Therapist‐supported Internet cognitive behavioural therapy for anxiety disorders in adults

Summary of findings for the main comparison. Therapist‐supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults

Therapist‐Supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults
Patient or population: patients with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: waiting list, attention, information, or online discussion group only control
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Waiting list, attention, information, or online discussion group only control	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	Study population		RR 3.75 (2.51 to 5.60)	866 (12 studies)	⊕⊕⊝⊝ low²
	14 per 100	53 per 100 (35 to 79)
	Moderate
	10 per 100	39 per 100 (26 to 58)
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean anxiety symptom severity at post‐treatment in the intervention groups was 1.06 standard deviations lower (1.29 to 0.82 lower)		2147 (28 studies)	⊕⊕⊝⊝ low^3,4	A standard deviation of 0.80 or greater represents a large difference between groups⁵
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.75 standard deviations lower (0.98 to 0.52 lower)		1496 (19 studies)	⊕⊕⊝⊝ low^4,6	A standard deviation of 0.80 or greater represents a large difference between groups⁵
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.47 standard deviations higher (0.38 to 0.57 higher)		1639 (23 studies)	⊕⊕⊕⊝ moderate⁶	A standard deviation of 0.50 represents a moderate difference between groups⁵
Adverse events at post‐treatment not reported	Study population		Not estimable	0 (0)	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
	See comment	See comment
	Moderate

Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	Study population		Not estimable	0 (13)	See comment	Studies reported high overall treatment satisfaction for therapist‐supported ICBT
	See comment	See comment
	Moderate

The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement. ² Downgraded for risk of bias (‐1) primarily because four of the included studies did not blind their outcome assessors to participants' group assignment and due to lack of blinding of participants and study therapists. Downgraded for publication bias (‐1) because only 12 studies reported this outcome. Not downgraded for inconsistency (0) because heterogeneity was reduced following subgroup analysis by anxiety disorder. ³ Downgraded for risk of bias (‐1) primarily due to minor concerns with selective outcome reporting, incomplete outcome data, baseline imbalances in a few studies, and lack of blinding of participants and study therapists. ⁴ Downgraded for inconsistency (‐1) because the heterogeneity amongst the included studies was quite high. This may be explained by the variety of anxiety disorders investigated and differences in the treatment details; however, the number of studies that could be included in subgroup analyses was not sufficient to provide useful reasons for this heterogeneity. ⁵ According to Cohen's (1969) interpretation of effect sizes. ⁶ Downgraded for risk of bias (‐1) primarily because two studies included baseline imbalances in participant severity across study groups and due to lack of blinding of participants and study therapists.

Summary of findings 2. Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults

Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults
Patient or population: patients with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: unguided ICBT
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Unguided ICBT	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	See comment	See comment	Not estimable	54 (1 study)	⊕⊝⊝⊝ very low^2,3	Not pooled because only one study in this comparison for this outcome
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean disorder‐specific anxiety symptom severity at post‐treatment in the intervention groups was 0.22 standard deviations lower (0.56 lower to 0.13 higher)		312 (5 studies)	⊕⊝⊝⊝ very low^4,5,6	A standard deviation of 0.20 represents a small difference between groups⁷
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.28 higher (2.21 lower to 2.78 higher)		138 (2 studies)	⊕⊝⊝⊝ very low^3,4
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.07 standard deviations higher (0.37 lower to 0.5 higher)		199 (3 studies)	⊕⊝⊝⊝ very low^4,5,6	A standard deviation of 0.10 represents a small difference between groups⁷
Adverse events at post‐treatment not reported	Study population		Not estimable	0 (0)	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
	See comment	See comment
	Moderate

Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	See comment	See comment	Not estimable	0 (2 studies)	See comment	Studies generally reported higher satisfaction with therapist‐supported ICBT
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement. ² Downgraded for risk of bias (‐1) primarily because of lack of blinding of outcome assessors, participants, and study therapists. ³ Downgraded for imprecision (‐2) as there is only one or two studies within the comparison for this outcome. ⁴ Downgraded for risk of bias (‐1) due to lack of blinding of participants and study therapists. ⁵ Downgraded for inconsistency (‐1) as the heterogeneity amongst the included studies was quite high. ⁶ Downgraded for imprecision (‐1) as there is a limited number of studies included in the comparison for this outcome. ⁷ According to Cohen's (1969) interpretation of effect sizes.

See Characteristics of included studies; Characteristics of excluded studies

Summary of findings 3. Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults

Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults
Patient or population: adults with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: face‐to‐face CBT
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Face‐to‐face CBT	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	Study population		RR 1.09 (0.89 to 1.34)	365 (4 studies)	⊕⊕⊝⊝ low^2,3
	41 per 100	44 per 100 (36 to 54)
	Moderate
	45 per 100	49 per 100 (40 to 61)
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean anxiety symptom severity at post‐treatment in the intervention groups was 0.06 standard deviations higher (0.25 lower to 0.37 higher)		450 (7 studies)	⊕⊕⊝⊝ low^4,5	There was no significant difference between groups
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.06 standard deviations higher (0.42 lower to 0.55 higher)		343 (6 studies)	⊕⊕⊝⊝ low^4,5	There was no significant difference between groups
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.26 standard deviations higher (0.06 to 0.45 higher)		392 (5 studies)	⊕⊕⊝⊝ low^2,3	A standard deviation of 0.20 represents a small difference between groups⁶
Adverse events at post‐treatment ‐ not reported	See comment	See comment	Not estimable	‐	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	Study population		Not estimable	0 (2)	See comment	Studies reported high overall treatment satisfaction across both conditions
	See comment	See comment
	Moderate

The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement at post‐treatment. ² Downgraded for imprecision (‐1) primarily due to small sample size. ³ Downgraded for risk of bias (‐1) due to lack of blinding of participants and study therapists. ⁴ Downgraded for risk of bias (‐1) primarily because two included studies provided incomplete outcome data (though sensitivity analyses suggest no difference in findings when these studies are excluded) and due to lack of blinding of participants and study therapists. ⁵ Downgraded for inconsistency (‐1) primarily due to unexplained heterogeneity. ⁶ According to Cohen's (1969) interpretation of effect sizes.

Background

Description of the condition

Individuals with anxiety disorders experience excessive anxiety (fear or worry) which is disproportionate to actual threat or danger and significantly interferes with normal daily functioning. Anxiety disorders can include a range of physical (for example, trembling, tense muscles, rapid breathing), cognitive (for example, worries, difficulty concentrating), emotional (for example, distress, negative affect, irritability), and behavioural (for example, difficulty sleeping, hyperarousal) symptoms. Often those with anxiety disorders develop maladaptive strategies to lessen anxiety, such as avoidance (Health Canada 2002; Wilson 2006) or substance use (Stewart 2008). Studies from Canada (Statistics Canada 2004), the USA (Kessler 2005a), Australia (Slade 2007), Nigeria (Gureje 2006), and Europe (ESEMeD/MHEDEA 2000 Investigators 2004) suggest that 6% to 18% of adults experience an anxiety disorder every year. Moreover, rates of remission within one year are low, that is, from 33% to 42% across specific anxiety disorders (Robins 1991).

There are many types of anxiety disorders, including panic disorder (PD), agoraphobia, social phobia, post‐traumatic stress disorder (PTSD), acute stress disorder, generalized anxiety disorder (GAD), obsessive compulsive disorder (OCD), and specific phobia. These are diagnosed according to criteria outlined in the Diagnostic and Statistical Manual of Mental Disorders (DSM IV‐R) (APA 2000) or the International Classification of Diseases (ICD 10) (WHO 1999). Anxiety disorders often co‐occur with each other (Kessler 2005a) as well as with mood disorders (Fava 2000) and substance abuse or dependence (Stewart 2008). They tend to have an early onset (Kessler 2005b) and chronic course (Bruce 2005). Anxiety disorders also have a major economic impact; for instance, costs of direct treatment, unnecessary medical treatment, and work absences or lost productivity amount to more than USD 40 billion per year in the United States (DuPont 1996; Greenberg 1999). Studies have shown significantly higher annual per capita medical costs for primary care patients with social phobia than for those with no mental health diagnosis (GBP 11,952 and EUR 2957 respectively) (Acarturk 2009); primary care patients with PD versus those with a chronic somatic condition (EUR 10,269 versus EUR 3019) (Batelaan 2007); and primary care patients with GAD as compared to those without GAD (USD 2375 versus USD 1448) (Revicki 2012).

Description of the intervention

Accumulating research supports the efficacy of CBT in the treatment of anxiety disorders (Bisson 2007; Hunot 2007; Norton 2007; Stewart 2009) and anxiety symptoms (Deacon 2004). As its name suggests, CBT includes both cognitive as well as behavioural interventions or techniques. It has no one 'founder' and now exists in many different forms. Its roots, however, lie largely in the work of Aaron Beck (Beck 1979). While pharmacotherapy (most commonly, benzodiazepines or selective serotonin reuptake inhibitors) has been shown to be effective in the treatment of anxiety disorders, meta‐analyses and review articles suggest that CBT is as effective in the acute phase of anxiety and may be more effective than pharmacotherapy or a combination of both treatments in the long term (Westra 1998; Otto 2000; Otto 2005; Pull 2007). Moreover, some anxiety medications pose significant risk for addiction (McNaughton 2008) or serious side effects, or both (Buffett‐Jerrot 2002; Choy 2007).

Unfortunately, certain barriers (for example, time constraints, transportation problems, stigma, long waiting lists, a lack of sufficiently qualified clinicians) continue to limit access to CBT (Alvidrez 1999; Young 2001; Mohr 2006). Many of these barriers are particularly relevant for those living in rural communities (Yuen 1996; Rost 2002; Hauenstein 2006). National surveys in Canada (Statistics Canada 2004) and the US (Kessler 2004) suggested that less than one third (only 32% and 20%, respectively) of those with a current psychiatric disorder received some form of treatment in the past year. In a Canadian sample, only 11% of individuals with an anxiety disorder had received treatment (Ohayon 2000). Increasingly, efforts are being made to improve access to CBT on a large scale, particularly for those groups who are most at risk due to lack of services (for example, the UK‐based National Health Service 'Improving Access to Psychological Therapies' (IAPT) programme launched in 2006) (Department of Health 2008). A distance delivery approach wherein CBT is delivered over the Internet with a therapist providing support by telephone or e‐mail is one way to minimize treatment barriers and increase access to care while still delivering empirically‐supported treatment. Such an approach could increase access to mental health professionals for those in rural areas, facilitate treatment for those of limited mobility, and increase patient confidentiality (that is, by engaging in treatment from home clients do not 'risk' being seen at mental health clinics) and privacy (for example, a degree of visual anonymity). The widespread availability of the Internet makes this type of intervention feasible and worth consideration. Recent systematic reviews of computer‐ and Internet‐based treatment for mental health problems suggest largely that these types of treatment are more effective than a waiting list control and equally effective as face‐to‐face psychotherapy in treating anxiety and depression symptoms (Spek 2007; Bee 2008; Cuijpers 2009; Reger 2009; Cuijpers 2010).

How the intervention might work

Therapist‐supported ICBT should work to treat anxiety in the same manner as conventional face‐to‐face CBT. The underlying principles of CBT posit that psychopathology, or emotional disturbances, are the result of cognitive distortions and maladaptive behaviour. Whereas there are hypotheses about the relative importance of cognitive and behavioural techniques, as well as suggestions that the strong collaborative working relationship between the therapist and client are key to the success of CBT, the exact mechanisms of action in CBT are not yet well understood (Olatunji 2010). It is thought that disorder‐specific symptoms develop as a result of a particular pattern of dysfunctional cognitions in combination with a specific set of behaviours that serve to exacerbate these dysfunctional cognitions further (Beck 2005). As such, CBT works to improve symptoms by treating these maladaptive cognitions and behaviours.

In essence, cognitive techniques and behaviour modification strategies are used to identify, evaluate, and challenge underlying maladaptive thoughts and beliefs. As an example, it is thought that catastrophic thoughts about the outcomes of experiencing arousal‐related physiological sensations, as well as inaccurate predictions about the probability of these dangerous outcomes, and avoidance of situations that may induce these sensations contribute to the development and maintenance of PD (Clark 1986; Barlow 1988). Accordingly, CBT for panic uses cognitive restructuring techniques to teach individuals to identify and challenge their maladaptive cognitions and beliefs. This is combined with the use of gradual, repeated exposure to feared sensations to help individuals revise their perceptions of threat and reduce their fear of these arousal‐related physiological sensations (Landon 2004). A similar description of the CBT model could be provided for the other anxiety disorders (for example, social phobia) (Heimberg 2002). Whereas the underlying cognitive and behavioural principles are evident in the CBT interventions for each of the anxiety disorders, current forms of CBT also target core components of a particular disorder and, as such, specific models of CBT now exist for each disorder, which modify and adapt CBT principles to fit disorder‐specific symptoms (for example, specific phobia (Ost 1997); OCD (Salkovskis 1985; Foa 2010); PD (Clark 1986; Casey 2004); social phobia (Heimberg 2002); GAD (Dugas 2007); PTSD (Ehlers 2000).

