Effectiveness of Guided Internet‐Based Cognitive Behavioral Therapy for Adult Anxiety and Depression in Routine Care: An Observational Study

ABSTRACT

Introduction

Internet‐based cognitive behavioral therapy (iCBT) could help bridge the gap in treatment provision for mental disorders. iCBT is efficacious for the treatment of anxiety and depression in RCTs. However, more research is needed to translate findings from controlled trials to natural clinical settings. Additionally, more research is needed on predictors for treatment outcome in iCBT to guide allocation of treatment resources.

Method

Data originated from a routine care guided iCBT clinic and covered 1475 adults treated for either mild‐moderate depression (n = 719), panic disorder (n = 376), social phobia (n = 276), or specific phobia (n = 104). Joint models were used to examine treatment effects and predictors. Effect estimates were supported by effect sizes (Cohen's d) and calculations of the reliable change index (RCI).

Results

All four treatments showed significant reductions on their primary outcome measure at each assessment point, according to the joint models (depression, PHQ‐9: −1.35, 95% CI: −1.44; −1.27; panic disorder, PDSS‐SR: −0.96, 95% CI: −1.04; −0.88; social phobia, SIAS: −0.96, 95% CI: −1.25; −0.67; specific phobia, FQ Main Phobia: −0.25, 95% CI: −0.33; −0.16), and effect sizes were moderate to large from baseline to the last observation (depression, d = 0.87; panic disorder, d = 0.62; social phobia, d = 0.80; specific phobia, d = 0.47). In total, 26.7% of patients improved according to RCI, and 27.0% recovered at last observation. Higher baseline symptom severity was significantly associated with the extent of improvement for all programs. Similarly, baseline comorbid severity was associated with faster improvements on primary symptoms for depression and panic disorder. Lower age, being in a relationship, and studying increased the rate of improvement for panic disorder.

Conclusion

iCBT treatments for depression, panic disorder and social phobia were effective. For specific phobia, effects were smaller but still significant. Future studies should investigate process variables, theoretically relevant predictors or full prediction models to enable impactful predictions.

1. Introduction

Mental disorders are severely undertreated, with treatment coverage across countries ranging from 8% to 48% (Alonso et al. 2007; Evans‐Lacko et al. 2018; Moitra et al. 2022; Thornicroft et al. 2017). Furthermore, among those receiving treatment, only between 3% and 23% receive minimally adequate treatment (Moitra et al. 2022; Thornicroft et al. 2017). Delivering treatments online could help bridge this gap, as waiting time can be reduced (Catarino et al. 2023) and access to treatment improved (Cuijpers 2017).

Internet‐based cognitive behavioral therapy (iCBT) is an adaptation of face‐to‐face (ftf) cognitive behavioral therapy (CBT). iCBT users access a secure online platform where treatment content may be available in text, audio, or video format (Andersson 2016). The treatment content of iCBT is based on standard CBT treatment protocols and includes topics like psychoeducation, cognitive restructuring, and behavioral strategies (Andersson 2016). iCBT programs typically have a sequential structure where patients must complete one session before moving on to the next. Levels of support range from fully self‐guided interventions to guided iCBT, where a therapist provides feedback on exercises (Andersson 2016). Guided iCBT is more effective than self‐guided interventions (Karyotaki et al. 2021). iCBT is used for a range of different disorders, but most commonly for anxiety and depression (Andersson 2016).

Guided iCBT has been thoroughly examined in controlled clinical settings, and it is efficacious for both anxiety (g = 0.62–1.31) (Andrews et al. 2018; Pauley et al. 2023) and depression (g = 0.67) (Karyotaki et al. 2021; Andrews et al. 2018). Additionally, guided iCBT seems to have effects comparable to ftf CBT (g = −0.14–0.14) (Andrews et al. 2018; Pauley et al. 2023; Andersson et al. 2014). However, the effectiveness of iCBT, that is, when it is implemented in routine care settings (Streiner 2002), has not received the same attention (Andrews et al. 2018; Etzelmueller et al. 2020; Andersson et al. 2013). Studies investigating effectiveness do show positive results (g = 0.94–1.18) (Andrews et al. 2018; Etzelmueller et al. 2020). However, additional effectiveness research is important to determine how findings translate from controlled trials to routine care.

Not all individuals respond equally to iCBT (Arndt et al. 2020), and it is essential to identify which characteristics make iCBT a better fit for certain individuals. This could help tailor interventions and guide treatment allocation. However, predictor analyses of treatment outcomes in iCBT have yielded limited or unclear results (Haller et al. 2023; Sextl‐Plötz et al. 2024). Few effectiveness studies examine predictors (Edmonds et al. 2018; Flygare et al. 2020; Mathiasen et al. 2018; Ritola et al. 2022), and only two had large samples to support the anlayses (Edmonds et al. 2018; Ritola et al. 2022). More large‐scale studies are needed to clarify which characteristics influence the effectiveness of iCBT in routine care.

1.1. Objectives

In this observational study, we investigated the effectiveness and predictors for effect in guided internet‐based cognitive behavioral therapy for adults with mild‐moderate depression, panic disorder, social phobia, or specific phobias. Specifically, we examined:

1. The overall effect of each treatment program.

2. Predictors for treatment effects based on sociodemographic variables and baseline symptoms.

3. Associations between treatment completion and treatment effect.

4. Treatment effects on comorbid symptoms.

2. Method

2.1. Design

This observational study used data from the Danish nationwide psychiatric clinic, Internetpsykiatrien, and covered patients who self‐referred for treatment between November 14^th, 2019, and December 31^st, 2022. Data was extracted for the Personae project (Centre for Digital Psychiatry n.d.) and subsequently transferred to this project.

2.2. Ethics

The Mental Health Services in Southern Denmark approved the study (file no. 22/37157). The Regional Ethical Committee evaluated the study and declared it exempt from ethical approval (file no. 20202000‐38). In accordance with local legislation, The Regional Council of Southern Denmark approved data extraction without requiring informed consent from participants (file no. 23/28472). Data was stored and handled in accordance with the European General Data Protection Regulation (GDPR).

2.3. Setting: A Nationwide Danish Psychiatric Clinic

The clinic is a nationwide psychiatric clinic delivering iCBT situated in the Mental Health Services in the Region of Southern Denmark. The clinic offers guided iCBT for adults with mild to moderate depression, agoraphobia, panic disorder, social phobia, and specific phobias.

Applicants self‐refer through an online screening on the clinic's website. Afterward, a psychologist conducts a video‐based anamnestic interview to evaluate treatment eligibility. During the interview, the Mini International Neuropsychiatric Interview (Sheehan et al. 1998) is used to make a diagnostic assessment.

2.3.1. The Clinic's Inclusion Criteria

Eighteen years or older, living in Demark, and meeting ICD‐10 (World Health Organization 1992) diagnostic criteria for (a) mild or moderate depressive episode, (b) agoraphobia, (c) panic disorder, (d) social phobia, or (e) specific phobia.

