This naturalistic study explores differences in pre-treatment characteristics across referral pathways by comparing GP-referred and self-referred patients receiving guided ICBT in specialized routine care mental health services. Specifically, we compare patients with depression and anxiety disorders, focusing on demographic characteristics, socioeconomic factors, and social support. In line with previous concerns about the demographic profile of those utilizing self-referral, we expect to observe differences that may shed light on the types of patients reached through each referral pathway.

This was a naturalistic cross-sectional study using data from a single pre-treatment assessment and conducted in a specialized routine care mental health clinic. The study compared two groups of patients: those referred by general practitioners (GP-referred) and those who self-referred, all of whom received guided internet-delivered cognitive behavioral therapy (guided ICBT) for moderate depression, panic disorder, or social anxiety disorder.

Participants were recruited from the eCoping clinic at Haukeland University Hospital in Bergen, Norway. Since 2013, the eCoping clinic has provided guided ICBT as part of its routine care for panic disorder (PD) (32) and social anxiety disorder (SAD) (33). Guided ICBT for moderate depression was introduced in 2015 (34). For the current study we present data collected at pre-treatment only. Data includes demographic information and information on the use of healthcare services covering 6 months prior to treatment start. GP-referred patients were enrolled between September 2014 and May 2019, while self-referred patients were enrolled between September 2016 and May 2019. Thus, the majority of both groups were recruited during an overlapping period (2016–2019), and all patients were assessed using identical inclusion criteria and procedures.

In Norway's specialized mental healthcare services, specialists assess patients’ symptoms and impairments to determine if they meet the criteria for legal access to this level of care (35). Primary healthcare services in Norway are for patients with mild to moderate conditions, while specialized healthcare is reserved for those with moderate and severe conditions. In the eCoping clinic, specialists in clinical psychology reviewed all referrals in accordance with the national priority guidelines to determine eligibility for specialized healthcare treatment (35).

Inclusion criteria for all study patients in the eCoping study were: 1) being 18 years of age or older, 2) if using antidepressants, being on a stable dosage over the previous four weeks, 3) being fluent in oral and written Norwegian. Exclusion criteria were: 1) current suicidal ideation, 2) current psychosis, 3) current substance abuse, 4) using benzodiazepines daily, 5) immediate need of other treatment, and 5) no access to the internet. Patients who were too mild or too severely ill were redirected elsewhere in the system, while patients who met the treatment criteria received an appointment for a face-to-face consultation with an ICBT therapist.

All patients granted therapy participated in a face-to-face consultation including a structured clinical diagnostic interview using the 6th version of the Mini-International Neuropsychiatric Interview (MINI) (36). In cases of an unclear diagnosis, the therapists would do further assessment in accordance with their clinical discretion. Included patients were diagnosed with either moderate depression, PD, or SAD. Written informed consent was obtained from all patients prior to data collection. All self-report symptom measures and demographic questionnaires in the present study were administered via the internet and collected at pre-treatment. All digitalized versions of the assessment tools used in the eCoping study had previously been validated for their psychometric properties (37).

Descriptive statistics include percentages, means, and standard deviations. Bivariate analyses included cross-tabulations of categorical variables using chi-square tests, ordinal regression of ordinal outcome variables, and t-tests for continuous outcome variables. We examined whether the observed bivariate associations were sensitive to adjustment in multivariable models. Multiple logistic regression analyzed associations between the outcome variable referral pathway and multiple predictors: demographic and socioeconomic information (Tables 1, 2), health care utilization 6 months prior to treatment start (Table 3), and the information source from where patients had learned about the treatment (Table 4). These four regression models were integrated into one main model. Explained variance was estimated with Nagelkerke. Temporal difference in inclusion period was controlled for by restricting analyses to 2016–2019, the main model was re-estimated on the restricted time interval. Logistic regression results include odds ratio (OR) and 95% confidence intervals (CI). Effect sizes (ES) for differences between GP- and self-referred patients were reported using Cohen's d, providing a standardized measure based on pooled standard deviations (47). Internal consistency was assessed with Cronbach's alpha. All analyses were conducted using IBM SPSS version 29 (48). The primary analyses assume missingness completely at random (MCAR) (49). A sensitivity analysis using multiple imputation (MI) under the missing at random (MAR) assumption was conducted to evaluate the impact of missing data.

Two recent studies from our research group based on the same patient cohort examining treatment effectiveness and predictors of treatment outcomes, reported no differences in pre-treatment symptom severity or illness duration between GP-referred and self-referred patients (42, 43). The present study replicated these findings. This lack of difference in pre-treatment symptom levels contrasts with previous research in primary care settings, which found higher pre-treatment symptom levels in GP-referred patients compared to those who self-referred (20). However, the study sample partially originated from a clinic designated at reaching specific populations. Our result also differs from findings in specialized care, where self-referred patients have been found to report higher symptom levels (18), and more comorbid diagnoses than those referred by GPs (19). However, symptom severity in our study was assessed after the initial face-to-face consultation that included the diagnostic MINI interview. Thus, the lack of difference between the referral pathway groups likely reflects that the ICBT therapists conducting the assessments applied consistent inclusion criteria and did not favor one referral pathway over the other. The lack of difference may also reflect the symptom level required to be eligible for treatment at the specialized level (35), and the redirection of patients too far above or below this threshold.