ICBT therapists would be expected to draw on these models in the same manner as face‐to‐face CBT therapists. Typically, ICBT involves the client following a written treatment program available on the Internet in conjunction with receiving therapist support, either via telephone calls, texts, or e‐mail (Andersson 2006). The intervention involves content that mimics that of face‐to‐face CBT, therapist‐client contact (albeit through non‐traditional means), and the client engaging in further 'homework' outside of the session. As such, we anticipated that ICBT will work in the same way and as well as traditional face‐to‐face CBT.

Why it is important to do this review

Recently, research into ICBT has elicited considerable interest from within the scientific and clinical communities. With advances in modern communication technologies and their widespread availability, this type of treatment is quickly becoming a more realistic option. These advances have come at a time when long waiting lists and a lack of treatment availability stand in stark contrast to the growing emphasis on the importance of mental health and provision of evidence‐based treatments. A desire to pursue Internet treatment as a viable option to increase access to treatment is growing. The importance of ensuring that the decision to promote such treatment is grounded firmly in high quality evidence is therefore paramount.

The present review asked whether therapist‐supported ICBT is efficacious in treating anxiety, and if it is as efficacious as face‐to‐face CBT. Past meta‐analyses have reviewed the efficacy of ICBT for anxiety symptoms (Spek 2007). A number of reviews that have included ICBT have looked more broadly, however, at health problems in general (Barak 2008; Bee 2008) or all computer‐based interventions (Cuijpers 2009; Reger 2009; Andrews 2010). Moreover, many of these reviews have not focused on the role of therapist involvement (for example, Cuijpers 2009; Reger 2009; Andrews 2010). Ultimately, as the field of ICBT is growing quickly, an updated review on therapist‐supported ICBT is needed. The findings of this review will be helpful in guiding the path of future research in this field away from continued replication of established findings and toward addressing gaps in the literature and considering the next steps in ICBT implementation.

There is a Cochrane Review on media‐delivered CBT and behavioural therapy (BT) (self‐help) for anxiety disorders (Mayo‐Wilson 2013). Mayo‐Wilson's review answers questions about the efficacy of delivering CBT to clients in non‐traditional formats, including via the Internet. In the protocol of their review, Mayo‐Wilson specified that they would not include studies with therapist contact. With a post‐protocol change, they revised their review to include studies that involved therapist contact with the qualifier that the interventions must be able to be delivered stand‐alone without therapist contact. With this in mind, the focus of their review remains largely on self‐help therapies in which therapist involvement is not necessary and treatment is largely client‐driven. Mayo‐Wilson did not conduct analyses separating out those interventions with and without therapist contact. As such, a meta‐analysis with a particular emphasis on the efficacy of therapist‐supported ICBT is needed, particularly as at this point there remains conflicting evidence of the comparable efficacy of self‐help and therapist‐supported interventions (for example, Spek 2007; Titov 2008c; Berger 2011). The present review considered the specific efficacy of therapist‐supported ICBT in comparison to each of a waiting list control (that is, no treatment), traditional face‐to‐face CBT, and self‐help interventions and as such will fill a gap in the literature and answer current calls for research in this area (Reger 2009). The protocol for the present review can be found in the Cochrane Library (Olthuis 2011).

Objectives

To assess the effects of therapist‐supported ICBT on remission of anxiety disorder diagnosis and reduction of anxiety symptoms in adults as compared to waiting list control, unguided CBT, or face‐to‐face CBT. Effects of treatment on quality of life and patient satisfaction with the intervention were also assessed.

Methods

Criteria for considering studies for this review

Types of studies

We included parallel group randomised controlled trials (RCTs), cross‐over, and cluster randomised trials.

Types of participants

Participant characteristics

We included studies of adults (over 18 years of age; no upper limit).

Diagnosis

Participants with a primary diagnosis of an anxiety disorder according to the DSM‐III (APA 1980), DSM‐III‐R (APA 1987), DSM‐IV (APA 1994), DSM‐IV‐TR (APA 2000), ICD‐9 (WHO 1979) or ICD‐10 (WHO 1999) diagnostic criteria.

We included studies that focused on or adequately reported subgroup information for any of the following anxiety disorders: panic disorder (PD) with or without agoraphobia, agoraphobia without a history of panic, social phobia (social anxiety disorder), post‐traumatic stress disorder (PTSD), acute stress disorder, obsessive compulsive disorder (OCD), specific phobia, generalized anxiety disorder (GAD), and anxiety disorder not otherwise specified. Included studies used diagnoses determined using a validated diagnostic instrument, for example, the Structured Clinical Interview for DSM‐IV‐TR Axis I Disorders (SCID‐I) (First 2002).

Setting

We included studies in which treatment entailed participants engaging in the treatment from their homes and therapists located at primary care settings, university laboratories, community mental health clinics, or private practice clinics. Participants could be treatment‐seeking community members responding to media advertisements for study participation or they could be referred to the study by a health professional.

Co‐morbidities

We included studies of participants with co‐morbid diagnoses (for example, major depressive disorder, substance abuse) only if they had been diagnosed with a primary anxiety disorder. We did not include studies of participants reporting anxiety symptoms that did not meet criteria for an anxiety disorder (for example, participants with a clinical presentation of major depressive disorder who reported subthreshold anxiety symptoms or participants scoring high on measures of anxiety symptoms but who were not assessed for a DSM diagnosis).

Types of interventions

Experimental interventions

Cognitive behavioural therapies

We included studies that investigated the efficacy of a therapist‐supported Internet cognitive behavioural therapy (CBT), behavioural therapy (BT), or cognitive therapy (CT) intervention for anxiety, defined as the following.

BT interventions must have been designed to change the behaviours that result from maladaptive anxiety‐related cognitions (we included interventions including, but not limited to, exposure, desensitization, and behavioural experiments).
CT must have been focused on elements of cognitive restructuring of irrational or maladaptive anxiety‐related cognitions.
CBT interventions consisted of some combination of the elements of CT and BT.

Whereas psychoeducation often is an important part of CBT, we did not consider psychoeducation alone to be a sufficient CBT intervention unless it included some of the other treatment components described here.

Internet interventions

To be considered an Internet intervention, CBT must have been delivered over the Internet through the use of web pages or e‐mail, or both. Crucially, Internet interventions must have included therapist support but this interaction could not be face‐to‐face. However, we included interventions that involved an initial face‐to‐face intake or interview session or an initial session to orient clients to the Internet delivery method or to engage in treatment planning, or a combination of these. Thus, therapist support must have occurred via e‐mail or the telephone, or both. Including only interventions that could be delivered entirely by distance methods reflected a primary motive for conducting this review, to find ways to increase access to treatment for those who may not be able to visit provider centres. While it was possible that Internet‐based interventions that provided some support in a face‐to‐face setting could be just as effectively restructured to be delivered completely by distance, it was more rigorous to include only studies that provided evidence specifically on the efficacy of Internet CBT delivered completely via distance methods. We did not select interventions based on their length, or the number or duration of sessions.

Comparator interventions

Waiting list, attention, information, or online discussion group only control condition (no intervention for participants beyond weekly status monitoring by research personnel or accessing online non‐treatment related disease information or discussion groups)
Unguided CBT (i.e., self‐help CBT with no therapist support)
Conventional face‐to‐face CBT interventions (including individual or group CBT delivered in a traditional face‐to‐face format)

Types of outcome measures

Primary outcomes

Efficacy of therapist‐supported ICBT in leading to clinically important improvement in anxiety as determined by a diagnostic interview, for example, the SCID‐I (First 2002) or the Anxiety Disorders Interview Schedule (ADIS‐IV) (DiNardo 1994) or a defined cut‐off on a validated scale, for example, the Yale Brown Obsessive Compulsive Scale (YBOCS) (Goodman 1989). In case the Clinical Global Impression scale change or improvement items (CGI) (Guy 1976) were used, we employed a score of 1 = 'very much' or 2 = 'much improved' to indicate clinically important improvement.
Efficacy of therapist‐supported ICBT in leading to reduction in anxiety symptom severity measured by scores on a validated, observer‐rated instrument, for example, the Hamilton Anxiety Rating Scale (Hamilton 1959), or a validated self‐report measure of: (a) disorder‐specific symptoms, for example, the Social Phobia Scale (SPS) (Mattick 1998), and (b) anxiety symptoms in general, for example, the Beck Anxiety Inventory (BAI) (Beck 1991).

Secondary outcomes

Quality of life as assessed by either measures of quality of life, for example, the Quality of Life Inventory (QOLI) (Frisch 1992), or measures of disability, for example the Sheehan Disability Scales (SDS) (Leon 1997) as increasing disability entails decreased quality of life. While research suggests that quality of life and disability are distinct but somewhat overlapping constructs (Hambrick 2003), quality of life measures have not often been conceptually or operationally distinguished from measures of disability, resulting in considerable overlap amongst indices of quality of life and disability (Mogotsi 2000). With this in mind, we anticipated an overlapping conceptualization of these two constructs in the included studies and included both types of measures within the meta‐analysis in order to capture all possible information about treatment outcome related to quality of life.
Participant satisfaction with the intervention. Participant satisfaction tends to be measured uniquely across different studies using a mix of qualitative and quantitative indices. In anticipation of this, we evaluated participants' satisfaction with the intervention of interest as compared to the comparator interventions in a qualitative manner.
Adverse events, in whatever manner reported by study authors.

Timing of outcome assessment

We performed separate analyses based on different periods of assessment: immediately post‐treatment and at one follow‐up period at least six months post‐treatment but not more than one year. When studies reported more than one follow‐up assessment point, we used the longest follow‐up period so as to provide the best estimate of the long‐term outcomes of the intervention.

Hierarchy of outcome measures

For primary outcomes, separate meta‐analyses were conducted for the two outcomes. The clinically important improvement in anxiety outcome measures were selected according to the following hierarchy, based on availability in a particular study: (1) diagnostic interview, (2) cut‐off on a validated scale, (3) CGI scores. For reduction in anxiety symptom severity, the outcomes of available observer‐rated and self‐report measures were statistically combined and a mean score was created across the measures within a particular study. Measures of variance for this mean score were created by combining standard deviations across studies according to the method described by Borenstein 2009. This method requires that the correlation between two measures be known; as such, in the case that this correlation was not known, the measures with better psychometric properties were included in the analysis.

For secondary outcomes, quality of life outcome measures were treated in the same way as anxiety symptom severity measures. Due to the qualitative nature of the other secondary outcome, participant satisfaction with the intervention, a hierarchy of outcome measures was not required.

Search methods for identification of studies

We used several methods to identify both published and unpublished studies for possible inclusion in this review (see below). We did not restrict studies to those reported in any particular language; however, we conducted searches in English and initiated contact with authors in English.

Electronic searches

The Cochrane, Depression, Anxiety and Neurosis Review Group's Specialised Register (CCDANCTR)

The Cochrane Depression, Anxiety and Neurosis Group (CCDAN) maintains two clinical trials registers at their editorial base in Bristol, UK, a references register and a studies‐based register. The CCDANCTR‐References Register contains over 39,500 reports of RCTs in depression, anxiety, and neurosis. Approximately 60% of these references have been tagged to individual, coded trials. The coded trials are held in the CCDANCTR‐Studies Register and records are linked between the two registers through the use of unique Study ID tags. Coding of trials is based on the EU‐Psi coding manual. Please contact the CCDAN Trials Search Co‐ordinator for further details. Reports of trials for inclusion in the Group's registers are collated from routine (weekly), generic searches of MEDLINE (1950 to date), EMBASE (1974 to date) and PsycINFO (1967 to date); quarterly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), and review‐specific searches of additional databases. Reports of trials are also sourced from international trial registers via the World Health Organisation (WHO International Clinical Trials Registry Platform (ICTRP), ClinicalTrials.gov, drug companies, and the handsearching of key journals, conference proceedings, and other (non‐Cochrane) systematic reviews and meta‐analyses.

Details of CCDAN's generic search strategies can be found on the Group‘s website (http://cmd.cochrane.org/), or available on request from the Trials Search Co‐ordinator (email: [email protected]).

To date, three searches have been run for this review. Two for the first published version (with searches to 12 April 2013 and 25 September 2014, Appendix 1) and the latest search to 16 March 2015 (listed below). All findings have now been fully incorporated into the meta‐analyses.

1. CCDANCTR‐Studies and CCDANCTR References:

The CCDANCTR was searched all years to 16 March 2015 using the following (amended) search strategy and results de‐duplicated against those retrieved previously (using the Cochrane Register of Studies (CRS) software).

