The study is a multicentric, retrospective observational study. We analyzed data from patients who used the electronic mental health platform Minddistrict at 2 psychiatric hospitals during their inpatient treatment. Minddistrict is a platform founded in the Netherlands in 2008, and it provides a transdiagnostic catalog structured into modules, self-help training, and diaries, drawing on cognitive behavioral therapy principles. The platform’s modules primarily target psychoeducation, cognitive restructuring, and behavioral activation. It allows for individualized treatment by assigning suitable interventions tailored to each patient’s needs. Patients can monitor behavioral patterns and track therapy progress through diaries or independently select interventions from the self-help catalog. Minddistrict additionally offers various communication channels, including video calls and chat functions. Access to Minddistrict was granted to selected and interested patients by their therapists, who created user accounts on their behalf. After the account creation, patients could engage with the platform through therapist-guided modules, which required therapists to unlock new content after certain steps were completed. Additionally, patients had the option to use diaries, where they could document and reflect on their experiences independently and without therapist involvement. Minddistrict additionally offers various communication channels, including video calls and chat functions. The use of the video calls and diary features was excluded from this study because these components do not involve structured BT interventions.

The analysis includes inpatients from 2 psychiatric hospitals: Clienia Schlössli and Clienia Littenheid, which accessed the electronic mental health platform between January 2020 and May 2024. Platform use was reassessed in August 2024 to capture a comprehensive overview of module engagement and log in frequency. Although it is possible that some patients completed additional modules or increased use after this cutoff, such cases are considered unlikely. Patients were excluded if no modules had been assigned by their therapists. Patient accounts with multiple log ins and without module assignments were excluded, as we assumed that the platform was only used for video calls rather than for structured BT in these cases.

As outlined in the Introduction section, this study investigates noninitiation, defined as the failure to begin the intervention, and dropout, the incomplete use of the assigned therapeutic modules. To enable this analysis, patients were categorized into the following groups:

A schematic overview of the patient categorization process is presented in Figure 1.

This study (project ID: 2024-00768) has been reviewed and approved by the Zurich Ethics Committee. This research was conducted as a retrospective cohort study; therefore, active informed consent was not required by the ethics committee. Patient data were pseudonymized prior to analysis. Identifying information was stored separately from the analytic dataset and accessible only to authorized study personnel.

Descriptive statistics include means and SDs for continuous variables, as well as numbers and percentages for categorical variables. Multivariable logistic regression models were used to identify factors associated with the extent of platform tool use based on a predefined set of variables, including age, sex, education, employment, civil status, disease severity (HoNOS or BSCL), treatment setting, and main diagnosis. Missing values were handled using multiple imputation by chained equations with the mice package in R [17]. Appropriate imputation methods were applied depending on the variable type, including logistic, polytomous, proportional odds regressions, and predictive mean matching. A total of 70 datasets were imputed for the whole dataset with 30 iterations, and 45 datasets were imputed for the stationary setting subset with 15 iterations. All analyses were carried out with R (version 4.4.3; R Foundation for Statistical Computing), and RMarkdown (Posit Software) was used for dynamic reporting.

This study identified 2 significant predictors of engagement with BT among inpatients: increasing age was associated with a lower risk of noninitiation, while comorbid anxiety disorder was associated with a lower risk of dropout. These findings are particularly noteworthy as they differ depending on whether noninitiation or dropout is examined, and these associations were observed in a heterogeneous patient population encompassing a broader range of diagnoses than that included in a previous study [10], which was limited to patients with anxiety disorders and depression. In contrast, our study examines a more heterogeneous patient population with a wider range of diagnoses, allowing a more comprehensive analysis of the BT’s applicability across different conditions. The statistical analysis reveals significant findings, which interestingly differ depending on whether we examine the issue of noninitiation or dropout.