We found no difference between the two referral pathway groups when exploring the demographics: gender, age, relationship, and parental status. The prevalence of female patients between referral pathways in our study aligns with findings from other studies comparing patients by referral pathway (18, 20), and with the gender distribution found across systematic reviews of ICBT (50–52). This pattern reflects the well-established gender disparity in help-seeking behavior within the mental health field. While being female is positively associated with help-seeking (53), males tend to avoid or delay seeking help (7, 54, 55). Men with low mental health literacy are the least likely to perceive a need for mental healthcare and often avoid seeking such help (7). In contrast, women with high mental health literacy are the least likely to experience unmet mental healthcare needs (56). Access to care is shaped by patients’ ability to perceive a need and to act on it, and these abilities may be influenced by factors such as education and social support (57). Although we did not measure health literacy directly, these factors are conceptually connected to literacy. In this context, the gender imbalance in our study suggests that current referral pathways may not sufficiently address barriers experienced by men. It is crucial to develop targeted strategies to increase the treatment's visibility and encourage help-seeking among this group.

Our result that self-referred patients had higher educational levels compared to GP-referred, aligns with previous research comparing patients by referral pathway (18). A systematic review of referral pathways across healthcare settings also found that individuals with higher educational levels are more likely to self-refer, with one study indicating that higher education was the strongest predictor of self-referral (23). In addition, several studies have found higher educational levels among patients in ICBT clinics (58–60), suggesting that this treatment format may attract individuals with higher education, regardless of referral pathway. Higher educational level emerged as a strong predictor of self-referral in the multiple regression analyses. However, the association between educational level and referral pathway disappeared in the main model. This suggests that educational level may be correlated with other patient characteristics included in the model and therefore does not independently account for differences in referral pathway. Notably, education might be regarded as a more distal causal factor compared to many of the other factors examined. Some of these factors may even be on the causal pathway between educational attainment and referral pathway. However, exploring such mediation is beyond the scope of the current study.

In the bivariate results we found a difference in level of social support, with those who self-referred reporting higher levels than the GP-referred. This result contributes to the overall pattern suggesting the two referral pathways may reach and be utilized by distinct patient populations. Moreover, the higher level of social support reported among self-referring patients may be related to differences in social, cognitive, and material resources, which are often distributed along a socioeconomic gradient (61). Social support can also shape help-seeking behavior by strengthening self-efficacy, for instance through verbal encouragement (e.g., being prompted to consider treatment) and observational learning (e.g., witnessing others benefit from it) (62). The multivariate results did not confirm the group difference in social support, thus the findings should be interpreted with caution.

We found that self-referred patients were more likely to be employed than GP-referred, whereas the two groups did not differ in their level of reported sick leave in the four weeks prior to treatment start. Both bivariate results partially aligns with previous research, where GP-referred patients had higher unemployment rates than self-referred (20), though findings on sick-leave between GP- and self-referred patients have been mixed (18). The relation between employment status and self-referral in general also provides mixed findings, with some studies showing that employed individuals are more likely to self-refer, while others find no differences (23). The mixed findings were replicated in our multivariate analyses, where the predictive values of being employed disappeared.

The reported differences in sources of income between referral pathways in our study may likely reflect broader socioeconomic gradients. Patients reporting to rely on welfare benefits or work assessment allowances were more often GP-referred, whereas those reporting stable income sources (e.g., employment, student support) were more likely to belong in the self-referred group. This pattern suggests that financial security may enhance autonomy in navigating healthcare systems, while economic vulnerability increases reliance on structured referral processes. However, these results likely reflect broader socioeconomic advantages, as financial security often co-occurs with higher education, health literacy, and personal resources that facilitate healthcare navigation (63). The findings related to differences in receiving welfare benefits or work assessment allowances may also partly reflect the structured nature of the GP-referral pathway. Receiving welfare benefits requires contact with a GP, and some benefits additionally require attending active treatment (64). The reported differences in income sources between referral the pathways in our study may also be relevant for understanding health outcomes, as education shapes access to stable employment and financial security, which in turn are linked to better health outcomes (63).

