We conducted a pragmatic 2-arm, open-label, multisite, randomized controlled clinical trial. We applied noninferiority design criteria, including an active comparator (TAU) and a preset noninferiority margin [28]. The E-COMPARED study was conducted across 8 European countries: France, Germany, the Netherlands, Poland, Spain, Sweden, Switzerland, and the United Kingdom, ensuring geographical coverage across Europe. Denmark was subsequently added due to the country’s emergent interest in bCBT. We conducted the RCTs in settings where both bCBT and TAU for MDD treatment could be provided. In Denmark, France, the Netherlands, and Switzerland, the study was conducted within specialized mental health services, while in Germany, Poland, Spain, Sweden, and the United Kingdom, it was conducted within primary care. The results were reported in accordance with the CONSORT (Consolidated Standards of Reporting Trials) guidelines for noninferiority trials [29]. We refer to the protocol for a detailed description of the study [27].

The inclusion criteria were age ≥18 years, meeting DSM-IV (Diagnostic and Statistical Manual of Mental Disorders [Fourth Edition]) diagnostic criteria for MDD via the Mini International Neuropsychiatric Interview (MINI; excluding sections M [anorexia nervosa], N [bulimia nervosa], and P [antisocial personality disorder]) [30], and depressive symptoms (Patient Health Questionnaire-9 [PHQ-9] score ≥5) [31]. Patients were permitted to receive pharmacotherapy in addition to bCBT at baseline or during treatment, if deemed necessary by a health care professional, in line with routine clinical practice. At the 12-month follow-up, phone-based MINI interviews were conducted by independent, trained interviewers who were masked to the assigned group. The exclusion criteria included currently being at high risk of suicide (MINI section C), psychiatric comorbidity (substance dependence, bipolar disorder, psychosis, and obsessive-compulsive disorder), currently undergoing psychological treatment for depression, language comprehension barriers, lack of a stable internet connection or access to a computer, and either the absence of a compatible smartphone or an unwillingness to loan one for bCBT.

Recruitment took place between February 2015 and December 2016; data collection closed on December 31, 2017. Patients were approached by mental health professionals in health care settings, who then referred them to the research team for eligibility assessment. In Germany and Sweden, recruitment leaflets were placed in general practitioner waiting rooms. Sweden also used social media platforms to recruit patients who had contacted primary care services about depressive complaints. Multimedia Appendix 1 provides an overview of the recruitment procedures for each country, along with any additional procedures carried out.

Ethical approval and trial registrations were obtained at the national level. Participants provided written informed consent, authorizing the E-COMPARED partners to share and analyze the anonymized dataset and to report the results in publications. Participants did not receive compensation for study participation. Ethical approval for the trials has been obtained locally in each country: Ethics Committee of the Region of Southern Denmark (S-20150150); France: Comité de protection des personnes, Île de France V (15033-n 2015-A00565-44); Germany: Ethik Kommission DGPsychologie, Universität Trier (MB 102014); The Netherlands: METC VUMC (2015.078); Poland: Komisja ds. Etyki Badań Naukowych (10/2014); Spain: Comisión Deontológica/Comité Ético de Investigación en Humanos (H1414775276823); Sweden: Regionala etikprövningsnämnden (2014/428-31); Switzerland: Kantonale Ethikkommission Bern (001/2015); and United Kingdom: NRES Committee London-Camden and King’s Cross (15/LO/0511).

After the baseline assessment, randomization of patients was conducted centrally at VU Amsterdam (Netherlands) by an independent researcher (EK) using automated Random Allocation Software (Department of Anaesthesia, Isfahan University of Medical Sciences). The allocation, stratified by country, was 1:1. Block randomization was used with variable block sizes, ranging from 2 to 14 allocations per block. In the United Kingdom, the process was outsourced to a local independent researcher for logistical reasons, which required the randomization outcome to be available immediately. Furthermore, stratification by recruitment location was applied at the national level for France (11 centers), Switzerland (3 centers), the Netherlands (3 centers), and Sweden (2 cities). Researchers and clinicians were blinded to the randomization scheme, but blinding of treatment allocation was impossible because both groups were evident to the therapists and patients.