Previous studies have demonstrated that nonparticipation in general psychotherapy does not occur at random over time. The risk is particularly high at the beginning of the therapeutic process, often even before the first therapy session [18]. This finding is consistent with our study, where approximately one-third of the participants (94/278, 33.8%) were classified as belonging to the noninitiation group. This pattern suggests that not initiating a therapy may be driven not by the intervention itself but by other underlying factors. Therefore, it is essential to monitor this vulnerable phase closely and provide enhanced support precisely at this critical phase.

A striking finding of this study is the overall low uptake of the intervention. Despite being offered to a preselected group of interested patients, only 5 (1.8%) out of 278 patients completed all the assigned modules. One-third of patients never initiated the platform at all, and of those who did, most did not complete their assigned modules, resulting in a mean module completion rate of approximately 24% (mean 0.9, SD 2.0 out of 3.8, SD 3.4 assigned modules). These findings are consistent with the broader literature on digital interventions, which has repeatedly documented high rates of nonuse and early disengagement [5,6]. Several factors may help explain this low overall uptake. First, the absence of standardized follow-up procedures likely played a relevant role. Without systematic monitoring of patient engagement, low levels of platform use may have gone unnoticed or unaddressed by therapists [4]. Second, the lack of standardized criteria for module assignment may have contributed to low completion rates. Therapists in a qualitative study conducted at the same institution reported uncertainty about which modules were suitable for which patients and emphasized that active integration and follow-up are essential for meaningful engagement and that module assignment alone is insufficient unless supported by structured guidance [4].

Furthermore, the findings suggest that increasing age has a protective effect against noninitiation, indicating that older patients may be more likely to begin the intervention after modules have been assigned by the therapist. Previous research investigating internet-based cognitive behavioral therapy has shown that older adults tend to complete more modules and remain engaged over longer periods compared to younger users [19-22]. These studies suggest that older individuals may benefit from greater persistence, show more stable use patterns, and show potentially higher motivation, resulting in better adherence and lower dropout rates in digital interventions. One hypothesis could be that older individuals may adhere slightly better to digital interventions because they are more likely to be retired and therefore may have more time available to engage with therapy. However, this explanation does not apply to this study, as all participants were treated under inpatient conditions and therefore had access to comparable time resources. This suggests that factors other than time availability may underlie the higher adherence observed among older patients. A possible explanation is that younger users may approach digital interventions in a more exploratory way while lacking a strong initial commitment to the intervention. If immediate benefits are not apparent, they may disengage quickly. In contrast, older adults may begin treatment with more realistic expectations and greater persistence. Their decision to engage with digital therapy, despite potential barriers such as unfamiliarity with technology or initial skepticism, may be due to a high baseline level of intrinsic motivation. Younger individuals, for whom digital tools are more embedded in daily life, may face lower entry thresholds, resulting in a more heterogeneous group with varying levels of motivation, including those with only low therapeutic interest. Additionally, older adults who agree to participate in digital treatment may possess other characteristics associated with treatment success, such as higher education, more cognitive resources, or greater health-related self-efficacy.

Our findings indicate that patients with comorbid anxiety disorders in inpatient settings have a lower likelihood of discontinuing therapy. Anxiety disorders frequently cooccur with other psychiatric conditions, making it important to investigate their potential influence on the chance of dropouts. The prior evidence regarding the relationship between comorbid anxiety and psychotherapy dropout has been mixed and inconsistent. Studies have examined anxiety as a predictor of dropout both in general psychotherapy settings and specifically in the context of internet-based interventions. Some studies suggest that comorbid anxiety is associated with lower dropout rates [23], while others report the opposite, indicating a higher likelihood of early termination [24]. The evidence remains inconclusive. This raises the question of why patients with anxiety disorders in our study were less likely to drop out. One possible hypothesis relates to personality characteristics. Evidence suggests that individuals with anxiety disorders more frequently display cluster C personality traits [25]. These traits are associated with high conscientiousness, and a strong sense of duty can have a link to improved treatment adherence. Such individuals may be more willing to follow therapeutic guidance and complete assigned modules due to their conscientious traits. Another possible explanation for the lower dropout rate among patients with comorbid anxiety is that the structured and predictable nature of BT aligns well with their needs. The clear framework and guidance of online modules may support adherence by complementing traits commonly associated with anxiety. Additionally, the platform’s transdiagnostic content may target essential mechanisms of anxiety, such as rumination and avoidance, which may increase the perceived personal relevance of the material. In addition, anxious patients also tend to closely monitor their own progress, which may support ongoing participation, particularly in digital formats that involve self-tracking or structured exercises. It is also possible that individuals with anxiety disorders require additional effort to engage in social situations, including those encountered during face-to-face psychotherapy. psychotherapy. Working on a digital platform may reduce the potentially distressing effects of direct interpersonal interaction, which could make digital therapy a more appealing and manageable option for them. Finally, the higher frequency of contact points in BT may lead to a sense of safety and continuity that helps reduce uncertainty.

