The search was not limited by date, language, or publication status. We contacted first authors of all included publications for additional information on further (un)published trials and specific study information (see Köhnen et al [15] for details on the literature search/strategy).

See study protocol [15] for more details on eligibility criteria. Our inclusion criteria were (1) at least 80% of sample having a diagnosed unipolar depression (assessed by criteria of a formal classification system or by conducting a diagnostic interview [eg, F32.x, F33.x, or F34.1 according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision]) with any comorbidities in the acute treatment phase for depression and consisted of adults aged 18 years and older, (2) intervention was at least partly delivered through technical devices (eg, telephone, smartphone, computer), (3) intervention was based on an explicit psychotherapeutic theory, and (4) study was an individual or cluster RCT.

Our exclusion criteria were (1) participants were solely diagnosed by applying cutoff scores on symptom severity scales or when they had a depressive episode in the course of a bipolar disorder, (2) concurrent conditions (either somatic or mental) were the focus of the intervention, or (3) intervention provided solely psychoeducational content, patient decision aids, or depression management tools or focused exclusively on medication adherence.

Since we placed a special focus on application formats in this review, they are presented visually in Figure 1. We applied a rather broad definition for blended treatments, since we included all studies that provided any type of f2f treatment tailored to depression (eg, psychotherapy, medication, depression specific general practitioner care) in addition to TBIs irrespective of the study’s definition/label. In contrast, trials concurrently providing TAU in addition to TBIs were not considered blended treatments (but considered for the comparison TBI vs TAU) if TAU consisted of systematically offered generic treatments (eg, general practitioner care for all participants) that were not specifically tailored to depression. Since RCTs for blended treatment may be delivered in different designs (eg, superiority, noninferiority) resulting in content-related heterogeneity of interventions (eg, fewer therapeutic contacts), we decided to conduct meta-analyses separately.

The study flowchart is presented in Figure 2. Electronic searches yielded 20,603 records. After deduplication, 15,546 records were screened by title and abstract. Two reviewers (MK, SL) independently screened the first 100 records for inclusion. Since the interrater reliability for this sample was found to be high (98%), only one reviewer (MK) screened the remaining records in the course of the title/abstract screening. The second reviewer (SL) assessed publications labeled unclear by the first reviewer. Selected full-text articles (n=901) were subsequently assessed for inclusion by 2 independent reviewers (MK, MD). Discrepancies were resolved by discussion with a third reviewer (SL). In total, 241 publications representing 143 trials (83 completed studies and 60 ongoing studies awaiting further classification) fulfilled all inclusion criteria for the TIDECA study [11]. Of those, 78 completed studies assessed the acute treatment phase.

See Köhnen et al [15] for detailed information on extracted data and extraction procedure.

Risk of bias was independently assessed by 2 reviewers (from a group of 5 reviewers: MK, EW, MD, SL, TS) following Cochrane guidance (including the following domains for RCTs: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other bias) [16]. In line with a previous operationalization [17], we specified the domain other bias using the following 3 categories: insufficient treatment adherence, allegiance bias, and attention bias. Selective outcome reporting was categorized as unclear risk (trial registration or study protocol were missing or there was a deviation in one secondary outcome) or high risk (there were deviations in one primary or ≥2 secondary outcomes that could not be justified by the study authors). Disagreements were resolved by discussion or by consulting another reviewer (SL). Interrater reliability for risk of bias ratings was calculated to be 74%.

Meta-analyses were computed applying random effects models [18] since we assumed that heterogeneity regarding the sample, treatment, and methodological features of the included studies would be best captured by assuming that moderately diverging study-specific effect estimates are distributed around a grand mean [19]. Results were visually displayed as forest plots.

Continuous data (posttreatment and long-term depression severity) were analyzed as standardized mean differences (SMDs). Dichotomous data ([any] dropouts) were analyzed using the risk ratio (RR). We calculated 95% confidence intervals for all estimates. In addition, we computed 95% prediction intervals (PIs) for meta-analysis (when possible) capturing the range in which the effect of a new study (in a different setting) is expected; PIs can be very imprecise when only a few studies are considered [20].