RCTs can determine whether a treatment effect is statistically significant and quantify its magnitude, but they do not necessarily capture the clinical relevance of the findings. Several approaches exist to assess clinical importance. In this study, we adopt the minimal clinically important difference from the patient perspective, as proposed by Cuijpers and colleagues [32], who identified a threshold of Cohen d=0.24. Given our noninferiority design, we applied a more conservative margin of d=0.20 for the primary outcome (PHQ-9). Under this criterion, bCBT is considered noninferior to TAU in clinical terms if the upper bound of the confidence interval does not exceed this predefined margin (1-sided tested). The required sample size was 1052 patients; to allow for dropouts, we aimed to enroll 1200 patients. This calculation was based on 90% power (type II error ≤0.10) and the requirement that the upper bound of the 2-sided 95% CI would not exceed the prespecified noninferiority margin.

The bCBT protocols used in this study were based on ftfCBT protocols for MDD. The bCBT protocols integrated face-to-face sessions with structured internet-based treatment modules and ecological momentary assessments (EMAs) via smartphone to monitor daily mood, cognition, activity, social interaction, and sleep. During these face-to-face sessions, the core principles of CBT were applied, alongside mandatory internet modules at all participating sites (Table 1). These mandatory modules focused on psychoeducation, cognitive restructuring, behavioral activation, and relapse prevention, as well as homework assignments. These internet-based modules were discussed in subsequent face-to-face sessions. The mandatory internet modules comprised 75% of the total number of modules. Sites also had the option to add extra modules (25%), such as emotional regulation or coping skills training, to facilitate adaptation to setting-specific practices, while maintaining protocol standardization across sites. At least one-third of the total number of sessions were to be held face-to-face, while another one-third were to be held online to ensure sufficient face-to-face and online modules.

Table 1 summarizes, by country, the setting, duration, intensity, the alternation between face-to-face sessions and online modules, and additional modules for the bCBT group. As we followed routine care procedures, 5 countries used their own bCBT platforms: Denmark, the Netherlands, Spain, Sweden, and Switzerland. Four countries—France, Germany, Poland, and the United Kingdom—implemented a translated version of the Dutch Moodbuster platform for their patients because no bCBT platforms were available in these countries. All these platforms complied with General Data Protection Regulation requirements [33]. bCBT was delivered within primary care in 4 countries, while in Spain, it was provided both within primary care and outpatient specialized mental health care due to constraints within the original specialized mental health care setting. In the other 3 countries, it was delivered within specialized outpatient mental health care settings. bCBT began with a face-to-face session in all the countries. Within primary care, the treatment duration of bCBT ranged from 6 to 13 weeks, comprising 3-7 face-to-face sessions and 5-10 internet modules. Within specialized mental health care services, the treatment duration was longer, typically from 12 to 20 weeks, with 8-12 face-to-face sessions and 7-10 internet modules. All countries alternated face-to-face sessions with online modules, except for Spain (which used 2 face-to-face sessions, 4 internet modules, and a final face-to-face session). The total number of ftfCBT sessions ranged from 3 to 6 in primary care settings and from 8 to 12 in specialized mental health care settings, and sessions were conducted in accordance with guidelines for treating moderate to severe MDD [34]. These face-to-face sessions lasted between 45 and 60 minutes each. The internet modules required 30-45 minutes of patients’ time and needed 3-5 minutes daily to answer the EMA questions. The therapists used the platforms to monitor their patients’ progress and to communicate securely with them. They were required to spend approximately 15 minutes each week per patient to review their online homework assignments. CBT therapists comprised licensed professionals and supervised psychology trainees; supervision by a licensed psychologist specializing in CBT applied only in primary care settings.

