Participants were recruited between March 2011 and March 2012 through newspaper, magazine, and radio announcements as well as via social networks and reports on websites for people with eating disorders. Potential participants completed a self-report screening battery before treatment. The inclusion criteria included female gender and at least one of the following eating disorder symptoms once or more per week within the past 4 weeks (assessed with the Short Evaluation of Eating Disorders [40]): purging, dieting, or excessive exercise. Additionally, a self-reported BMI greater than 15 kg/m² and less than 30 kg/m² was required. Participants who indicated no weight control behaviors at least once per week within the past 4 weeks and participants who reported binge eating only with the absence of any compensatory behaviors were excluded. Furthermore, those with severe depression (a score of 35 or more on the Center for Epidemiologic Studies Depression Scale, CES-D [41]), risk of suicide (assessed with items designed by the authors [13]), severe self-harming behavior (assessed with items designed by the authors [13]), psychotic disorders (a score of 13 or more on the Dutch Screening Device for Psychotic Disorder [42]), dissociative symptoms (a score greater than 8 on the Somatoform Dissociation Questionnaire [43]), substance abuse (a score of 10 or more on the Alcohol Use Disorders Identification Test [44] or the Drug Use Disorders Identification Test [45]), or in current psychotherapy treatment were excluded from the study (for more information on the inclusion and exclusion criteria, see [13]). The final sample consisted of 179 participants who commenced the online program (ie, completed the baseline assessment and received an invitation for the first session).

The online program ESS-KIMO (“Klärendes Internetprogramm zur Steigerung der Veränderungsmotivation bei Essstörungen” or “Internet program to enhance motivation to change in eating disorders”) aims to enhance motivation to change in women with symptoms of anorexia nervosa or bulimia nervosa. The methods and results of the ESS-KIMO have been published elsewhere [13]; this is a secondary analysis of the program. It is based on the transtheoretical model of behavior change [46] and uses the principles of motivational interviewing [47]. It contains 6 weekly online sessions comprising evidence-based interventions to enhance motivation to change [48,49], which are often used in conjunction with motivational interviewing. All aspects of the program and the study took place online and participants remained completely anonymous apart from providing an email address at which they could be contacted. The ESS-KIMO was designed based on the results of previous research [1]. It consisted of individual sessions, a closed website with screening for inclusion and exclusion criteria, and individualized feedback (from RvB or KH) for the required writing tasks in each session. In this feedback, the authors used methods and principles of motivational interviewing, such as selectively reflecting participants’ change or confidence talk and trying to enhance participants’ perception of discrepancies between their current (eating-disordered) behavior and their long-term values and goals.

In general, all sessions dealt with the participants’ ambivalence regarding behavior change. The session content included (1) the typical reasons for and against change in women with eating disorders (eg, participants were asked to make a list of pros and cons); (2) physical and psychological consequences of eating disorders and the eating disorder’s congruity or incongruity with life goals; (3) the eating disorder’s impact on the person’s daily life; (4) self-esteem and its link to the eating disorder; (5) a summary of the aspects that were considered important by the participant regarding behavior change; and (6) information regarding treatment. The content of each session (eg, information given) and the therapeutic tasks were standardized and were the same for each participant. All participants received an invitation to the next session via email 1 week after completing the previous session. During this time, an author (RvB or KH) wrote individualized feedback for the participant’s answer to the writing task. If participants did not log into the program during the 2 days following their invitation, a reminder email was sent on the third day. Up to 3 reminders were sent, each after 3 days. The effectiveness of the program was evaluated in a randomized controlled trial, in which the ESS-KIMO showed promising results for the completers of the program in terms of an improvement in motivation to change and eating disorder symptomology as well as an enhancement of self-esteem [13].

Ethical approval for the study was obtained by the German Psychological Society (Deutsche Gesellschaft für Psychologie, DGPs). Individuals interested in participating received information about the study and were informed that they could withdraw from the study at any time. Participants also received an email address so that they could contact one of the authors (RvB or KH) if they needed additional support. Participants were also given the telephone numbers of emergency contacts in Germany, such as the crisis helpline.

Several authors have pointed out the importance of clearly defining dropout [16,22]. Because we were interested in the predictors of treatment adherence in women with eating disorders, we only investigated dropout from treatment (treatment dropout) and not from the study (study dropout). That is, we only included participants who had completed all baseline measurements before treatment. Each dropout was patient-initiated; we did not recommend withdrawal from the program at any time. We defined dropout for a particular participant in a particular session when this participant completed the prior session but did not finish the current session during a 4-week time period. We then deleted the participant’s account because of our data privacy protection policy.