2.3.2. The Clinic's Exclusion Criteria

Acute risk of suicide, meeting ICD‐10 (World Health Organization 1992) diagnostic criteria for substance‐related disorders, schizophrenia and related disorders, bipolar affective disorder, obsessive‐compulsive disorder, post‐traumatic stress disorder (with no prior successful treatment), or personality disorder (with no prior successful treatment), currently receiving another psychological treatment, not having access to a personal computer and internet connection, or inadequate understanding of spoken and written Danish.

2.4. Participants

Patients who self‐referred during the inclusion period and started treatment were included in this study. Returning patients were excluded since they might influence the treatment effects.

2.5. Interventions

The clinic has four guided iCBT programs targeting (1) depression, (2) panic disorder and/or agoraphobia, (3) social phobia, and (4) specific phobias. All programs consist of six to seven sessions, last 12 weeks, and include CBT elements like psychoeducation, goal setting, cognitive restructuring, behavioral activation (depression), exposure (anxiety), and relapse prevention. The programs contain written information and illustrations visualizing CBT models. Therapists provide written feedback following each session, and patients can ask questions through an asynchronous chat. In rare cases, therapists might provide a brief phone conversation to help a patient progress. The treatments are described in the supplementary materials

2.6. Measures

Patients responded to the following questionnaires as part of routine treatment at the clinic. The clinic uses additional questionnaires described in the supplementary materials.

2.6.1. Sociodemographic Information

Age, sex, civil status, number of children, education, and primary source of income.

2.6.2. Patient Health Questionnaire‐9 (PHQ‐9)

(Kroenke et al. 2001) measures depressive symptoms. It consists of nine items rated on a Likert scale from 0–3. The total score ranges from 0–27. The PHQ‐9 has shown good psychometric properties (Kroenke et al. 2001; Lowe et al. 2004; Wittkampf et al. 2007), and it has been validated in Danish (Pedersen et al. 2016). We used a cutoff of 10 or higher as indicative of depression (Kroenke et al. 2001), and a change of five or more as an indicator of reliable change (Lowe et al. 2004).

2.6.3. Generalized Anxiety Disorder‐7 Item Scale (GAD‐7)

(Spitzer et al. 2006) measures general symptoms of anxiety (Spitzer et al. 2006; Lowe et al. 2008). It consists of seven items rated on a Likert scale from 0–3. The total score ranges from 0–21. The GAD‐7 has shown good psychometric properties (Spitzer et al. 2006; Lowe et al. 2008; Beard and Bjorgvinsson 2014).

2.6.4. Social Interaction Anxiety Scale (SIAS)

(Mattick and Clarke 1998) measures fear of social interactions. It contains 20 items rated on a five‐point Likert scale ranging from 0–4. The total score ranges from 0–80. The SIAS has shown good psychometric properties (Mattick and Clarke 1998). It is a useful tool for screening purposes as well as evaluating treatment outcomes for social anxiety disorder (Brown et al. 1997; Peters and LOhoo 2000). We used a cutoff of 37 or higher as indicative of social phobia (Peters and LOhoo 2000).

2.6.5. Panic Disorder Severity Scale‐Self Report (PDSS‐SR)

(Houck et al. 2002) measures panic disorder and agoraphobia symptoms. It consists of seven items answered on a five‐point ordinal scale from 0–4. Higher scores indicate more severe symptoms. The total score ranges from 0–28. The PDSS‐SR has shown good internal consistency, sensitivity to change, and correlation with the clinician‐administered interview version (Houck et al. 2002). We used a cutoff of eight or higher as indicative of panic disorder/agoraphobia (Shear et al. 2001).

2.6.6. Fear Questionnaire (FQ)

(Marks and Mathews 1979) measures phobias with six subscales divided across 24 items. This study only uses the FQ main phobia subscale, which consists of just one item measuring the severity of the primary phobia on a scale from 0–8.

2.7. Assessment Points

The screening questionnaire was administered via SurveyXact, while the remaining questionnaires were delivered through the treatment platform Minddistrict. The clinic placed the baseline assessment after the first introductory session to ensure patients had a positive start to the treatment before receiving questionnaires. During treatment, questionnaires were scheduled as additional “sessions” between the treatment sessions, meaning patients complete them between finishing one session and starting the next. After the final session, patients received a post‐treatment questionnaire. If a patient discontinued treatment, no further assessments were administered. See Figure 1 and Table 1 for more details.

FIGURE 1.

Flowchart of the assessment points for all four treatments. ¹Due to a mistake in the clinic's questionnaire setup, 64 patients in specific phobia treatment did not receive the FQ in the questionnaire at baseline. ²Some assessment points have been removed or added with updates to the treatment programs. Therefore, not all patients received these questionnaires, and these assessment points have larger amounts of missing data than the subsequent assessment points. FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

TABLE 1.

Structure of the four treatment programs.

Depression	Panic disorder	Social phobia	Specific phobia
Screening questionnaire: PHQ‐9, GAD‐7	Screening questionnaire: PDSS‐SR, PHQ‐9, GAD‐7	Screening questionnaire: SIAS, PHQ‐9, GAD‐7	Screening questionnaire: FQ, PHQ‐9, GAD‐7
MINI	MINI	MINI	MINI
Session 1—intro to online treatment & treatment goals	Session 1—Intro to online treatment	Session 1—intro to online treatment	Session 1—Intro to online treatment
Baseline questionnaire: PHQ‐9, GAD‐7	Baseline questionnaire: PDSS‐SR, PHQ‐9, GAD‐7	Baseline questionnaire: SIAS, PHQ‐9, GAD‐7	Baseline questionnaire: FQ, PHQ‐9, GAD‐7
Session 2.1—Intro to CBT	Session 2—Intro to CBT & treatment goals	Session 2—Intro to CBT & treatment goals	Session 2—intro to CBT & treatment goals
Session 2.2—Intro to CBT	PDSS‐SR	SIAS	FQ
Session A1—Rumination (optional)	Session 3.1—anxiety loop	Session 3.1—anxiety loop I	Session 3—anxiety loop
Session A2—Rumination (optional)	PDSS‐SR a	Session 3.2—anxiety loop II	FQ
PHQ‐9	Session 3.2—Fight or flight	SIAS	Session 4—cognitive restructuring
Session 3.1—self‐monitoring	Session 3.3—components of anxiety	Session 4—self focus	FQ
PHQ‐9 a	PDSS‐SR	SIAS a	Session 5—psychoeducation about exposure
Session 3.2—scheduling activities	Session 4.1—anxiety hierarchy	Session 4.1—self focus exercise I	FQ
PHQ‐9	PDSS‐SR a	Session 4.2—self focus exercise II	Session 6—anxiety hierarchy
Session 4.1—automatic & alternative thoughts	Session 4.2—interoceptive exposure	Session 4.3—self focus reflection	FQ a
PHQ‐9 a	PDSS‐SR a	SIAS	Session 6.1—anxiety exposure
Session 4.2—behavioral experiments	Session 5.1 —understanding agoraphobia	Session 5—catastrophizing	Session 6.2—anxiety exposure
Session 4.3—behavioral experiments	PDSS‐SR	SIAS	Session 6.3—anxiety exposure
PHQ‐9	Session 5.2—anxiety hierarchy	Session 6—anxiety hierarchy	FQ
Session 5.1—rules & assumptions	PDSS‐SR a	SIAS a	Session 7—relapse prevention
Session 5.2—rules & assumptions	Session 5.3—anxiety exposure	Session 6.1—anxiety exposure	Post treatment questionnaire: FQ, PHQ‐9, GAD‐7
PHQ‐9	Session 5.4—anxiety exposure	Session 6.2—anxiety exposure
Session 6—relapse prevention	Session 5.5—anxiety exposure	Session 6.3—anxiety exposure
Post treatment questionnaire: PHQ‐9, GAD‐7	PDSS‐SR	SIAS
	Session 6—relapse prevention	Session 7—relapse prevention
	Post treatment questionnaire: PDSS‐SR, PHQ‐9, GAD‐7	Post treatment questionnaire: SIAS, PHQ‐9, GAD‐7

