Nowadays, almost all people worldwide experience increased stress. In the last few years, many studies have found an enormous increase of stress in adults, teenagers, and children [1]. Especially in Western countries, the rise in workload has resulted in a rapid growth of the number of employees experiencing psychological problems related to occupational stress [2]. In 2006, an international survey revealed that approximately 75% of the general population in developed countries reported stress on a daily basis. In addition, 44% of Americans surveyed in 2010 specified that they had experienced a growth in stress over the past 5 years [3].

There are few findings known for a Web-based intervention in primary care because most eHealth interventions for stress have been evaluated in workplace settings. Stress could be perceived as such a minor problem that it does not require any treatment or professional assistance [4]. However, there is no doubt that chronic stress clearly is a risk factor for a wide range of mental and physical health problems, such as metabolic syndrome [5], diabetes [6], cardiovascular disease [7,8], ischemic stroke [9], and depression [10-12]. Internet-based interventions have shown to be effective in community and clinical settings, including the treatment of depression [13-16], sleep disorders [17], weight reduction [18], smoking cessation [19], and stress reduction [20-27]. Some studies have also failed to find any effects on stress [28-30]. A meta-analysis showed that cognitive behavioral interventions are more effective in stress reduction than other techniques, such as relaxation techniques, multimodal programs, and organization-focused interventions [31]. Additionally, it has been noted on the basis of several trials that the effect sizes from Internet-based stress management programs were close to estimations of face-to-face cognitive behavioral interventions [31,32]. To use health care resources at an optimal level, graded treatment systems represent attempts to improve the efficiency and access to mental health. In a first attempt, low-cost interventions are offered. For those who are not sufficiently helped by the initial low-cost intervention, more intensive and costly interventions are then used in a second step [33]. In addition to the well-established intensive and costly interventions [34], the need to implement and to verify interventions with low financial and accessibility thresholds is still demanded [35,36]. Therefore, a Web-based program was developed that combines an individually tailored strategy for stress reduction with automated advice and feedback elements based on cognitive behavioral therapy (see Multimedia Appendices 1 and 2). In the cluster randomized trial reported subsequently, we investigated whether adult primary care patients who wanted stress reduction and used a fully automated 12-week Web-based coaching program did reduce their stress more effectively than with usual care by general practitioners (GPs).

Originally, 92 GPs were interested in participating and were randomized, but 16 GPs withdrew early after randomization (7 GPs from the intervention and 9 GPs from the control group) and 41 GPs (25 GPs from the intervention and 16 GPs from the control group) did not recruit any participants for the study (Figure 3). Altogether, 93 patients were recruited by 35 GPs (40 patients by 14 GPs in the intervention group; 53 patients by 21 GPs in the control group) between April 18, 2011 and July 1, 2013. In all, 45 of 93 (60%) participants were female and the mean age was 42.2 years (SD 11.5). Overall, 15 participants had incomplete data in the intervention group, 11 did not show up for the measurement at 12 weeks, 3 participants had incomplete follow-up data, and 1 participant had incomplete baseline data. In the control group, 4 participants had missing values at 12 weeks. For 74 participants (25 from the intervention and 49 from the control group), information on PSQ was available both at baseline and after 12 weeks. The proportion of noncompleters (intervention: 15/40; control: 4/53) was significantly higher in the intervention group than in the control group (χ² ₁=12.6, P<.001). The intervention and control groups were similar at enrollment regarding gender, age, employment status, family status, and physical activity (Table 1).

Stress levels decreased in both groups from baseline to follow-up (Table 2). In the CC analysis, overall PSQ scores were reduced by mean 13.1 (SD 13.9, P=.02) points in the intervention group and mean 13.7 (SD 14.8, P<.001) in the control group. In the ITT analysis, reductions were by mean 8.2 (SD 12.7, P=.07) and mean 12.6 (SD 14.7, P<.001), respectively. Group differences within both analyses were nonsignificant. After adjustment for baseline differences between the groups, overall PSQ scores remained nonsignificant for the CC population (mean difference –0.3, 95% CI –7.1 to 6.5, P=.93) and for the ITT population (mean difference –4.2, 95% CI –10.2 to 1.3, P=.13). The secondary analysis using a generalized estimating equation also showed a nonsignificant result (P=.45).

The results from the secondary subscales (worries, tension, joy, and demands) also revealed no significant group differences for either the CC or ITT analyses (Table 3). The ITT analysis revealed no significant differences for worries (mean difference –4.6, 95% CI –10.6 to 1.4, P=.13), tension (mean difference 1.0, 95% CI –8.4 to 6.3, P=.78), joy (mean difference 3.6, 95% CI –3.0 to 10.2, P=.28), and demands (mean difference –2.8, 95% CI –9.7 to 4.2, P=.44).

Adverse events from 2 participants were documented. In the intervention group, one participant reported family and workplace problems, whereas in the control group one participant specified an adverse event without further details. The authors did not consider that these adverse events were directly related to the intervention.

