Abstinence rates ranged between 12% [43] and 46.2% [42] for unguided CBIs (waiting list: 10% to 20% [43]), between 16.6% [45] and 46.6% [45] for guided CBIs (waiting list: 0% [45] to 20.7% [48]), and between 21% [62] and 66.7% [61] for therapist-delivered treatments (waiting list: 0% [60]; face-to-face: 25.8% to 53.3% [62]). In studies evaluating CBIs for subthreshold eating disorders, abstinence rates were between 37% [74] and 45.1% [75] (waiting list: 8% [74] to 26.9% [75]; bibliotherapy: 8% to 34% [74]). Mobile interventions showed abstinence rates between 29.4% [84] and 37.8% [85] (control: 18.1% [85]). Abstinence rates for individual studies are shown in Multimedia Appendix 3.

In the following, results of further relevant aspects of TBIs are presented. Note that not all studies mentioned so far provided data regarding these points.

Twenty studies provided information about non-take-up rates. These ranged from 2.9% [80] (carers) to 50% [82] (carers) with an average of 20.1%. However, reasons for non-take-up were rarely reported. Paxton et al [67] and Hoyle et al [81] indicated that persons did not initiate treatment due to technical difficulties or difficulties accessing a computer or the website, due to seeking alternative treatment, or due to not having enough time to take part in the intervention.

Murray et al [47] found no differences between those who started the program and those who did not, except for the expected personal usefulness of this kind of intervention. In the study by Leung et al [56], patients who took up the intervention had a lower BMI and were generally more educated. There were no differences in age or severity of eating disorder. Leung et al [77] reported that participants working through the motivational enhancement exercises had a slightly higher stage of change than those who did not take part in these.

Twenty-six studies provided data about compliance rates (we defined compliance as the full completion of the intervention regardless of whether participants completed posttreatment and/or follow-up assessment, and we only extracted data from studies that explicitly reported information on this). Between 18.4% [52] and 95.5% [51] completed the program (mean compliance rate in the TBI conditions: 57.6% of patients that were included for uncontrolled studies or RCTs).

Pretorius et al [57] found that higher baseline scores on eating concerns were associated with higher compliance in terms of number of completed sessions. Fichter et al [63] reported that patients with less pronounced compensatory behaviors and a lifetime mood disorder showed better adherence. The results by Leung et al [56] suggested that the motivational stage of change in patients and their treatment expectations predict compliance particularly well. These authors also named reasons for discontinuing: motivational difficulties, not enough energy and time for the program, loss of interest, or nonresponse.

Most of the studies found no differences between completers and noncompleters in sociodemographic or clinical variables or in baseline scores [55,66,67,73,75,77,84]. Serdar et al [70] reported that noncompleters had higher baseline scores in thin-ideal internalization. Ljotsson et al [49] found that completers showed less bulimic episodes at baseline. Likewise, Carrard et al [50] showed that completers and noncompleters differed in frequency of binges and vomiting.

Satisfaction was either measured by a satisfaction scale or assessed by qualitative comments. The majority of participants were satisfied with the programs and rated modules as pleasant, easy, and useful [46,71,72,75,78,96]. Patients also said that they would recommend the programs to others [54,72,74,78,97]. Some patients stated that they preferred participating in a program offered via the Internet over face-to-face treatment [71,73]. However, there were also patients who would recommend it only as an adjunct to face-to-face psychotherapy [52].

Follow-up dropout rates in the studies included in this review ranged from 4.7% [69] to 84.8% [46]. In terms of predictors of dropout, results were inconsistent. There were studies where no predictors could be identified [43], whereas others found that dropout could be predicted by higher anxiety scores, a lower hyperactivity, a lower minimum BMI, and lower reward dependence scores on the Temperament and Character Inventory-Revised (TCI-R [98]) [53]. Graham and Walton [46] reported that dropouts were characterized by higher scores on the “drive for thinness” EDI-3 scale. Fichter et al [63] observed that lack of time or not wanting to deal with the eating disorder prevented patients from taking part in the follow-up measurement. Bauer et al [85] found that follow-up dropouts did not differ from those who completed follow-up assessment in any of the baseline characteristics.