The present study aimed to test 14 variables as potential predictors of dropping out of the program. These were the EDE-Q scales Dietary Restraint, Eating Concerns, Weight Concerns, and Shape Concerns; participants’ BMI; the frequency of the core eating disorder behaviors binge eating, purging, and excessive exercise; the 4 URICA scales precontemplation, contemplation, action, and maintenance; depressed mood assessed with the HSCL-25; and participants’ age. As recommended [39], we used survival analysis (ie, Cox regression) to test the predictors. Because 91 participants dropped out, the number of events per predictor variable (EPV) was much lower than the recommended 10 EPV [63]. Therefore, we used the LASSO to identify relevant predictors [64] because this method yields reliable estimates in such scenarios [65]. Because this method is not invariant to linear scaling, we standardized the individual variables before the analysis. The variability of the LASSO estimates is typically assessed using a simple bootstrapping approach [66]. Specifically, we drew (with replacement) 1000 pseudosamples from our full sample and calculated the LASSO. For each pseudosample, the optimal lambda was calculated by using a crossvalidation procedure with 50 folds. The smallest lambda was then chosen to calculate the LASSO estimates, and the resulting estimates were recorded. Descriptive statistics (mean, SD, 95% CI) were used to inspect these estimates. The software R was used for data analysis [67].

The core demographic characteristics of participants at the beginning of the program are shown in Table 1. The maximum BMI reported (30.47 kg/m²) is based on participants’ responses to the EDE-Q directly before the beginning of the intervention, whereas eligibility for the study (ie, a maximum BMI of 30) was determined based on participants’ earlier responses to the screening battery. Some participants differed in their answers between the screening and preassessment time points given the time lag between these assessments (eg, participants in the waiting-list condition waited 8 weeks until the beginning of the program).

Of the 179 participants who completed the first session, 91 (50.8%) dropped out at some point during the program. On average, the participants completed mean 4.92 (SD 2.4) out of the 6 sessions. The dropout was equally distributed across the sessions (Figure 1), with approximately 10% of the participants dropping out after each session.

As shown in Table 2, the LASSO method identified the EDE-Q subscale Shape Concerns, frequency of vomiting within the past 28 days, frequency of binge eating within the past 28 days, and the HSCL-25 depression score with Cox regression as relevant predictors of dropout within the full sample. Results indicated that participants with high levels of Shape Concerns and depression as well as frequent binge eating episodes and vomiting were more likely to drop out of the program (see Figure 2). All other variables, including age, the URICA scores, BMI, and the EDE-Q scales Restraint, Eating Concerns, and Weight Concerns were not significantly related to dropout.

By using bootstrapping to estimate the variability of these results, we found that the 4 variables identified as predictors in the full sample were also the only variables that were identified in more than 50% of the bootstrapping samples (see Figure 3). Across samples, the frequency of binge eating, the frequency of vomiting, the EDE-Q subscale Shape Concern, and depressive mood were identified as predictors of participants remaining in the program. Importantly, the bootstrapped confidence intervals of these estimates did not include zero, indicating that these are larger than expected by chance effects alone.

The primary aim of the present study was to identify predictors of dropping out of an online program for women with eating disorders. We found that 50.8% (91/179) of the participants terminated their participation in the program prematurely. Our analysis identified 4 indexes of symptom severity that predicted dropout, namely the EDE-Q Shape Concerns subscale, frequency of vomiting within the previous 28 days, frequency of binge eating within the previous 28 days, and the HSCL-25 depression subscale. Neither motivation to change nor age predicted dropout.

The overall level of symptom severity in the present sample was high when comparing the scores of the present sample with the published norms of the EDE-Q [51]. This shows that a large proportion of participants suffered from significant eating disorder pathology, although diagnoses were not made by means of structured clinical interviews. Therefore, the sample seems to be suitable for drawing assumptions about women with eating disorders in general.

Similarly, the overall level of the dropout rate is comparable to that from other Web-based programs in general [17]. The current program was completely anonymous, which might have made it easier for participants to prematurely terminate their participation. Future research should evaluate the importance of anonymity in Web-based interventions for individuals with eating disorders and determine its role in dropout.

The dropout rate was equally distributed across the 6 sessions. However, it remains unclear whether the reasons for early versus late dropout are identical [22,39]. It is possible that early dropout is primarily because of participants’ dissatisfaction with the program, whereas late dropout may also be because of progress withdrawal (ie, participants feeling that they have already benefited enough) [22,39,68]. Speculative reasons for program withdrawal in the study may have included feeling dissatisfied with the motivational content and the absence of practical advice and strategies for overcoming the eating disorder, feeling overwhelmed by negative emotions when dealing with one’s own reasons for and against the eating disorder, or having resolved motivational issues faster than in 6 sessions [22]. Because dropout was distributed evenly across the 6 sessions, the specific content of any particular session can be discarded as a cause of dropout.

Although the high dropout rate is fairly typical for Web-based interventions [19], a high dropout rate is related to poorer outcome [16], meaning that it is important to understand the factors predicting early termination. Concerning the predictors of dropout, various indexes of symptom severity (ie, the EDE-Q Shape Concerns subscale as well as the frequency of binge eating and vomiting) predicted premature discontinuation of the Internet program. The finding regarding Shape Concerns is consistent with results from research in women with eating disorders in face-to-face settings [25,26] as well as previous Web-based interventions [6], which also showed that higher body dissatisfaction makes treatment dropout more likely. The finding that the higher the frequency of binge eating and vomiting, the more likely participants were to terminate treatment early is also in-line with previous findings. Accordingly, Kahn and Pike [26] and Woodside and colleagues [28] found that patients with the binge eating/purging type of anorexia nervosa were also more likely to drop out of face-to-face treatments. The commonly high levels of impulsivity in this patient group may also facilitate treatment dropout [28]. Because higher symptom severity is associated with longer duration of illness, it may reflect that these patients are also more reluctant to confront their maladaptive behavior [25]. The relationship between higher eating disorder pathology and premature termination of the online treatment program observed in the present study suggests that the online modality may be insufficient for those with more severe eating disorders.