Note: Current structure of the treatment programs in use at the clinic. For a comparison between current and older versions see the supplementary materials.

Abbreviations: FQ, Fear Questionannire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; MINI, Mini International Neuropsychiatric Interview; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^a In previous versions of the treatment programs, there were additional assessment points. These have been removed in the current version.

2.8. Statistics

2.8.1. Categorizations and Transformations of Variables

Completion was defined as finishing the core treatment content within 16 weeks of starting. For depression, this meant completing four of six sessions; for panic disorder, five of six sessions; and for social and specific phobias, six of seven sessions.

We categorized education as low (primary and high school 0–12^th grade), medium (vocational and short further education, < 3 years), high (intermediate, 3–4 years, and long further education, ≥ 5 years), or other.

For the predictor analyses, we grouped age in 4‐year increments starting at 18 (18–21; 22–25; etc.) and baseline SIAS scores in ten‐point increments (0–9; 10–19; etc.). We did this to avoid a scaling issue with the regression coefficients, which resulted in the models failing to converge. However, by rescaling the variables, we were still able to make some inferences regarding the predictive value of these variables.

For some categorical variables, the groups became too small to report on due to data protection regulations (social phobia: retired), or the small groups caused convergence issues for the models (panic disorder: other education; specific phobia: other education, retired, other income). Therefore, we merged these groups with the most appropriate alternative group. For social phobia, retired was merged with other income for descriptive statistics and analyses. For panic disorder and specific phobia, other education was merged with low education in the analyses. For specific phobia, retired and other income was merged with on welfare in the analyses.

2.8.2. Primary Analyses

The primary analysis was an intention to treat‐analysis using all data points from included patients. We estimated symptom change using joint modeling (Tsiatis and Davidian 2004), which integrates two linked sub‐models to address the relationship between longitudinal and time‐to‐event data. This involves (1) A mixed‐effects sub‐model for longitudinal data (symptoms over time) and (2) a Cox proportional hazards sub‐model for time‐to‐event data (risk of dropout).

Joint modeling simultaneously analyzes the longitudinal and time‐to‐event data, which may have dependencies. Thus, the model handles missing data, by accounting for the correlation between the longitudinal outcomes and the time‐to‐event process. This reduces the potential for bias in estimating symptom change, even if data is missing not at random (Tsiatis and Davidian 2004). For example, if deterioration leads to dropout, a mixed‐effects model would overestimate the change because replies from deteriorated patients are missing later in the treatment. The joint model adjusts the estimates to account for this relationship. (See (Levine et al. 2015) for an example of how this type of modeling might influence change estimates in a trial.)

The mixed‐effects sub‐model estimated symptom changes over time, using assessment points to represent time, which was nested within patients. Assessment points and patients were treated as random effects (random intercept and slope). If a model failed to converge, we switched to Nelder‐Mead optimization.

The Cox proportional hazards sub‐model estimated the risk for dropout. The first missing assessment point was considered the time of dropout.

We only reported results for the joint models' mixed‐effects part. For the estimate of overall symptom change, we were interested in the slope, which indicates the average change in the symptoms for each assessment point during treatment.

As sensitivity analyses, we checked if (1) program version and (2) patients below cutoff for questionnaires at screening affected slope estimates.

We assessed model fit with Q‐Q normality plots and plots of fitted values versus standardized residuals to check whether error terms were normally distributed and whether the models showed heteroscedasticity.

2.8.3. Effect Sizes

We estimated effect sizes using Cohen's d with pooled standard deviations. Separate effect sizes were calculated using screening and baseline as reference points. To handle missing data for Cohen's d and reliable change (see 2.8.6. below) we used last observation carried forward (LOCF). Additionally, we calculated effect sizes for those completing the post‐treatment questionnaire.

2.8.4. Predictors for Outcome

We used the same joint modeling approach for the predictor analyses. We were interested in the interactions between the predictors and assessment points, as these indicate how the predictors change the slope estimate. A negative interaction indicates that the predictor causes a steeper slope (faster improvements). First, we included one variable as a predictor in each joint model to identify significant predictors in unadjusted models. Afterward, we conducted adjusted models for each predictor that had a significant interaction with assessment points in any of the treatments. The adjusted joint models included all other predictors with significant interactions in any of the treatments.

2.8.5. Effect on Comorbid Symptoms

For comorbid symptoms, we did not have time‐series data on patients. Therefore, we used mixed‐effects models with assessment points as a categorical variable with three levels: (1) screening (reference level), (2) baseline (after session 1), and (3) post‐treatment.

2.8.6. Reliable Change

We estimated improvement, recovery, reliable recovery, and deterioration rates for all patients using LOCF. These were defined as: (1) improvement: decrease on the questionnaire ≥ reliable change index (RCI), (2) recovery: decrease from a screening score above cutoff to a score below cutoff, (3) reliable recovery: meeting criteria for both improvement and recovery, and (4) deterioration: increase on questionnaire ≥ RCI.

Because PDSS‐SR, SIAS, and FQ do not have predefined RCIs, we calculated these using the formula $1,96\ast \left(SD1\ast \sqrt{2\ast \left(1-r\right)}\right)$ (Jacobson and Truax 1991). SD1 is the standard deviation at screening, and r the internal reliability (Martinovich et al. 1996) at screening (FQ Main Phobia uses test‐retest reliability from the original study (Marks and Mathews 1979), since it is not possible to calculate Cronbach's alpha for a single item). RCI was ≥ 5 for PDSS, ≥ 12 for SIAS, and ≥ 2 for FQ Main Phobia. The RCI for PHQ‐9 was defined as ≥ 5 based on previous studies (Lowe et al. 2004).

