We conducted a randomized, controlled efficacy trial in women with DSM-IV BN. The trial protocol, summarizing details on study design and intervention, has been published elsewhere [41]. Over a 4½-year period, patients were screened and recruited from 13 psychosomatic hospitals offering specialized inpatient treatment for eating disorders throughout Germany. Patients were eligible for inclusion if they were aged 17 years or older, fulfilled DSM-IV-TR criteria for BN at hospital admission, and had successfully completed inpatient treatment, defined by a reduction of binge eating and compensatory behaviors by at least 50% based on the past 2 weeks before discharge compared with admission. Patients were excluded if their BMI had dropped below 17.5 kg/m² during the inpatient treatment or if local hospital staff regarded them as unfit to participate in a Web-based program (eg, due to psychotic symptoms, acute suicidality, severe personality disorders, or language barriers).

Assessment points were as follows: prebaseline (T0; hospital admission), baseline (T1; hospital discharge), postintervention (T2; 9 months after baseline), and 9-month follow-up (T3; 18 months after baseline). Patients received up to 80€ for participating in all assessments, whereas there were no monetary incentives for using the Web-based intervention.

The study was approved by the local ethics committee of TU Dresden and by ethics committees of all other federal states in which participating hospitals were located. Written informed consent was obtained from all patients and—in case of minors—their legal guardians. The study was conducted according to the Declaration of Helsinki and Good Clinical Practice principles.

Quality-control methods comprised case-report forms, independent data management, on-site monitoring, and documentation of adverse and severe adverse events. Data management was provided by the independent Clinical Trials Center in Marburg and included regular checks for consistency and plausibility, and queries if inconsistencies or missing data became evident.

Concealed randomization was carried out centrally by the independent Clinical Trials Center in Marburg after patients had been enrolled in the study and had given informed consent in a ratio of 1:1. The randomization was done with permuted blocks stratified by Center. Patients and psychologists involved in the moderation of the aftercare program could not be masked to intervention allocation. Assessors who carried out T0-T3 diagnostic assessments were blind to intervention allocation and neither involved in the moderation of the intervention nor in the final data analyses.

Within the first 2 weeks after hospital admission, potentially eligible patients were contacted by hospital staff and informed about the study. Patients willing to participate were interviewed over the phone by trained interviewers at TU Dresden to confirm a DSM-IV diagnosis of BN and to assess comorbidity and received login data to access the password protected Web-based trial platform. During the week before or after discharge, patients were interviewed and asked to complete the Web-based assessments again. Patients who had reduced their core BN symptoms by 50% compared with hospital admission were then randomized to the IN@ or TAU condition.

We designed IN@ to target the maintenance of inpatient treatment gains and reduce relapses after discharge. The intervention was based on principles of cognitive behavioral treatment for BN [14] and covered topics such as, eating behaviors and core bulimic symptoms, healthy exercise, body image, self-esteem, emotional and social skills. Table 1 provides an overview of the program content. Additional interactive features of the program were a monitoring log for bulimic symptoms, a body image, and a personal diary. Three clinical psychologists trained in behavior therapy for eating disorders guided the program, that is, provided individualized email feedback to entries in diaries and offered up to 9 monthly real-time individual chats of approximately 1 hour per participant. The program consisted of 11 Web-based sessions over 9 months with fortnightly sessions scheduled during the first 2 months after hospital discharge and monthly sessions thereafter.

The program’s home page also provided an overview of all participating hospitals and CVs of the program moderators once patients had logged in to facilitate the credibility of the intervention. Screenshots of the intervention are provided in Multimedia Appendices 1 and 2.

Patients assigned to the TAU group were assessed at all assessment points but did not receive additional treatment recommendations from the research team. However, most hospitals recommend some form of outpatient treatment following inpatient treatment. For ethical reasons, we did not interfere with these recommendations and subsequent decisions on concomitant treatment but documented it at all assessment points for both treatment arms.