To the best of our knowledge, this study is the first study to investigate a Web-based stress-reduction intervention in primary care. We found that the fully automated Web-based coaching program was not effective for achieving stress reduction compared to usual care. The mean PSQ score decreased in both groups without a significant group difference. Thus, this trial could not show any advantages compared to usual care. Nevertheless, previous findings revealed that stress reduction can be delivered effectively via the Internet [20-27]. Most computer-based interventions for stress have been evaluated in workplace settings [21,22,27,29,43]. For example, Ruwaard and colleagues [43] demonstrated that an Internet-based cognitive behavioral treatment of work-related stress was more effective in reducing stress than a waiting control group. Few studies have evaluated the impact of a Web-based intervention in the general population [26,44]. However, the content of the evaluated interventions and the methodological approaches offered great variability. Zetterqvist and colleagues [26] found that an Internet self-help intervention for relaxation training, exercises (cognitive and behavioral restructuring), and information could be effective in reducing symptoms of stress. Drozd et al [44] demonstrated from a RCT that a Web-based intervention based on mindfulness and metacognitive exercises lead to a reduction of stress. Both studies recruited their participants through webpages or newspaper articles. This might be due to the involvement of a different study population compared to this study sample for which recruitment was carried out by GPs. In addition, the condition and contents of usual care in general practice are not equivalent to a waiting list or a simple online information offer. Therefore, the results from the 2 studies mentioned previously are not directly comparable with this study; furthermore, it is unlikely that this study could discover greater group differences than these previous ones. This is due, firstly, to the comparison of usual care instead of a waiting list or simple online information. Secondly, it might be possible that the mere participation in the control group with the advice from the GP to reduce stress started an autonomous process that led to a reduced level of stress even without exact knowledge about usual care. Another reason why the findings from this study are inconsistent with previous findings is because they were collected in different settings and there might be a “black box” phenomenon or a lack of understanding about how and why some interventions work and others do not. The particular setting and the realization of the intervention may be crucial for their effectiveness. To date, there is insufficient knowledge about the impact of different implementations of Web-based interventions. Due to limited funds, the implementation of this study was designed quite basically. The shortcomings caused by this may have had an influence on the findings from this study. Therefore, the diversity of different kinds of implementation of Web-based interventions should be addressed more in further studies.

One meta-analysis showed that mindfulness can have a broad range of health benefits [45]. Chiesa and colleagues [46] stated that mindfulness-based stress-reduction interventions are generally effective. Another meta-analysis found that cognitive behavioral interventions are more effective than other interventions [31]. Wilhelmsen et al [47] illustrated within a qualitative study that Internet-based cognitive behavioral interventions may add a structured agenda to consultations and simultaneously empower patients. Otherwise, they have shown how challenging and complex it is to conduct an Internet-based cognitive behavioral intervention deployed from GPs in primary care. In summary, current evidence for stress reduction shows that cognitive behavioral interventions seem to be the most effective treatment for a Web-based approach. To this end, further studies are necessary to investigate different modalities of Web-based interventions to learn more about the black box phenomenon. In addition, this trial confirmed the well-known problem that Web-based interventions are often accompanied by a high attrition rate [48]; the significantly higher proportion of noncompleters in the intervention group underlines this fact.

One strength of this study was the embedding of the study in a realistic primary care setting. However, some important methodological aspects for the interpretation of the study results need to be considered. First of all, the randomization of this study was conducted at the GP level before individual participants were included. Thus, physicians knew whether they recruited patients for the intervention or the control group, which could lead to bias. Secondly, due to the highly variable cluster sizes the statistical analysis of our data was not straightforward. Classical linear mixed models taking the cluster design into account could not be used because of numerical problems. Therefore, we used a simple Student t test (which ignores intracluster correlation) and an additional multilevel analysis (which performs inadequately when cluster sizes differ) as the sensitivity analysis. Third, according to our power calculations, the target number of participants was not reached due to slow recruitment of participants; the study had to be stopped at a certain point, which may have reduced the study’s ability to detect significant differences between the groups. Fourth, participating GPs were self-selected, training and supervision were very basic, and other implementation components, such as administrative support, were not available due to limited funding. Fifth, due to strict data privacy requirements we could not access the automatically documented data about the extent participants accessed and used the program. Sixth, the proportion of participants without follow-up values was definitely higher in the intervention than in the control group. This could be because participants in the control group received a small financial incentive, whereas those in the intervention group did not. Therefore, participants in the intervention group might have been less willing to make an additional practice visit after completing the program than those in the control group. Finally, the content of usual care was not further evaluated. The practitioners for the control group were asked to change nothing in their usual way of counseling and to treat their participants in the same manner as usual. There was no additional documentation of the counseling provided.

Our findings suggest that this tested Web-based coaching program was not effective for achieving stress reduction compared to usual care. The change from baseline was similar to usual primary care. The limited statistical power and the high dropout rate may have reduced the study’s ability to detect significant differences between the groups. Further randomized controlled trials are needed to investigate larger populations, the long-term outcomes, and the content of usual primary care.