All participants in the control group received active treatment. TAU comprised routine care for patients with MDD in the treatment settings from which they were recruited (primary care or specialized mental health care). Six countries delivered ftfCBT within routine care (Denmark, France, the Netherlands, Poland, Switzerland, and the United Kingdom). In Sweden, Germany, and partly in Spain, TAU consisted of generic general practitioner care, which might have included watchful waiting, referral to a mental health specialist, psychotherapy (in Sweden and Germany), pharmacological treatment, or a combination of these. In the ftfCBT settings, therapists’ professional qualifications were similar to those in the bCBT groups.

All measures were self-reported and completed online (except for the MINI, which was assessed by independent interviewers blinded to the groups). Primary and secondary outcomes (including costs) were assessed at baseline and at the 3, 6, and 12-month follow-up periods. The MINI was assessed only at baseline and at the 12-month follow-up. We followed routine care RCT protocols in the participating countries, which resulted in primary care studies being shorter in duration and lower in intensity than those conducted in specialized mental health care settings. Posttreatment assessments meant that all participants could have completed their treatment 6 months after baseline. To enable an overall comparison, we therefore defined the 3-month assessment as a during-treatment measurement, the 6-month assessment as a posttreatment measurement, and the 12-month assessment as a follow-up measurement. The outcomes and time points are presented in Table 2. The reliability of the questionnaires is presented in Multimedia Appendix 2. Cronbach α ranged from 0.70 to 0.98, indicating good overall reliability.

The PHQ-9 was the primary outcome at posttreatment. It has demonstrated good psychometric properties in both primary and specialized care populations [30,35,36]. The total score ranges from 0 to 27, with 0‐4 indicating no to minimal depression, 5‐9 indicating mild depression, 10‐14 indicating moderate depression, 15‐19 indicating moderately severe depression, and 20‐27 indicating severe depression. Paper-based and online versions of the PHQ-9 yield comparable psychometric properties [37]. The PHQ-9 was also used to assess reliable deterioration. Participants were classified as having reliably deteriorated when the increase in PHQ-9 score exceeded the Reliable Change Index (RCI) threshold (|RCI|>1.96) [38]. See Multimedia Appendix 3 for a detailed description of the deterioration calculation.

The Quick Inventory of Depressive Symptomatology Self-Report-16 (QIDS-SR 16) was used as a second measure to assess the level of depressive complaints [39]. The MINI [31] was used to evaluate MDD and comorbid anxiety disorders at baseline and at the 12-month follow-up to assess depression-free episodes (remission).

Each participating country collected data in accordance with a predefined data codebook, which was then uploaded to a centralized database (Research Department of GGZ inGeest Mental Health Care). The encrypted data, free of identifiable patient and therapist information, complied with General Data Protection Regulation regulations [33].

Analyses were based on the intention-to-treat (ITT) principle. We initially considered a per-protocol analysis, as ITT approaches in noninferiority trials may dilute between-group differences by including nonadherent participants, potentially biasing results toward noninferiority [48]. However, we ultimately adopted an ITT analysis because per-protocol can compromise the benefits of randomization and introduce selection bias. In addition, ITT better reflects routine clinical practice, where nonadherence is common, thereby enhancing generalizability to routine care settings. It also captures real-world effectiveness, aligns with health economic evaluation objectives, and accounts for variability in adherence and outcomes.