Studies with multiple treatment groups were considered by combining data from interventional study arms (ie, pooling of means and standard deviations for continuous data and summing up sample sizes and people with events for binary data) when possible to avoid a unit-of-analysis error [16].

In cases of missing or unclear data, we contacted the corresponding authors. Intention-to-treat (ITT) analyses were used when reported by the included studies. When ITT data were not reported, we used the analysis defined as primary by the authors of the trial. Data on dichotomous outcomes were excluded from data analysis if there were no events in either study arm, since the direction and magnitude of a potential effect is not indicated [16].

We assessed statistical heterogeneity in the included studies by using a Cochran Q test and quantified it using the I² statistic [21]. As defined in the study protocol [15], we considered I² values of 50% or more as indicators of relevant statistical heterogeneity requiring further exploration. If indicated, we explored heterogeneity either quantitatively by means of a priori (see Köhnen et al [15]) and post hoc subgroup analyses (if the number of studies was sufficient [≥10]) or narratively (if only a few studies were available [<10]). We tested for possible reporting biases and small-study effects using visual examination of funnel plots (when useful). Possible control interventions and comparisons of interests were prespecified in our protocol [15] and used to structure our results section. All meta-analyses were computed by using Review Manager 5.4 (Cochrane Collaboration); descriptive data (eg, mean age of included participants) and PIs were calculated using Excel 2013 (Microsoft Corp).

Overall, the selected studies (n=78) included 13,180 participants ranging from 14 to 1089 per study. The mean age of participants was 45.15 (SD 12.01) years, and two-thirds (8029/11981, 67.01%) were female. TBIs in the included studies were delivered as stand-alone interventions (61/78; 78%), blended treatments (12/78; 15%), collaborative care (3/78; 4%), or stepped care trials (2/78; 3%). Duration of TBIs ranged from 1 week to 52 weeks, with most interventions lasting between 6 weeks and 12 weeks (median treatment length of 8 weeks). Interventions of 8 weeks’ duration were the most frequent (26/89; 29%) in the included studies (see Multimedia Appendix 2 [22-99] for baseline diagnoses). TBIs were based on 13 therapeutic rationales with most (83/101, 82.2%) based on CBT approaches (see Multimedia Appendix 3 for details). Concerning the applied technical medium, most TBIs were delivered via the internet (55/101, 54.5%), followed by telephone (12/101, 11.9%), offline computer programs (8/101, 7.9%), and videoconferencing tools (3/101, 3.0%). Additionally, 22.8% (23/101) of interventions applied more than one technical medium (internet-based treatment plus telephone support was most frequently [17/101, 16.8%] combined). The most common source of risk of bias was nonblinding of participants and personnel, selective reporting, and other bias (especially due to insufficient treatment adherence; Figure 3; see Multimedia Appendix 4 [22-99] for details).

11 RCTs tested blended treatments against different f2f treatments. Six RCTs were identified combining TBIs with f2f psychotherapy versus f2f psychotherapy alone. In these trials, TBIs were delivered in addition to outpatient psychotherapy [26,52,97], inpatient psychotherapy [99], and psychotherapy treatment sessions where the setting was not specified [59,87]. Two RCTs were identified comparing a TBI in addition to medication versus medication alone [53,61], and 2 RCTs tested a TBI with f2f TAU against TAU [28,68]. Additionally, we identified one RCT [69] where blended treatment (f2f CBT and internet-based CBT) was provided alongside TAU (psychiatric treatment) compared to TAU. Overall, 8 superiority [26,28,53,61,68,69,97,99] and 3 noninferiority trials [52,59,87] applying blended treatments were identified.

Three RCTs were identified applying TBIs, which were tested against usual care arms [43,79,80], in the context of a collaborative care approach. TBIs delivered in the context of collaborative care trials yielded lower posttreatment (SMD –0.20, 95% CI –0.36 to –0.04; I²=0%) and long-term (12 months: SMD –0.23, 95% CI –0.39 to –0.07; I²=0%) depression severity compared to usual care arms (see Figure 11).