2.8.7. Software

R version 4.4.1 (R: A Language and Environment for Statistical Computing 2024) and R Studio version 2023.03.1 were used for all analyses. We used the following packages: pastecs (pastecs 2024) for descriptive statistics, JM (Rizopoulos 2010) for joint modeling, nlme (nlme 2023) for mixed‐effects models, and tidyverse (Wickham et al. 2019) and jsonlite (The Jsonlite Package: A Practical and Consistent Mapping Between JSON Data and R Objects. 2014) for filtering, organizing, and visualizing data.

3. Results

The sample consisted of 1475 adults. Most were women (69.4%) and in a relationship (62.4%). Almost half the sample had high education (42.8%) and were employed (46%). Table 2 reports additional descriptive statistics.

TABLE 2.

Descriptive statistics for patients in four routine care internet‐based cognitive behavioral therapy programs.

Variable a	All N = 1475	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobias N = 104
Sex, female	1024 (69.4%)	501 (69.7%)	255 (67.8%)	183 (66.3%)	85 (81.7%)
Age	33 (12.0)	34 (12.3)	33 (11.4)	30 (10.2)	37 (14.6)
Civil status
Singe	554 (37.6%)	296 (41.2%)	117 (31.2%)	120 (43.3%)	21 (20.2%)
In a relationship, living alone	187 (12.7%)	94 (13.1%)	46 (12.3%)	33 (11.9%)	14 (13.5%)
In a relationship, living together	733 (49.7%)	328 (45.6%)	212 (56.5%)	124 (44.8%)	69 (66.3%)
Number of children	0.8 (1.1)	0.8 (1.1)	0.8 (1.1)	0.5 (1.0)	1.0 (1.1)
Education
Low	479 (32.5%)	224 (31.2%)	119 (31.7%)	113 (40.8%)	23 (22.1%)
Medium	312 (21.2%)	164 (22.8%)	92 (24.5%)	44 (15.9%)	12 (11.5%)
High	631 (42.8%)	303 (42.1%)	150 (40.0%)	113 (40.8%)	65 (62.5%)
Other	53 (3.6%)	28 (3.9%)	14 (3.7%)	7 (2.5%)	4 (3.8%)
Primary source of income
On welfare	136 (9.2%)	80 (11.1%)	33 (8.8%)	19 (6.9%)	4 (3.8%)
Employed	679 (46.0%)	317 (44.1%)	190 (50.7%)	119 (43.0%)	53 (51.0%)
Educational stipend	415 (28.1%)	187 (26.0%)	100 (26.7%)	104 (37.5%)	24 (23.1%)
Retired	50 (3.4%)	31 (4.3%)	10 (2.7%)	NA b	6 (5.8%)
Other	195 (13.2%)	104 (14.5%)	42 (11.2%)	35 (12.6%)	17 (16.3%)
Symptoms at screening
PHQ‐9	14.0 (6.0)	17.1 (4.2)	11.4 (5.9)	11.8 (5.4)	7.2 (5.8)
GAD‐7	11.5 (4.8)	11.6 (4.5)	12.5 (4.8)	11.3 (4.5)	7.5 (5.9)
PDSS‐SR	—	—	14.8 (4.9)	—	—
SIAS	—	—	—	45.1 (12.4)	—
FQ main phobia	—	—	—	—	6.7 (1.5)

Abbreviations: FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^a Categorical variables are reported as n (%); Continuous variables are reported as mean (standard deviation).

^b This group was too small to report on due to data protection regulations. Therefore, it was merged with Other.

3.1. Objective 1: Overall Treatment Effect

The joint models indicated a significant effect on symptom scores for each assessment point. Depression (−1.35, 95% CI: −1.44; −1.27), panic disorder (−0.96, 95% CI: −1.04; −0.88), and social phobia (−0.96, 95% CI: −1.25; −0.67) all showed about one point decrease on their respective score for each completed assessment. Specific phobia showed a smaller decrease (−0.25, 95% CI: −0.33; −0.16). See Table 3 and Figure 2. The program versions did not influence the treatment effects. Excluding patients below cutoff at screening barely changed the estimates for depression (−1.39) and panic disorder (−0.99). For social phobia, the change was somewhat bigger (−1.30).

TABLE 3.

Change in symptom scores for each completed assessment (time point) starting at screening and ending with post‐treatment assessment.

	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobias N = 104
Outcome	PHQ‐9	PDSS‐SR	SIAS	FQ main phobia
Time points	10	11	10	9

Variable	Beta (95% CI)
Estimated change per assessment point	−1.35 (−1.44; −1.27) ***	−0.96 (−1.04; −0.88) ***	−0.96 (−1.25; −0.67) ***	−0.25 (−0.33; −0.16) ***

Note: All models are joint models consisting of a mixed‐effects submodel and a Cox proportional hazards submodel. We report on the estimates of the mixed‐effects part of the joint model. Significant results are highlighted with bold text.

Abbreviations: CI, Confidence Interval; FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^*** p < 0.001.

FIGURE 2.

Symptom developments of primary and secondary symptoms for all four treatment programs. FQ= Fear Questionnaire; GAD‐7 = Generalized Anxiety Disorder‐7 item scale; PDSS‐SR = Panic Disorder Severity Scale‐Self Report; PHQ‐9 = Patient Health Questionnaire‐9; SIAS = Social Interaction Anxiety Scale.

Using the baseline assessment point as the reference level, all programs showed moderate to large effect sizes at the last observation (d = 0.47–0.87) and large effects for those completing the post‐treatment questionnaire (d = 1.09–1.50). When using the screening questionnaire, effect sizes increased for all programs, except for social phobia. Table 4 reports all the effect sizes.

TABLE 4.

Effect sizes for all four treatment programs.

Treatment	Baseline (after session 1)		LOCF		Post‐treatment		Screen—LOCF	Baseline—LOCF	Screen—post	Baseline—post
	n	M (SD)	n	M (SD)	n	M (SD)	d	d	d	d
Depression	719	14.9 (4.6)	719	10.5 (5.5)	191	7.7 (5.6)	1.35	0.87	2.08	1.50
Panic disorder	375	12.4 (4.96)	375	9.2 (5.2)	101	5.2 (4.8)	1.11	0.62	1.99	1.47
Social phobia	277	49.9 (13.0)	277	37.6 (17.4)	72	35.5 (14.0)	0.50	0.80	0.75	1.09
Specific phobia	50 a	6.4 (2.1)	104	5.3 (2.4)	31	3.4 (2.2)	0.73	0.47	2.00	1.41

Abbreviations: d, Cohen's d; LOCF, Last observation carried forward; M, Mean; Post, Post‐treatment; SD, Standard deviation.

