervention participants’ adherence to, and acceptance of, the intervention.

A 2-armed randomized controlled trial was conducted with 150 participants, comparing an internet and app-based intervention with feedback on demand (StudiCare Stress) to a waitlist control group (WCG). Both conditions had full access to treatment as usual (TAU). The sample size allowed to detect effect sizes of d=0.41 with a power (1− β) of 0.80 with alpha of .05 and was based on a meta-analysis on internet-based interventions for college students, which reported an SMD of 0.73 for stress but lower effects for depression outcomes (SMD=0.43) [17]. A sample size of 150 was therefore chosen to also detect significant changes for secondary outcomes in this study such as depression.

Assessments took place at baseline (T1), posttreatment (T2; 7 weeks), and 3 months after baseline (T3; see Figure 1). Self-report data were collected using a Web-based assessment tool (Advanced Encryption Standard, 256-bit encryption). All procedures involved in the study were consistent with the generally accepted standards of ethical practice. The study was approved by the University of Erlangen-Nuremberg ethics committee (Erlangen, Germany; 322_15 B). The trial is registered in the German Clinical Trials Register (DRKS00010212 [24]). This study describes the main effectiveness analysis of the intervention; we also assessed moderator and mediator variables, which are listed in the trial registration (see Multimedia Appendix 1) and will be analyzed and reported in due length elsewhere.

Inclusion criteria were (1) elevated levels of perceived stress (Perceived Stress Scale 4-item version, PSS-4≥8 [25]; representing a level of stress one SD=2.92 above the mean of 4.49 in a large student sample [25]), (2) enrollment in a German-speaking university at the beginning of the training, (3) age ≥18 years, (4) internet access, (5) willingness to provide self-report data at all assessment points, and (6) informed consent. Exclusion criteria were (1) self-reported diagnosis of dissociative symptoms or psychosis in the past or (2) considerable risk for suicide (Beck Depression Inventory item 9 >1; “I feel I would be better off dead” or “I would kill myself if I had the chance”). Individuals showing an elevated risk for suicide were given detailed information about treatment options and were asked to see a physician or psychiatrist as soon as possible.

Participants were recruited via university press reports, student counseling services, and social media platforms. Potential participants declared interest in partaking in the study by filling out a Web-based registration form on the study website.

Individuals who declared interest in participating received an information letter along with an informed consent sheet and were asked to provide an email address for their intervention platform profile. Applicants were informed that withdrawal from the study was possible at any time, did not go along with any negative consequences, and all collected case data could be deleted on request during the study. Interested participants were asked to complete the written informed consent form and fill out the Web-based screening questionnaire.

Individuals meeting all the inclusion and none of the exclusion criteria were invited to fill out the baseline assessment. After completion, individuals were randomly allocated to either the IG or the WCG. Randomization took place at a ratio of 1:1 and a block size of 2 using an automated computer-based random integer generator (Randlist, Datinf GmbH, Tübingen, Germany) and was performed by a researcher not otherwise involved in the study. Participants could not be blinded to study conditions; yet, during the randomization process, the allocation was concealed from participants, researchers involved in recruitment, and e-coaches.

The primary outcome was perceived stress as measured by the PSS-4 [25]. The PSS-4 assesses the degree to which individuals evaluate their lives as stressful, especially regarding how uncontrollable and overloading relevant aspects of life are perceived. The PSS-4 comprises 4 items (Item 1: “How often have you felt you were unable to control the important things in your life?”; Item 2: “How often have you felt confident about your ability to handle your personal problems?”; Item 3: “How often have you felt that things were going your way?”; Item 4: “How often have you felt difficulties were piling up so high that you could not overcome them?”), yielding a score between 0 and 16. Participants rated their level of perceived stress within the last 2 weeks on a 5-point Likert scale (0= never; 4= very often). A two-factor structure has been commonly found for the PSS [32-35], with positively framed items representing perceived coping self-efficacy and negative items reflecting hopelessness, the latter being a strong predictor for depression [36]. Higher scores on the PSS have shown to have good predictive validity for several adverse health outcomes [36-38]. Despite its brevity, the PSS-4 has been found to have acceptable to good psychometric properties [39,40]. The scale has a good level of internal consistency in this study as indicated by a Cronbach alpha of .83.