Descriptive statistics were used to assess baseline characteristics of the participants (Table 3). Multiple imputation was conducted using chained equations with predictive mean matching until the loss of efficiency was less than 5%, resulting in 20 complete datasets [49]. Linear mixed modeling [50] estimated the mean levels of the outcome variable for the TAU group and the intervention effects at 3 time points, using 3 time dummies (3-, 6-, and 12-month measurements) and 3 time dummy-by-group interaction terms; hypothesis tests for these interaction effects were conducted 1-sided, in line with the noninferiority framework. Alongside these fixed effects (for the main estimates), the model included random effects (random intercepts and time slopes for countries) and allowed for correlated residuals (for measurements within individuals). Missing observations for the CSQ and SUS were not imputed, as the answers to these questions were highly subjective. This reduces the risk of bias due to incorrect imputation [51]; AD use at the 3-, 6-, and 12-month assessments was also not imputed. Estimated intervention effects (EE) were presented on the corresponding scales for the 4 outcomes. Cohen d and 95% CIs [52] were used to quantify between-group treatment effect sizes for both the PHQ-9 and the QIDS. Effect sizes were classified as very small (d<0.2), small (0.2-0.5), medium (0.5-0.8) or large (>0.8). MINI-D (Depression) and MINI-A (Anxiety) were presented as log-odds ratios using the Hasselblad and Hedges logit method [53]. A “leave-one-out” sensitivity analysis assessed the robustness of the overall effect by systematically excluding one country at a time and recalculating the effect size for the primary outcome [54]. To assess the impact of baseline AD use on treatment outcomes during treatment and follow-ups, we extended the linear mixed modeling by adding a third factor (AD use: yes/no). This allowed estimation of marginal mean PHQ-9 scores at different time points for the bCBT and TAU groups within strata of AD use, as well as effect sizes. We calculated a reliable deterioration in the PHQ-9 [38] during treatment (3 months), posttreatment (6 months), and at the 12-month follow-up point. The nonimputed dataset assessed the number of patients who deteriorated, while the log ratios and corresponding effect sizes were estimated using generalized linear (logistic) mixed models. Stata (version 14.2; StataCorp LLC) was used to conduct the analyses.

Cost-effectiveness analyses were conducted from the health care perspective, as this perspective best aligned with our hypothesis. This perspective is also recommended by the National Institute for Health and Care Excellence in the United Kingdom [56]. Health-related quality of life for the cost-effectiveness analyses was measured using the EQ-5D-5L [57]. EQ-5D-5L health states were converted into utility scores using the Dutch tariff [58], and quality-adjusted life years (QALYs) were calculated by multiplying utilities by patient time spent in each health state, with linear interpolation applied between transitions.

Within the multiple imputation procedure, predictive mean matching was used to account for the skewed distribution of cost data. Mixed-effects regression models with repeated measurements clustered within participants were used to estimate differences in overall PHQ-9 scores and the number of MDD cases prevented over the 12-month follow-up period. Differences in QALYs and costs were analyzed using linear regression models.

Incremental cost-effectiveness ratios (ICERs) were calculated by dividing the difference in costs between the 2 conditions by the corresponding difference in effects. Statistical uncertainty surrounding the ICERs was assessed using bias-corrected and accelerated bootstrapping with 5000 replications and visualized in cost-effectiveness planes. Cost-effectiveness acceptability curves were constructed to estimate the probability that bCBT was cost-effective compared with TAU across a range of willingness-to-pay (WTP) thresholds. The WTP threshold represents the maximum amount society is willing to pay for 1 additional unit of health gain. No established WTP threshold exists for improvements in PHQ-9 scores. For QALYs, however, a maximum WTP threshold of €50,000 per QALY gained was applied, in accordance with the Dutch Health Care Institute’s recommendation for interventions targeting conditions with an estimated disease burden of 0.48. Finally, a sensitivity analysis was conducted in which differences in costs and effects were adjusted for potential confounders, including baseline health care costs, age, sex, and treatment setting (primary vs secondary care) [55,56]. Analyses were conducted in Stata (version 14.2), and reporting adhered to the CHEERS (Consolidated Health Economic Evaluation Reporting Standards) [59].