The primary outcome was defined as abstinence from any core BN symptoms (binge eating, vomiting, laxative abuse, abuse of diuretics or other medication to control weight, and driven exercise) in the past 2 months before postintervention (T2) and not having resumed inpatient treatment after T1. Information on the primary outcome was derived both from the Structured Interview for Anorexia and Bulimia nervosa (SIAB-EX) and the weekly symptom checklists. If there was contradictory information from these two sources, the patient was classified as nonabstinent. Secondary outcomes were abstinence from any core BN symptoms in the 2 months before follow-up (T3) as defined earlier, the frequencies of episodes of binge eating, vomiting, and all compensatory behaviors per week in the past 3 months before postintervention (T2) and follow-up (T3), and recovery defined as not fulfilling DSM-IV-TR diagnostic criteria for an eating disorder anymore at postintervention (T2) and follow-up (T3). As abstinence rates are strongly influenced by definitions of abstinence [17], we chose to align our definition of abstinence with previous recommendations [17]. Regarding the frequencies of binge eating and compensatory behaviors as well as recovery, we aligned our definitions with the DSM-IV-TR criteria.

Because hospitals were located nationwide and patients lived in and returned to different parts of Germany, all interview assessments were carried out over the telephone. At all assessment points, trained assessors blinded to patients’ group assignment assessed eating disorder-specific and general psychopathology. We applied the SIAB-EX [42] to measure eating disorder symptomatology and general psychopathology and ascertain the diagnosis of BN at all assessment points. The interview was slightly modified by adding continuous items measuring the absolute frequencies (rather than categorical items of frequency spans) of binge eating and compensatory behaviors over periods of 2 weeks, 2 months, and 3 months. At prebaseline, we also used the Structured Clinical Interview for DSM-IV axis I mental disorders (SCID-I [43]) to measure psychiatric comorbidity. All interview documentations were monitored and checked for plausibility by the study coordinator and the independent monitor, and assessors received regular feedback based on this monitoring.

Questionnaire assessments (for further moderator as well as mediator analyses) and the weekly symptom checklists were integrated into the Web-based platform that also hosted the intervention. Patients in both the intervention and the TAU group were prompted regularly to report the frequencies of all core BN symptoms.

At T2 and T3, we also assessed the amount of any additional in- and outpatient treatment patients had utilized during the IN@ intervention and the follow-up period (in-, out-, day-patient therapy, days of treatment, number of outpatient treatment sessions, and number of therapists).

Sample size calculations were based on average rates of patients who do not remain abstinent of binge eating and compensatory behaviors after having achieved abstinence during treatment [17]. We assumed an average abstinence rate of 15% for TAU and of 35% for IN@, that is, we considered a difference of 20 percent points between groups, a clinically relevant reduction in core BN symptomatology. Using standard sample size formulae for the comparison of two rates by Fisher exact test at an alpha-level of 5% (two-sided) and statistical power of 85%, we calculated the minimum sample size to be n=90 for each group or a total of N=180 patients. Assuming a dropout rate of 30% at postintervention, a total of N=258 patients had to be included in the study.

We analyzed the primary outcome by intention to treat (ITT), which included all patients who had been randomized, using logistic regression analyses adjusted for prebaseline binge eating frequency, frequency of compensatory behaviors, BMI, SIAB general psychopathology, and baseline abstinence. If no data (from SIAB-EX or weekly symptom checklist) were available for a patient or if data from both sources were contradictory, she was considered nonabstinent (worst case imputation). The primary outcome was also analyzed using the complete cases (CCs) that included only women who had completed postintervention assessments and the per-protocol sample (PP) which included all women of the IN@ group who had completed postintervention interview assessments and had actively engaged with the intervention, that is, opened at least 25% of program pages or participated in at least two one-to-one chats. For the TAU group, the PP was identical with the CC. Abstinence at follow-up was analyzed in the same manner. Group differences were analyzed by logistic regression analyses, adjusting for prebaseline binge eating frequency, frequency of compensatory behaviors, BMI and SIAB general psychopathology, and baseline abstinence.

We analyzed the frequencies of episodes of binge eating, vomiting, and of all compensatory behaviors per week in the past 3 months preceding baseline (T1, discharge), postintervention (T2), and follow-up (T3) using mixed effects model to account for the nested data structure of three observations across time within individual participants [44]. Longitudinal mixed effects modeling is the method of choice for intent-to-treat analyses to assess clinical trial outcomes [45]. Multiple imputation before conducting longitudinal mixed effects modeling does not add value and is not necessary [46,47]. Episodes of binge eating, vomiting, and other compensatory behaviors were measured as count data and modeled as log-linked dependent variables to estimate intervention effects on change in rate from baseline to T2 or T3.