The results of this study could lead to 2 possible approaches. On the one hand, it might be beneficial to provide additional support to patients with a high risk of dropout or noninitiation to ensure successful completion of the therapy, for instance, by giving special attention to these patients. On the other hand, one could argue that these patients may not benefit from this therapy form and that alternative therapeutic approaches should be prioritized for them. Considering the wide range of available therapy options, certain approaches may simply be better suited for specific conditions. Wentzer et al [26] aimed to develop a tool to systematically assess factors critical for successful outcomes in BT, focusing on aspects such as computer access, internet skills, acute crises, and patient motivation. Similarly, Karyotaki et al [27] applied a personalized medicine approach in a systematic review to improve therapy allocation for patients with depression. Using sociodemographic characteristics and baseline symptom severity as predictors, the authors developed an interactive decision-support tool to guide treatment allocation among guided digital therapy, unguided digital therapy, and treatment as usual. While these studies concentrated on either practical barriers or depression-specific treatment optimization, this study extends this approach by including a broader patient population across multiple diagnoses. Such efforts are essential to optimize both treatment outcomes and resource use. In the future, standardized tools could help predict dropout risk, thereby enabling more targeted and effective implementation of BT. Accurate identification of suitable patients may enhance therapeutic efficiency and reduce health care costs. However, the findings of this study should be interpreted within their specific context. Our analysis focuses exclusively on adherence to BT, without assessing treatment outcomes. It is conceivable that some patients completed modules without having any clinical benefit, for example, to meet expectations from their therapists. Nevertheless, previous research has shown that homework adherence is positively associated with therapeutic success [28]. In line with this, we assume that module completion in BT also contributes to improved outcomes. By identifying risk factors for noninitiation and dropout, this study offers valuable insights into an underexplored area of digital mental health. It supports a more personalized and cost-efficient approach to psychiatric care by helping to match BT more precisely to patients who are likely to benefit. It expands the existing evidence base and contributes to the advancement of targeted, evidence-based treatment strategies.

Our findings must be interpreted considering several limitations. A limitation is the heterogeneity of the data, collected over an extended period (January 2020 to May 2024), with considerable variability in therapist training, application methods, and patient selection. No standardized procedures were applied, and information on selection criteria is missing. Importantly, as access to the Minddistrict platform was granted selectively by therapists to interested patients, no systematic data were collected on patients who were not offered BT during the study period. This precludes any formal comparison between included and nonincluded patients, and the extent and direction of the resulting selection bias therefore remains unknown. Future prospective studies should consider collecting data on all potentially eligible patients, regardless of whether they are ultimately offered BT, to allow for such comparisons.

Moreover, therapist-related variables are incomplete. Prior research shows that the extent and consistency of guidance significantly affect adherence [29,30] and that therapists’ attitudes toward digital tools play a role in patient engagement. The therapeutic relationship is also a known protective factor against dropout [31]. The absence of standardized training and varying application strategies may have contributed to differences in adherence. Additional influencing factors such as self-efficacy, expectations, digital affinity, and therapeutic alliance were not assessed. Finally, while the study reveals associations between predictors and dropout or noninitiation, it does not clarify underlying mechanisms. Qualitative studies could offer further insights, and future prospective studies with standardized protocols and more homogeneous samples are recommended. These limitations should be considered when interpreting the findings.