Several studies reported predictors of outcome of TBIs in eating disorders. A better state of general psychological health was found to predict a better outcome [54]. In relation to primary symptomatology, Marrone et al [99] identified reduction of binge eating as the best indicator of a positive treatment outcome (abstinence) at the 1-year follow-up, whereas for short-term follow-up (3 months), it was the reduction of purging behavior. Moreover, higher scores on the EDI “perfectionism” scale and on the Eating Attitudes Test (EAT) [100], as well as a higher minimum BMI significantly correlated with a better outcome [53]. Johnston et al [44] reported that patients who benefited the most from their intervention were those with high body shame at baseline. Fichter et al [63] found lower scores on the Structured Interview for Anorexic and Bulimic Syndromes (SIAB-EX [101]) “compensatory behavior” subscale and higher spontaneity to be predictive of successful relapse prevention. Mezei et al [65] identified a higher ratio of words related to family of origin, a higher BMI, and lower binging and emotional distress at baseline to be significantly associated with change in EDE-Q total. Furthermore, better compliance was associated with a better outcome also. Adherence and more spontaneity were also identified as long-term predictors for a favorable course, as were higher scores on the “ineffectiveness” scale of the EDI-2. In the “SALUT” program, improvement increased with the number of diary entries and the number of steps patients completed [50,54]. Even in email therapy, patients who displayed more commitment in terms of number of words written had better outcomes [59,60]. Paxton et al [67] and Serdar et al [70] also found a correlation between the number of sessions and outcome, whereas Zabinski et al [71] found no such correlation. Finally, Cardi et al [86] reported that the mean use of the vodcasts was correlated with the BMI change.

Six eating disorder studies presented information about therapists’ time and efforts in guided CBIs. It ranged between 45 minutes and 135 minutes per patient [47,79,80,86,102]. Therapists in the study by Leung et al [56] invested approximately 5-10 minutes per email. The intervention study of Ruwaard et al [74] included 25 scheduled therapist feedback moments, which took approximately 13 hours to complete. Regarding content, emails from therapists were predominantly supportive (95.4%) with only 14.7% of the emails containing at least 1 CBT comment and 13.6% at least 1 technical comment [102].

Only 1 study on videoconferencing for eating disorder patients investigated therapeutic alliance. Therapists experienced differences between the delivery methods in terms of adherence to therapeutic tasks, adherence to therapeutic goals, and therapeutic bond, whereas patients did not [103].

Only 1 study compared cost-effectiveness of face-to-face CBT and CBT delivered via telemedicine [104]. Efficacy was found to be comparable, whereas costs of telemedicine were lower, although still substantial.

Three studies provided information about adverse events occurring in the reviewed TBI studies. Sánchez-Ortiz et al [48] as well as Robinson and Serfaty [60] reported that no major adverse events occurred in their studies. Fichter et al [63] reported that 11 of 258 (4.3%) AN patients showed adverse events (BMI<13.5 during the course of the trial). Fortunately, no serious adverse event (acute suicidality, suicide attempts, or death) was identified.

Interest in research on TBIs for AN and BN has increased during the last decade. Forty studies whose outcome results were published in 45 papers fulfilled the inclusion criteria of this systematic review. Most studies dealt with guided CBIs (“Overcoming Bulimia,” “SALUT”) or programs for the prevention and early intervention of eating disorders. One has to consider that many of these treatments are derived from (guided) self-help interventions based on other media (eg, self-help manuals such as “Overcoming Bulimia”) that have been evaluated by a considerable number of researchers in the past and whose results have been summarized in a current meta-analysis [105]. Other approaches were rarely investigated: only 2 studies looked into videoconferencing and email therapy. Merely 3 studies evaluated unguided CBIs and 2 researched the potential of CBIs in relapse prevention.

With regard to BN, guided CBIs led to improvements in the core symptoms of binging and purging and global eating disorder psychopathology. Patients receiving guided CBIs improved more than controls. Guided CBI was shown to be as effective as guided bibliotherapy. Initial findings suggest that treatment results can also be maintained at follow-up. Furthermore, videoconferencing showed promise in treating patients with BN. Unfortunately, this approach has only been evaluated in 1 RCT so far.

With regard to AN, CBIs might be used for relapse prevention. However, only 1 study has empirically evaluated this kind of intervention in AN patients so far. Several case reports by Yager [106-108] suggest that CBIs (eg, email) may be effective when used as a therapeutic adjunct in the outpatient treatment of AN. In his opinion, CBIs increase adherence, patient satisfaction, and therapeutic alliance, and he did not observe any negative effects. Therefore, he suggests including email contact in the regular therapy process to enhance weekly sessions, to monitor daily food intake, or for crisis intervention.