The percentage of women who did not fulfill diagnostic criteria for any eating disorders at T2 and T3 were compared between the groups by Fisher exact test.

Moderator analyses were performed for abstinence (from binge eating and compensatory behaviors) at postintervention and follow-up using mixed effects model. Because IN@ represents a novel aftercare approach and findings on patient characteristics as predictors of outcome for ED treatment are inconclusive, we decided to employ an exploratory approach as proposed by the MacArthur Foundation [48,49]. Accordingly, in a first step, we calculated correlations (Kendall’s Tau) between all available prebaseline and baseline variables and abstinence at postintervention and follow-up. A variable was considered a potential moderator if the correlation between the variable and abstinence at postintervention or follow-up was >.20. In a second step, each candidate moderator variable was entered separately into a mixed effects model. A variable was considered a moderator if there was a significant group × variable interaction effect on either postintervention or follow-up abstinence. Finally, if abstinence was moderated by more than one variable, all moderators were entered in a mixed effects analysis to determine the final prediction model. Statistical analyses were performed using SPSS 22 (IBM, Armonk, NY) and HLM7 (Scientific Software International, Inc., Skokie, IL).

Between 2007 and 2012, about 2500 patients were admitted for eating disorder treatment to collaborating hospitals and assessed for eligibility. A total of 431 women gave written informed consent to participate in the trial, of which 364 were reached for the prebaseline assessments, and of those, 315 fulfilled DSM-IV-TR criteria for BN (Figure 1).

Inpatient treatment comprised CBT-oriented (22%) and mixed (psychodynamic with CBT-elements; 78%) treatment approaches. At T1, 274 patients completed the baseline assessment and of those, 253 patients had reduced their core BN symptoms by at least 50% and were thus included in the trial (which would still result in 84% power under the initial planning assumptions); 126 were randomized to IN@ and 127 to TAU. At T2, 88 patients of the intervention group and 79 patients of the control group had completed postintervention assessments resulting in a dropout rate of 30% in the IN@ group and 38% in TAU. At T3, 82 patients of the IN@ group and 81 of the TAU group had completed follow-up assessments resulting in an overall dropout rate of 34.9% in IN@ and 36.2% in TAU.

Table 2 summarizes baseline characteristics of all patients. There were no relevant baseline differences between groups regarding BMI, age, illness duration, frequency of core BN symptoms, and frequency of any prior (in- and outpatient) psychiatric or psychotherapeutic treatment. However, a history of AN was more prevalent in the TAU group (45.6% vs 32.5%), whereas a current anxiety disorder was more prevalent in the IN@ group (46.0% vs 32.3%) and a current substance use disorder was more prevalent in the TAU group (6% vs 1%). Approximately 85.3% of patients had already received some kind of psychiatric or psychotherapeutic treatment before hospital admission.

Overall, patients’ BMI was in the normal range. Patients reported illness duration of 7.2 years on average and reported high levels of binge eating and vomiting episodes. In the 2 weeks preceding hospital discharge (baseline), the average number of objective binge eating and vomiting episodes per week had dropped considerably and 54.9% of patients were abstinent from binge eating and compensatory behaviors. Current or lifetime comorbidity was high.

In total, 19 (15.1%) intervention participants never logged on to the program. Of the remaining 107 women, 46 (36.5%) accessed at least half of the intervention content. On average, participants opened 36% of the program pages (Median: 28%) and accessed 5 of 11 sessions (Median: 4 sessions). Furthermore, 47 women (37.3%) took part in at least one live chat. More detailed information on adherence and therapist time invested will be reported separately.

The analysis of the primary outcome showed that at postintervention, about 1 in 5 patients in the ITT sample reported abstinence from binge eating and any compensatory behaviors during the previous 2 months, with no significant difference between the intervention group (21.4%) and the TAU group (18.9%; Table 3). Additional analyses also revealed no significant differences between the groups: In the CC sample (n=167), abstinence rates were 31.0% in the intervention group and 28.2% in the TAU group (P=.53). In the PP sample (n=68), the abstinence rate was 30.1% (P=.76). Abstinence rates at follow-up (a secondary outcome) remained stable, with no significant difference between the intervention group (22.2%) and the TAU group (17.3%).