Unless otherwise specified, all outcomes were measured for a retrospective time frame of 2 weeks. All measures were administered in German. When no German translation was available, scales were translated independently by two of the researchers (MH and SHA), who then compared and discussed the translations to resolve disagreement.

Descriptive statistics were used for process evaluation. To assess overall user satisfaction across various domains, item data provided by the CSQ-8 was examined individually. Acceptance of intervention modules was analyzed using the module feedback of the IG. Adherence was assessed by analyzing intervention completion rates tracked within the intervention platform. Finally, we analyzed the proportion of participants who accessed the diary app and requested automated short messages via SMS.

The study flow can be found in Figure 1. Participants who were lost to follow-up at T2, T3, or both assessments did not differ significantly from participants who adhered to the protocol on any baseline characteristic (all P>.05). Table 1 summarizes detailed baseline characteristics of study participants. The majority (76.8%, 106/138) of the participants indicated that they had not consulted a physician, psychotherapist, or counselor for their health-related problems and may thus be considered first-time help-seekers. Descriptive data including all 3 assessment points for all outcomes is depicted in Table 2. Both study arms did not differ significantly (all P>.05) on any characteristic at baseline.

Table 3 summarizes the results of the ITT analyses for the secondary outcomes. ANCOVAs revealed significant effects (P<.05) in favor of the IG for the majority of outcomes at both assessment points, with effect sizes ranging from d=0.33 (95% CI 0.01-0.65) for productivity (T2) to d=0.82 (95% CI 0.49-1.15) for emotional exhaustion (T2). No statistically significant effect was found for perfectionism (F_1,147=0.38, P=.53) at T2, but at T3 (P<.001). Resilience (T2: F_1,147=1.69, P=.17; T3: F_1,147=2.94, P=.08), self-compassion (T2: F_1,147=2.97, P=.09; T3: F_1,147=1.46, P=.23), and self-esteem (T2: F_1,147=0.15, P=.70; T3: F_1,147=1.36, P=.25) did not differ significantly between both study arms at both assessment points.

More than three-fourths of the participants (77.3%; IG: 58/75; WCG: 58/75) showed symptoms above the cut-off for clinically relevant symptoms of depression at baseline. Between-group effect sizes for depression in this subgroup were moderate to large, both for T2 (d=0.67, 95% CI 0.34-1.00) and T3 (d=0.73, 95% CI 0.40-1.06). Treatment response was achieved by 36 (62%; T2) and 33 of 58 participants (57%, T3) in the IG compared with 14 of 58 participants (24%; T2 and T3) in the WCG, resulting in an NNT to achieve one additional treatment response in the IG compared with the WCG of 2.64 (95% CI 1.83-4.70) for T2 (χ²₁=17.0, P<.001) and 3.05 for T3 (95% CI 2.02-6.28, χ²₁=12.9, P<.001).

The results of the completer analyses were similar to the ITT analyses, with moderate to large between-group effect sizes for the primary outcome at T2 (IG: mean=6.72, SD 2.86; WCG: mean=9.32, SD 3.16; F_1,88=18.60, P<.001; d=0.85, 95% CI 0.44-1.27) and T3 (IG: mean=6.41, SD 2.84; WCG: mean=8.65, SD 3.43; F_1,88=13.41, P<.001; d=0.69, 95% CI 0.29-1.10). In contrast to the main analysis, however, resilience had increased significantly in the IG compared with the WCG at T2 (F_1,88=8.56, P=.004; d=0.46, 95% CI 0.06-0.86).

Results of this study indicate moderate to large intergroup effects for the reduction of perceived stress and other relevant health- and college-related outcomes, as well as substantial effects in individuals with clinically relevant symptoms of depression, which were highly prevalent in our recruited sample. No significant effects were found for self-compassion, perfectionism (T2), resilience, and self-esteem.