Figure 1 illustrates the study’s participant flow. Of 1755 participants assessed for eligibility, 953 started treatments, of whom 118 were excluded from our analysis. Ten participants withdrew before randomization, and for 108, the 12-month assessment would be beyond December 31, 2017 (the funder-determined time frame). Thus, the ITT sample included 835 participants, with 423 in bCBT and 412 in TAU. The total study dropout rates for the sample were 17.5% at 3 months after baseline assessment, 26.7% after 6 months, and 33% after 12 months.

Table 3 summarizes the participants’ baseline sociodemographic, clinical, and cost profiles. Multimedia Appendix 4 presents participant counts and baseline characteristics for each country. The mean age of the total group was 39.17 (SD 13.18) years, with 67% female and half of the sample having a high level of education. The employment rate was 61.80%. The mean PHQ-9 score for the total group was 15.28 (SD 4.7), indicating moderately severe symptoms. Over half (420/835, 50.30%) of the participants had a comorbid anxiety disorder. At baseline, on average, 42.77% (352/823) of the participants used ADs, with no significant differences between the 2 groups. This percentage remained stable over time, averaging 42.65% (351/823) at 3 months, 38.40% (316/823) at 6 months, and 40.70% (335/823) at 12 months, with nonsignificant differences between the groups. At prerandomization, 56.41% (471/835) of the participants preferred bCBT, 16.41% (137/835) preferred TAU, and the remainder reported no preference. The bCBT group scored significantly higher on the credibility and expectancy scales for treatment than the TAU group, indicating moderate to high levels on both measures. Mean health care costs per patient 4 weeks prebaseline were €1172 (SE €185) for the bCBT group and €1338 (SE €183) for the TAU group.

Table 4 and Figure 2 show that, for the primary outcome (PHQ-9), bCBT was noninferior to TAU at 3 (Figure 2A), 6 (Figure 2B), and 12 (Figure 2C) months, as all upper bounds of the confidence intervals lay within the noninferiority margin of ≥0.20. A “leave-one-out” sensitivity analysis also showed noninferiority of the primary outcome at the 3- and 6-month assessments. However, at 12 months, the “leave-one-out” analysis revealed that excluding Germany yields inferior results. This indicates that the overall effect estimate is sensitive to this omission and suggests potential heterogeneity across settings. Together, these findings highlight uncertainty in the 12-month results and suggest that the observed effects may be context-dependent.

Table 7 presents several disaggregated health care cost categories and total health care costs per patient for bCBT and TAU at the 12-month assessment. These costs did not differ significantly between the 2 groups, except for home care costs, which were significantly higher for TAU than for bCBT. However, the average total health care costs per patient were slightly higher for the bCBT group than for the TAU group (estimated mean difference per patient €315, 95% CI −582 to 1163). This difference was primarily driven by the estimated bCBT intervention costs.

Table 8 presents the cost-effectiveness analysis results at the 12-month follow-up from a health care perspective. The mean overall change in PHQ-9 score over the 12-month period was small but significantly higher for bCBT than for TAU. Within the main analysis, the ICER for the PHQ-9 score was €643 per point of improvement, meaning that to gain a single point of improvement in the PHQ-9 score, an investment of €643 is required for bCBT compared with TAU.

The cost-effectiveness plane shows that in 73% of the bootstrapped simulations, bCBT was both more effective and more expensive (Table 7: northeast quadrant of the cost-effectiveness plane). Figure 4 present the cost-effectiveness planes, plotting differences in costs against differences in effects (PHQ-9 (Figure 4A), presence of MDD (Figure 5A), and QALYs (Figure 6A) between the 2 conditions combined with the cost-effectiveness acceptability curves (Figure 4B, Figure 5B, and Figure 6B) representing the probability that bCBT is cost-effective compared with TAU across a range of WTP thresholds (WTP for PHQ-9, presence of MDD, and QALYs). As no universally accepted thresholds exist for these outcomes (except for QALYs), WTP values were varied from €0 to €50,000 to reflect a broad range of decision-making scenarios.