The frequency of binge eating episodes (Table 4) increased significantly in both groups after hospital discharge. At postintervention, the frequency of binge eating episodes in the intervention group was 28% lower than in the TAU group (3.4 vs 4.7 episodes per week), but the difference was not statistically significant. At follow-up, there was no difference between the groups (4.2 vs 4.1 episodes per week).

The frequency of vomiting episodes (Table 3) also increased significantly in both groups after hospital discharge (baseline). At postintervention, it was 46% lower in the intervention than in the TAU group (4.3 vs 7.9 episodes per week); this difference was statistically significant. At follow-up, the frequency of vomiting episodes in the intervention group was 41% lower than in the TAU group (5.1 vs 8.7 episodes per week), but the difference was no longer significant. The frequency of episodes of all compensatory behaviors (Table 3) increased significantly in both groups after hospital discharge (baseline). Although at postintervention the frequency of episodes of all compensatory behaviors in the intervention group was 30% lower than in the TAU group (6.8 vs 9.8 episodes per week), the difference was not statistically significant. At follow-up, the frequency of episodes of all compensatory behaviors in the intervention group was 35% lower than in the TAU group (7.2 vs 11.1 episodes per week), but the difference was not statistically significant.

At postintervention, 37 (43.5%) women in the IN@ group and 30 (36.5%) in the TAU group who provided data (CCs set) did not meet diagnostic criteria for any eating disorder anymore; the difference was not significant (P=.43); whereas, 25 women (29.4%) in the IN@ group and 37 women (45.1%) in the TAU group met criteria for full syndrome BN, 19 (22.3%) in the IN@ group and 9 (11.0%) in the TAU group met criteria for atypical BN, 1 woman in the IN@ group had crossed over to binge eating disorder and 1 woman in the TAU group had crossed over to binge-eating/purging type AN.

At follow-up, 44 (55.6%) women in the IN@ group and 39 (45.8%) in the TAU group who provided data did not meet diagnostic criteria for any eating disorder anymore; the difference was not significant (P=.27). However, 19 women (24.1%) in the IN@ group and 28 women (32.9%) in the TAU group met criteria for full syndrome BN; 14 women (17.7%) in the IN@ group and 17 (20.0%) in the TAU group met criteria for atypical BN; 2 women in the IN@ group had crossed over to binge eating disorder and 1 woman in the TAU group had crossed over to binge-eating/purging type AN.

The following variables were identified as candidate moderators for abstinence at postintervention and follow up: number of diagnostic criteria for BN still present at baseline (hospital discharge), BMI at baseline, abstinence from binge eating and compensatory behaviors, frequency of episodes of binge eating, vomiting and all compensatory behaviors in the 2 weeks preceding baseline, and SIAB subscale scores at baseline.

Mixed model analyses revealed a significant group × variable interaction effect only for abstinence from binge eating both at postintervention and follow-up (Table 5).

Within the subgroup of patients who had failed to achieve abstinence at baseline, women in the intervention group had 4.93 (95% CI 1.90-12.80) times greater odds of abstinence at postintervention and follow-up time points, on average (P=.001).

At the end of the intervention period, 13 women (14.8%) in the IN@ group and 13 women (16.5%) in the TAU group reported clinically relevant suicidal thoughts, with no difference between the two groups (Χ²₄=3.8; P=.44). One of these women from the TAU group reported suicidal thoughts for the first time at the end of the intervention period; all other women had also reported suicidal thoughts at prebaseline or baseline. Four women (two from each group) reported suicide attempts during the intervention period. Three of those women had comorbid major depression.

