We conducted and reported the health-economic evaluation in agreement with the Consolidated Health Economic Evaluation Reporting Standards statement [17] and the guidelines from the International Society for Pharmacoeconomics and Outcomes Research [18]. We conducted the economic evaluation with a 12-month time horizon from both a societal and a public health care perspective alongside a 2-armed pragmatic randomized controlled trial in Germany to establish the cost-effectiveness and cost-utility of an indicated Web-based guided self-help intervention in conjunction to usual care for people with sD as compared with enhanced usual care (ie, Web-based psycho-education in addition to treatment as usual). Whereas the societal perspective included all costs and consequences regardless by whom they were incurred, the health care system perspective only considers direct medical costs. Full details of the trial design can be found elsewhere [19]. The study was approved by the medical ethics committee of the University of Marburg (reference number AZ 2012-35K) and registered under DRKS00004709 in the German clinical trial registry.

In total, 406 participants were included in the study. Participants were recruited from March 2013 to March 2014 from the general population via a large German health insurance company and through newspaper articles, on-air media, and related websites. We chose this open recruitment strategy as it reflects the clinical practice for this type of intervention, thus enhancing the trial’s ecological validity. As we conducted a pragmatic trial, the use of antidepressant medication was allowed as part of care-as-usual. However, participants needed to be on a stable dose for at least four weeks to be able to enter the study. Textboxes 1 and 2 present participant eligibility for inclusion and exclusion in the study.

Randomization took place at an individual level and was conducted centrally by an independent statistician not otherwise involved in the study using an automated computer-generated random numbers table. Block randomisation, of size 2, was used to ensure equal sample sizes across both conditions. Details about the randomization procedure can be found elsewhere [16]. Study participants were not masked to their treatment allocation because of the nature of the intervention. SCID interviewers were, however, unaware of participants’ randomization status. Steps taken to maintain blinding are described in detail elsewhere [16]. In case of evidence for blinding breakdown, the interviewer was changed to the second outcome interview. The research staff conducting SCID interviews were not otherwise involved in the study.

All study participants had unrestricted access to care-as-usual (CAU). CAU for sD entails visits to the general practitioner (GP), but no treatment provided by mental health specialists. If depressive symptoms deteriorate, the German S3-Guideline/National Disease Management Guideline Unipolar Depression recommends psychotherapy and the prescription of antidepressant medication [21]. In our pragmatic study, we did not interfere in CAU. Instead, we maintained a naturalistic CAU condition to represent current clinical practice as far as possible. It should also be noted that health care use was measured in detail (see Measures), implying that we now can describe CAU in great detail (see Table 1).

Self-reported measures (ie, EuroQol and SF-12) were collected at baseline, posttreatment (6 weeks after randomization), and 6- and 12-month follow-up using a secured Web-based assessment system (AES, 256-bit encrypted). The SCID interviews at baseline and 6- and 12-month follow-up were conducted by telephone.

We used the Trimbos and iMTA questionnaire for costs associated with psychiatric illness (TiC-P) [32,33] for collecting data on health care utilization and productivity losses in patients with mental health conditions. We adapted the TiC-P for use in Germany and used it at baseline and at 6- and 12-month follow-ups. We computed health care costs, the patient’s out-of-pocket costs, the costs for informal care provided by the patient’s family and friends, and costs stemming from productivity losses due to absenteeism and lesser productivity while at work (presenteeism). Costs were expressed in Euro and indexed for the year 2013 (index factor 1.04) based on the German consumer price index [34]. Costs were converted to pound sterling (£) using the purchasing power parities reported by the Organization for Economic Cooperation and Development. For the reference year 2013, €1 was equated to £0.85.

The study was powered to demonstrate a risk reduction of 10% between study conditions as statistically significant in a survival analyses with alpha<.05 (2-tailed), a power of (1−beta)=.80 using survival analysis, and accounting for a 20% dropout (calculated using PASS 12). However, the study was not powered to statistically test differences in health economic outcomes. Therefore, we took a probabilistic decision-making approach for our health economic inferences [39]. We did not discount costs and effects because the analysis was restricted to a 12-month follow-up period.