With regard to adolescents with eating disorders, research findings suggest that CBIs may be a treatment option for bulimic patients. However, results should be confirmed in RCTs and replicated by other research groups before widely recommending it.

Furthermore, CBIs may also be considered in the prevention and early intervention of eating disorders as well as for supporting carers of eating disorder patients. Finally, preliminary evidence suggests that mobile interventions are useful for patients with eating disorders in relapse prevention or as an adjunct to therapy (eg, symptom monitoring).

Efficacy results of TBIs are in-line with studies evaluating CBIs in other mental health disorders. For example, medium effects at posttreatment were also shown in a meta-analysis of CBIs for depressive disorders [16]. Videoconferencing—the intervention that is closest to traditional face-to-face therapy—showed very promising results. Videoconferencing was also shown to yield similar clinical outcomes as face-to-face therapy with regard to a variety of other disorders [109]. Although this intervention provides no benefit for the therapist in terms of time or cost-effectiveness, it might be an attractive alternative treatment for patients living far away from a specialist eating disorder therapist. CBIs for carers of eating disorder patients are a relatively new field of research. In-line with Hu et al [110] who showed that Internet-based interventions were able to reduce stress and to improve well-being of caregivers for others with medical conditions, we found that CBIs may also be helpful for carers of patients with eating disorders.

Although efficacy results in the reviewed studies are promising, high rates of non-take-up, noncompliance, and dropout in the reviewed studies severely hamper the validity of study results. Therefore, findings must be interpreted with some caution. The result that only slightly more than half of the participants (57%) completed the offered TBI parallels the report by Waller and Gilbody [111] that only half of patients completed a full course of CBI for anxiety and depression. However, as adherence and satisfaction are important determinants of therapy outcome, it will take additional efforts to improve acceptance of TBIs for participants. This may be accomplished by educational advertising and by developing specific, individually tailored programs that appeal to participants searching for help. To increase adherence, future interventions should feature the following characteristics: a strong theoretical foundation, perceived personal relevance, tailoring, persuasive technologies, credibility, social networking, and common “push factors,” including human support and/or periodic prompts (by email or telephone), as suggested by Murray [112].

Guidance may be essential for both compliance and outcome of CBIs [113,114]. In the field of research on CBIs for eating disorders, only 3 studies directly compare unguided and guided CBI. Huon [45] found more improvement in the 2 guided groups. In contrast, Murray et al [115] did not find any significant differences between brief therapist guidance and minimal researcher guidance. However, this study was not an RCT, but rather compared the results of 2 consecutive cohorts. Hoyle et al [81] got mixed results on whether further guidance is useful. Eating disorder patients see guidance as a helpful and important element of the intervention [54,97].

Studies of CBIs for various mental health disorders showed that guidance augments efficacy [27]. Palmqvist et al [114] calculated a meta-analytic correlation of ρ=.75 between time invested by therapists and outcome. Outcomes equivalent to face-to-face therapy were also reported [116,117].

Overall, details about guidance are unsatisfactorily reported in most studies on CBIs for eating disorders. Information about the type and qualifications of coaches, about the type of support, as well as the timing, frequency, and overall amount of contact is often missing. However, all this important information is required to be able to compare the different programs and their efficacy and to get a clearer picture of how much and what kind of contact is needed to optimize the interventions.

The issue of human contact in guided CBIs also raises concerns about whether online therapy can establish any meaningful therapeutic alliance [118]. The only study in the field of CBIs for eating disorders suggests that videoconferencing establishes an adequate therapeutic alliance [113]. However, there are no findings regarding its quality when nonverbal cues are completely lost, such as in email guidance.

A systematic review investigating therapeutic relationships in e-therapy for mental health suggests that a therapeutic alliance equivalent to that in face-to-face therapy can be established in Internet-based therapy [119]. However, this topic is underresearched because just 11 (1.3%) of 840 reviewed studies reported data on this matter. Furthermore, Andersson et al [120] speculated that despite high alliance ratings, these relationships may not be as important as in face-to-face therapy. In conclusion, much more research is necessary to really understand similarities and differences as well as special features of therapeutic relationships in both delivery methods.