^a Due to a mistake in the clinic's questionnaire setup, 64 patients in specific phobia treatment did not receive the FQ in the questionnaire at baseline.

3.2. Objective 2: Predictor Analyses

For the predictor analyses, we only report on the interactions. Symptom severity at screening was the only covariate that significantly affected the rate of improvement for all treatments (Depression = −0.10, 95% CI: −0.12; −0.08. Panic disorder = −0.06, 95% CI: −0.07; −0.05. Social phobia = −0.68, 95% CI: −0.88; −0.47. Specific phobia −0.06, 95% CI: −0.12; −0.01). Higher symptoms were associated with faster improvements (e.g., for depression, an increase of one point on the PHQ‐9 at screening was associated with an additional decrease of −0.10 points at each assessment point). Comorbid symptoms were also associated with a faster improvement, but only for depression (−0.03, 95% CI: −0.05; −0.01) and panic disorder (−0.03, 95% CI: −0.04; −0.03). In the adjusted models, panic disorder was the only treatment, where sociodemographic variables significantly affected the rate of improvement. For panic disorder, being in a relationship was associated with a faster response (in a relationship, living alone: −0.47, 95% CI: −0.66; −0.29. In a relationship, living together: −0.16, 95% CI: −0.28; −0.04), as was receiving an educational stipend (i.e., being a student) (−0.29, 95% CI: −0.46; −0.11), while older age was associated with a slower response (0.05, 95% CI: 0.03; 0.06). For depression, being retired was associated with a slower response in the unadjusted model (0.49, 95% CI: 0.11; 0.88), but this association disappeared in the adjusted model. Table 5 reports all the prediction estimates.

TABLE 5.

Estimates of how sociodemographic variables and severity at screening affect the rate of symptom change.

	Depression N = 719		Panic disorder N = 375		Social phobia N = 277		Specific phobias N = 104/50 d
Outcome	PHQ‐9		PDSS‐SR		SIAS		FQ main phobia
Variable	Unadjusted beta	Adjusted beta (95% CI)	Unadjusted beta	Adjusted beta (95% CI)	Unadjusted beta	Adjusted beta (95% CI)	Unadjusted beta	Adjusted d beta (95% CI)
Age
Estimated change per assessment point	−1.35***	−1.29 (−1.44; −1.13)***	−1.17***	−1.14 (−1.22; −1.06)***	−0.76**	−0.75 (−1.26; −0.23)**	−0.23**	−0.27 (−0.43; −0.11)**
Age	−0.09	−0.03 (−0.11; 0.06)	−0.20*	−0.10 (−0.18; −0.02)*	−0.48	0.16 (−0.19; 0.51)	−0.04	0.01 (−0.07; 0.09)
Estimated change… × age	−0.001	−0.001 (−0.03; 0.03)	0.04 ***	0.05 (0.03; 0.06) ***	−0.05	−0.09 (−0.20; 0.03)	−0.004	0.001 (−0.02; 0.02)
Sex
Estimated change per assessment point	−1.35***	—	−0.88***	—	−1.18***	—	−0.29**	—
Sex, female	0.29	—	−0.58	—	2.87	—	0.39	—
Estimated change… × sex, female	−0.01	—	−0.09	—	0.32	—	0.06	—
Civil status
Estimated change per assessment point	−1.35***	−1.27 (−1.40; −1.13)***	−0.78***	−0.79 (−0.88; −0.70)***	−1.08***	−1.13 (−1.59; −0.67)***	−0.34***	−0.35 (−0.52; −0.17)***
Single (reference level)	—	—	—	—	—	—	—	—
In a relationship, living alone	−0.91	−0.51 (−1.23; 0.21)	1.59	−0.02 (−0.84; 0.79)	0.22	0.24 (−2.43; 2.90)	−1.12	−0.63 (−1.53; 0.27)
In a relationship, living together	−0.27	−0.14 (−0.60; 0.32)	−0.38	−0.53 (−1.08; 0.02)	−1.45	0.50 (−1.21; 2.21)	−0.27	−0.13 (−0.79; 0.52)
Single (reference level)	—	—	—	—	—	—	—	—
Estimated change… × in a relationship, living alone	0.15	0.01 (−0–28; 0.31)	−0.49 ***	−0.47 (−0.66; −0.29) ***	0.46	0.33 (−0.65; 1.31)	0.14	0.13 (−0.16; 0.41)
Estimated change… × in a relationship, living together	−0.02	−0.05 (−0.24; 0.31)	−0.20 *	−0.16 (−0.28; −0.04) **	0.13	0.05 (−0.55; 0.65)	0.10	0.09 (−0.11; 0.30)
Children
Estimated change per assessment point	−1.30***	—	−0.96***	—	−0.89***	—	−0.27***	—
Number of children	−0.06	—	−0.29	—	−0.58	—	−0.09	—
Estimated change… × number of children	−0.07	—	−0.003	—	−0.14	—	0.02	—
Education
Estimated change per assessment point	−1.32***	—	−1.04***	—	−0.73**	—	−0.25**	—
Low (reference level)	—	—	—	—	—	—	—	—
Medium	−0.22	—	−1.63*	—	1.88	—	0.19	—
High	−1.14**	—	−1.32*	—	−3.53*	—	0.41	—
Other	−1.25	—	NA c	—	−0.31	—	NA	—
Low (reference level)	—	—	—	—	—	—	—	—
Estimated change… × medium	−0.06	—	0.16	—	−0.53	—	−0.06	—
Estimated change… × high	−0.05	—	0.13	—	−0.24	—	0.02	—
Estimated change… × other	0.25	—	NA c	—	−1.12	—	NA	—
Primary source of income
Estimated change per assessment point	−1.38***	−1.31 (−1.57; −1.04)***	−0.89***	−0.80 (−0.95; −0.66)***	−0.64	−0.95 (−2.14; 0.23)	−0.54*	−0.33 (−0.51; −0.16)***
On welfare (reference level)	—	—	—	—	—	—	—	—
Retired	−1.54	−0.13 (−1.59; 1.33)	−1.12	−0.02 (−1.48; 1.43)	NA c	NA c	−2.27	NA c
Educational stipend	0.71	0.76 (−0.02; 1.54)	0.10	−0.11 (−0.98; 0.76)	4.11	2.51 (−0.99; 6.01)	−1.72	−0.76 (−1.56; 0.03)
Employed	−0.40	0.08 (−0.62; 0.79)	−1.69	−0.50 (−1.30; 0.31)	2.10	2.43 (−0.97; 5.83)	1.21	−0.15 (−0.79; 0.49)
Other	0.63	0.18 (−0.68; 1.03)	−0.18	−0.26 (−1.26; 0.74)	5.52	2.66 (−1.22; 6.53)	−1.82	NA c
On welfare (reference level)	—	—	—	—	—	—	—	—
Estimated change… × retired	0.49 *	0.30 (−0.24; 0.85)	−0.25	−0.01 (−0.27; 0.24)	NA c	NA c	0.12	NA c
Estimated change… × educational stipend	0.05	−0.05 (−0.35; 0.25)	−0.23	−0.29 (−0.46; −0.11) **	−0.20	−0.01 (−1.29; 1.27)	0.31	0.15 (−0.11; 0.41)
Estimated change… × employed	−0.02	−0.02 (−0.31; 0.27)	−0.03	−0.08 (−0.24; 0.08)	−0.52	−0.24 (−1.51; 1.03)	0.29	0.07 (−0.15; 0.28)
Estimated change… × other	0.13	0.20 (−0.16; 0.56)	−0.01	0.30 (0.17; 0.23)	−0.33	−0.11 (−1.53; 1.27)	0.38	NA c
Severity at screening
Estimated change per assessment point	0.40*	0.41 (0.10; 0.72)**	−0.03	0.003 (−0.14; 0.15)	2.33***	2.32 (1.22; 3.42)***	0.12 d	0.15 (−0.20; 0.50)
Primary symptoms a	0.90***	0.85 (0.80; 0.90)***	0.84***	0.79 (0.74; 0.84)***	9.09***	8.80 (8.19; 9.40)***	0.88***	0.87 (0.71; 1.03)***
Estimated change… × primary symptoms a	−0.10 ***	−0.10 (−0.12; −0.08) ***	−0.06 ***	−0.06 (−0.07; −0.05) ***	−0.66 ***	−0.68 (−0.88; −0.47) ***	−0.06 *	−0.06 (−0.12; −0.01) *
Comorbid symptoms at screening
Estimated change per assessment point	−1.07***	−0.90 (−1.15; −0.66)***	−0.59***	−0.55 (−0.64; −0.46)***	−1.29***	−1.50 (−2.16; −0.84)***	−0.35***	−0.32 (−0.45; −0.19)***
Comorbid symptoms b	0.43***	0.14 (0.09; 0.19)***	0.40***	0.13 (0.09; 0.17)***	0.69***	0.25 (0.10; 0.41)**	−0.05	0.02 (−0.03; 0.07)
Estimated change… × comorbid symptoms b	−0.02 *	−0.03 (−0.05; −0.01) ***	−0.03 ***	−0.03 (−0.04; −0.03) ***	0.03	0.04 (−0.01; 0.09)	0.01	0.01 (−0.005; 0.02)