#1 (anxiety or *phobi* or PTSD or post‐trauma* or “post trauma*” or posttrauma* or (stress and disorder*) or panic or OCD or obsess* or compulsi* or GAD):ti,ab,kw
#2 (therap* or train*):ti,ab
#3 (acceptance* or assertive* or brief* or commitment* or exposure or group or implosive or “problem solving” or problem‐solving or "solution focused" or solution‐focused or schema):ti,ab,kw
#4 (psychotherap* or *CBT* or cognitive or behavio* or “contingency management” or “functional analys*” or mindfulness* or “mind training” or psychoeducat* or relaxation or “role play*”):ti,ab,kw,ky,mh,mc,emt
#5 ((#2 and #3) or #4)
#6 (computer* or distance* or remote or tele* or Internet* or web* or WWW or phone or mobile or e‐mail* or email* or online* or on‐line or videoconferenc* or video‐conferenc* or "chat room*" or "instant messaging" or iCBT):ti,ab,kw
#7 (#1 and #5 and #6)
#8 (internet* or online or web*):ti
#9 (anxi* or *phobi* or panic or GAD or "general* anxiety" or OCD or obsess* or PTSD or *trauma* or "stress disorder*"):ti
#10 (assisted or administer* or administr* or coach* or guided or guidance or *therapist* or ((telephone or email) next (support or assist*))):ti,ab
#11 (#8 and #9 and #10)
#12 (#7 or #11)

2. International Trial Registries

Trial registries were searched (to 16 March 2015) to identify unpublished and/or ongoing studies. These included ClinicalTrials.gov and the WHO International Clinical Trials Registry Portal (ICTRP).

Searching other resources

Reference lists

We examined the reference lists of previous related meta‐analyses (Spek 2007; Bee 2008; Cuijpers 2009; Reger 2009; Andrews 2010; Cuijpers 2010) and of articles selected for inclusion in the present review.

Personal contacts and correspondence

We contacted experts in the field, including principal authors of RCTs in the field of ICBT for anxiety, via e‐mail and asked them if they were aware of any further studies which meet the present review’s inclusion criteria.

Unpublished studies

In order to search for unpublished studies, we searched international trial registries including via the WHO ICTRP (http://apps.who.int/trialsearch/) and ClinicalTrials.gov (www.clinicaltrials.gov) to March 2015.

Data collection and analysis

Selection of studies

In collaboration with the CCDAN Trials Search Co‐ordinator, one review author (JVO) conducted searches of electronic databases and reference lists and contacted authors in order to locate potential trials to be included in the review. Two review authors (JVO and KMB) independently assessed the titles and abstracts of the resulting lists of studies for relevance. We then obtained full articles for potentially relevant abstracts. Both review authors independently assessed the identified trials to determine eligibility as outlined in Criteria for considering studies for this review. We collated and compared assessments. In the case of disagreement with respect to trial eligibility, we made the final decision by discussion and consensus, if necessary with the involvement of another member of the review group (MCW or SHS, or both).

Data extraction and management

We independently extracted data from the included studies regarding methodology and treatment outcomes, and recorded the data using a data extraction spreadsheet designed by one of the review authors (JVO). If the included trials did not provide complete information (for example, details of dropout, group means and standard deviations), we contacted the primary investigator by e‐mail to attempt to obtain unreported data to permit an intention‐to‐treat (ITT) analysis. We contacted other investigators as needed.

Two review authors (JVO and KMB) independently extracted the following data from each trial report:

description of trial, including primary researcher and year of publication;
characteristics of trial methodology, including the diagnostic criteria employed, participant inclusion and exclusion criteria, the screening instrument(s) used, the inclusion or exclusion of co‐morbidity, the receipt of other interventions simultaneously, and the number of centres involved;
characteristics of participants, including age, gender, primary diagnosis, any co‐morbid diagnoses, and duration of primary symptoms;
characteristics of the intervention (for both the experimental and comparator interventions), including intervention classification (i.e., CBT, BT, CT), content and components (e.g., psychoeducation, relaxation training, exposure, cognitive restructuring), method of delivery of therapist support (e.g., telephone, e‐mail), duration, amount of therapist and experimenter contact, and number of participants randomised to each intervention; and
outcome measures employed, as listed in Types of outcome measures, as well as the dropout rates for participants in each treatment condition and whether the data reflected intention‐to‐treat (ITT) analyses with last observation carried forward (LOCF) or another method.

We subsequently recorded data in RevMan 5.3 data tables (RevMan 2014).

Main planned comparisons

We planned to compare each of the outcomes of interest, at post‐treatment and 6 to 12 month follow‐up, for each of the following comparisons:

therapist‐supported ICBT versus waiting list, attention, information, or online discussion group only control,
therapist‐supported ICBT versus unguided CBT, and
therapist‐supported ICBT versus face‐to‐face CBT.

Assessment of risk of bias in included studies

We assessed the risk of bias in each included study using the Cochrane Collaboration’s 'risk of bias' tool (Higgins 2011a). We assessed the following six areas for risk of bias.

Sequence generation: was the allocation sequence of participants adequately randomised?
Allocation concealment: was the allocation sequence adequately concealed from participants as well as those involved in the enrolment and assignment of participants?
Blinding: were participants, study personnel, and those assessing outcomes kept unaware of participants’ allocation to a study condition throughout the course of the investigation?
Incomplete outcome data: were there incomplete data for the main or secondary outcomes (e.g., due to attrition)? Were incomplete data adequately addressed?
Selective reporting: was the study free of suggestions of selective reporting of outcomes (e.g., reporting of a subset of outcomes on the basis of the results)?
Other potential threats to bias: was the study free of any other problems (e.g., early stopping, baseline imbalance, cross‐over trials) that could have introduced bias?

We did not assess risk of bias related to therapist experience and qualifications. Evidence in the field as to the impact of therapist experience on treatment outcomes remains mixed (for example, Hahlweg 2001; Andersson 2012; Norton 2014), as such, it would be inappropriate to impose bias on a study based on a characteristic we are unsure would actually introduce bias. In addition, we did not assess risk of bias related to therapist allegiance. This was because: (a) all studies investigated CBT, and (b) it would have been impossible to know if researchers were allied with a particular type of delivery method.

Two review authors (JVO and KMB) independently assessed risk of bias for each included study. We resolved disagreements by consensus and discussion with a third review author (MCW or SHS) where necessary. If further information about a particular trial was required to assess its risk of bias, we contacted the primary investigator of the relevant study. We created 'risk of bias' tables describing the information outlined above, as reported in each study. These tables also include a judgement on the risk of bias, made by the review authors for each of the six areas, based on the following three categories: (1) low risk of bias, (2) high risk of bias, and (3) unclear or unknown risk of bias.

Measures of treatment effect

Dichotomous outcomes

We analysed our only dichotomous outcome, clinically important improvement in anxiety (yes or no) (as measured by no longer meeting diagnostic criteria on a diagnostic interview, no longer meeting a designated cut‐off on a validated scale, or meeting the criteria for very much or much improved on the CGI) using risk ratios (RRs) and 95% confidence intervals (CIs) within studies.

Continuous outcomes

As most studies that were selected for inclusion used different measures to assess sufficiently similar constructs, we compared continuous outcomes (that is, general and disorder‐specific anxiety symptoms, quality of life) by calculating the standardized mean difference (SMD) and its 95% CI. However, when all of the studies within a meta‐analysis used the same measure to assess an outcome (for example, if all studies within a meta‐analysis used the BAI to assess general anxiety symptoms), we compared continuous outcomes by calculating the mean difference (MD) to facilitate the interpretation of the clinical relevance of the findings.

Most included studies used more than one measure to assess each of the continuous outcomes. Thus, a mean score was created across the measures included within each study. Measures of variance for this mean score were created by combining standard deviations across studies according to the method described by Borenstein 2009. This method requires that the correlation between two measures be known; as such, on the rare occasion when this correlation was not known and could not be identified in prior literature the measure in question was excluded from analyses. This occurred in five instances (Klein 2006, Richards 2006, and Kiropoulos 2008 for the Body Vigilance Scale; Andersson 2009 and Andersson 2013 for the Fear Survey Schedule III). A sixth study simply included too many measures to be meaningfully combined (Berger 2014) and so the Brief Symptom Index was used as a proxy to index disorder‐specific symptoms.

To combine measures of quality of life and disability into one outcome, we reversed the scores of the disability measures (that is, by subtracting mean scores from the measure total scores) to align them with the quality of life measures.

Endpoint versus change data

We anticipated that we might encounter some studies that reported analyses based on changes from baseline and other studies that reported analyses based on final values. We planned to present the two types of analysis results in separate subgroups to avoid confusion for readers and, where appropriate, to combine both types of scores in the final results. Despite these plans, none of the included studies reported change data so we used endpoint data in all meta‐analyses.

Skewed data

We dealt with skewed data according to the guidelines in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a) and Higgins 2008. In order to conduct the final analysis, transformed or untransformed data had to be obtained for all studies because log‐transformed and untransformed data cannot be combined in meta‐analyses (Higgins 2011a). In the case that a limited number of studies included in one meta‐analysis presented log‐transformed data, we back‐transformed these data and included untransformed data in the meta‐analysis. We then conducted a sensitivity analysis excluding any studies that presented transformed data.

Unit of analysis issues

Parallel group randomised controlled trials (RCTs)

In some parallel group RCTs, participants randomly assigned to a waiting list, attention, information, or online discussion group only control were permitted to pursue the active treatment after their period on the waiting list was complete. To analyse dichotomous and continuous data for these trials, we only included data from participants before they crossed over to their second treatment condition; in other words, only data from the original comparison (waiting list, attention, information, or online discussion group only control versus therapist‐supported ICBT) was used in the meta‐analyses.

Cross‐over trials

When included studies were cross‐over trials, we planned to include only data from the first phase of the trial.

Cluster randomised trials

When cluster randomised trials had accounted for clustering within their analyses (through the use of multilevel modelling or general estimating equations, for example) we planned to include data directly in the meta‐analyses. For studies that failed to appropriately account for clustering, we planned to impute the data based on the number of clusters reported in each intervention group, the size of each cluster, summary statistics, and an estimate of intracluster correlation. We also planned to exclude cluster trials with a high risk of bias (that is, where clustering was not accounted for in analyses) from sensitivity analyses.

Multiple intervention arms

When multiple intervention arms met our inclusion criteria, we planned to combine eligible groups to create a pair‐wise comparison following the procedure outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). We planned to conduct sensitivity analyses excluding any studies with multiple intervention arms that did not report all intervention comparisons.

Dealing with missing data

We used data from ITT analyses whenever they were reported by study authors. In 21 studies, authors employed a LOCF method to address missing data with the assumption that participants who were missing data following randomisation (that is, dropouts) did not respond to treatment. Of the remaining studies, two studies used multiple imputation methods to create ITT data (Kok 2012; van Ballegooijen 2013). Seven studies used a mixed effects models approach in an ITT approach to deal with missing data (Bergstrom 2010; Berger 2011; Hedman 2011; Paxling 2011; Andersson 2012a; Andersson 2012b; Silfvernagel 2012). Two studies did not include ITT data (Andersson 2009; Andersson 2013). One study did not report whether they used ITT data (Greist 2012).

Because included studies did not report individual participant data, if authors did not provide ITT analyses in their manuscript we contacted the primary investigator by e‐mail to attempt to obtain unreported data to permit an ITT analysis. When we did not receive responses from study authors we simply included their reported, non‐ITT, continuous outcome data in the analysis. This was the case for two studies (Andersson 2009; Andersson 2013). For dichotomous outcomes, we were able to impute ITT data by assuming that participants who had dropped out did not meet the target event (that is, clinically important improvement in anxiety). We conducted sensitivity analyses excluding studies for which ITT data were not available (either from the published manuscript or from study authors) to determine the extent to which missing data influenced effect sizes.

If included trials did not provide complete information (that is, group means, standard deviations, and sample size), we contacted the primary investigator by e‐mail to attempt to obtain unreported data. We contacted other study investigators as needed. The only sources for outcome data were the original published report or author correspondence. If standard deviations were not available from the authors, we planned to calculate these using other data reported in the article, including t‐values, CIs, and standard errors. If that was not possible, we planned to impute standard deviations from other investigations using similar measures and populations.

Assessment of heterogeneity

We tested the extent of statistical heterogeneity in meta‐analyses using the I² statistic (Higgins 2002), which calculates the percentage of variability due to heterogeneity rather than chance. According to the guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions, I² values may be interpreted as follows:

0% to 40% might not be important;
30% to 60% may represent moderate heterogeneity;
50% to 90% may represent substantial heterogeneity; and
75% to 100% represents considerable heterogeneity (Higgins 2011a).

We interpreted the importance of these I² values in consideration of the magnitude and direction of effects and the strength of evidence for heterogeneity (as indexed by the P value from the Chi² test). If there was evidence of heterogeneity, we first re‐checked the data for accuracy. We considered sources of heterogeneity according to the pre‐specified subgroup and sensitivity analyses listed in Subgroup analysis and investigation of heterogeneity.

Assessment of reporting biases

Where there were sufficient numbers of trials to make such a plot meaningful (that is, at least 10 included studies (Higgins 2011a)) we constructed funnel plots to determine the possible influence of publication bias. We planned to enhance funnel plots with contour lines delineating areas of statistical significance (as suggested by Peters 2008) to assist in the differentiation of asymmetry due to publication bias or other causes.