Cost-effectiveness planes illustrate the uncertainty around the ICER across the 4 quadrants. In Figure 4, 73% of bootstrap replications fall in the northeast quadrant, indicating that bCBT is not only more effective than TAU but also more costly (PHQ-9). It also shows that, for a 1-point improvement on the PHQ-9 and a WTP of €0, the probability of cost-effectiveness was low (0.25), rising to 0.95 at a WTP of €3,800.

A similar pattern was observed for preventing an MDD diagnosis at follow-up (12 months; Table 7). The ICER for preventing 1 MDD case was €4381, indicating that this amount would need to be invested in bCBT rather than TAU to prevent an MDD diagnosis. As shown in Figure 5, 74% of the cost-effectiveness pairs fall within the northeast quadrant (more effective and more expensive), while the remaining 26% are in the southeast quadrant (more effective and less expensive). Figure 5 also shows that the probability of cost-effectiveness ranges from 0.25 at a WTP of €0 to 0.76 at a WTP of €10,000 for every MDD episode prevented.

For QALYs, the difference between bCBT and TAU was small and not statistically significant (Table 7), with considerable uncertainty indicated by the spread of bootstrap replications across all quadrants. The ICER indicates a cost of €36,729 per QALY gained. At a €50,000 threshold, the probability that bCBT is cost-effective is 0.55 (Figure 6). In the adjusted analysis, the cost difference between bCBT and TAU increased to €421 (Table 7), but this difference was not statistically significant, and the overall results were consistent with the main analysis.

Our hypotheses were that bCBT for patients with MDD would be clinically noninferior to TAU and more cost-effective from a health care perspective. The results showed that while bCBT was noninferior to TAU, it was not cost-effective. The first part of our hypothesis was confirmed, as all the differential effects observed for the primary outcome (PHQ-9) in the comparison between bCBT and TAU fell within the conservative noninferiority margin (Cohen d=0.20). Depressive symptoms showed a significant decline up to the posttreatment assessment, and this improvement was maintained at the 12-month follow-up. The posttreatment effects for both groups also aligned well with the benchmark (d=1.19 [60]) for the treatment of depression with CBT in routine care settings. Additionally, the long-term effects observed in both groups are comparable with those found in meta-analyses of CBT for depression [61,62]. The bCBT group also had a lower risk of experiencing an MDD episode than the TAU group at the 12-month follow-up (OR 0.67, 95% CI 0.45-0.99). AD use was evenly distributed between the 2 groups (approximately 40%) from baseline to the 12-month follow-up, with no significant differences in proportions. Subgroup analyses indicate that participants with and without baseline AD use benefited equally from treatment. These findings are comparable with those of a meta-analysis by Cuijpers and colleagues [63], which reported a similar percentage of AD use in psychotherapeutic RCTs and found that the effects of psychotherapy for adult depression were not concurrently mediated by pharmacotherapy use. Adverse treatment effects were limited, as indicated by the reliable deterioration rates (less than 5%) for the overall group up to the 12-month follow-up. These rates compare favorably with those found in a meta-analysis of deterioration rates within psychotherapies for depression [64].

Patients’ acceptance of treatment, in terms of both the credibility and expectancy of therapy, was significantly higher for bCBT than for TAU, as was patients’ satisfaction with treatment. This may be partly explained by the higher baseline preference for the bCBT group, which, in turn, was influenced by patients’ willingness to participate in the study. An additional explanation for these higher satisfaction ratings may be that the internet-based components of bCBT better support patients’ self-management than traditional face-to-face psychotherapy, as patients are less reliant on therapists only during their treatment. This may also partly explain why, although patients and therapists valued the working alliance as strong in both conditions, patients in the bCBT group gave higher ratings than those in the TAU group. Conversely, therapists rated the therapeutic alliance significantly higher in TAU than in bCBT, except on the bond subscale. This difference may reflect therapists’ preference for face-to-face contact, including access to nonverbal cues, as well as established professional norms [65]. Nevertheless, the observed difference was relatively small, and other studies have not found such disparities [16]. Overall, these findings suggest that a strong working alliance—encompassing both human and technological components—can be established between patients and therapists in bCBT, although many participants in our study were relatively new to this mode of treatment delivery. The high alliance ratings reported by patients in both conditions are consistent with findings from the meta-analytic review by Flückiger et al [66]. For a more detailed analysis of the association between working alliance and treatment outcomes for E-COMPARED, see the study by Doukani et al [67].