While evaluating the clinical outcomes, we reported all analyses in accordance to the CONSORT statement [40] (Multimedia Appendix 2). Data were analyzed on an intention-to-treat basis meaning that all participants were included in the analyses as randomized. To this end, we used Cox proportional hazard regression analyses to test differences in time to onset of MDD (in weeks) between intervention and control group. Concurrent use of antidepressants was included as covariate into the Cox proportional hazard model (post hoc). As the use of antidepressants was not a predictor of the outcome, it was excluded from the final model.

To account for missing data in cost and utility data, we used the regression imputation procedure in Stata version 13 (StataCorp) to obtain required predicted values. Predictors of outcome and dropout were identified by (logistic) regression analysis. Identifying predictors of outcome helped us to obtain the most likely values of the outcome whereas identifying predictors of dropout allowed us to correct for bias that might arise by differential loss-to-follow-up. We did not impute hospitalization costs because only 3 participants (0.7%) were hospitalized during the 12-month follow-up period leading to instable imputations. Therefore, we reassessed the impact of hospitalization costs on outcomes in a sensitivity analysis. At baseline, mean EQ-5D utility values were the same in intervention and control group (both groups: 0.74, SD=0.15). Therefore, no baseline adjustments were made when calculating QALYs. Differences in QALYs and DFYs between the intervention and control groups were assessed using independent samples t-tests.

In the cost-effectiveness analysis, the incremental cost-effectiveness ratio (ICER) was based on the incremental costs per unit of effect (DFY or QALY) gained. The corresponding equation is ICER=(Costs_INT– Costs_CTR)/(Effects_INT– Effects_CTR), where Costs are the annual per-participant costs and Effects are the DFYs (QALYs) in intervention and control group (subscripted with INT and CTR, respectively). Sampling uncertainty in the ICER was handled using nonparametric bootstrapping by resampling patient-level data to generate 2500 simulations of the ICER. We bootstrapped the SURE model (seemingly unrelated regression equations; sureg command in Stata) to allow for correlated residuals of the cost and effect equations. Bootstrapping was used to obtain confidence intervals for cost-effectiveness ratios based on the percentile method, since parametric techniques are inappropriate for use on skewed variables and ratios. The bootstrapped ICERs were plotted on a cost-effectiveness plane with effects along the horizontal axis and costs along the vertical axis. A cost-effectiveness acceptability curve (CEAC) was graphed to assess the probability of the intervention being cost-effective at varying willingness-to-pay (WTP) ceilings. All analyses including uncertainty analyses were performed using Stata version 13.

In total, 406 participants were enrolled in the study (N_INT=202; N_CTR=204). At the posttreatment stage, 366 participants (90.1%, 366/406) were still participating. At 6- and 12-month follow-up, 325 (80.1%, 325/406) and 286 (70.4%, 286/406) participants completed the questionnaires, respectively. The CONSORT flowchart (of the participants through the trial) can be found elsewhere [16]. Dropout rates did not differ between experimental and control conditions except for the 12-month follow-up. Here, dropout was higher in the intervention group (χ²₁ = 8.4, P=.004). Study dropout was not associated with baseline depressive symptom severity or any sociodemographic factor. Participants’ characteristics at baseline are presented in detail elsewhere [16]. In brief, the average participant was female aged 45 years with an above average level of education and employed.

The mean depression-free survival time within the 12-month trial period was 43 weeks (95% CI 41-46) in the intervention group and 37 weeks (95% CI 36-40) in the control group [16], corresponding to 0.82 DFYs and 0.70 DFYs, respectively. The incremental effectiveness of 0.82−0.70=0.12 DFYs was statistically significant (95% CI 0.05-0.18; t₄₀₄=3.37, P<.001). Between-group differences in EQ-5D, QALY gains were not statistically significant (intervention group: 0.78, SD=0.14 vs control group: 0.77, SD=0.13; t₄₀₄=−0.99, P=.32), but incremental SF-6D QALY gains differed significantly between study groups (intervention group: 0.71, SD=0.08 vs control group: 0.67, SD=0.07; t₄₀₄=−4.40, P<.001).