Data synthesis

We combined data using an inverse‐variance random‐effects model due to expected variation in the characteristics of the interventions investigated and participant populations. We combined dichotomous outcome measures by computing a pooled risk ratio (RR) and 95% CI. We combined continuous outcomes when means and standard deviations were available. When sufficiently similar continuous outcomes were measured differently across studies we calculated an overall standardized mean difference (SMD) and 95% CI. However, as indicated previously, when outcomes were measured similarly across studies we used a mean difference method. We used the RevMan 5.3 software for data synthesis.

Subgroup analysis and investigation of heterogeneity

We conducted subgroup analyses but interpreted these with caution due to the risk of false positive conclusions. We planned to perform the following subgroup analyses:

gender of participants;
type of anxiety disorder (i.e., PD with or without agoraphobia, agoraphobia without a history of panic, social phobia (social anxiety disorder), PTSD, acute stress disorder, OCD, specific phobia, GAD, and anxiety disorder not otherwise specified);
amount of therapist contact, designated as low (90 min or less), medium (91 to 299 min), or high (300 min or more);
type of CBT (i.e., BT, CT, or CBT); and
research group (i.e., the laboratory from which the study was generated).

We were not able to conduct a subgroup analysis based on gender of participants as none of the included studies distinguished outcomes based on this participant variable. We also were not able to conduct a subgroup analysis based on type of CBT. Only three studies (Andersson 2009; Kok 2012; Andersson 2013) had a stronger focus on BT, as compared to CT or CBT; and no studies examined a CT‐only intervention. For the final subgroup analysis by research group, seven research groups were identified: a group each in Sweden, Switzerland, and the USA; and two distinct groups in Australia and the Netherlands.

Sensitivity analysis

We conducted sensitivity analyses to determine the extent to which observed pooled effect sizes depend on the quality of the design characteristics of studies. We planned to conduct the following sensitivity analyses:

exclusion of studies with a designation of high risk of bias for one or more of the categories as outlined in Assessment of risk of bias in included studies;
exclusion of cluster randomised trials where clustering was not appropriately accounted for in analysis;
exclusion of studies with multiple intervention arms with selective reporting of intervention comparisons;
exclusion of studies with a somewhat more active waiting list control condition (i.e., attention, information, or online discussion group only control)
exclusion of studies with imputed standard deviations for continuous outcomes;
exclusion of studies with back transformed data for continuous outcomes;
exclusion of studies not reporting: (a) dichotomous, and (b) continuous outcomes according to the ITT principle;
exclusion of studies with continuous outcomes analysed using LOCF; and
assuming treatment dropouts were responders for dichotomous outcomes.

Summary of findings

Summary of findings tables were created to present the main findings of the review. We imported meta‐analytic data from RevMan into GRADEprofiler version 3.6 to create summary of findings tables for each of the three most clinically relevant comparisons: ICBT with therapist support versus waiting list control, ICBT with therapist support versus unguided ICBT, and ICBT with therapist support versus face‐to‐face CBT. The summary of findings tables present meta‐analytic outcomes for each of the continuous and dichotomous outcomes at post‐treatment and summarize the number of studies and participants included in each analysis. In addition, GRADEprofiler allowed us to rate the quality of the evidence for each outcome for each comparison considering: (a) risk of bias, (b) inconsistency, (c) indirectness, (d) imprecision, and (e) publication bias.

Results

Description of studies

Results of the search

We screened a total of 1736 citations and selected 38 studies with 3214 participants for inclusion.

In detail, searches of the CCDANCTR (all years to 16 March 2015) retrieved a total of 1265 records, including manuscripts in peer‐reviewed journals, conference abstracts, and clinical trial registrations. Secondary search methodologies ‐ including searching the reference lists of eligible studies, contacting experts in the field and conducting additional searches of trial registries ‐ identified a further 471 records. After de‐duplication and following a brief screening of the titles and abstracts, 244 full‐text articles were retrieved for a more detailed evaluation of eligibility.

The PRISMA flow diagram shown in Figure 1 outlines the study selection process and broad reasons for exclusion. Studies were excluded if: (a) participants did not meet diagnostic criteria for an anxiety disorder, as assessed by study authors (population), (b) the intervention of interest was not ICBT, did not involve a therapist, or included too much face‐to‐face therapist contact (intervention), (c) the comparator was not appropriate given our selection criteria (comparator), (d) the trial was not randomised or did not use adequate diagnostic measures (methods), or (e) the trial was ongoing (ongoing). After consolidating references into studies, 38 were eligible for inclusion in the meta‐analyses.

Figure 1

PRISMA diagram of the search process.

E‐mail correspondence for supplemental data was exchanged with Dr Tomas Furmark (Furmark 2009a; Furmark 2009b), Dr Per Carlbring (Carlbring 2001; Carlbring 2006; Carlbring 2007; Carlbring 2011), Dr Nickolai Titov (Titov 2008a; Titov 2008b; Titov 2008c; Titov 2009; Titov 2010; Titov 2011), Dr Britt Klein (Klein 2006; Richards 2006; Kiropoulos 2008), Dr Wouter van Ballegooijen (van Ballegooijen 2013), Dr. Thomas Berger (Berger 2014), and Dr. Jill Newby (Newby 2013).

Included studies

See Characteristics of included studies for details of individual studies and Table 1 for a summary table of the characteristics of the included studies.

See: Summary of findings for the main comparison Therapist‐supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults; Summary of findings 2 Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults; Summary of findings 3 Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults

Table 1. Summary of included studies table

Study	Diagnosis and Co‐morbidity	Participant Characteristics *(M age, age range, sex, country of residence)*	Co‐Use of Medication	N	Intervention Type & Therapist Duration Contact		Comparison	Assessment Points	Outcomes
Andersson et al (2009)	Specific Phobia, Spider Type co‐morbidity not reported	M age=25.6 (4.1) 18‐65 years 84.8% women Sweden	Not reported	27	IBT with email: 4 wks; 5 online modules	M total time spent per participant = 25 min	Orientation and 1 3‐hour live exposure session	post‐treatment 12‐month follow‐up	specific phobia sx; general anxiety sx
Andersson et al (2012a)	Social Phobia co‐morbidity included but not reported	ICBT M age=38.1 (11.3) WLC M age=38.4 (10.9) 19‐71 years 61% women Sweden	13.7% using medication	204	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	Online Discussion Group	post‐treatment	diagnostic status; social phobia sx; QOL; general anxiety sx;
Andersson et al (2012b)	GAD 22.2% Social Phobia, 19.8% PD, 3.7% OCD, 23.5% MDD	ICBT M age=44.4 (12.8) IPDTM age=36.4 (9.7) WLC M age=39.6 (13.7) 19‐66 years 76.5% women Sweden	32.1% using medication	81	ICBT with email: 8 wks; 8 online modules	M total time spent per participant = 92 min (SD=61)	(1) Waiting List Control (2) IPDT: 8 wks; 8 online modules	post‐treatment	diagnostic status, GAD sx; general anxiety sx; QOL
Andersson et al. (2013)	Specific Phobia, Snake Type comorbidity not reported	M age=27.2 (8.1) 19‐54 years 84.6% women Sweden	Not reported	30	IBT with email: 4 wks; 4 online modules	M total time spent per participant = 25 min	Orientation and 1 3‐hour live exposure session	post‐treatment 12‐month follow‐up	specific phobia sx; general anxiety sx
Berger et al (2009)	Social Phobia 26.9% co‐morbid Axis I disorder	M age=28.9 (5.3) 19‐43 years 44.2% women Switzerland, France, Belgium	Excluded	52	ICBT with email: 10 wks; 5 online modules	M=5.5 emails from participant weekly emails from therapist	Waiting List Control	post‐treatment	social phobia sx; treatment satisfaction
Berger et al (2011)	Social Phobia 38% co‐morbid Axis I disorder; 12% PD, 10% Specific Phobia, 2% GAD, 22% MDD/Dysthymia, 2% ED	M age=37.2 (11.2) 19‐62 years 53.1% women Switzerland	7.4% using medication	81	ICBT with email: 10 wks; 5 online modules	M=6.16 (SD=4.56; range=1‐17) emails from participant M=12.44 (SD=2.85; range=6‐17) emails from therapist	(1) Unguided ICBT 10 weeks; 5 online modules (2) Step‐up on demand ICBT	post‐treatment 6‐month follow‐up	diagnostic status; social phobia sx; treatment satisfaction
Berger et al. (2014)	33.3% PD with or without Agoraphobia 85.6% Social Phobia 25% GAD 37.1% PD with or without Agoraphobia, Social Phobia, or GAD, 13.6% MDD, 15.9% Specific Phobia, 5.3% OCD, 12.1% other Axis I disorder	M age=35.1 (11.4) 18‐65 years 56.1% women Switzerland, Germany, Austria	14.4% using medication	132	ICBT with email: 8 wks; 8 online modules	M=6.53 (SD=7.2; range=0‐36) emails from participant M=12.6 (SD=4.6; range=8‐35) emails from therapist	(1) Waiting List Control (2) Tailored ICBT: 8 wks; 8 online modules	post‐treatment 6‐month follow‐up	diagnostic status; anxiety sx; general anxiety sx; treatment satisfaction
Bergstrom et al (2010)	15.4% PD 84.6% PD with Agoraphobia co‐morbidity not reported	ICBT M age=33.8 (9.7) GCBT M age=34.6 (9.2) 18 years or older 61.5% women Sweden	45% using medication; 34% SSRI/SNRIs, 13% BZ, 24% BZ derivatives or neuroleptics; 5% TCAs	104	ICBT with email: 10 wks; 10 online modules	M=11.3 (SD=4.3) emails from therapist M total time spent per participant = 35.4 min (SD=19)	10 weekly 2‐hour sessions of GCBT	post‐treatment 6 month follow‐up	diagnostic status; PD sx; QOL
Carlbring et al (2001)	PD co‐morbidity included but not reported	M age=34 (7.5) 21‐51 years 71% women Sweden	64% using medication; 44% SSRIs, 10% BZ, 5% TCAs, 5% beta‐blockers	41	ICBT with email: 7‐12 wks; 6 online modules	M reciprocal emails = 7.5 (SD=1.2; range=6‐15) M total time spent per participant = 90 min	Waiting List Control	post‐treatment	diagnostic status; PD sx; QOL; general anxiety sx; treatment satisfaction
Carlbring et al (2005)	49% PD 51% PD with Agoraphobia 49% Anxiety Disorder; 6% MDD	M age=35.0 (7.7) 18‐60 years old 71% women Sweden	30.6% SSRIs, 8.2% BZ, 6.1% TCAs, 6.1% beta blockers	49	ICBT with email: 10 wks; 10 online modules	M reciprocal emails =15.4 (SD=5.5; range=4‐31) M total time spent per participant =150 min	10 weekly 45‐60 min sessions of individual CBT	post‐treatment 12‐month follow‐up	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL
Carlbring et al (2006)	PD co‐morbidity included but not reported	M age=36.7 (10) 18‐60 years 60% women Sweden	54% using medication	60	ICBT with email & phone: 10 wks; 10 online modules	M reciprocal contacts = 13.5 (SD =4.4; range=7‐29) M time spent per participant per week = 12 min M length phone call = 11.8 min (range= 9.6‐15.6)	Waiting List Control	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL; treatment satisfaction
Carlbring et al (2007)	Social Phobia co‐morbidity included but not reported	ICBT M age=32.4 (9.1) WLC M age=32.9 (9.2) 18‐60 years 64.9% women Sweden	Included but not reported	60	ICBT with email & phone: 9 wks; 9 online modules	M time spent per participant per week = 22 min M length phone call = 10.5 min (SD= 3.6)	Waiting List Control	post‐treatment	social phobia sx; general anxiety sx; QOL
Carlbring et al (2011)	9% PD 22% PD with Agoraphobia 39% Social Phobia 20% GAD 13% ADNOS 2% OCD, 2% PTSD, 20% MDD, 7% mild depression; 15% Dysthymia	M age=38.8 (10.7) 22‐63 years 76% women Sweden	26% using an antidepressant or anxiolytic	54	ICBT with email: 10 wks; 6‐10 online modules	M time spent per participant per week = 15 min	Attention Control 10 wks of posts in an online support forum	post‐treatment	diagnostic status; anxiety sx (broadly); QOL; general anxiety sx
Furmark et al (2009a)	Social Phobia co‐morbidity not reported	ICBT M age=35 (10.2) WLC M age=35.7 (10.9) Bib M age=37.7 (10.3) 18 years or older 67.5% women Sweden	13.9% using medication	120	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	(1) Biblio‐therapy: 9 wks; 9 lessons (2) Waiting List Control	post‐treatment	social phobia sx; general anxiety sx; QOL
Furmark et al (2009b)	Social Phobia co‐morbidity not reported	ICBT M age=34.9 (8.4) Bib M age=32.5 (8.5) Applied Relaxation M age=36.4 (9.8) 18 years or older 67.8% women Sweden	6.7% using medication	115	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	(1) Biblio‐therapy: 9 wks; 9 lessons (2) Biblio‐therapy and discussion group: 9 wks; 9 lessons (2) Internet‐based applied relaxation: 9 wks; 9 online modules	post‐treatment	social phobia sx; general anxiety sx; QOL
Greist et al. (2012)	OCD 32% Anxiety Disorder, 31% Mood Disorder, 7% Substance Disorder, 7% ADHD, 2% ED	M age=38.34 (13.93) 18 years or older 63% women USA	19.5% using medication	87	ICBT with phone: 12 wks; 9 online modules	Weekly phone calls	(1) Unguided ICBT: 12 wks; 9 online modules (2) Lay Coaching ICBT: 12 wks; 9 online modules	post‐treatment	OCD sx; QOL; treatment satisfaction
Hedman et al (2011)	Social Phobia 47.5% Anxiety Disorder, 15.1% MDD	ICBT M age=35.2 (11.1) GCBT M age=35.5 (11.6) 18‐64 years 35.7% women Sweden	19.8% SSRIs, 4.8% SNRIs	126	ICBT with email: 15 wks; 15 online modules	M=17.4 emails per participant M time spent per participant per week = 5.5 min (SD=3.6)	15 weekly 2.5‐hour sessions of GCBT	post‐treatment 6 month follow‐up	diagnostic status; social phobia sx; QOL; general anxiety sx
Ivarsson et al. (2014)	PTSD comorbidity not reported	M age=46 (11.7) 21‐67 years 82.3% women Sweden	Included but not reported	62	ICBT with email: 8 wks; 8 online modules	M time spent per participant per week = 28 min (SD=19.8)	Attention Control: sent weekly question on wellbeing, stress, sleep	post‐treatment	diagnostic status; PTSD sx; general anxiety sx; QOL
Johnston et al (2011)	20.6% PD with or without Agoraphobia 34.4% Social Phobia 45% GAD 29% Anxiety Disorder, 9.2% Affective Disorder, 32.1% both disorders	M age=41.62 (12.83) 19‐79 years 58.8% women Australia	29% using medication	139	ICBT with email & phone: 10 wks; 8 online modules	M=8.83 (SD=3.29) emails per participant M=7.54 (SD=2.43) phone calls per participant M total time spent per participant = 69.09 min (SD=32.29)	Waiting List Control	post‐treatment	disorder‐specific sx; general anxiety sx; QOL; treatment satisfaction
Kiropoulos et al (2008)	41.9% PD 58.1% PD with Agoraphobia 72.1% co‐morbid Mood, Anxiety, Somatoform, or Substance Disorder	M age=38.96 (11.13) 20‐64 years 72.1% women Australia	47.7% using medication	86	ICBT with email: 6 wks, 6 required & 2 optional online modules	M=18.24 (SD=9.82) emails from therapist M=10.64 (SD=8.21) emails from participant M total time spent per participant = 352 min (SD=240)	12 weekly 1‐hour sessions of individual CBT	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL; treatment satisfaction
Kok et al. (2012)	41% PD with Agoraphobia 17% Agoraphobia 53.3% Social Phobia 83.5% Specific Phobia Comorbidity included but not reported	M age=34.6 (11.7) 18 years or older 61% women Netherlands	43% using medication	212	IBT with email: 5wks, 8 online modules	weekly contact by therapist	Waiting List Control: also sent self‐help book without instructions	post‐treatment	phobia sx; general anxiety sx; treatment satisfaction
Newby et al. (2013)	84% GAD Comorbidity included but not fully reported; 56% MDD	M age=44.3 (12.2) 21‐80 years 77.8% women Australia	40.4% using medication	100	ICBT with email and phone: 10 wks; 6 online modules	M total time spent per participant = 23.37 min (SD=12.15)	Waiting List Control	post‐treatment	GAD sx; QOL; treatment satisfaction
Nordgren et al. (2012)	31% PD with or without Agoraphobia 8% Agoraphobia 32% Social Phobia 10% GAD 19% ADNOS 21% Anxiety Disorder, 43% Mood Disorder, 1% Hypochon‐driasis	ICBT M age=35 (13) WLC M age=36 (12) 19‐68 years 63% women Sweden	26% using medication	100	ICBT with email: 10 wks; 7‐10 online modules	M time spent per participant = 15 min/week	Attention Control: sent weekly questions on wellbeing	post‐treatment	anxiety sx; general anxiety sx; QOL
Paxling et al (2012)	GAD co‐morbidity included but not fully reported; 22.5% MDD	M age=39.3 (10.8) 18‐66 years 79.8% women Sweden	37.1% using medication	89	ICBT with email: 8 wks; 8 online modules	M total time spent per participant = 97 min (SD=52)	Waiting List Control	post‐treatment	GAD sx; general anxiety sx; QOL
Richards et al (2006)	21.9% PD 78.1% PD with Agoraphobia 22% Social Phobia, 13% GAD, 9% Specific Phobia, 6% PTSD, 9% MDD, 6% Hypochondriasis, 3% Somatization	M age=36.59 (9.9) 18‐70 years 68.8% women Australia	15.6% anti‐depressants, 12.5% BZ, 9.4% antidepressants and BZ	23	ICBT with email: 8 wks, 6 online modules	M=18 (SD=6.5) emails from therapist M=15.3 (SD=12.8) emails from participant M total time spent per participant =376.3 min (SD=156.8)	Information Only Control Weekly status updates to clinician and access to online non‐CBT info	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL
Robinson et al (2010)	GAD co‐morbidity included but not reported	M age=46.96 (12.7) 18‐80 years 68.3% women Australia	Included but not reported	101	ICBT with email and phone: 10 wks; 6 online modules	M =33.2 (SD=4) emails/calls per participant M total time spent per participant = 80.8 min (SD=22.6)	Waiting List Control	post‐treatment	diagnostic status; GAD sx; QOL; treatment satisfaction
Silfvernagel et al (2012)	7% PD 83% PD with Agoraphobia 16% Social Phobia 19% GAD 2% ADNOS 32% co‐morbid disorder	M age=32.4 (6.9) 20‐45 years 65% women Sweden	47% using medication	57	ICBT with email: 8 wks; 6‐8 online modules	M time spent per participant = 15 min/week	Waiting List Control	post‐treatment	diagnostic status; PD sx; general anxiety sx; QOL
Spence et al (2011)	PTSD 62% MDD, 33% Social Phobia, 31% PD with or without Agoraphobia, 26% GAD; 17% OCD	M age=42.6 (13.1) 21‐68 years 81% women Australia	60% using medication	44	ICBT with email & phone: 8 wks; 7 online modules	M=5.39 (SD=3.54) emails per participant M=7.87 (SD=2.56) phone calls per participant M total time spent per participant = 103.91 min (SD=96.53)	Waiting List Control	post‐treatment	diagnostic remission; PTSD sx; QOL; general anxiety sx; treatment satisfaction
Tillfors et al (2008)	Social Phobia co‐morbidity included but not reported	ICBT M age=32.3 (9.7) ICBT+exposure M age= 30.4 (6.3) 19‐53 years 78.9% women Sweden	Included but not reported	38	ICBT with email: 9 wks; 9 online modules	M=35 min per participant per week	ICBT with email (9 online modules) + 5 live 2.25‐hour exposure sessions; 9 wks	post‐treatment 12‐month follow‐up	social phobia sx; general anxiety sx; QOL; treatment satisfaction
Titov et al (2008a)	Social Phobia co‐morbidity included but not reported	M age=38.13 (12.24) 18‐72 years 59% women Australia	29% using medication	105	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 125 min (SD=25)	Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2008b)	Social Phobia co‐morbidity included but not reported	M age=36.79 (10.93) 20‐61 years 62.96% women Australia	25.9% using medication	88	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 126.76 min (SD=30.89)	Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2008c)	Social Phobia co‐morbidity included but not reported	M age=37.97 (11.29) 18‐64 years 61.05% women Australia	25.9% using medication	98	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 168 min (SD=40)	(1) Unguided ICBT 10 wks; 6 online modules (2) Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2009)	GAD co‐morbidity included but not reported	M age=44 (12.98) 18 years or older 76% women Australia	29% using medication	48	ICBT with email & phone: 9 wks, 6 online modules	M =23.7 emails, 5.5 instant messages, and 4.1 calls per participant M total time spent per participant = 130 min	Waiting List Control	post‐treatment	diagnostic status; GAD sx; QOL; treatment satisfaction
Titov et al (2010)	26.9% PD with Agoraphobia 29.5% Social Phobia 43.6% GAD 28.2% Anxiety Disorder, 20.5% Affective Disorder, 26.9% both disorders	M age=39.5 (13) 18 years or older 67.9% women Australia	47.4% using medication	86	ICBT with email & phone: 8 wks; 6 online modules	M =23.6emails from therapist M total time spent per participant = 46 min (SD=16)	Waiting List Control	post‐treatment	diagnostic status; disorder‐specific anxiety sx; QOL; treatment satisfaction
Titov et al (2011)	10% PD with or without Agoraphobia 11% Social Phobia 28% GAD 51% MDD 81% had a co‐morbidity	M age=43.9 (14.6) 18‐79 years 73% women Australia	54% using medication	74	ICBT with email & phone: 10 wks; 8 online modules	M=5.45 (SD=3.57) emails per participant M=9.35 (SD=2.96) phone calls per participant M total time spent per participant = 84.76 min (SD=50.37)	Waiting List Control	post‐treatment	disorder‐specific sx; general anxiety sx; QOL; treatment satisfaction
Van Ballegooijen et al (2013)	78% PD with or without Agoraphobia 14% Agoraphobia co‐morbidity included but not reported	M age=36.6 (11.4) 18‐67 years 67.5% women Netherlands	Included but not reported	126	ICBT with email: 12 wks; 6 online modules	M total time spent per participant = 1 to 2 hours	Waiting List Control	post‐treatment	PD sx; general anxiety sx
Wims et al (2010)	PD with or without Agoraphobia 21% Social Phobia, 31% GAD, 10% OCD, 7% PTSD, 21% MDD	M age=42.08 (12.29) 18 years or older 76% women Australia	31% using medication	59	ICBT with email: 8 wks; 6 online modules	M =7.5 emails from therapist M total time spent per participant = 75 min	Waiting List Control	post‐treatment	diagnostic status; PD & agoraphobia sx; QOL
Wootton et al. (2013)	OCD 26.9% Social Phobia, 40.4% GAD, 15.4% PD, 11.5% PTSD, 38.5% MDD	ICBT M age=39.93 (12.57) Bib M age=35.55 (9.68) WLC M age=38.58 (10.51) 18‐64 years 75% women Australia	61.5% using SSRIs	56	ICBT with phone: 8 wks, 5 online modules	M=15.05 (SD=3.93) phone calls per participant M total time spent per participant = 88.63 min (SD=46.41)	1) Guided bibliotherapy: 8 wks, 5 lessons 2) Waiting List Control	post‐treatment	diagnostic status; OCD sx; general anxiety sx; treatment satisfaction

Notes: All data in the above table represent only that included in/relevant to the present review.

ADNOS = anxiety disorder, not otherwise specified; Bib = Bibliotherapy; BZ = benzodiazepine; ED = eating disorder; GAD = generalized anxiety disorder; GCBT = group cognitive behavioural therapy; IBT = internet‐based behavioural therapy; ICBT = internet‐based cognitive behavioural therapy; IPDT = internet‐based psychodynamic therapy; MDD = major depressive disorder; PD = panic disorder; QOL = quality of life; SNRI = serotonin‐norepinephrine re‐uptake inhibitor; SSRI = selective serotonin re‐uptake inhibitor; sx = symptoms; TCA = tricyclic antidepressant; VCBT = videoconferencing cognitive‐behavioural therapy; WLC = waiting list control.

Design

All of the 38 included studies were parallel group RCTs. For studies in which participants in the waiting list, attention, information, or online discussion group only control were given the opportunity to complete the treatment after their time on the waiting list, only data from the original comparison were used in the meta‐analyses. There were no cross‐over or cluster randomised trials.

Ten studies included multiple intervention arms: two (Titov 2008c; Furmark 2009a) compared the intervention of interest to two eligible comparators (a waiting list, and unguided CBT) so were included in multiple meta‐analyses (ICBT versus waiting list control, and ICBT versus unguided CBT), and eight (Richards 2006; Furmark 2009b; Robinson 2010; Berger 2011; Johnston 2011; Berger 2014; Greist 2012; Kok 2012) included a third treatment arm not relevant to the present review.

Sample sizes

Sample sizes of included studies ranged from 21 (12 in the intervention arm, 9 in the comparator arm (Richards 2006)) to 212 participants (105 in the intervention arm, 107 in the comparator arm (Kok 2012)). The average study sample size was 85 participants. In most studies there was an equal distribution of participants between the treatment and control arms. Only 2 studies had < 30 participants, 16 studies had 30 to 60 participants, 9 studies had 60 to 90 participants, and 9 studies had 90 to 140 participants, with 2 outliers at 204 participants (Andersson 2012a) and 212 participants (Kok 2012).

Setting

Included studies came from one research group in Sweden (18 trials), two groups in Australia (Klein et al.: 2 trials; Titov et al.: 12 trials), two groups in the Netherlands (Kok et al.: 1 trial; van Ballegooijen et al.: 1 trial), a research group in Switzerland (3 trials), and one in the USA (1 trial).