Only a few RCTs have directly evaluated the clinical effectiveness of bCBT for MDD compared with active face-to-face treatment controls, limiting direct comparisons with our study. These trials recruited outpatients with MDD, compared bCBT with ftfCBT or face-to-face psychotherapy, and had broadly comparable treatment durations and intensities (12‐20 weeks). Their findings are generally consistent with ours. For example, Berger et al [68] reported that CBT supplemented with internet-based modules was superior to ftfCBT at 12 weeks posttreatment (moderate effect) [68]. Similarly, Thase et al [69] demonstrated the noninferiority of integrated bCBT compared with ftfCBT 4 months after baseline assessment, whereas Kooistra et al [23] found no significant differences in depressive symptoms between the 2 treatment conditions. Berger et al [68] and Kooistra et al [23] also reported similarly low deterioration rates, good therapeutic alliance, and high treatment acceptability, with no significant between-group differences. These findings should, however, be interpreted cautiously because the available trials were relatively small and therefore provided limited certainty regarding treatment effects.

Since completion of our study, Schaeuffele et al [70] reported the results of the large pragmatic PsyTOM trial conducted in Germany, which compared a transdiagnostic blended psychotherapy intervention with routine face-to-face psychotherapy for patients with a broad range of mental health disorders. Based on a superiority design, the authors found no evidence that blended treatment resulted in greater clinical improvement or fewer face-to-face sessions up to 6 months after randomization [70]. Although these findings differ from ours, direct comparisons should be made cautiously because the interventions and study designs differed substantially. Whereas E-COMPARED evaluated a structured, integrated bCBT protocol specifically designed for patients with MDD, PsyTOM evaluated a highly flexible transdiagnostic blended intervention delivered across multiple therapeutic orientations. In addition, PsyTOM included patients with a wide range of mental health disorders, did not restrict inclusion to formally diagnosed disorders, and allowed therapists considerable flexibility in determining both treatment duration and the use of digital components. The PsyTOM authors therefore suggested that the absence of superiority may reflect the flexibility and heterogeneity of the intervention rather than the ineffectiveness of blended treatment itself.

From a health care perspective, bCBT was associated with nonsignificantly higher health care costs than TAU over the 12-month follow-up period. Although differences in clinical outcomes were modest, they consistently favored bCBT, with slightly greater improvements in PHQ-9 scores and fewer MDD episodes. Consequently, whether bCBT can be considered cost-effective depends on how much policymakers are willing to pay for these additional health gains. Evidence on the cost-effectiveness of bCBT compared with active face-to-face treatment remains scarce. This is not unexpected, as relatively few pragmatic trials have evaluated blended care under routine clinical conditions, and economic evaluations have only occasionally been conducted alongside these trials. The few available studies have also reported inconsistent findings, with some reporting lower health care costs primarily because blended care reduced therapist time, whereas others found no significant cost differences. These inconsistencies likely reflect differences in study design, intervention characteristics, health care settings, cost perspectives, and economic evaluation methods rather than true differences in the cost-effectiveness of bCBT [23,69].