The benefits of this intervention were larger than those found in previous trials evaluating internet-based stress interventions in college students [19,20], albeit somewhat smaller than the reported overall effect of technology-delivered skill training interventions on perceived stress [18], and comparable to internet-based stress interventions in general, as reported in a recent meta-analysis, with a pooled standardized mean difference of d=0.64 (95% CI 0.50-0.79; perceived stress) in guided internet- and mobile-based interventions [66]. The study further contributes to current literature by showing that targeting perceived stress in students does not only result in better mental health–related outcomes and well-being but can also have a substantial beneficial impact on college-related outcomes which, to the best of our knowledge, have not been investigated so far.

Participants’ adherence to the intervention was satisfying, and the intervention was well accepted among the large majority of students. Participant feedback on the length of specific modules, however, suggests that participants may have spent more time than anticipated on some of the modules. This may indicate that some modules could be shortened to attempt to further improve adherence. Whether shortening might, in fact, result in higher adherence to the intervention, however, is not fully clear. Earlier research has reported higher adherence rates for shorter interventions, albeit focusing on the number of modules used [67]. Whether this effect also holds true for the length of specific modules remains unknown. Shortening some of the modules might potentially optimize adherence but may also compromise the intervention’s overall efficacy due to less potentially beneficial information or techniques being conveyed and trained. It has been argued that various ways in which participants prioritize provided content may lead to positive outcomes, and the ability to progress through interventions at one’s own pace might represent a key asset of internet-delivered treatment [68]. Qualitative interviews conducted with participants of this intervention suggest that the elective mini-modules for various student-relevant topics were very well accepted [69]. Providing larger amounts of content in a flexible way, allowing participants to tailor the intervention to their specific needs, could, therefore, be a promising approach to optimize intervention usage patterns [70]. However, research is warranted to test whether this might further increase adherence.

This study has some limitations. First, women were overrepresented in the study sample, as frequently seen in preventive interventions. Second, study dropout in the IG at 3-month follow-up was relatively high and larger in the IG. Albeit being a common limitation in clinical trial research [71], and with differential dropout rates having been reported before for internet-based stress intervention trials in tertiary education students [72,73], this restricts the generalizability of our findings on long-term effects. Attrition analysis, however, did not result in any significant baseline differences between dropout and nondropout cases, which may be an indicator that results were not overly biased due to unequal dropout [74]. Third, because of ethical reasons, participants had full access to treatment-as-usual. Thus, we cannot rule out potential cointervention effects due to utilization of health services. Finally, because of feasibility and ethical reasons, such as not denying one half of the sample access to the intervention they sought after, participants in the IG of this study were compared with a WCG to assess effects of the training. The influence of treatment and change expectancies have been discussed as an artifact in clinical evaluation trials using WCGs because they potentially discourage participants with delayed access to treatment to initiate health-related behavior changes, and thus lead to accentuate effects [75].

Most college students with depression do not seek treatment through conventional health care channels [76], and attitudinal barriers, such as fear of stigmatization, have been shown to have a large impact on treatment utilization [77]. As our findings indicate that (1) a large number of students in this sample did not use conventional treatment options before, (2) the majority of students who were willing to use this intervention reported clinically significant symptoms of depression, and (3) among this group, the treatment response was favorable; future studies should explore whether internet-delivered stress interventions, labeled as providing “support for coping with academic stress,” might potentially attract students with symptoms of depression who would not use formal mental health treatment and whether they can reduce the incidence of depressive disorders [8]. Future studies should thus investigate the utility of internet and mobile-based interventions in affected students, that is, with symptoms of major depression, as an indicated preventative or early intervention approach to narrow the treatment gap and improve academic functioning [4].

In conclusion, internet- and mobile-based interventions could be an acceptable, effective, and potentially cost-effective approach to reduce the negative consequences associated with college-related stress.