One strength of the present review is that we considered the whole spectrum of care from prevention, early intervention, treatment, relapse prevention of eating disorders, to interventions for carers of eating disorders. Furthermore, by including a wide range of study designs beyond just RCTs, a broad range of new types of TBIs were included. This made it possible to give a comprehensive up-to-date picture of the dynamic field of TBIs in eating disorders. Moreover, a wide range of aspects relevant to TBIs, such as acceptance, efficacy, predictors of outcome, need of guidance, and therapeutic alliance, were reviewed. This review is limited by the fact that a meta-analysis could not be performed due to the enormous heterogeneity of studies. Furthermore, diagnoses in studies were made in sometimes more but also less rigorous ways (informal clinical interview, semistructured clinical interview, questionnaire using DSM criteria), thereby reducing the chances that all studies had the same diagnostic threshold. Unfortunately, no pure samples of BN patients were available in most studies. Instead, patients with EDNOS or binge eating disorder were also included, so that results must be interpreted with some caution.

There are also several methodological limitations in the studies discussed in this review. We mainly followed the coding for weaknesses as suggested by Newman et al [121] (see table in Multimedia Appendix 1). Sixteen studies were uncontrolled and overall, 55% RCTs with either a control group or 13 with an active treatment as a comparator were still too few. For 11 studies, no follow-up data were available and in 28 studies outcomes were determined only through self-report and not by blinded interviewers. Sample sizes were predominantly moderate and rarely included more than 100 participants; 14 studies had sample sizes of less than 50. In most cases, no power calculation was reported. Fifteen studies did not report on additional psychological treatment, as did 22 studies on pharmacological treatment. Twenty-two studies failed to present information on adherence or quality checks in the treatment as did 18 studies on therapist training regarding TBIs. Most studies did not explicitly investigate outcome predictors or the active ingredients of mostly complex interventions. Outcome measures varied considerably making it hard to compare different study results.

Consequentially, a number of challenges for future research arise that are detailed subsequently.

There is need for more high-quality RCTs that adhere to the Consolidated Standards of Reporting Trials (CONSORT) statement [122], with sufficient sample sizes and long-term follow-ups to investigate stability of effects. Future studies should compare TBIs for AN and BN patients not only concerning waiting list or no treatment comparisons, but also with respect to active conditions such as face-to-face psychotherapy and traditional (manual-based) self-help interventions. Furthermore, different modes of TBIs as well as TBIs with different amounts of guidance should be compared with each other to derive information about comparative effectiveness. Future research should also evaluate the helpfulness of TBIs in preparing patients for face-to-face therapy, and explore the potential of combining face-to-face therapy and TBIs. Outcomes of evaluations should be preferably assessed by independent, blinded assessors and not by self-reports alone. To facilitate comparability of treatments, standardized outcome measures should be used. In addition, abstinence and compliance should be defined consistently across studies. Furthermore, reporting of studies should be improved. Studies should include information about additional psychological and/or pharmacological treatments to control for outside effects. Additionally, information about adherence or quality checks on treatment needs to be consistently reported. In addition, detailed descriptions of guidance should be given including information on qualification and on training of the coach/therapist. Modality, frequency, and intensity of guidance, as well as information about therapist’s time and efforts should also be specified. To improve the methodological quality and the reporting of studies, researchers should follow the guidelines for executing and reporting Internet intervention research [123].

Many questions regarding the optimal delivery method of TBIs, as well as the optimal dose in terms of frequency, intensity, and duration of interventions, still await qualified answers. Research shows that adding more therapy components does not lead to better results [124]. In this context, Andersson and Titov [125] pointed out that it is important to consider patients’ capacity and to avoid overloading. Studies should also try to find out how much time patients have to spend on a treatment program to achieve maximal benefits [24]. Similarly, questions regarding optimal guidance remain unresolved: research needs to identify the usefulness of clinical contact [24] and the ideal frequency and kind of provided support [125]. The issue of what background the therapist should ideally have and how much face-to-face contact a therapist needs to have in order to be empathetic, interested, present, and thus therapeutically effective, also needs clarification [126]. In case of engaging TBIs of adequate quality, therapist expertise may not be as important as in face-to-face therapy [125]. Furthermore, future studies should also include checking adherence to the intervention protocol of therapists who provide guidance and the quality of their guidance.

