Using a 2-arm randomized controlled design, 264 participants were randomly allocated (at a ratio of 1:1 and a block size of 2) to the iSMI or to a WLC group. Both groups had full access to treatment as usual.

Participants 18 years and older were included if they were currently employed and scored 22 or above on the PSS-10. Due to the fact that the PSS is not a diagnostic measurement and there is no official cut-off available, we decided to use one standard deviation (SD 6.2) above the mean (PSS-10=15.3) in a large working population [33] as a cut-off value to choose participants with an elevated level of stress. We excluded any applicants who were at risk of suicide (Beck Suicide Item >1; [34]) or self-reported to have been previously diagnosed with dissociative or psychotic symptoms. Participants were recruited in Germany from January to October 2013 in the general working population through newspaper articles and announcements by the ministry of education. Primarily, they were recruited through a large German health insurance company. The intervention addressed employees who were frequently stressed or exhausted, who felt that problems were increasingly difficult to handle, and who struggled to cope with difficult emotions. The intervention was advertised in the print membership magazine of a large German health insurance company. It was stated that, in cooperation with a university, the health insurance company offered this online training to employees, provided they were suitable for study inclusion. Moreover, the advantages of the intervention such as 24/7 availability, personal e-coaching, and participation free of charge were delineated. Those interested in participating had to provide an email address and a first and last name that could be pseudonyms if desired. Individuals received a link to the online screening questionnaire via email. Provided they were eligible, applicants had to submit their signed informed consent via regular post or scanned via email. Upon receipt of the informed consent, participants had to complete all baseline questionnaires. Subsequently, they were randomized into either the intervention or the waitlist control group.

The iSMI GET.ON Stress is based on Lazarus’s transactional model of stress [26]. This intervention applied both well-established problem solving and more recently developed emotion regulation strategies. Important principles for health behavior change such as goal setting, action planning, and coping planning were followed. The iSMI consisted of seven sessions and a booster session provided 4 weeks after training completion. Following psycho-education (Session 1), the participants learned a 6-step procedure to systematically solve problems (Sessions 2-3). In Sessions 4-6, the participants were introduced to emotion regulation techniques (muscle- and breathing relaxation, acceptance of negative emotions, and self-support in difficult situations). Session 7 included a plan for the future. The iSMI was specifically tailored to employees; this was reflected in the wording of the intervention, the example characters provided throughout the training, as well as in optional short informational material related to typical stress-related topics (eg, psychological detachment from work, time management, sleep hygiene, worrying, and organization of breaks during work) provided alongside the intervention. The application of exercises was strongly recommended. The participants were advised to complete 1-2 sessions per week. The program included interactive exercises, audio/video files, and downloadable material and was presented on a secured Web-based platform. Upon activation of the account through the research team, participants used their email address and a self-chosen password to log on. Within 48 hours after session completion, an e-coach provided approximately three quarters of a page of written, non-therapeutic feedback intended to increase adherence and motivation. The e-coaches reported that the average time spent per feedback was 30 minutes. In the event of non-completion of a session, they also sent reminders. Each e-coach had a degree in psychology and followed a standardized manual on feedback writing. Fidelity and adherence to the feedback manual was ensured by providing extensive coaching on feedback writing and by employing a psychotherapist who provided supervision for the e-coaches. The participants could receive automatic text messages on their mobile phone along with the iSMI (eg, short relaxation exercises: “Relax your muscles in your hands and arms for 3 seconds now. Follow your breathing and each time you breathe out, relax a little more.”) and were given the choice of either light (1 text message every other day) or intensive support (2-3 text messages per day) according to personal preferences. The text message coach was part of the intervention, aimed at reminding participants to practice and increasing the adherence to the intervention [9]. A more detailed description of the iSMI can be found in the protocol of the trial [35]. Screenshots of the intervention are available in the Multimedia Appendix 1.

All questionnaires were self-assessed online at baseline (T1), 7 weeks (T2, post-treatment), 6 months (T3), and 12 months (T4, iSMI group only) after randomization. The WLC group received access to the intervention following T3.

The primary outcome was the level of perceived stress as measured by the PSS-10 [36]. As this scale is based on Lazarus’s transactional model of stress, it fits well with the theoretical basis of the intervention. We further decided to employ a general stress scale as previous research in a similar intervention for employees showed that work-related and non-work-related problems are equally often indicated and addressed [30]. The items were answered using a 5-point Likert scale (0=never; 1=almost never; 2=sometimes; 3=fairly often; 4=very often; range 0-40) referring to the past week. Cronbach alphas for this scale have been reported to range from .78 to .91 [37] and was .70 at T1, .90 at T2, .90 at T3, and .91 at T4 in this study.