We also found that participants with higher levels of depressive mood were more likely to drop out of our program. This finding is in accordance with previous findings from face-to-face settings for eating disorders [29,30] and a study on a Web-based treatment program for women with eating disorders [6]. However, other studies have reported a lack of association between depression and dropout from Web-based programs for women with eating disorders [37,38]. It seems plausible that an impact of depression which emerges in conventional therapies could also be relevant in Web-based treatment because the contents of the Web-based treatments were the same as the contents of face-to face therapies (ie, mostly consisting of cognitive behavioral therapy and motivational interviewing; eg, [5,36]). Participants in the present study showed high depression scores, whereas the participants of other Web-based interventions for eating disorders reported only moderate depressive symptoms [37,38]. Thus, it is possible that the effect of depression on treatment dropout only emerges for participants with relatively severe depressive symptomatology, for whom programs such as ours may be inadequate because they fail to target depressive features [13]. It is important to note that although we excluded participants with severely elevated depressive symptoms, our cutoff point of 35 on the CES-D is still relatively high [69], meaning that participants with elevated depressive symptoms were included in the study. Moreover, the fact that depressive mood was predictive of treatment termination although participants with extreme scores were a priori excluded from the study underscores that even moderate levels of depression may constitute a contraindication to engaging in such programs.

It is also noteworthy that several of the tested variables were not significant predictors of treatment dropout. In accordance with previous studies [6,38], we did not find any effect of age on dropout although the age range in our study was relatively wide, which should have facilitated the detection of any existing age effect. Furthermore, we did not find any effect of motivation to change as assessed by the URICA on treatment retention, despite the fact that motivation to change seems to be a stable predictor in face-to-face therapies [31,32,70]. Given that our program dealt exclusively with participants’ motivation to change and aimed specifically at enhancing motivation, it may have been helpful to negate the impact of low motivation on treatment adherence. It is plausible that motivation to change is a more robust predictor of treatment adherence in interventions without a motivational focus, such as conventional cognitive behavioral therapy [71]. In contrast, in most studies, motivation to change was not investigated as a predictor of treatment retention in face-to-face programs aimed at enhancing motivation to change [72,73]. An exception to this is the study by Allen and colleagues [74] who directly investigated the role of motivation to change in predicting treatment dropout from a motivational program. They did not find any effect on most of the measures, with the exception that higher scores on the precontemplation scale were predictive of dropout. To sum up, motivation to change seems to be a more robust predictor of treatment retention in interventions targeting change rather than motivation to change. This may be because motivational programs help to increase participants’ motivation; hence, eliminating the adverse effects of low motivation on dropout. Furthermore, the limited variance in motivation to change in this study may have made it more difficult to detect any effect of motivation. The limited variance in the URICA scores in our study is comparable to those from other studies on motivational enhancement programs [75,76], with the highest scores on the contemplation scale. It is perhaps not surprising that participants were primarily ambivalent about change given that participants would have needed a minimum amount of motivation to register for the program and they believed that they could benefit from a motivational program.

Several limitations should be kept in mind when interpreting the results of the present study. The complete anonymity and lack of direct contact made it impossible to obtain clinical diagnoses. Although the high scores on the EDE-Q indicate a clinical level of symptom severity, because we were unable to distinguish between people with different eating disorder diagnoses, we were also unable to test for diagnosis as a predictor of dropout. Furthermore, we used a global measure of motivation to change, which has been criticized in the context of eating disorders [77] because a patient’s motivation to change her eating disorder may vary between different symptoms. For example, motivation to change binge eating is usually higher than motivation to change dietary restriction [78]. Therefore, some authors argue that it is not adequate to assess a patient’s motivation to change globally for the eating disorders as a whole [79]. It is possible that we were unable to detect an effect of motivation to change because of the global assessment of the construct.

Nevertheless, this is the first study investigating a wide range of predictors of dropout from a Web-based program using adequate methodology in a large sample of women suffering from a wide range of eating disorder pathology.

This study identifies symptom severity and depressive symptoms as predictors of dropout from a Web-based program targeting motivation to change in women with symptoms of an eating disorder. In contrast, retention in the program was not dependent on age or initial motivation. These findings have the potential to guide further development of Web-based treatment programs in terms of tailoring online-delivered therapies to the specific needs of patients. For example, women who indicate high levels of depression could be offered an additional module that addresses affective disturbance. Furthermore, women who suffer from more severe eating disorder pathology might need more therapeutic guidance or additional contact via telephone or in a face-to-face setting to help them to continue the program. Although high dropout rates remain problematic in online treatments, there is still a large proportion of patients (particularly those with less severe eating disorders and depressive symptoms) who complete this easily accessible means of delivering therapeutic interventions.