Note: All models are joint models consisting of a mixed‐effects submodel and a Cox proportional hazards submodel. We report on the estimates of the mixed‐effects part of the joint model. Significant results are highlighted with bold text.

Abbreviations: FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^a Primary symptoms were different for each disorder and used different questionnaires: Depression used PHQ‐9, panic disorder used PDSS‐SR, social phobia used SIAS, specific phobia used FQ Main Phobia, and anxiety grouped used GAD‐7.

^b Comorbid symptoms were different for depression and anxiety disorders. Depression used GAD‐7, and all anxiety disorders used PHQ‐9.

^c This group was merged with another group because the small sample size in the group caused convergence issues.

^d Due to a mistake in the clinic's questionnaire setup, 64 patients in specific phobia treatment did not receive the FQ in the questionnaire at baseline. In order to include baseline severity in the prediction models, we had to exclude these missing cases. Thus, the sample size in the adjusted analyses and in the “severity at screening” analysis is only n = 50.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

3.3. Objective 3: Completion

Completers showed slower improvements compared with non‐completers for depression (0.58, 95% CI: 0.42; 0.74) and panic disorder (0.22, 95% CI: 0.07; 0.37). Contrarily, completers in social phobia treatment improved faster than non‐completers (−1.38, 95% CI: −1.91; −0.84). See Table 6 and Figure 3.

TABLE 6.

Differences between completers and non‐completers on symptoms at screening and rate of change.

	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobias N = 104
Outcome	PHQ‐9	PDSS‐SR	SIAS	FQ main phobia
Variable	Beta (95% CI)
Estimated change per assessment point	−1.69 (−1.82; −1.56) ***	−1.07 (−1.19; −0.95) ***	−0.57 (−0.92; −0.22) **	−0.19 (−0.35; −0.03) *
Completion
Non‐completer (reference level)	—	—	—	—
Completer	−1.82 (−2.50; −1.14) ***	−0.17 (−1.14; 0.80)	3.22 (−0.31; 6.75)	0.70 (−0.18; 1.58)
Estimated change… × completion
Non‐completer (reference level)	—	—	—	—
Estimated change…× completer	0.58 (0.42; 0.74) ***	0.22 (0.07; 0.37) **	−1.38 (−1.91; −0.84) ***	−0.11 (−0.29; 0.07)

Note: All models are joint models consisting of a mixed‐effects submodel and a Cox proportional hazards submodel. We report on the estimates of the mixed‐effects part of the joint model. Significant results are highlighted with bold text.

Abbreviations: FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

FIGURE 3.

Symptom developments of primary and secondary symptoms for all four treatment programs comparing completers with non‐completers. Red lines represent non‐completers, and blue lines represent completers. Completion was defined as completing central treatment content within 16 weeks of starting treatment. FQ, Fear Questionnaire; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PDSS‐SR, Panic Disorder Severity Scale‐Self Report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

3.4. Objective 4: Effect on Comorbid Symptoms

The effects on comorbid symptoms were significant for all programs, from screening to baseline and to post‐treatment. See Table 7. All treatments showed large effects from screening to post‐treatment (d = 0.90–1.24) on co‐morbid symptoms and moderate to large effects from baseline to post‐treatment (d = 0.59–0.98). See Table 8.

TABLE 7.

Change in comorbid symptoms from screening to first assessment during treatment and post‐treatment.

	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobias N = 104
Outcome	GAD‐7	PHQ‐9	PHQ‐9	PHQ‐9
Variable	Beta (95% CI)
Screening (reference level)	—	—	—	—
Baseline (after session 1)	−0.67 (−0.96; −0.38) ***	−2.93 (−3.39; −2.47) ***	−2.42 (−2.91; −1.92) ***	−2.40 (−3.20; −1.60) ***
Post‐treatment	−5.08 (−5.68; −4.49) ***	−6.64 (−7.45; −5.84) ***	−5.17 (−6.16; −4.17) ***	−4.60 (−5.62; −3.57) ***

Note: Estimates from linear mixed‐effects models. Significant results are highlighted with bold text.

Abbreviations: CI, Confidence Interval; GAD‐7, Generalized Anxiety Disorder‐7 item scale; PHQ‐9, Patient Health Questionnaire‐9.

^*** p < 0.001.

TABLE 8.

Effect sizes for comorbid symptoms for all four treatment programs with screening as the reference.