As the current evidence base for Web-based stress-management is limited and the effect sizes vary considerably between trials (from non-significant to moderate-large), we decided to use a conservative approach in order to also detect small effect sizes. We relied on a meta-analysis on traditional stress management interventions [13] and expected an effect size of d=0.35. Therefore, based on an alpha of .05 (two-tailed test), and a power of 80%, a sample size of 132 participants per group was necessary.

The applied random integer list was created by an independent researcher using a Web-based randomization program (Randlist). The participants were informed about the randomization outcome via email.

All analyses are reported according to the Consolidated Standards of Reporting Trials (CONSORT) statement regarding eHealth [50] using intention-to-treat (ITT) procedures (see Multimedia Appendix 2). Additionally, per-protocol and study completers-only analyses are reported. A significance level of .05 (two-sided) was used for all analyses. Analyses were performed using IBM SPSS version 22.

A total of 450 individuals were screened for eligibility, and 186 were excluded primarily because they scored below the cut-off (136/450), because of a lack of informed consent/baseline (30/450), or other reasons (20/450). The study flow is illustrated in Figure 1.

The demographic variables for all study participants are displayed in Table 1.

The average age of the participants was 43.3 years (SD 10.2). The sample was primarily female (193/264, 73.1%), married or in a relationship (160/264, 60.6%), and highly educated (203/264, 76.9%). Most participants were employed full-time (204/264, 77.3%); their average working experience was 18.1 (SD 11.1) years; and they were working in various sectors, most frequently in the social sector (97/264, 36.7%). Only a small percentage of participants had previously taken part in any health training (34/264, 12.9%). Having received psychotherapy was indicated by 95 (36.0%) of the 264 participants and currently being in psychotherapy by 16/264 (6.1%). Table 2 summarizes all means and SDs for the iSMI and WLC groups.

Overall, 8.0% (21/264) of participants at T2, 10.6% (28/264) of participants at T3, and 30.3% (40/132; iSMI only) of participants at T4 did not provide follow-up data for the primary outcome. A somewhat higher dropout rate was observed for the iSMI group (T2: 16/132, T3: 17/132) compared with the WLC (T2: 5/132, T3: 11/132). Thereby, groups significantly differed at T2 (χ ² ₁=6.26; P<.05), but not at T3. Participants who did not provide follow-up data did not differ in a meaningful way from those who provided data, neither on baseline stress scores or any other baseline outcomes, with the exception of worrying (P<.05). Little’s overall test of randomness indicated that data were missing completely at random. Thus, multiple imputations to estimate missing values could be performed [56].

In the WLC condition, 37 participants (28.0%) indicated at T2 that they had received other help within the previous 7 weeks (eg, psychotherapy, health training other than the iSMI) as opposed to 24 participants (18.2%) in the iSMI condition. No significant differences were found in stress levels between those participants who had received help and those who had not.

Table 3 also shows the results of the ITT analyses for secondary outcomes for mental health, work-related health, and skills/competencies. The ANCOVAs showed highly significant between-group effects for almost all outcomes at both assessment points; all significance levels were P<.001 apart from work engagement at T3 (P=.002) and the physical health component of quality of life at T3 (P=.04). Between-group effects were not significant for absenteeism and presenteeism.

At T2, the majority of effect sizes were in the range of moderate (eg, d=0.40 for insomnia) to large (eg, d=0.75 for emotional exhaustion) apart from work engagement for which a small effect size was obtained (d=0.23). At T3, almost all effect sizes became more pronounced apart from the comprehension subscale of the ERSQ-27, which only slightly decreased (from d=0.55 at T2 to d=0.51 at T3). Thereby, large effect sizes were found at T3 for depression (d=0.95), anxiety (d=0.84), emotional exhaustion (d=0.86), and emotion regulation skills regarding general distress (d=0.87). The additional measurements taken at T3 yielded effect sizes of d=0.68 for the mental health component and d=0.20 for the physical health component of quality of life, as well as d=0.16 for absenteeism and d=0.01 for presenteeism.

The primary aim of this study was to evaluate the efficacy of a guided iSMI for employees. For this purpose, a two-arm, waitlist-controlled randomized trial was conducted. The results indicate that the training is highly effective in reducing employee stress levels in the short term (d=0.83) and long term (d=1.02) compared with the levels observed in a waitlist control group. Reduced stress levels in the iSMI group could be maintained up to 12 months. Significant medium to large between-group effects were also found for relevant secondary outcomes concerning mental health (eg, depression), work-related health (eg, emotional exhaustion), and stress-related skills (eg, emotion regulation competencies). High levels of client satisfaction and adherence were observed, and the study dropout rate was low.