At baseline, mean total costs were €483 (£411) in the intervention group and €528 (£449) in the control group, which is only a small difference of €45 (£38), indicating that randomization had been well balanced. Table 1 presents the 12-month accumulated per-participant costs for various cost categories by study condition. The mean health care costs were higher in the intervention group as compared with the control condition. This difference can largely be explained by differences in the costs of the intervention as compared with the control condition with Web-based psycho-education (€299 vs €10). Patient and family’s out-of-pocket costs and the costs stemming from changes in productivity losses remained quite similar between study groups. Mean total costs, as seen from both the health care and societal perspective, were hence slightly higher in the intervention group as compared with the control group (societal perspective: €143 (£121); health care perspective: €136 (£116).

Table 2 shows the incremental cost, effects, and cost-effectiveness ratios (based on 2500 bootstrap simulations) for the main analysis and the sensitivity analyses and from both the societal and health systems perspective. From a societal perspective, the intervention resulted in a greater mean health benefit (0.12 DFY gained) achieved at higher mean total costs (€134; £114) as compared with enhanced CAU.

The cost-effectiveness plane, representing the 2500 bootstrap replications, is shown in Figure 1. Most (62%) of the bootstrapped ICERs fell in the north-east quadrant, indicating a 62% probability that the intervention produces greater health, but at greater costs than enhanced CAU. The remaining 38% of ICERs fell in the south-east quadrant, indicating a 38% probability that the intervention dominates enhanced CAU because additional health gains are obtained for lesser costs. In other words, at a willingness-to-pay (WTP) of €0, the probability that the intervention must be regarded as more cost-effective than CAU is 38%.

However, at a WTP of €7350 (£6248), €9680 (£8228), and €20,000 (£17,000) for gaining a depression-free life-year, the intervention’s probability of being more cost-effective than CAU rises to 90%, 95%, and 99% (Figure 2).

The ICER based on QALY gains showed a small health benefit (0.01 QALYs gained) for higher mean costs (€134; £114). As seen in Figure 3, most of the simulated ICERs fell in the north-east quadrant (49%; see also Table 2).

The intervention’s probability of dominating enhanced CAU was 35% when taking the societal perspective. Assuming a willingness-to-pay of €20,000 (£17,000) for gaining a QALY, this probability rose to 60% (Figure 4).

From a public health care perspective, the health benefit (0.12 DFY gained) was also achieved at higher costs (€135; £115). In the cost-effectiveness plane, 83% of simulated ICERs fell in the north-east quadrant with the remaining 17% in the south-east quadrant (Table 2). The intervention’s probability to be more cost-effective than enhanced CAU increased to 51% at a WTP of €1250 (£1063) and became 90%, 95%, and 99% at a WTP of €3140 (£2669), €3,920 (£3332), and €6000 (£5100), respectively. From the societal perspective, the ICER from the public health care perspective based on QALY gains showed a small health benefit (0.01 QALYs gained) for higher mean costs (€135). In the cost-effectiveness plane, 68% of the simulated ICERs fell in the north-east quadrant (Table 2). The intervention’s probability of dominating enhanced CAU was 16% when considering health care costs. Assuming a willingness-to-pay of €20,000 (£17,000) for gaining a QALY, this probability was 64%.

Using the SF-6D resulted in a greater incremental QALY gain in favor of the intervention group (0.70 QALY, SD 0.08) as compared with the control group (0.67 QALY, SD 0.07), which was statistically significant (t₄₀₄=4.40, P<.001). This agrees with available evidence that the EQ-5D suffers from a ceiling effect in less severe diseases [41]. At a WTP of €20,000 for gaining one QALY the probability of being cost-effective was 84% (societal perspective) and 99% (health care perspective). Hospital costs were higher in the control group so excluding these costs resulted in higher ICERs (Table 2). From a societal perspective and at a WTP of €20,000, the intervention’s probability of being cost-effective then became 53%.

Our study was set out to evaluate the cost-effectiveness and cost-utility of a Web-based guided self-help intervention to prevent the onset of MDD in adults suffering from sD in comparison with usual care enhanced with Web-based psycho-education. Main outcomes were DFYs and QALYs.