Treatment	Baseline (after session 1)		Post‐treatment		Screening—post	Baseline—post
	n	M (SD)	n	M (SD)	d	d
Depression	719	10.9 (4.6)	191	6.4 (4.7)	1.14	0.98
Panic disorder	375	8.4 (5.3)	101	4.5 (3.9)	1.24	0.79
Social phobia	277	9.4 (5.2)	72	6.3 (4.9)	1.05	0.61
Specific phobia	104	4.9 (4.4)	31	2.5 (2.9)	0.90	0.59

Note: The depression program used the generalized anxiety disorder‐7 item scale (GAD‐7) to measure co‐morbid anxiety symptoms. The anxiety programs used the patient health questionnaire‐9 (PHQ‐9) to measure co‐morbid depressive symptoms.

Abbreviations: d, Cohen's d; M, Mean; SD, Standard deviation.

3.5. Clinically Significant Change

A total of 26.7% patients improved during treatment. Of the patients above cutoff at baseline (n = 949/1371, 69.2%), 27.0% recovered, and 21.8% reliably recovered. Deterioration occurred in 5.8% of patients. A large proportion of patients were already below cutoff at baseline (30.8%). See Table 9 and Figure 4. Because such a large group was below cut‐off at baseline, we decided to provide estimates for improvement rates based on the screening questionnaire as well. These results are reported in the supplementary materials.

TABLE 9.

Number of patients exhibiting clinically significant changes.

	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobia N = 50 a	Total N = 1421
Improved b	210 (29.2%)	93 (24.8%)	60 (21.7%)	16 (32.0%)	379 (26.7%)
Recovered c , d	163 (31.2%)	53 (24.1%)	40 (19.3%)	NA e	256 (27.0%) f
Reliably recovered c , d	133 (25.5%)	44 (20.0%)	30 (14.5%)	NA e	207 (21.8%) f
Deteriorated	23 (3.2%)	39 (10.4%)	17 (6.1%)	3 (6.0%)	82 (5.8%)
Below cut‐off at baseline	197 (27.4%)	155 (41.3%)	70 (25.3%)	NA e	422 (30.8%) f

Abbreviations: FQ Main Phobia, Fear Questionnaire, Main Phobia Subscale; PDSS‐SR, Panic Disorder Severity Scale‐Self report; PHQ‐9, Patient Health Questionnaire‐9; SIAS, Social Interaction Anxiety Scale.

^a Due to a mistake in the clinic's questionnaire setup, 64 patients in specific phobia treatment did not receive the FQ in the questionnaire at baseline.

^b Reliable improvement for each questionnaire was defined as a decrease of PHQ‐9 ≥ 5; PDSS‐SR ≥ 6; SIAS ≥ 10; FQ Main Phobia ≥ 2.

^c Cutoffs for recovery for each questionnaire were defined as: PHQ‐9 < 10; PDSS‐SR < 8; SIAS < 37.

^d Percentages are based on the number of patients above cutoff at baseline. Not the full sample. Depression, n = 522; Panic disorder, n = 220; Social Phobia, n = 207.

^e There are no estimates for recovery for specific phobia because the FQ does not have a specified cutoff value to identify specific phobias.

^f Specific phobia not included.

FIGURE 4.

Distribution of baseline scores compared to scores at last observation. The solid diagonal line and blue dots indicate no change. The upper dashed line and red dots indicate deterioration. The lower dashed lines and green dots/triangles indicate improvement. The blue dots between the dashed lines indicate no change (neither improved nor deteriorated). The horizontal red dashed line indicates the cutoff for the questionnaire. Triangles indicate recovery from a screening score above cutoff. Green triangles indicate reliable recovery (recovered and improved).

Around 25% of non‐completers still showed improvements (see Table 10). This number was a little higher for depression (28.9%) and panic disorder (28.8%), but quite a bit lower for social phobia (15.9%). For social phobia and specific phobia, the proportion of improvers was much higher among completers than non‐completers, but for panic disorder the proportion of improvers was lower among completers.

TABLE 10.

Proportion of improvers and non‐improvers split between completers and non‐completers.

	Depression N = 719	Panic disorder N = 375	Social phobia N = 277	Specific phobia N = 50	Total N = 1421
Non‐completers (% of N)	418 (58.1%)	229 (61.1%)	208 (75.1%)	19 (38.0%)	874 (61.5%)
Improvers (% of non‐completers)	121 (28.9%)	66 (28.8%)	33 (15.9%)	5 (26.3%)	225 (25.7%)
Non‐improvers a (% of non‐completers)	297 (71.1%)	163 (71.2%)	175 (84.1%)	14 (73.7%)	649 (74.3%)
Completers (% of N)	301 (41.9%)	146 (38.9%)	69 (24.9%)	31 (62.0%)	547 (38.5%)
Improvers (% of completers)	89 (29.6%)	27 (18.5%)	27 (39.1%)	11 (35.5%)	154 (28.2%)
Non‐improvers a (% of completers)	212 (70.4%)	119 (81.5%)	42 (60.9%)	20 (64.5%)	393 (71.8%)

^a Non‐improvers consist of both participants classified as “deteriorated” and as “no change”.

4. Discussion

4.1. Summary of Findings

Our first objective was to examine the overall effectiveness of routine care iCBT for depression, panic disorder, social phobia, and specific phobia. For all programs, symptoms decreased significantly for each completed assessment. From baseline to last observation, we observed large effects for depression (d = 0.87) and social phobia (d = 0.80), a moderate effect for panic disorder (d = 0.62), and a small effect for specific phobia (d = 0.47). Improvement and recovery were both observed for about a quarter of the patients (26.7% and 27.0% respectively). However, a large proportion (30.8%) were already below cutoff at baseline, which made further improvements as defined by the RCI unlikely. (A large proportion of improvement seems to have occurred between screening and baseline, since the proportion of improvers based on the screening questionnaire was 57.2%. See supplementary materials.) Effect sizes using the screening questionnaire were also larger for all treatments except for social phobia. Thus, it is worth noting, that the assessment interview and intro treatment session also seem to have substantial effects and might have deflated effects measured from the baseline assessment and onward.

The second objective was to investigate potential predictors of treatment response. Severity at screening was the only predictor that was significant for all programs. For depression and panic disorder, comorbid symptom severity was also significant. Both indicated improved treatment response with higher severity of symptoms. These findings are not surprising, since higher baseline scores provide more room for improvement and extreme scores tend to move toward the sample mean over time (Barnett et al. 2005). However, it could also be interpreted that iCBT seems to work well regardless of severity. Panic disorder was the only treatment that had significant predictors in the adjusted models. Being in a relationship, being a student (receiving an educational stipend), and younger age were all associated with a larger symptom reductions per assessment point. For depression, was associated with smaller symptom reduction per assessment point, but this association disappeared in the adjusted models, indicating that it was confounded by one of the other variables.