Whereas researchers and treating clinicians were located at university‐affiliated hospitals or mental health centres, participants received the intervention of interest in their home. Treatment took place over the Internet and by telephone. Face‐to‐face CBT, when included in a trial, was conducted in a psychiatric setting (for example, hospital, mental health clinic).

Participants

Participants were men and women over 18 years of age. The average mean age of study participants was 37.3 years. Women represented an average of 67.7% of participants in each study. The ethnicity of participants was not reliably reported. For most studies, participants were recruited via media advertisements or a recruitment website (33 studies); in a minority of studies participants were recruited via clinic referrals (Bergstrom 2010; Hedman 2011; Kok 2012; Nordgren 2012). One study recruited using both methods (Greist 2012).

All included participants qualified for one of the following anxiety disorder diagnoses: social phobia (11 trials), PD with or without agoraphobia (8 trials), GAD (5 trials), PTSD (2 trials), OCD (2 trials), and specific phobia (2 trials). The eight remaining studies included participants with a range of anxiety disorder diagnoses. Thirty‐two trials included participants with co‐morbid diagnoses and six studies did not report on their inclusion and exclusion criteria. Among all studies, regardless of their inclusion or exclusion of co‐morbidities, 29 studies excluded participants who scored above a certain threshold on a measure of depressive symptoms, for example, above 30 on the Montgomery‐Asberg Depression Rating Scale (MADRS) (Svanborg 1994), and 33 studies excluded participants who endorsed suicidal ideation, for example, on the MADRS suicide item, with the rationale that they were unclear about how to handle this high risk participant via a distance treatment. Nineteen studies excluded participants with substance misuse or dependence problems and 19 studies excluded participants with active psychosis with the rationale that these problems could interfere with anxiety treatment.

Thirty‐five trials included participants who were using psychiatric medication (including selective serotonin reuptake inhibitors, serotonin norepinephrine reuptake inhibitors, benzodiazepines, benzodiazepine derivatives, neuroleptics, tricyclic antidepressants, beta‐blockers) concurrent with study participation (Berger 2009 excluded those using medication, and Andersson 2009 and Andersson 2013 did not report on co‐use of medication). Participants were typically included only if they had been at a stable dose for a certain time period (one to three months) preceding the study. Six studies (Carlbring 2001; Carlbring 2005; Carlbring 2006; Greist 2012; Newby 2013; van Ballegooijen 2013) included participants engaged in another type of psychological therapy concurrent with study participation, one of which had no participants that met this characteristic (Carlbring 2006).

Interventions

Experimental interventions

Thirty‐five studies included in the present review tested ICBT, while three studies investigated Internet‐based BT with a focus on exposure (Andersson 2009; Kok 2012; Andersson 2013). Interventions involved participants following 4 (Andersson 2013) to 15 (Hedman 2011) online treatment modules (mean = 7; median = 7; mode = 6). Twenty‐nine studies provided e‐mail support from therapists only, two studies provided telephone support from therapists only (Greist 2012; Wootton 2013), and seven studies provided therapist support by email and telephone (Titov 2009; Robinson 2010; Titov 2010; Johnston 2011; Spence 2011; Titov 2011; Newby 2013). Seven studies also included participation in an online discussion forum (Tillfors 2008; Furmark 2009a; Furmark 2009b; Bergstrom 2010; Spence 2011, Titov 2011; Andersson 2012a).

Interventions ranged in length from 4 (Andersson 2009; Andersson 2013) to 15 weeks (Hedman 2011) (mean = 9; median = 9; mode = 10). The degree of therapist involvement in the included interventions was widely variable; the average total time spent by a therapist with a participant ranged from a minimum of 23 minutes (Newby 2013) to a maximum of 376 minutes (Richards 2006), with the overall mean = 128 minutes and median = 120 minutes. Five studies (Berger 2009; Berger 2011; Greist 2012; Kok 2012; Berger 2014) did not report therapist contact time. Similarly, among studies that reported this information, the average number of contacts made by study therapists (email and/or telephone) ranged from a minimum of 5 (Kok 2012) to a maximum of 33 (Robinson 2010), with the overall mean = 15 contacts and median = 14 contacts.

Of the 38 included studies, 13 specified that treatment was provided by a clinical psychologist, 9 specified that treatment was provided by clinical psychology graduate students in training, and 12 specified treatment was provided by both. The four remaining studies described treatment providers as therapists trained by the treatment founder who also provided supervision (Andersson 2009), a clinical psychologist and a clinical social worker (Greist 2012), a practice manager supervised by a clinical psychologist (Newby 2013), and a psychiatry registrar (Wims 2010). Clinical psychology graduate students providing therapy were enrolled in masters or doctoral psychology programs as required for them to practice in their country. Of those studies in which licensed clinical psychologists delivered the treatment, eight specified that clinicians were provided with supervision from an expert in the field. Similarly, of those studies in which clinical psychology students provided treatment, 16 specified that supervision from an expert in the field was provided. Details on the experience and training of study therapists, if provided, can be found in the Characteristics of included studies section.

Comparator Interventions

Twenty‐eight studies compared therapist‐supported ICBT to a waiting list, attention, information, or online discussion group only control. All but eight of these studies included strict waiting list control conditions with no treatment provided to participants and assessments occurring after the designated waiting list period. Of the remaining eight studies, the control condition in one study included a weekly self‐report assessment but no intervention (Furmark 2009a) and in two studies participants completed weekly questions on stress, wellbeing, and/or sleep (Nordgren 2012; Ivarsson 2014). The control condition in two studies (Richards 2006; van Ballegooijen 2013) provided basic non‐treatment disease‐related information to participants and one of these studies (Richards 2006) included weekly status check‐ins by phone. In one study (Kok 2012), participants in the control condition were sent a self‐help book but not provided with any information on how to use it. Finally, the control condition in two studies permitted participants to engage in an online discussion group (Carlbring 2011; Andersson 2012a).

Five studies compared therapist‐supported ICBT to unguided CBT (that is, self‐help). Finally, seven studies compared therapist‐supported ICBT to traditional, face‐to‐face group or individual CBT. This number of studies adds up to more than the total number of included studies because two studies included more than one comparator.

Outcomes

Primary outcomes

Each of the included studies reported on the efficacy of therapist‐supported ICBT. Seventeen studies assessed participants post‐treatment for clinically important improvement in anxiety (a dichotomous outcome) and three studies reassessed this outcome at a follow‐up of 6 to 12 months later. Each of the included studies reported on participants' disorder‐specific anxiety symptom severity using a validated self‐report or observer‐rated instrument (a continuous outcome) at post‐treatment. Eight studies assessed anxiety symptom severity at a follow‐up of 6 to 12 months later. Twenty‐six of the included studies also measured participants' symptoms of general anxiety using validated self‐report instruments at post‐treatment. Seven studies assessed general anxiety at a follow‐up of 6 to 12 months later. Please see Table 1 and Characteristics of included studies for more details of outcome assessment.

It was rare for studies to report adverse events. In fact, adverse events could only be assumed from measures of participants' symptom deterioration during the study or reasons for participant dropout related to the treatment.

Secondary outcomes

Twenty‐nine studies measured quality of life at post‐treatment, while six studies included quality of life as an outcome at 6 to 12 month follow‐up.

Participant satisfaction with treatment was indexed by 19 studies at post‐treatment. A variety of different measures of treatment satisfaction were used ranging in degrees of comprehensiveness and complexity. Across different measurement approaches, participants were most commonly asked to indicate their overall satisfaction with the treatment program, their satisfaction with particular portions of the treatment program (for example, therapist correspondence, Internet modules), and their satisfaction with the pace of the treatment program. Of the 19 studies, only four reported treatment satisfaction for both the experimental and comparator interventions; the remaining trials compared the experimental intervention to a waiting list control, which did not lend itself to an evaluation of satisfaction.

Excluded studies

Studies were excluded for a variety of reasons (see Characteristics of excluded studies and Figure 1, the former of which lists a number of studies that were most like the included studies but differed in important ways that prevented inclusion). Studies were frequently excluded because the intervention was: (a) not distance‐based, (b) distance‐based but included more than two sessions of face‐to‐face contact between therapist and participant, (c) not delivered by a therapist (that is, was a self‐help program), or (d) not CBT. Similarly, studies were excluded if participants did not meet our criteria because they had subclinical anxiety symptoms or an anxiety disorder was not their primary diagnosis. We also excluded a number of studies because a closer look showed that they were not RCTs or did not compare the intervention of interest to a comparison group that met the eligibility criteria.

Ongoing studies

There is a total of nine ongoing studies: Bishop 2012; Clark 2012; Rollman 2012; Titov 2012; Tulbure 2012; Lindner 2013; Miclea 2014, together with Carlbring 2012 and Richards 2014, which we previously classified as awaiting classification (communication with the study authors revealed that these latter two studies are yet to be completed).

A study previously listed as ongoing (Kok 2012) has since been completed and entered as an included study in this review update. Results of two other studies (also previously listed as ongoing) revealed that they no longer qualified for inclusion and have been moved to the excluded studies section (von Essen 2008; Andrews 2012b).

Studies awaiting classification

There are now just two studies awaiting classification compared to 14 listed in the previous version of this review (those identified from the precision maximizing update search of the CCDANCTR Registers in September 2014 (Appendix 1).

The changes (documented below) leave Schreuders 2008 awaiting classification, together with a new study (Tabari 2013) identified from the search of 16 March 2015. These studies appear to qualify for entry in this review but we have been unable to obtain additional data from the trialists who have unfortunately failed to responded to our enquiries.

Studies previously listed as awaiting classification which have now been analysed and entered as included studies include: Greist 2012; Nordgren 2012; Andersson 2013; Newby 2013; Berger 2014; Ivarsson 2014.

We previously listed the following studies as awaiting classification, but the results of the latest search update reveal they no longer qualify for inclusion and so we have moved them to the excluded studies section: Berger 2012; Andrews 2011b; Andrews 2012a. These studies were abandoned due to lack of recruitment.

A closer look at the references of Andrews 2011c and Andrews 2011d (previously awaiting classification) revealed that they were secondary reports of Newby 2013 and Titov 2010 respectively.

Risk of bias in included studies

Results of the risk of bias assessments of included studies are summarized succinctly in Figure 2 and Figure 3. Overall, the risk of bias in the included studies was low, with some notable exceptions related to the nature of clinical trials of psychological treatments.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

The majority of included studies (n = 35) used an adequate method of randomisation, primarily an online random number generator, to avoid selection bias. The three remaining studies reported that participants were randomised but did not describe the randomisation procedure.

Most study authors (n = 32) did not adequately report allocation concealment. The remaining six studies reported allocation concealment procedures that would have minimized the risk of selection bias (for example, random assignment was maintained by an independent research team member not involved in other study aspects who gave randomisations to participants just prior to treatment commencement).

Blinding

The blinding of participants and study personnel is difficult when investigating the efficacy of psychological treatments. Unlike pharmacological trials in which medication type can be concealed, it is very difficult to blind participants to the characteristics of the treatment they are receiving as they are active participants. Similarly, it is impossible to blind study therapists to the treatment they are delivering as they take an active role in its execution. As such, each of the included studies was rated as having a high risk of bias because participants and personnel were not blind to treatment assignment. Because this study characteristic was a limitation across studies, we did not conduct sensitivity analyses based on the characteristic. We did however, downgrade study evidence based on this risk of bias in order to remain consistent with the use of GRADE across other disciplines and conditions. With interventions that are difficult or impossible to blind, there are strategies to limit related biases (e.g. participants or therapists who do not favour interventions, intervention providers in both groups believing they are delivering the ‘best’ treatment, and measuring potential impact by asking patients in all groups if they believe they received the ‘effective’ treatment). Downgrading gives an opportunity to highlight potential biases in observed effects and make recommendations for future studies to limit (or at least explore potential impact of) this source of bias.

We indexed blinding of outcome assessment separately for self‐report versus observer or interview‐rated outcome measures. As participants were not blind to their treatment condition in the included studies, self‐report outcomes measured in all of the included studies were not blinded. Nineteen studies measured outcomes using observer‐rated instruments. In 12 of these studies, interviewers who were blind to the treatment condition conducted the outcome assessments ensuring a low risk of bias. Of the remaining seven studies, two were compromised by participants who too frequently revealed their treatment condition to interviewers (Berger 2011; Berger 2014) and five used at least one interviewer who was aware of participants' random assignment (Richards 2006; Wims 2010; Spence 2011; Titov 2011; Wootton 2013).

Incomplete outcome data

Attrition bias was not a significant issue in 33 of the included studies. These 33 studies used an ITT analysis by either carrying forward the last observations, imputing missing values, or using mixed models analyses to control for outcomes lost to attrition. Moreover, rates of attrition were often quite similar between treatment conditions. Two studies did not use an ITT approach and as such may have been biased due to attrition (Andersson 2009; Andersson 2013). An additional two studies did use ITT analyses but had large attrition that may have biased the findings despite the use of ITT analyses (Kok 2012; van Ballegooijen 2013). One study did not report on their study dropout or data analytic approach and so was rated as having an unclear risk of bias (Greist 2012). We investigated the effect of these studies using sensitivity analyses.