Significantly more DFYs were gained in the intervention group as compared with the control group. The probability that the intervention is deemed to be cost-effective depends on the willingness-to-pay (WTP) for a depression-free year. Assuming that the intervention is considered to be cost-effective if its likelihood of being cost-effective is greater than 90%, a WTP of €7350 and above for gaining a DFY would make the intervention cost-effective. While these DFY health gains were not mirrored in EQ-5D QALY gains, they were observed in SF-6D QALY gains with the latter being more sensitive to change in milder conditions such as sD and therefore not suffering as much from a ceiling effect as the EQ-5D does [31]. If society would be willing to pay €20,000 (£17000) for gaining a QALY, the probability of being cost-effective will be 64% for gaining an EQ-5D QALY and 84% for gaining a SF-6D QALY.

Both burden of disease studies and cost of illness studies [1,2] attest to the importance of cost-effective interventions that can reduce the burden of depressive disorders on a wider scale. Recent meta-analytic evidence suggests that it is possible to prevent the onset of major depression using psychological interventions [8]. Results of our study showed that a guided eHealth intervention can successfully reduce the incidence of diagnosed major depression [16]. In addition, some evidence indicates that bibliotherapeutic self-help interventions for the prevention of depression represent good value-for-money [15]. However, economic evaluations in the field of depression prevention mainly relied on health-economic modelling [12,13,42] with direct evidence stemming only from 2 randomized controlled trials [10,11]. Although some trial-based economic evaluations of computerized cognitive-behavioral interventions for treating depressive symptoms exist (ie,) [43,44], to our knowledge, this is the first trial-based economic evaluation of a Web-based intervention to prevent the onset of major depression in an adult population with sD. Results from our trial adds to the converging evidence pointing to the effectiveness and cost-effectiveness (sometimes even costs-savings) of depression prevention across a range of outcomes.

This study has some limitations. First, the time horizon of this study was limited to 12 months. A recent meta-analysis showed a small positive association between effects of preventive interventions during the first months of follow-up, indicating that with passing months, the intervention effects get somewhat larger. However, meta-regression analyses suggested that the effects of interventions are lower at longer follow-up periods of 1 to 2 years [8]. Assuming diminishing long-term effects, the cost-effectiveness of this particular intervention will also decline. However, only few studies had longer follow-up periods than 2 years. Thus, more research with longer follow-up periods are needed to assess the long-term effectiveness and cost-effectiveness of preventive (Web-based) interventions. Second, we did not assess lifetime history of MDD at baseline, and therefore we cannot be sure if we prevented first-ever onsets of MDD or MDD recurrences. Future studies should thus clarify whether Web-based guided self-help interventions are cost-effective both for the prevention of first depression onset and the prevention of recurrence. Third, costs were assessed with the help of self-reports and may suffer from under-reporting. However, the structured questionnaire used in this study can be considered as valid instrument for recall periods up to 3 months [45]. Finally, the trial has been conducted in a highly-educated sample. Evidence suggests that better adherence is predicted by higher education [46]. In our trial, only 2% of participants were low educated. Hence, we cannot predict the uptake of such an intervention in less educated people or among people with a lower socioeconomic status. However, we used an open recruitment strategy in our trial mimicking the way how people will be recruited for eHealth interventions in the future, thus providing ecological validity to the sample on which this study is based. In other words, the trial sample reflects the population segments that are interested in engaging in a Web-based intervention. However, one conclusion drawn from this trial is that not all people who are in need of psychological interventions could be reached via the Internet. The applicability of Web-based interventions is related to (1) the acceptance of such interventions by the target population (ie, preferences for different treatment modalities, such as face-to-face interventions) and (2) the availability of technical requirements (ie, reliable access to the Internet).