The posttest effect size in stress reduction found in this study is larger than what has been found in other iSMI trials. Available RCTs on iSMI with employees show mixed between-group effects for stress at posttest, ranging from non-significant [22] to moderate effect sizes (eg, d=0.74; [18]). Several reasons could explain the large effect sizes found in this study. First, the intervention used a guided format. Guided iSMIs [16-18] appear to be more effective than unguided interventions [21,22,57,58], a result that is known from Web-based interventions for other mental health problems [59]. Second, the theoretical basis of the intervention was confined to two evidence-based components. Research on face-to-face interventions suggests that interventions with fewer treatment components are superior to those using more components [12]. Problem-solving training that has already been successfully introduced in other Web-based interventions to manage depressive symptoms [30] was combined here with evidence-based emotion regulation techniques including the acceptance of emotions and compassionate self-support based on the Affect Regulation Training (ART) [32]. Third, the use of mobile components to flexibly introduce training components into daily life in real time may have reinforced a regular application of the intervention exercises and therefore the efficacy of the training. The text messages were particularly popular among the participants, and given the choice of receiving the text messages, the vast majority requested this mobile component. However, we did not assess the actual engagement and future studies should compare the efficacy of the intervention with and without the mobile phone component. Fourth, the level of intervention adherence, which is regarded as leading to better treatment outcomes [60], was relatively high. Unfortunately, comparisons with adherence levels of other iSMIs are difficult as few studies have reported this information so far. Compared with the available intervention completion rates (eg, 38.5%, [14]; 44.0%, [61]; 88.2% [19]), the percentage of participants completing the intervention in this study was higher range (70.5%). Considerable efforts were undertaken to increase adherence through methods that are generally considered to be effective, including human support [62], interactive exercises [63], tailoring of the intervention [64], and reminders [65] via mobile phone. Finally, the effects may have further stabilized at T3 through the booster session, as booster sessions can be successful in maintaining treatment outcomes [66].

With regard to long-term follow-up, this work is the first study of an intervention combining Web-based and mobile components that focuses on stress reduction in employees to assess the effects compared to a control group over a longer time period (ie, 6 months). The results show that this type of intervention can have large long-term effects. Moreover, the results of this study also compare favorably to the limited existing evidence on the long-term effectiveness of iSMIs in populations other than employees, including studies finding non-significant [23] and small to moderate effect sizes (d=0.37 [14]) at the 6-month follow-up point. In addition, the results found for stress show similar effect sizes to those found for face-to-face interventions (d=0.73 [12]). Thus, iSMIs may be a useful alternative to traditional interventions.

With regard to the relationship of treatment intensity and outcome, no clear conclusions can be drawn from this study. Although intervention completers showed significantly lower stress levels at posttest, this effect was not maintained at the later follow-up points. Participants receiving text message support were also not doing significantly better as compared to those who did not receive any messages. Future research may benefit from further information on the amount of time spent on the intervention exercises in between sessions and the actual engagement with the text messages.

The following limitations of this trial must be acknowledged. First, for feasibility reasons, only self-report measures were assessed. Although the replacement of self-report measures with physiological measures is not recommended in occupational stress research [67], a combination of both could produce further valuable insights. Second, because this study was in the setting of indicated prevention, these results only account for participants showing relatively high baseline scores. The current sample was severely distressed and showed high baseline scores on all measures. Thus, no conclusions can be drawn regarding participants with lower stress levels (eg, in a universal prevention setting). Third, with regard to the generalizability of results, the fact that participants self-selected into the trial, the majority were female, and individuals working in the social sector were slightly overrepresented needs to be taken into consideration. Fourth, to determine the added value of the mobile component providing real-time support and encouragement, direct comparison studies would be needed comparing the intervention with and without mobile support. Fifth, the fact that the effects on physical health and work engagement were smaller than the effects on the other outcome measures demands an explanation. It is possible that Web-based interventions do not produce meaningful differences on these outcomes. Alternative explanations for the small effect sizes include that, for physical health, a rather global measure (SF-12) was applied and it may be more promising to use more specific, stress-related health measures. For work engagement, the original response categories were not adapted to the study period and the outcome measure may therefore not have been as sensitive to change. Finally, it is important to acknowledge that some improvements were also observed in the WLC group over time; in fact, this pattern has previously been found in other trials [17,31].

Future research should replicate the results of this trial and investigate the moderators of outcome and adherence. It is also of interest whether the coaching time spent on each individual (up to 4 hours) could be reduced without losses in treatment effects, thereby resulting in more economical versions of iSMI. Moreover, future research should test iSMIs against the gold standard in the field (ie, face-to-face interventions) and assess which training format works best for which type of participant and under what circumstances. Although both formats may be equally effective, they may work differently on participants with varying personal characteristics and Web-based interventions may be more advantageous in terms of efficiency and costs.

This trial contributes to the limited evidence base on the feasibility and efficacy of Web-based and mobile-supported stress management interventions and is among the first studies to include a longer follow-up period. The iSMI presented herein proved feasible and highly effective in improving perceived stress and other mental and work-related health indices in employees in the long term. These results indicate that this iSMI could be a valuable alternative to face-to-face trainings. Web-based interventions for coping with stress should be further evaluated as such interventions have the potential to improve the mental health of individuals on a large scale.