Selective reporting

Twenty‐four of the included studies had been registered as clinical trials allowing for a more accurate analysis of selective reporting. Of these 24 studies, 16 reported on all outcomes outlined in the trial registration. For seven of the studies, one outcome outlined in the trial registration was not reported in the final manuscript (Titov 2008a; Titov 2008b; Titov 2008c; Berger 2009; Johnston 2011; Kok 2012; van Ballegooijen 2013) and they were rated as having an unclear risk of bias. One study (Titov 2010) had many outcomes indicated in the trial registration that were not reported in the final manuscript and was rated as having a high risk of bias. Those studies that were not registered reported results for each of the outcomes they measured, as described in their method; however, given the lack of trial registration or protocol publication, these studies were rated as having an unclear risk of bias.

Other potential sources of bias

Three of the included studies had a high risk of bias due to differences in baseline severity between treatment groups (Richards 2006; Titov 2011; Newby 2013). Seven studies did not report any evaluations of differences in baseline severity and so were rated as having unclear risk of bias in this domain (Tillfors 2008; Andersson 2009; Bergstrom 2010; Paxling 2011; Greist 2012; Nordgren 2012; Andersson 2013). One study reported differences in age and marital status between study groups (Robinson 2010) and another reported differences in psychotropic medication use at baseline (Kok 2012); as it was unclear if this would have an effect on study results, these studies were rated as having unclear risk of bias in this domain.

Effects of interventions

Primary and secondary outcomes are reported by comparison below. Because adverse events were so rarely reported, they are not reported by comparison but are instead reported here. Only five studies included a measure that allowed for the assessment of participant deterioration over the course of treatment, for example, the CGI (Guy 1976). Andersson 2012a and Titov 2011 each identified one participant in the treatment condition who had deteriorated over the course of the study, but in neither case could their deterioration be linked to the treatment itself. Carlbring 2011 reported that no participants in their treatment condition had deteriorated. Hedman 2011 found one to two participants had deteriorated in each of the ICBT and face‐to‐face CBT conditions, but there was no difference between conditions. Ivarsson 2014 found that while eight participants in their control group showed a deterioration, only two participants in the treatment group showed a deterioration.

1. Therapist‐supported ICBT versus waiting list, attention, information, or online discussion group only control

Twenty‐eight studies compared therapist‐delivered distance CBT with a waiting list, attention, information, or online discussion group only control: Carlbring 2001; Carlbring 2006; Richards 2006; Carlbring 2007; Titov 2008a; Titov 2008b; Titov 2008c; Berger 2009; Furmark 2009a; Titov 2009; Robinson 2010; Titov 2010; Wims 2010; Carlbring 2011; Johnston 2011; Paxling 2011; Spence 2011; Titov 2011; Andersson 2012a; Andersson 2012b; Kok 2012; Nordgren 2012; Silfvernagel 2012; Newby 2013; van Ballegooijen 2013; Wootton 2013; Berger 2014; Ivarsson 2014. See Table 2 for subgroup analysis details.

Table 2. Subgroup analyses. Comparison 1: therapist‐supported ICBT versus waiting list control

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	2	42	39	RR, M‐H, Random	18.32 [2.50, 134.18]	3
		ii) Social Phobia	1	102	102	RR, M‐H, Random	6.00 [2.64, 13.62]	‐‐
		iii) GAD	3	91	94	RR, M‐H, Random	2.58 [1.48, 4.51]	36
		iv) PTSD	2	54	50	RR, M‐H, Random	3.48 [1.78, 6.80]	0
		v) OCD	1	15	14	RR, M‐H, Random	14.06 [0.88, 225.47]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	3	129	134	RR, M‐H, Random	3.76 [1.68, 8.43]	53
	b. By Therapist Contact
		i) High	1	12	9	RR, M‐H, Random	6.92 [0.42, 114.19]	‐‐
		ii) Medium	6	155	153	RR, M‐H, Random	4.34 [2.43, 7.76]	31
		iii) Low	4	204	202	RR, M‐H, Random	4.13 [1.73, 9.82]	70
	c. By Research Group
		i) Sweden	5	210	215	RR, M‐H, Random	5.12 [2.63, 9.98]	38
		ii) Australia‐Klein	1	47	48	RR, M‐H, Random	2.10 [1.44, 3.04]	‐‐
		iii) Australia‐Titov	5	149	140	RR, M‐H, Random	3.33 [1.88, 5.91]	44
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	62	69	RR, M‐H, Random	2.30 [1.41, 3.74]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	6	165	158	SMD, Random	‐1.52 [‐2.56, ‐0.48]	93
		ii) Social Phobia	8	336	325	SMD, Random	‐1.38 [‐1.63, ‐1.13]	48
		iii) GAD	6	194	200	SMD, Random	‐0.80 [‐1.19, ‐0.42]	69
		iv) PTSD	2	54	50	SMD, Random	‐0.78 [‐1.38, ‐0.17]	55
		v) OCD	1	15	17	SMD, Random	‐0.63 [‐1.35, 0.08]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	7	319	314	SMD, Random	‐0.81, [‐1.20, ‐0.41]	81
	b. By Therapist Contact
		i) High	1	12	9	SMD, Random	‐0.80 [‐1.70, 0.11]	‐‐
		ii) Medium	14	472	468	SMD, Random	‐1.34 [‐1.69, ‐0.99]	82
		iii) Low	10	420	416	SMD, Random	‐0.83 [‐1.15, ‐0.50]	78
	c. By Research Group
		i) Sweden	12	464	471	SMD, Random	‐1.50 [‐1.88, ‐1.11]	85
		ii) Australia‐Klein	1	12	9	SMD, Random	‐0.80 [‐1.70, 0.11]	‐‐
		iii) Australia‐Titov	12	365	350	SMD, Random	‐0.84 [‐1.06, ‐0.61]	50
		iv) Netherlands‐Kok	1	104	106	SMD, Random	‐0.16 [‐0.44, 0.11]	‐‐
		v) Netherlands‐van Ballegooijen	1	63	63	SMD, Random	‐0.30 [‐0.66, 0.05]	‐‐
		vi) Switzerland	2	75	65	SMD, Random	‐1.04 [‐1.63, ‐0.46]	59
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	4	126	122	SMD, Random	‐0.74 [‐1.35, ‐0.13]	78
		ii) Social Phobia	3	171	170	SMD, Random	‐0.64 [‐0.85, ‐0.42]	0
		iii) GAD	2	67	71	SMD, Random	‐1.91 [‐3.57, ‐0.26]	94
		iv) PTSD	2	54	50	SMD, Random	‐0.63 [‐1.02, ‐0.23]	0
		v) OCD	1	15	17	SMD, Random	‐1.05 [‐1.79, ‐0.30]	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	7	319	314	SMD, Random	‐0.49 [‐0.69, ‐0.29]	31
	b. By Therapist Contact
		i) High	1	12	9	SMD, Random	0.07 [‐0.79, 0.94]	‐‐
		ii) Medium	10	326	324	SMD, Random	‐0.92 [‐1.32, ‐0.52]	83
		iii) Low	6	266	261	SMD, Random	‐0.61 [‐0.82, ‐0.39]	25
	c. By Research Group
		i) Sweden	11	426	427	SMD, Random	‐0.94 [‐1.29, ‐0.59]	82
		ii) Australia‐Klein	1	12	9	SMD, Random	0.07 [‐0.79, 0.94]	‐‐
		iii) Australia‐Titov	4	103	95	SMD, Random	‐0.59 [‐0.88, ‐0.30]	0
		iv) Netherlands‐Kok	1	104	106	SMD, Random	‐0.21 [‐0.48, 0.07]	‐‐
		v) Netherlands‐van Ballegooijen	1	63	63	SMD, Random	‐0.39 [‐0.74, ‐0.04]	‐‐
		vi) Switzerland	1	44	44	SMD, Random	‐0.91 [‐1.23, ‐0.59]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Primary outcomes

1.1 Clinically important improvement in anxiety

Twelve studies assessed clinically important improvement in anxiety at post‐treatment after therapist‐supported ICBT versus a waiting list, attention, information, or online discussion group only control. A meta‐analysis with 433 treatment participants and 433 controls yielded a RR of 3.75 (95% CI 2.51 to 5.60; Analysis 1.1) in favour of the experimental intervention, with substantial heterogeneity (I² = 50%). These results did not change significantly following sensitivity analyses according to active waiting list control conditions, high risk of bias (ROB), or assuming dropouts were treatment responders. Results continued to favour the experimental intervention (though the size of the effect varied) following subgroup analyses by anxiety disorder (social phobia: 1 trial; PD: 2 trials; GAD: 3 trials; PTSD: 2 trials; OCD: 1 trial; mixed anxiety: 3 trials), amount of therapist contact (high: 1 trial; medium: 6 studies; low: 4 studies; not reported: 1 trial), or research group (Sweden: 5 trials; Australia‐Klein et al.: 1 trial; Australia‐Titov et al.: 5 trials; Switzerland: 1 trial).

1.2 Reduction in disorder‐specific anxiety symptom severity

All 30 studies that compared therapist‐supported ICBT to a waiting list, attention, information, or online discussion group only control assessed disorder‐specific anxiety symptoms at post‐treatment. Taken together, these 30 studies included 1083 treatment participants and 1064 control participants. Meta‐analytic findings showed a significant SMD of ‐1.06 (95% CI ‐1.29 to ‐0.82; Analysis 1.2; see Figure 4) in favour of the experimental condition, with considerable heterogeneity (I² = 83%). These results did not change significantly following sensitivity analyses according to active waiting list control conditions or high ROB. One study, Titov 2010, included three separate anxiety disorder subgroups that completed disorder‐specific measures so this study was entered as three studies in this meta‐analysis: Titov 2010 GAD; Titov 2010 Panic; and Titov 2010 Social Phobia. Results continued to favour the experimental intervention following subgroup analyses by anxiety disorder (social phobia: 7 trials; PD: 5 trials; GAD: 5 trials; PTSD: 2 trials; mixed anxiety: 8 trials), amount of therapist contact (medium: 14 trials; low: 10 trials; unreported: 3 trials), or research group (Sweden: 11 trials; Australia‐Klein et al.: 1 trial; Australia‐Titov et al.: 12 trials; Switzerland: 2 trials; Netherlands‐Kok et al.: 1 trial; Netherlands‐van Ballegooijen et al.: 1 trial). Exceptions were the single trials in each of the categories of OCD (Wootton 2013), high therapist contact (Richards 2006), Australia‐Klein group (Richards 2006), Netherlands‐Kok group (Kok 2012), and Netherlands‐van Bellegooijen group (van Ballegooijen 2013) did not emerge as significantly favouring the experimental intervention.

Figure 4

Forest plot: therapist‐supported ICBT versus waiting list control for anxiety symptom severity at post‐treatment.

1.3 Reduction in general anxiety symptom severity

Nineteen studies assessed participants' general anxiety after therapist‐supported ICBT (752 treatment participants) versus a waiting list, attention, information, or online discussion group only control (744 controls). Data analysis resulted in a SMD of ‐0.75 (95% CI ‐0.98 to ‐0.52; Analysis 1.3) showing a significantly greater decrease in general anxiety following the experimental intervention, with considerable heterogeneity (I² = 78%). Results were consistent following sensitivity analyses according to active waiting list control conditions and high ROB. Results continued to favour the experimental intervention following subgroup analyses by anxiety disorder (social phobia: 3 trials; PD: 4 trials; GAD: 2 trials; PTSD: 2 trials; OCD: 1 trial; mixed anxiety: 7 trials), amount of therapist contact (medium: 10 trials; low: 6 trials; unreported: 2 trials), or research group (Sweden: 11 trials; Australia‐Titov et al.: 4 trials; Switzerland: 1 trial; Netherlands‐van Ballegooijen et al.: 1 trial). Again, exceptions here were that the high therapist contact trial (Richards 2006), the 1 Australia‐Klein trial (Richards 2006), and the 1 Netherlands‐Kok trial (Kok 2012) did not significantly favour the experimental intervention.

Secondary outcomes

1.4 Quality of life

Twenty‐three studies reported on participants' quality of life following therapist‐supported ICBT (826 treatment participants) versus a waiting list, attention, information, or online discussion group only control (813 controls). Analysis resulted in a SMD of 0.47 (95% CI 0.38 to 0.57; Analysis 1.4) in favour of the experimental intervention, with minimal heterogeneity (I² = 0%) that may not be important. Results did not change significantly following sensitivity analyses according to active waiting list control conditions or high ROB.

1.5 Participant satisfaction with the intervention

A comparison of treatment satisfaction was not warranted as authors expectedly did not report on the satisfaction of participants in the waiting list, attention, information, or online discussion group only controls. Seventeen studies reported on participants' satisfaction with treatment. Overall, participants reported a high level of satisfaction with the intervention, with roughly 90% of participants across these studies reporting being very or mostly satisfied with the treatment. Several studies reported that over 90% of participants found the quality of the online treatment modules and their correspondence with a therapist to be excellent or good. When measures of treatment satisfaction were used, participants' scores generally reflected being somewhat to very satisfied with the intervention (e.g., M = 3.34 on the Client Satisfaction Questionnaire ‐ 8, Berger 2014; M = 25.64 out of a possible 30 on a combined measure of treatment satisfaction, Newby 2013).