Current guidelines on depression treatment (such as the NICE guideline and the Dutch Multidisciplinary Guideline for Depressive Disorder) recommend low-intensity psychosocial interventions (ie, computerized cognitive behavioral therapy) to manage (persistent) sD symptoms and mild-to-moderate depression [22]. Our study supports this recommendation by showing that an eHealth intervention may not only restore health in people with sD, but in addition reduces the risk of developing a major depressive disorder. Findings from our study also add that delivering cognitive-behavioral therapy over the Internet has a high probability of being cost-effective to prevent the onset of new depressive episodes. Given low participation rates in face-to-face preventive services and the potential to scale up Web-based interventions to efficiently alleviate the disease burden caused by MDD, it would be worthwhile to integrate such a Web-based intervention into routine practice. However, there are some opportunities and risks that need to be taken into account when scaling up this intervention. First, the feedback provided by an individual trainer on the Web in this Web-based intervention hinders scaling up the intervention. A recent review of randomized controlled trials showed that unguided interventions can also be effective (with lower adherence rates compared to guided interventions) [47]. To be more precise, in our study, out of the 202 participants who were initially assigned to the intervention, 138 (68.3%, 138/202) were intervention completers. This compares favorably with an unguided Internet-based intervention for the treatment of sD that was completed by 48.3% (49/102) of participants [48]. Providing guidance may not only affect the outcome and cost-effectiveness of the intervention but also the target group’s willingness to use such an intervention and thereby influencing the effects of such an intervention at population level. It is therefore not possible to predict the effects of this particular Web-based intervention at population level when it is offered without guidance by an eCoach. In addition, there are no guarantees that adherence and (by proxy) effectiveness will be maintained if Web-based preventive interventions are scaled up in the population. For example, Christensen et al reported that less than 1% of public registrants using a preventive intervention delivered openly on a website completed all modules [49]. Second, an unanswered question refers to how Web-based interventions could be rolled out to the population. For example, in the UK and the Netherlands, Internet-delivered cognitive behavioral therapy (CBT) is already prescribed by GPs [22]. In addition, Web-based interventions could be promoted independently at the GP’s (ie, promotion videos in the GP’s waiting rooms). However, a challenge in preventing depression is that most individuals at risk of developing a major depression do not show up in primary care. Therefore, innovative approaches are needed to reach these groups, for example through a systematic mental health screening of all people in specific settings (ie, occupational setting or universities) and to motivate those at risk to engage in preventive interventions (ie, acceptance facilitating interventions) [50]. However, such strategies do not guarantee uptake either. Third, the costs of the Web-based intervention were calculated without considering economies of scale effect. Economies of scale refer to the reduction in the cost per treatment of an intervention as a result of increasing the number of clients. Economies of scale arise because many of the costs associated with the Web-based intervention are fixed and not dependent on the number of clients (ie, hosting the intervention on a server) and thus increasing the intervention output reduces the fixed cost per treatment. Hence, we assume that the cost of IT infrastructure (ie, the per-client cost of servers to host the intervention) will be cheaper per additional client if the intervention is offered on a larger scale. Economies of scale might also reduce variable costs (ie, therapist’s support per participant) because therapists become more efficient through better organization and experience. However, the same technical resources available in the research setting (ie, reasonable Internet connections) may not be available when the intervention is scaled up. Finally, for some individuals a self-help approach might not be sufficient [51]. Some individuals may feel unable to apply psychotherapeutic self-help strategies. Some techniques could be inappropriately implemented by participants without guidance by an eCoach. One could argue that it is easier to observe and react to early signs of deterioration in face-to-face interventions as compared with Web-based interventions. Another potential negative effect of self-help interventions could be a delay in help-seeking leading to a further deterioration of symptoms, if the initial low-intensity self-help intervention should not be sufficient. Hence, multiple approaches to reach the target population are needed in successful depression prevention programs. Thus, future studies should evaluate the preventive effects of unguided Web-based interventions on the onset of MDD and compare the cost-effectiveness of unguided and guided interventions. In addition, implementation studies should be conducted to obtain real world effects of such interventions and to gather knowledge about the willingness to use these interventions in specific population segments (ie, in low-educated people, in a rural setting, or among people with lower socioeconomic status).

Given the evidence for the efficacy of psychological interventions to prevent depression and the potential scalability and cost-effectiveness of Web-based interventions, large-scale dissemination of these interventions might be a promising strategy to alleviate depression’s disease burden in an affordable way and on a wide scale. However, before a nationwide dissemination could be considered, future studies need to evaluate the preventive effects of unguided Web-based interventions on the onset of MDD and compare the cost-effectiveness of unguided and guided interventions. Moreover, implementation studies are needed to obtain real world effects of such interventions and to gather knowledge about the willingness to use these interventions in specific population segments (ie, in low-educated people, in a rural setting, or among people with lower socioeconomic status).