Our third objective was to investigate the effect of completion on treatment response. Interestingly, non‐completers showed faster improvement than completers for both depression and panic disorder. Visual inspection of the graphs (Figure 3) indicates that for depression, non‐completers simply catch up after having slightly higher scores at screening. For panic disorder, non‐completers improve slightly more early on, but completers still improve more over the full course of treatment. Contrarily, completers in social phobia showed better treatment response compared with non‐completers. For depression and panic disorder, this could indicate that patients are more likely to discontinue because they feel their symptoms have been alleviated. However, for social phobia, patients might discontinue because they do not experience sufficient effect. The proportions of non‐completers that showed improvements support this interpretation, since around 29% of non‐completers showed improvements for depression and panic disorder, but only 16% of non‐completers showed improvements for social phobia. However, for all treatments, the majority of non‐completers did not show any improvements.

Lastly, we examined the effect on comorbid symptoms. All treatments showed significant effects with large effect sizes at post‐treatment.

4.2. Comparison to Previous Studies

The treatment effects observed in this study are comparable to other routine care iCBT programs (depression: g = 1.18; anxiety: g = 0.94) (Etzelmueller et al. 2020). Our effect sizes were lower than that when using the baseline (depression: d = 0.87; anxiety d = 0.47–0.80), but higher when using the screening (depression: d = 1.35; anxiety: d = 0.50–1.11). The baseline effects are most comparable to previous studies, since most studies use a baseline questionnaire after screening (Ritola et al. 2022; El Alaoui et al. 2015; Hedman et al. 2014, 2013; Ruwaard et al. 2012). However, it should be noted that the baseline assessment in this study occurs after the first intro session, which could deflate the estimates.

Rates of improvement and recovery were also comparable, but in the low end compared to previous routine care iCBT studies (34%–66.2% and 20%–61.6%, respectively) (Flygare et al. 2020; Ritola et al. 2022; El Alaoui et al. 2015; Hedman et al. 2014, 2013; Ruwaard et al. 2012; Gellatly et al. 2018; Nordgreen et al. 2019, 2018; Nordgreen, Gjestad, Andersson, and et al. 2018; Titov et al. 2018). The large proportion of patients already below cutoff at baseline in our sample were unlikely to show further improvements and might somewhat deflate the estimates. A meta‐analysis of face‐to‐face CBT for depression has shown improvement rates to be around 42% and recovery rates around 34% (Cuijpers et al. 2023). Improvement for depression in this study was slightly lower at 29%, but recovery rates were almost comparable at 31%. Lastly, the rates in this study are in the high end when compared to care‐as‐usual (18%–27% and 7%–22%) (Cuijpers et al. 2024). However, these two studies did not limit inclusion to mild to moderate depression.

Only a few other effectiveness studies have conducted predictor analyses. Our findings are in line with a study on transdiagnostic iCBT that showed higher severity and lower age was associated with larger improvements (Edmonds et al. 2018). However, another study on generalized anxiety disorder showed a reverse relationship between age and treatment response (Ritola et al. 2022). Interestingly, we did not find any associations between education and treatment response. Unlike the study on transdiagnostic iCBT (Edmonds et al. 2018).

Looking beyond effectiveness studies, prediction studies in iCBT show unclear or contradictory results (Haller et al. 2023; Sextl‐Plötz et al. 2024). The presence and directions of associations in different studies might be sensitive to the measurement scales and categorizations used across studies, as well as sample size. Our study contributes to the field with analyses based on large samples, especially for depression and panic disorder. Future studies should investigate more theoretically relevant predictors, as suggested by (Haller et al. 2023), and process variables from iCBT treatments, such as logins and messages written to the therapist (Linnet et al. 2022).

4.3. Methodological Considerations

Not having a control group limits the interpretation of our findings. However, this is part of the trade‐off for investigating the effect in a natural clinical setting compared with the controlled conditions in RCTs. Additionally, the results are comparable to findings from other routine care iCBT studies.

The timing of the baseline questionnaire, might have deflated our effect sizes and estimates of improvement. Therefore, we have reported effect sizes from both screening and baseline. This should give an indication of the full effect of the service provided by the clinic from screening, as well as a conservative estimate of the effect of the treatment program from baseline. The same issue is present in the RCI estimates. Therefore, we decided to report the number of cases below cutoff for their respective questionnaire at baseline. This illustrates that a large group had already improved to a point, where they were unlikely to show further improvements during treatment. Thus, the conservative results for number of improved and recovered participants, should also be interpreted in this light.

We were unable to provide details regarding the reasons for discontinuation. It would have been helpful to know, how many participants involved their therapist in their decision to discontinue, the reasons they provided, as well as how many simply stopped responding to their therapist's messages. Unfortunately, the clinic did not have a consistent data collection on these issues throughout the study period.

Lastly, the many analyses increase the risk of false positives, so the results should be interpreted cautiously. Especially results with significance levels close to the border (e.g., severity at screening for specific phobia, comorbid symptoms). However, the large sample for depression and panic disorder and the generalizable findings across disorders for severity at screening and comorbidity still strengthen these findings.

5. Conclusion

This study supports the effect of routine care iCBT for depression and anxiety based on large sample sizes. Higher severity of symptoms at screening was associated with faster improvements for all programs, as was comorbid symptoms for depression and panic disorder. Non‐completion had different effects depending on treatment program. For depression and panic disorder, non‐completers had a faster rate of improvement, while non‐completers had a slower rate of improvement for social phobia.

5.1. Significant Outcomes

1. Routine care iCBT is effective for the treatment of anxiety and depression.

2. Few baseline variables and sociodemographic variables affect treatment effect. The effects of significant predictors have limited clinical application.

5.2. Limitations

1. Since this was a study of routine care data, no control group was available.

2. Several predictor analyses were conducted, increasing the risk of false positives.

Author Contributions

Esben Kjems Jensen: conceptualization, data curation, formal analysis, funding acquisition, methodology, project administration, validation, visualization, writing–original draft, writing–review and editing. Mia Beck Lichtenstein: conceptualization, funding acquisition, supervision, writing–review and editing. Heleen Riper: conceptualization, supervision, writing–review and editing. Kim Mathiasen: conceptualization, funding acquisition, methodology, supervision, writing–review and editing.

Funding

This publication was funded by the fund to support clinical doctoral candidates in the Region of Southern Denmark under Grant Nos. 21/58106, the Psychiatric Research Fund in Southern Denmark under Grant Nos. A4180, and Jascha Foundation under Grant No. 2021‐0069.

Ethics Statement

The regional ethical committee in Southern Denmark evaluated the study and declared it exempt from ethical approval (file no. 20202000‐38).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Jensen, Esben K. , Lichtenstein Mia B., Riper Heleen, and Mathiasen Kim. 2026. “Effectiveness of Guided Internet‐Based Cognitive Behavioral Therapy for Adult Anxiety and Depression in Routine Care: An Observational Study,” International Journal of Methods in Psychiatric Research: e70058. 10.1002/mpr.70058.

Data Availability Statement

The authors have nothing to report.