Two RCTs using TBIs in the context of a stepped care approach were identified in the course of the search update. The studies were too heterogeneous for evidence syntheses, since one study tested a stepped care approach (first step: internet-based treatment, second step: telephone-based treatment) against telephone-based treatment alone [67], and the other study tested an internet-based intervention against a waitlist control group as a first step within a stepped care approach [78].

Our study found that when compared to different control conditions, TBIs were more effective not only when delivered as stand-alone interventions but also when they were delivered as blended treatments or in collaborative care trials for people with diagnosed depression. Dropout rates did not differ between TBI and control conditions; however, assessment of TBI acceptance was limited due to underpowered comparisons. In addition, relevant statistical heterogeneity was a common finding for most meta-analytical comparisons. We included 78 RCTs comprising different application formats (stand-alone interventions [61/78, 78%], blended treatments [12/78, 15%], and stepped care [2/78, 3%] or collaborative care trials [3/78, 4%]), interventions, technologies for intervention delivery, clinical populations, and control groups.

TBIs showed comparable effects to f2f treatments. Our findings are in line with a previous meta-analyses that found equivalent overall effects when comparing internet-based CBT to f2f treatment for mental disorders and somatic conditions on posttreatment symptom burden for studies on depressive symptoms specifically and for dropouts rates [100]. However, both results should be interpreted with caution, since both evidence syntheses were based on a limited number of studies.

When TBIs were tested against TAU controls, we found medium-to-small effects favoring TBIs concerning posttreatment and long-term depression severity. TAU was heterogeneous and consisted mostly of a mix of treatment options depending on the resources and routines of health care providers, general practitioner care, or care delivered in outpatient clinics. In addition, two-thirds of the studies included for this comparison also provided TAU in the TBI condition. Our results are in line with 2 previous meta-analyses that found a small effect favoring TBIs in comparison with TAU [101,102].

TBIs yielded beneficial medium-to-small effects on posttreatment and long-term depression severity when compared to attention placebo controls. To our knowledge, there is no previous meta-analysis available on this issue. However, the results are comparable to those comparing f2f psychotherapy with placebo [103] and pill placebo control groups [104].

We found a large effect in favor of the TBI group compared to waitlist controls for posttreatment and long-term depression severity. Our findings are in line with the only existing meta-analysis investigating TBIs in people with diagnosed depression [13]. This is not surprising, as there was a high overlap between the included studies. However, we were able to include more RCTs (+10) for the comparison of TBIs versus waitlist controls due to broader inclusion criteria and an updated literature search. Thus, our analysis emphasizes the robustness of the previous findings. However, the funnel plot on posttreatment depression severity was asymmetrical, with an emphasis on small studies depicting large differences in favor of TBIs compared to waitlist controls. Nevertheless, this is not a clear indicator of reporting bias because there are other sources (eg, heterogeneity, poor methodological quality) causing funnel plot asymmetry [16]. Between-study heterogeneity seems plausible to partly explain asymmetry, since we applied broad eligibility criteria and suspicious studies differed from the others in terms of population (postpartum depression) or publication year (1990), potentially resulting in more elevated differences.

Finally, TBIs did not result in lower posttreatment depression severity scores than no-treatment controls. This was not reasonable to expect, since no-treatment controls are comparable weak control groups, such as waitlist controls, which yield large effects when compared to TBIs [13]. Moreover, based on study reports, it cannot be ruled out that people allocated to the no-treatment control group made use of other health services for depression complaints (eg, care by a general practitioner), thus questioning whether true no-treatment controls were applied.

We identified a small effect favoring blended treatments delivered in a superiority trial design compared to f2f treatments concerning posttreatment depression severity. Meta-analysis on blended treatments delivered in a noninferiority trial design (ie, substantial shortening of f2f contacts) did not reveal differences in posttreatment or long-term depression severity or on dropout rates compared to f2f treatments. To the best of our knowledge, there is no previous meta-analysis investigating the effectiveness and acceptance of blended treatments in people with depression. Additionally, despite extensive discussions on their potential usefulness for mental health care [105,106], there is no uniform definition of blended care/treatment as they are operationalized in different ways and rationales for blended treatments are often missing [105]. This was also the case in our study, since the concept of combining a TBI with an f2f treatment was usually explained insufficiently or not at all. In the included studies, it appears that blended treatments were implemented based on the motto more is more (intensification of the therapeutic dose by providing add-on treatment following a superiority trial design). Nevertheless, future studies could define and investigate more sophisticated variants of blended treatments, since there are many useful possibilities to enrich onsite therapy by, for example, fostering preprocessing and postprocessing of sessions or for diagnostic purposes in everyday life (eg, self-monitoring) [106].