A high proportion of patients of both groups utilized psychotherapeutic or psychiatric outpatient treatment between baseline and postintervention as well as postintervention and 9-month follow-up (Table 6). Rates were almost identical in both groups for both time periods as were number of sessions (IN@: 44.9; TAU: 48.3). A small proportion of patients also utilized further inpatient treatment between hospital discharge and follow-up. Women of the intervention group who did not utilize any further treatment during the intervention period were more likely to terminate the IN@ intervention prematurely (7 out of 14 women [50.0%] opened less than 25% of program pages) than women who received additional CBT (7 out of 30 women [23.3%] opened less than 25% of program pages) or psychodynamic therapy (10 out of 26 women [38.5%] opened less than 25% of program pages). However, the type (CBT vs psychodynamic) or amount of TAU treatment was not significantly related to any of the reported outcomes.

This is one of the first studies to evaluate the efficacy of a program targeting maintenance or improvement of treatment gains achieved during inpatient treatment in patients with severe BN. On the basis of ITT analysis at T2, we found that individuals receiving IN@ did not differ significantly from individuals receiving TAU in rates of abstinence from binge eating and any compensatory behaviors (21.4% vs 18.9%) and these rates remained stable at follow-up. They were higher in the CCs (and PP) sample but also not significantly different between groups. Although the intervention resulted in a 28% lower frequency of binge eating episodes at posttreatment compared with the TAU group, this difference was not significant. However, at postintervention, the IN@ group achieved a 46% lower frequency of vomiting episodes and this difference was significant. At follow up, this frequency was still 41% lower than in the TAU group but the difference was no longer significant. Moderator analyses revealed that both at post-intervention and follow-up, women of the intervention group who still reported binge eating and compensatory behaviors after the inpatient treatment had higher odds of achieving abstinence at post-intervention and follow-up compared with women of the TAU group who still reported binge eating and compensatory behaviors after the inpatient treatment.

The results of this study have to be discussed in the light of several important points: (1) general lack of comparison of maintenance studies for patients with BN, (2) illness severity, (3) additional treatment utilization and health care system in general, and (4) intervention characteristics.

For the primary outcome, abstinence of binge eating and any compensatory behaviors, only one other study for patients with BN has demonstrated that an aftercare intervention yields a significant reduction in abstinence while one other study did not find any differences on a number of self-report measures between the groups. In the first study, a 16-week, SMS for patients with BN after hospital discharge was investigated [36]. This relatively short intervention yielded significantly larger abstinence and partial remission rates at 8 months after hospital discharge than our intervention but the SMS-study differs in terms of duration of intervention and follow up, diagnoses (more EDNOS cases), and symptom frequency from ours. Additionally, treatment utilization in this study was 8-9 sessions over the 8 months of the intervention and follow-up compared with 27-29 sessions over 9 months of the intervention in our study. Because this study [36] was carried out as part of the same health care system as our own study, patients were probably not as severely ill as patients included in our study. In the second study [39] published after the starting of our study, both the Web-based support group and the waiting list group improved during the 4-month intervention with no significant differences between groups. Neither abstinence rates nor follow-up data were reported.

A second important point for the interpretation of the results of this study regards illness severity of the included patient group, indicated by symptom frequency, illness duration, and previous in- and outpatient treatment utilization. Patients in our study reported an average frequency of almost 15 binge eating episodes and over 18 vomiting episodes per week during the preceding 3 months before hospital admission. Although patients treated as inpatients may have higher frequencies of binge eating and purging episodes in general, these numbers clearly exceed frequencies reported in both earlier and more recent intervention trials for BN (eg, [50-52]). Our sample may therefore represent a more severely ill sample of BN patients which may limit the generalizability of the results.

The third crucial issue to be considered in the interpretation of the results is the amount of additional treatment utilization. Our study was originally planned to compare the efficacy of an aftercare intervention with TAU. However, at the time of planning the study, it was unclear what the “usual” amount of further treatment utilized by patients after hospital discharge would be, that is, whether and to what degree patients are recommended to take up further treatment following their inpatient stay and to what degree they would follow this recommendation. In this study, treatment utilization was surprisingly high: 82% of both groups engaged in additional treatment with 2-3 sessions per month during the 9-month intervention period and even higher numbers over the 9-month follow-up period. For patients, as severely ill as in this study, a Web-based aftercare intervention may have been less acceptable as stand-alone treatment compared with face-to-face treatment.

