The aim of this study was to investigate the quality of the therapeutic alliance between patient and therapist in an online and face-to-face CBT for depression. To our knowledge, this was the first randomized controlled trial in this context. First, we examined whether the therapeutic alliance was comparable in both groups. Our results showed that the online and the face-to-face group differed significantly in only one subscale: therapists' ratings of the tasks subscale were significantly higher in the online group. This finding is in line with previous studies reporting that a strong working alliance, comparable to that formed in face-to-face settings, can also be established in online settings. The WAI mean scores in our study ranged from 5.39 to 6.22 (of a maximum of 7). These findings are comparable to data presented by Knaevelsrud and Maercker [22
], who reported mean scores ranging from 5.6 to 6.4 in Table of their article. Furthermore, authors using other scales or other versions of the WAI have also provided evidence for a comparably strong working alliance in online settings as in face-to-face therapy. Cook and Doyle [18
], for example, reported results for an online sample to be comparable with normative data from a representative sample in face-to-face therapy. Most of the participants in their sample presented with relationship issues, depression, anxiety, or grief. However, because of the small sample size and the non-randomized allocation of patients, these preliminary results should be interpreted with caution. In the same vein, Reynolds and colleagues [20
] reported ratings of the therapeutic alliance in an online setting to be similar to existing data from a face-to-face group. The participants in their study presented with depression, stress, anxiety, or childhood abuse. We were able to replicate the findings from both of these studies in a randomized controlled setting with a sample of depressive adults. The higher therapist ratings of the tasks subscale in the online group in our study may be attributable to the clear presentation and structuring of the tasks in the online mode, and to the opportunity to focus carefully on elaborated tasks. This fact may have positively influenced the agreement between clients and therapists on the therapeutic tasks.
Further, the drop-out rate in our study was relatively low. Seven (22%) participants in the online group and two (7%) participants in the face-to-face group discontinued the treatment. In general, drop-out rates in internet-based interventions are known to be problematic [5
]. However, the drop-out rates reported for studies involving internet-based interventions for depression over the last five years differ widely. For instance, Titov and colleagues [47
] reported that 11% of participants in a clinician-assisted internet-delivered CBT for depression did not complete post treatment questionnaires. In contrast, Spek and colleagues [48
] reported a drop-out rate of 66% for the intervention group of an internet-based CBT intervention study for subthreshold depression (individuals who did not complete post-test, did not start the intervention, or withdrew). In our sample, the attrition rates in the online group (22%) versus the face-to face group (7%) differed significantly, χ2
(1) = 4.737, p
.05. This may indicate that the more anonymous online therapeutic relationship is less stable than the face-to-face relationship. It is easier for patients in online treatment settings to stop therapeutic communication by simply "disappearing." A study of online romantic relationships revealed that avoidance behavior and discontinuity are more likely in online relationships than in face-to face relationships [49
Furthermore, we were interested in whether the therapeutic alliance predicted depression as outcome in the online or the face-to-face group. In both groups, only the clients' ratings of the working alliance were associated with depression at post-treatment (specifically, the composite score and tasks subscale in the face-to-face group at mid-treatment and, at post-treatment, the tasks subscale in the online group and the composite score and the tasks and goals subscales in the face-to-face group). It is worth noting that the correlations reported here are statistically significant, but only moderately high, ranging from r
= -.42 to r
= .52. These results are in line with findings on face-to-face psychotherapy. In a review article, Martin and colleagues [50
] reported a moderate but consistent relationship between the therapeutic alliance and outcomes of face-to-face psychotherapy. However, in the online group, only the working alliance at post-treatment was significantly associated with depression at post-treatment. This result replicates the findings of Knaevelsrud and Maercker [23
], who found no significant relationship between the working alliance at mid-treatment and PTSD change scores. Further, our data showed no significant relations between the BDI residual gain score and the working alliance in either group at mid- or post-treatment. Knaevelsrud and Maercker [22
] discussed the importance of investigating the working alliance at several stages of the therapeutic process to elucidate the relationship between working alliance and outcome. The authors suggested that the working alliance might be more an "additional indirect measure of outcome" than a predictor of treatment outcome.
The limitations of our study include the assessment of the working alliance and depression. As participants were first contacted online and later allocated at random to the online or the face-to-face group, all measures were administered as self-rated questionnaires in an online setting. Although this procedure has proven valid and reliable in various previous studies [18
], a structured clinical interview would have allowed a better quality of diagnosis of depression and the therapeutic relationship.
A further limitation is that we are unable to present follow-up data at the present time. Collection of follow-up data (after 3, 6 and 12 months) is still ongoing. Therefore, it remains an open question whether the working alliance at post-treatment predicts outcomes at follow-up.
Furthermore, the sample used in this study was small, relatively well educated and more than half of the participants already had experience of psychotherapy. Future studies should enroll larger and more heterogeneous samples. Another limitation of the study is the generalizability of our results. Due to our strict exclusion criteria regarding co-morbidity, suicide ideation, and psychosis, a number of applicants were excluded from the study. Our findings may therefore not be comparable with more naturalistic designs. Further research is needed to focus specifically on patients with co-morbidities.