TBIs delivered in the context of collaborative care yielded small effects on posttreatment and long-term depression severity when compared to TAU controls. However, findings should be viewed with caution, since only a few studies have been available until now, and investigated collaborative care approaches are heterogeneous. The identified posttreatment and long-term effects on depression severity are comparable to reported effects investigating collaborative care approaches without TBIs in comparison to usual care [107,108]. However, we do not know if and how much the technology-based component is involved in the effectiveness of these interventions, since collaborative care approaches are complex. Testing collaborative care approaches with and without a TBI component may help to determine the add-on benefit of this element and may be concurrently useful for a comparative cost-benefit analysis.

Our review was conducted in line with Cochrane standards [16] and reported following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [109]. Additionally, studies were selected according to prespecified criteria [15]. We conducted a highly sensitive literature search considering key databases, databases of grey literature, and clinical trial registries without limiting the literature search to language. However, because of the extensive literature search, we deviated from the study protocol by omitting the forward and backward reference search. We structured and synthesized evidence using prespecified comparisons defined in the study protocol covering different application formats of TBIs in the acute treatment phase.

We applied broad inclusion criteria [15] contributing to observed heterogeneity regarding interventions, technologies for intervention delivery, psychotherapeutic rationales, and clinical populations in the included studies. Unfortunately, we were not able to explain statistical heterogeneity quantitatively (eg, by subgroup analyses) for most comparisons, since often only a few studies were available. However, we tried to explore heterogeneity narratively in these cases. In addition, when heterogeneity of the included studies is present (ie, I²≠ 0), the CI covers a narrower range than the PI of the respective comparison. Thus, pooled effects (SMDs) should be interpreted with caution: It may be that even if the pooled effect is significant (ie, CI not crossing null), the corresponding PI covers the null effect, meaning that in a new study conducted in a different setting (eg, different population), null treatment effects or effects in the other direction (harmful) may occur [20,110].

Although some information on dropouts [11] or treatment adherence [111] is addressed by most RCTs in this field, a comprehensive assessment of TBI acceptance was only partially possible, since data on dropouts were either missing or not usable (eg, data were only provided for one arm) or meta-analytic calculations were not possible (when no dropouts occurred in both study arms).

Considering the risk of bias ratings when interpreting the results, we found that the most common source of risk of bias was nonblinding of participants and personnel, followed by selective reporting and other bias. However, blinding of study participants is rarely possible in trials on TBIs.

TBIs delivered as stand-alone interventions, blended treatments, or in collaborative care trials yield mostly beneficial effects in people with diagnosed depression. By investigating different application formats of TBIs, people being diagnosed with depression, and the long-term effectiveness of interventions, our results may be especially helpful to inform routine care. Given the potential transferability of our findings to routine care, we think that our findings may represent effectiveness (effectiveness under routine care), rather than efficacy (effectiveness under ideal conditions) of findings. Additionally, our results show a very consistent image of TBIs (it works), despite the clinical and methodological heterogeneity of the included studies.

However, there are still open questions that need to be addressed in future research. Even though dropouts are by far the most reported indicator for treatment acceptance/patient safety in studies with TBIs [11], data were often not usable for data synthesis resulting in underpowered comparisons for safety/acceptance assessment. Therefore, our findings with regard to this outcome should be interpreted with caution.

Additionally, safety assessments of TBIs considering different types of safety measures in people with diagnosed depression have not yet been conducted. Thus, to obtain a more comprehensive impression of the safety of TBIs, we suggest including all indicators according to Rozenthal et al [112] to evaluate negative events: (severe) adverse events, dropouts, nonresponse, novel symptoms, and unwanted events.