Finally, treatment utilization in this study may also be reflective of the specific health care context in which the study was conducted, that is, access to outpatient psychotherapeutic care and coverage of up to 80 sessions of outpatient psychotherapeutic treatment by health insurances in Germany. However, waiting times can be 6-12 months depending on regional differences. Because patients originated from different parts in Germany, the amount of further treatment utilization is even more surprising.

The characteristics of the intervention itself may also need to be discussed for the interpretation of the results. Adherence is a problem for Web-based interventions in general, also labeled as “law of attrition” [53]. Compared with targeted preventive interventions for eating disorders in student populations, in which adherence, defined as the number of opened pages and/or completed sessions, ranges between 50% and 80% [40], adherence to the IN@ intervention was lower. Participants opened on average 36% of the program pages; 15.1% of patients never logged on to the program after randomization and only 19.1% completed all 11 sessions of the program. However, the duration of preventive interventions is usually 8-10 weeks compared with 9 months in our study. For the included patients, the frequency of provided Web-based sessions especially during the first months may have been too low and the duration of the intervention and of individual sessions may have been too long. Future maintenance interventions may benefit from an overall shorter intervention duration and shorter, more frequent sessions which may also improve adherence to the intervention.

Although the intervention did not have a significant effect on abstinence rates at T2, the results of secondary outcomes, specifically vomiting rates, and moderator analyses may have important clinical implications: In the IN@ group, rates of vomiting episodes were almost half the rates of patients in the TAU group. From a clinical perspective, reducing vomiting episodes is often more difficult for patients than reducing binge eating, and while patients may have given up or reduced the frequency and quality of binge eating episodes considerably, vomiting is often the more persistent symptom. Because vomiting is also specifically associated with severe medical complications, its reduction during treatment is of essential clinical importance [54]. However, despite the marked reductions in vomiting rates during the intervention and follow-up, average rates were still above the diagnostically relevant threshold.

The intervention also proved to be specifically beneficial for patients who had not achieved abstinence at the time of hospital discharge. Thus, for a subgroup of potentially even more severely ill patients, participating in the aftercare intervention turned out to be specifically advantageous. In our study, only 54.9% of patients achieved abstinence of all BN symptoms during inpatient treatment in the past 2 weeks before baseline. Accordingly, a considerable proportion of patients were in need of further or more specific support to achieve full remission. Generally, the lack of studies examining short- and long-term effects of inpatient or residential treatments limits the comparability of these outcomes [55]. However, studies addressing long-term course and outcome of BN often demonstrate that relapses after the end of treatment are common and that patient status at posttreatment does not necessarily predict recovery status at follow-up [56]. The fact that patients in this study did not maintain improvements achieved during inpatient treatment and that a large proportion of patients did not improve markedly despite further treatment utilization and participation in IN@ may also demonstrate the need for a stepped—and more consistent and specialized—aftercare treatment provision. TAU after inpatient treatment is obviously provided irrespective of previous treatment gains or further needs in specific symptom domains. Although for some patients the aftercare intervention may have been sufficient to maintain their treatment gains, others may have needed even more intensive outpatient face-to-face treatment. To answer the question of who might need which amount of further Web-based or face-to-face treatment would require a change in current provision of health care delivery based on evidence rather than nonevidence based exploitation of existing health care structures.

Our study includes one of the largest sample sizes of patients of intervention trials for patients with BN, which clearly is a strength of the study. Follow-up duration is relatively long, BN symptoms and diagnoses were ascertained by a well-validated clinical interview, and the number of adverse events was not higher in IN@ compared with TAU. In a stepped-care model, the intervention could represent an economic alternative to face-to-face treatment if provided in a shortened version as first step and full alternative to TAU.

Limitations of the study are moderately higher dropout rates compared with face-to-face intervention trials and low adherence to the Web-based intervention. Future research efforts should therefore also be directed at adapting the intervention to increase adherence to the program and reduce dropouts.

Taken together, data from this study suggest moderate effects of IN@ for patients with severe and chronic BN. With regard to abstinence rates, IN@ did not show a marked superiority over TAU. However, the intervention had encouraging effects on vomiting rates and might also be particularly beneficial for women who have not achieved abstinence during previous inpatient treatment when compared with TAU.