Usually TBIs are complex interventions (ie, they consist of a number of interacting components) [127]. Therefore, it is essential to explore mechanisms of action of TBIs and how therapy can be optimized by identifying the active ingredients. In addition, mediators and moderators of therapeutic change in TBIs should be explored. Future research will also have to clarify who benefits, who drops out, and what factors increase compliance. Moreover, the impact of sociodemographic factors, such as sex, age, and/or education, needs to be elucidated.

Integrating more motivational components into programs and tailoring the treatment to the AN and BN patients’ needs in terms of severity of illness and comorbidity will be imperative to further improve interventions. Research on depression has shown that more severely affected patients respond better to tailored, rather than to nontailored treatment [128]. Patients’ preferences should also be considered when designing future interventions (eg, by providing a choice of different treatment modules). Therefore, it is necessary to conduct qualitative studies to explore patients’ opinions and wishes concerning TBIs and to explore acceptance.

More research on how to enhance the uptake and utilization of TBIs and how to promote its broader dissemination and implementation in routine care is necessary. TBIs should be integrated into existing health care systems and efforts should be made to enhance their acceptability and adoption by patients as well as by health care delivery teams [129]. Many patients and clinicians still adopt a skeptical attitude toward TBIs, a problem that can be addressed through education [125]. More pragmatic RCTs in naturalistic settings should be performed to evaluate effectiveness and cost-effectiveness of TBIs within a stepped-care program [130]. This kind of research is able to identify where TBIs fit best into existing treatment options and to economically integrate them into our health care system [129]. TBIs may be presented as a first step in a stepped-care process probably followed by or combined with other treatment options such as face-to-face psychotherapy [125]. Cavanagh et al [131] reported a naturalistic nonrandomized trial of CBI for anxiety and/or depression that found CBI to be effective even under routine conditions, suggesting good generalizability.

TBIs also have some disadvantages. Some patients may not be able to access technological interventions [132] or may feel uncomfortable using them [129] due to security issues [125] and privacy concerns [133]. Moreover, the lack of nonverbal cues (eg, in email therapy) can lead to a loss of important therapeutic information and to misinterpretations. Additionally, the lack of immediate exchange may cause a patient who expects immediate response to feel insecure and helpless [133]. Furthermore, patients’ safety cannot be guaranteed and, in case of a crisis, there are few options for action. Kiluk et al [134] advocate that future studies should focus more on potential adverse events in TBIs before a final recommendation for TBIs is made. So far, adverse events have rarely been considered in reports of TBIs for eating disorder patients. Therefore, this issue should be more systematically addressed. In addition, future studies should focus on the identification of nonresponders or patients that even deteriorate (eg, by symptom monitoring on a regular basis) and on the development of treatment strategies for these patients.

Last but not least, research needs to determine the limits of TBIs [126]. In-line with Proudfoot et al [135], TBIs for eating disorders may not be indicated in case of a BMI<17.5, severe vomiting, comorbid posttraumatic stress disorder, severe depression, or a psychotic illness.

Unfortunately, the evidence for TBIs in the treatment of AN and BN remains insufficient because many approaches were investigated only once, by 1 research group, or only in uncontrolled studies. Nevertheless, the initial results are encouraging. At this stage, unguided CBIs cannot be recommended for the treatment of AN and BN, whereas guided CBIs may be a promising treatment approach, especially for BN. Videoconferencing may also be an approach worth pursuing further in research as well as in practice. Furthermore, Internet-based relapse prevention for AN inpatients may be an effective way to stabilize treatment success and to bridge the gap between inpatient and outpatient therapy. Guided CBIs seem to be a promising approach even in the treatment of adolescents with BN. Efficacy of email therapy still remains to be seen. CBIs might also be considered for the prevention of eating disorders as well as to support carers of eating disorder patients. Furthermore, evaluation of mobile interventions should be further pursued.

Until now, many TBIs have only been used within the context of research studies and have not become part of routine health care. Before a widespread implementation of TBIs, it is imperative to ascertain that they are also feasible under naturalistic conditions and across settings.

In conclusion, one should stay open-minded about the integration of novel technologies that may enhance psychological prevention and treatment of AN and BN and their carers. TBIs can especially serve as a first step in a stepped-care model. However, patient compliance, which is essential for TBIs to work, is still a major challenge. Future research is needed before widely recommending TBIs for AN and BN.