We found that GP-referred patients overall reported to have used healthcare services more frequently than the self-referred in the 6 months prior to treatment start. This difference likely reflects the structured nature of the GP-referral pathway, where the patients often underwent multiple steps and were redirected from an outpatient clinic before reaching the eCoping clinic. Moreover, the logistic regression showed that GP-referred patients’ higher overall healthcare utilization was not an independent predictor when specific service types were considered. Our adjusted analyses indicate that it is specifically contact with public mental-health services, such as psychologists or psychiatrists at outpatient-clinic and community outpatient team that independently predicts GP-referral, rather than a general pattern of high service use. However, the use of community outpatient team did not retain its predictive value of GP-referral in the main model. In contrast, more self-referred patients reported to have consulted private psychologists, and this remained a strong independent predictor of self-referral. This pattern of health care utilization may be associated with greater autonomy and resource availability, such as higher education, social support, and financial security. These two distinct patterns, public service use predicting GP-referral and private psychology use predicting self-referral, suggest that socioeconomic factors may be associated with both the choice of referral pathway and specific types of healthcare utilization rather than overall volume alone. From a service perspective, this raises important questions. While self-referral may potentially streamline access and reduce reliance on public health resources, it could inadvertently widen disparities if individuals with fewer personal resources remain dependent on their GPs as a gatekeepers to specialized mental health services (24). The strong predictive value of GP and public mental health services for GP-referral underscores this point. As our study did not assess comorbidities or functional impairments, we cannot determine whether the reported differences are driven by clinical complexity or systemic navigation patterns. Moreover, we did not assess the number of times each service was used. To our knowledge, no previous studies have examined patients’ overall healthcare status or their healthcare use prior to entering digital mental health treatment, highlighting the novelty of our findings. Our results add nuance by showing that it is the specific patterns of prior healthcare utilization, rather than the overall quantity of service use, that distinguish patients across referral pathways.

Due to the study's set up, more than half of the GP-referred patients reported to have learned about the treatment from a healthcare professional at an outpatient clinic. While half of the self-referred patients learned about the treatment from family, friends, the internet, or through media, these information sources were only reported by a small minority of the GP-referred (13%). This difference may reflect the higher educational and social support levels reported among self-referred patients. Nearly 20% of the total sample learned about the treatment from university psychology services, underscoring their role in raising awareness about available treatment options. According to the adjusted results, learning about the treatment from family, friends, media, internet or psychology services at the university increased the likelihood of self-referral. Interestingly, a similar proportion in both referral-groups had learnt about the treatment from their GP. However, as self-referrers contacted the eCoping clinic directly, this suggests an unexplored third pathway where GPs advised patients to contact the eCoping clinic without completing the referral process. This third pathway is previously described as GP-initiated self-referral (65). This approach is common in Prompt Mental Health Care (PMHC) (16). A study from the Swedish ICBT-clinic found that symptom severity corresponds with the source of information about ICBT treatment: active sources like online searches, mental health webpages, or clinic referrals attract patients with higher symptom levels, while passive sources like newspaper ads or information from friends or family attract those with lower symptom levels (21). Although these results differ from ours, the emphasis on recruitment sources influencing patient characteristics aligns with our findings that different referral pathways may attract distinct patient populations (21). More broadly, if internet-delivered mental health interventions are promoted primarily through digital channels, promoting them this way may unintentionally exclude individuals with limited digital access or skills. Digital promotion approaches risks reinforcing existing health inequities and widening the digital divide (66).

This study has some limitations. First, as conducted in routine care, we have no data on individuals who were screened out, since data collection could not start before obtaining signed informed consent. This limits our ability to analyze differences between included and excluded patients across the two referral pathways but ensures that our findings reflect real-world clinical practice. Second, the study is exploratory, and although multivariable logistic regression was applied, the sample size relative to the number of predictors may limit the precision of estimations. However, the naturalistic routine clinical care context represents a strength of this study as a naturalistic design increases external validity, including the use of several established instruments with strong psychometric properties. To receive access to specialized mental healthcare services, all patients were diagnostically screened during their face-to-face consultation with the ICBT therapist. Those who did not meet the inclusion criteria, were judged too severely or not sufficiently ill, were excluded, resulting in a somewhat homogeneous sample. However, this strengthens internal validity by ensuring that comparisons are made within a well-defined clinical population. Another strength is that missing data in the variables were equally distributed between referral pathways. The results based on MI and the MAR assumption largely reproduced the results. However, one cannot rule out that missing data also may be missing not at random (MNAR). This assumption is not empirically testable (49).

Our results contribute to the overall pattern suggesting that the patients in our study represent two distinct populations divided by referral pathway. The results underscore the importance of implementing both GP- and self-referral pathways to specialized mental healthcare, as these options may enhance accessibility and expand the reach of services. Self-referrals may particularly appeal to individuals with higher levels of education and social support, while GP-referrals provide structured guidance for those who may face challenges navigating the healthcare system. To promote equitable access, outreach efforts should focus on individuals less likely to seek help independently, particularly those who may not fit the demographic profile typically associated with self-referral by highlighting the availability of accessible, effective services. An efficient referral system tailored for different patient groups is crucial not only for improving access to care but also for optimizing public healthcare resources and sustaining treatment capacity. Future research should explore the effectiveness of different referral pathways for diverse patients while addressing the social determinants that affect equity in mental health care access.