This study is the first large-scale noninferiority trial to date, examining the clinical effectiveness and cost-effectiveness of bCBT compared with face-to-face TAU, predominantly ftfCBT. Noninferiority trials within routine mental health care remain scarce because of the large sample sizes required (small noninferiority margins necessitate high precision), the need for evidence-based treatment comparators, and evaluations under real-world conditions [71]. We addressed this gap by achieving a substantial sample size, using active comparators, setting a clear noninferiority margin, and conducting our RCT within routine care practices. Dropout rates in the total sample ranged from 17.5% at 3 months to 33% at the 12-month follow-up. The proportion of participants lost at 3 months (17.5%) was comparable with the average dropout reported in randomized trials of individual psychotherapy for depression in routine care [72], whereas the higher attrition at 12 months is consistent with the challenges of maintaining participant engagement during long-term follow-up. Despite this, dropout warrants careful consideration in future studies, as it may affect the validity of the findings and may indicate challenges with treatment adherence. The multifaceted nature of our study, which comprised diverse settings and technical platforms, enhances the generalizability of our results to routine care settings. Generalizability is further strengthened by including patients with and without AD use at baseline, reflecting the characteristics of patients with MDD in routine clinical practice.

Our study is not without its limitations, however. The sample size was smaller than planned due to time and budgetary constraints. However, our ability to detect noninferiority based on the predefined margin for the primary outcome during and after treatment and follow-up suggests that the sample size was sufficient to assess noninferiority [73]. This noninferiority of bCBT compared with TAU was also supported by the secondary QIDS-16 outcome at the 3- and 6-month assessments, consistent with the findings for the primary outcome (PHQ-9). However, the 2 depression measures no longer yielded consistent findings at the 12-month assessment, suggesting that the longer-term treatment effects should be interpreted with some caution. Next, all participants met the diagnostic criteria for MDD (MINI) and had a PHQ-9 score above 5 at baseline. However, because treatment was delivered within routine health care settings, participants received bCBT or TAU in either primary or specialized mental health care services. Consequently, the duration and intensity of treatment varied across settings; for example, specialized mental health care generally provided more frequent and longer treatment sessions than primary care. This between-setting heterogeneity represents a potential source of confounding that is inherent to pragmatic trials. Nevertheless, leave-one-out sensitivity analyses indicated that the overall findings were robust and were not driven by any single study site. Only the noninferiority analysis of the 12-month primary outcome showed some uncertainty after exclusion of 1 site, suggesting that the longer-term findings should be interpreted with some caution. Although this heterogeneity may have influenced the observed treatment effects, it also reflects routine clinical practice and therefore enhances the external validity of the findings.

The inability to blind patients and therapists, a common feature of psychotherapeutic studies, may also have influenced our results [74]. However, the assessors of the MINI interview were blinded, and other assessments were based on self-report. It was not possible to assess the therapists’ treatment fidelity, which may also have influenced the treatment outcomes. Additionally, some therapists delivered both treatments, which was necessary because the RCT was conducted within routine care. This may also have influenced treatment outcomes due to a preference for, or greater familiarity with, one treatment over the other [75]. Given that bCBT was still relatively new for most therapists, this could indeed have been an issue. Our cost-effectiveness analyses may also have been hampered by the sample size calculation, which was based on a noninferiority clinical margin rather than on economic outcomes. Thus, the sample size may have been underpowered to detect real effects in the cost differences between the 2 groups.

To the best of our knowledge, this is the first large-scale RCT evaluating the clinical effectiveness, cost-effectiveness, and safety of bCBT for MDD in routine care, compared with TAU. The demonstration of clinical noninferiority positions bCBT as an additional acceptable treatment option for patients and therapists in routine mental health care. However, large-scale implementation will also depend on more robust evidence of its cost-effectiveness. This may improve as blended care becomes more established in practice, with increasing familiarity among patients and therapists and further optimization and standardization of delivery processes. Future studies are advised to support our outcomes. Studies are also recommended to investigate in greater depth the facilitators and contextual characteristics of bCBT for successful implementation in routine care. Future studies are also needed to explore alternative modalities of “blending,” such as face-to-face psychotherapy combined with videoconferencing, virtual reality, or AI-supported interaction between therapists and patients based on natural language models, alongside active and passive EMAs.