Only a few studies mentioned any problems or dissatisfaction with the intervention. Most notably, three studies reported that a majority of participants (70%) found the treatment moved too quickly (Carlbring 2006; Titov 2008a; Titov 2008b). Several studies reported small numbers of participants who had been dissatisfied with treatment: 3% dissatisfied with treatment (Carlbring 2006); 6% rated quality of therapist correspondence as neutral or somewhat dissatisfied, 1% rated quality of therapist contact as very dissatisfied (Titov 2008b); 11% dissatisfied with treatment (Berger 2009); 13% neutral or somewhat dissatisfied with treatment, 2% rated quality of therapist correspondence as unsatisfactory (Robinson 2010); 5% rated quality of therapist correspondence as unsatisfactory (Titov 2010); 16% neutral or somewhat dissatisfied with treatment, (Johnston 2011). Titov 2008c also reported that 7% of participants found that their confidence in their ability to manage their symptoms and their motivation to continue practicing their skills had not changed. Berger 2009 reported that one participant rated the self‐help modules as too difficult and one participant indicated that they did not understand the purpose of the self‐help modules.

2. Therapist‐supported ICBT versus unguided CBT

Five studies compared therapist‐supported ICBT with unguided CBT: Titov 2008c; Furmark 2009a; Furmark 2009b; Berger 2011; Greist 2012. See Table 3 for subgroup analysis details for this comparison.

Table 3. Subgroup analyses. Comparison 2: therapist‐supported ICBT versus unguided CBT

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Low	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
	c. By Research Group
		i) Sweden	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	4	127	126	SMD, Random	‐0.24 [‐0.69, 0.21]	68
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	1	31	28	SMD, Random	‐0.14 [‐0.65, 0.37]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	3	100	99	SMD, Random	‐0.18 [‐0.78, 0.41]	77
		iii) Low	1	27	27	SMD, Random	‐0.43 [‐0.97, 0.11]	‐‐
	c. By Research Group
		i) Sweden	2	69	69	SMD, Random	0.12 [‐0.22, 0.45]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	1	31	30	SMD, Random	‐0.82 [‐1.34, ‐0.29]	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	SMD, Random	‐0.43 [‐0.97, 0.11]	‐‐
		vii) USA	1	31	28	SMD, Random	‐0.14 [‐0.65, 0.37]	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Clinically Important Improvement in Anxiety at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	3	96	96	SMD, Random	‐0.30 [‐0.58, ‐0.01]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact at Follow‐up
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	SMD, Random	‐0.31 [‐0.65, 0.03]	3
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	SMD, Random	‐0.31 [‐0.65, 0.03]	3
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	SMD, Random	‐0.26 [‐0.80, 0.27]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Primary outcomes

2.1 Clinically important improvement in anxiety

Only Berger 2011 assessed clinically important improvement in anxiety after therapist‐supported ICBT versus unguided CBT. They reported that 16/27 participants receiving therapist‐supported ICBT and 15/27 participants completing unguided CBT no longer met the diagnostic criteria post‐treatment (Analysis 2.1).

2.2 Reduction in disorder‐specific anxiety symptom severity

The five studies that compared therapist‐supported ICBT to unguided CBT (that is, self‐help) assessed disorder‐specific anxiety symptoms at post‐treatment. Combined, these studies included 158 treatment and 154 control participants and resulted in a non‐significant SMD of ‐0.22 (95% CI ‐0.56 to 0.13; Analysis 2.2), with substantial heterogeneity (I² = 58%). At 6 to 12 month follow‐up, 3 studies reported on this outcome; a meta‐analysis of 96 treatment and 96 comparator participants resulted in a significant SMD of ‐0.30 (95% CI ‐0.58 to ‐0.01; Analysis 2.3) in favour of the experimental intervention with minimal but difficult to estimate heterogeneity (I² = 0%). No sensitivity analyses were required.

Results for this outcome at post‐treatment remained non‐significant following subgroup analyses by anxiety disorder (social phobia: 4 trials; OCD: 1 trial), therapist contact (medium: 3 trials; not reported: 2 trials), and research group (Australia‐Titov et al.: 1 trial; Switzerland: 1 trial; Sweden: 2 trials: USA: 1 trial). Subgroup analyses based on anxiety disorder were not warranted at follow‐up as all studies investigated social phobia.

Results for this outcome at follow‐up following subgroup analyses by therapist contact (medium: Furmark 2009a; Furmark 2009b; not reported: Berger 2011) resulted in a non‐significant difference with minimal but difficult to estimate heterogeneity (SMD ‐0.31, 95% CI ‐0.65 to 0.03; I² = 3%). The same was true for results for this outcome at follow‐up following subgroup analyses by research group (Switzerland: Berger 2011; Sweden: Furmark 2009a; Furmark 2009b; SMD ‐0.31, 95% CI ‐0.65 to 0.03; I² = 3%).

2.3 Reduction in general anxiety symptom severity

Only two studies assessed participants' general anxiety after therapist‐supported ICBT (69 treatment participants) versus self‐help interventions (69 comparator participants). Data analysis resulted in a non‐significant mean difference of 0.28 (95% CI ‐2.21 to 2.78; Analysis 2.4), with minimal but difficult to estimate heterogeneity (I² = 0%). A similar result was found at 12 month follow‐up with the same studies; the mean difference was 0.72 (95% CI ‐2.12 to 3.57; Analysis 2.5), with minimal but difficult to estimate heterogeneity (I² = 0%). No sensitivity analyses were required. Subgroup analyses all relevant characteristics of the included studies were the same.

Secondary outcomes

2.4 Quality of life

Three studies indexed quality of life of participants following therapist‐supported ICBT (100 treatment participants) versus unguided CBT (99 control participants). Data analysis resulted in a non‐significant SMD of 0.07 (95% CI ‐0.37 to 0.50; Analysis 2.6), with moderate to substantial heterogeneity (I² = 58%). At six to 12 month follow‐up, only two of these studies indexed quality of life of participants following treatment (69 treatment and 69 comparator participants), with meta‐analysis showing a similar non‐significant SMD of ‐0.19 (95% CI ‐0.53 to 0.14; Analysis 2.7), with minimal but difficult to estimate heterogeneity (I² = 0%). No sensitivity analyses were required.

2.5 Participant satisfaction with the intervention

Two studies indexed participant satisfaction with the intervention. Berger 2011 found that treatment satisfaction was significantly higher in the therapist‐supported ICBT condition as compared to the self‐help condition according to the Client Satisfacton Questionnaire (Attkisson 1982). Similarly, Titov 2008c found that a significantly greater number of participants in the therapist‐supported ICBT condition as compared to the self‐help condition were very or mostly satisfied with their treatment (no participants reported being dissatisfied with treatment). However, Titov 2008c reported no differences between conditions in perceptions of how logical the treatment was, participants' confidence in recommending the treatment to a friend, and the extent to which treatment had increased participants' confidence in managing their symptoms. Seven per cent of participants in the ICBT condition reported that the treatment had not changed their confidence in managing their symptoms or their motivation to keep practicing techniques they had learned.

3. Therapist‐supported ICBT versus face‐to‐face CBT

Seven studies compared therapist‐supported ICBT with face‐to‐face CBT: Carlbring 2005; Kiropoulos 2008; Tillfors 2008; Andersson 2009; Bergstrom 2010; Hedman 2011; Andersson 2013. See Table 4 for subgroup analysis details for this comparison.

Table 4. Subgroup analyses. Comparison 3: therapist‐supported ICBT versus face‐to‐face CBT

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	3	121	118	RR, M‐H, Random	1.06 [0.85, 1.32]	0
		ii) Social Phobia	1	64	62	RR, M‐H, Random	1.45 [0.77, 2.76]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	46	40	RR, M‐H, Random	1.11 [0.57, 2.15]	‐‐
		ii) Medium	1	25	24	RR, M‐H, Random	1.20 [0.85, 1.69]	‐‐
		iii) Low	2	114	116	RR, M‐H, Random	1.08 [0.72, 1.60]	34
	c. By Research Group
		i) Sweden	3	139	140	RR, M‐H, Random	1.09 [0.88, 1.36]	0
		ii) Australia‐Klein	1	46	40	RR, M‐H, Random	1.11 [0.57, 2.15]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	3	119	115	SMD, Random	0.29 [0.03, 0.54]	0
		ii) Social Phobia	2	83	80	SMD, Random	‐0.18 [‐0.92, 0.57]	76
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	27	SMD, Random	‐0.02 [‐0.56, 0.52]	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	2	63	55	SMD, Random	0.42 [0.05, 0.78]	0
		ii) Medium	1	25	24	SMD, Random	0.05 [‐0.51, 0.61]	‐‐
		iii) Low	4	140	143	SMD, Random	‐0.10 [‐0.53, 0.33]	64
	c. By Research Group
		i) Sweden	6	184	185	SMD, Random	‐0.03 [‐0.34, 0.28]	49
		ii) Australia‐Klein	1	44	37	SMD, Random	0.49 [0.05, 0.93]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	2	67	62	SMD, Random	0.42 [‐0.75, 1.60]	90
		ii) Social Phobia	2	83	80	SMD, Random	‐0.18 [‐0.49, 0.13]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.14 [‐1.07, 0.79]	64
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	2	61	56	SMD, Random	0.49 [‐0.57, 1.56]	86
		ii) Medium	1	25	24	SMD, Random	‐0.19 [‐0.75, 0.38]	‐‐
		iii) Low	3	90	87	SMD, Random	‐0.18 [‐0.59, 0.22]	29
	c. By Research Group
		i) Sweden	5	134	129	SMD, Random	‐0.17 [‐0.42, 0.07]	0
		ii) Australia‐Klein	1	42	38	SMD, Random	1.01 [0.54, 1.48]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Clinically Important Improvement in Anxiety at Follow‐up
	a. By Disorder
		i) Panic	2	75	78	RR, M‐H, Random	1.09 [0.93, 1.28]	0
		ii) Social Phobia	1	64	62	RR, M‐H, Random	1.15 [0.73, 1.83]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	1	25	24	RR, M‐H, Random	1.05 [0.87, 1.27]	‐‐
		iii) Low	2	114	116	RR, M‐H, Random	1.17 [0.92, 1.50]	0
	c. By Research Group
		i) Sweden	3	139	140	RR, M‐H, Random	1.10 [0.94, 1.27]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	2	75	78	SMD, Random	‐0.04 [‐0.36, 0.28]	0
		ii) Social Phobia	2	83	80	SMD, Random	‐0.39 [‐0.71, ‐0.08]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.09 [‐0.64, 0.46]	0
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	19	18	SMD, Random	‐0.11 [‐0.76, 0.53]	‐‐
		ii) Medium	1	25	24	SMD, Random	‐0.04 [‐0.60, 0.52]	‐‐
		iii) Low	4	140	141	SMD, Random	‐0.24 [‐0.48, 0.00]	3
	c. By Research Group
		i) Sweden	6	184	183	SMD, Random	‐0.20 [‐0.41, 0.00]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	1	25	24	SMD, Random	‐0.17 [‐0.74, 0.39]	‐‐
		ii) Social Phobia	2	83	80	SMD, Random	‐0.14 [‐0.45, 0.17]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.06 [‐0.61, 0.49]	0
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	19	18	SMD, Random	‐0.19 [‐0.83, 0.46]	‐‐
		ii) Medium	1	25	24	SMD, Random	‐0.17 [‐0.74, 0.39]	‐‐
		iii) Low	3	90	87	SMD, Random	‐0.10 [‐0.40, 0.19]	0
	c. By Research Group
		i) Sweden	5	134	129	SMD, Random	‐0.13 [‐0.37, 0.11]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Primary outcomes

3.1 Clinically important improvement in anxiety

Four studies assessed clinically important improvement in anxiety at post‐treatment after therapist‐supported ICBT (185 treatment participants) versus face‐to‐face CBT (180 comparator participants). Meta‐analysis yielded a non‐significant RR of 1.09 (95% CI 0.89 to 1.34; Analysis 3.1), with minimal heterogeneity that may not be important (I² = 0%). At 6 to 12 month follow‐up, the results of 3 studies that reported on clinically important improvement in anxiety, with 139 treatment and 140 comparator participants, resulted in a non‐significant RR of 1.10 (95% CI 0.94 to 1.27; Analysis 3.2), again with minimal heterogeneity (I² = 0%). Results did not change significantly following a sensitivity analysis assuming dropouts were treatment responders. Results for this outcome remained non‐significant following subgroup analyses by anxiety disorder (social phobia: 1 trial; PD: 3 trials) , therapist contact (high: 1 trials; medium: 1 trial; low: 2 trials), and research group (Sweden: 3 trials: Australia‐Klein et al.: 1 trial).

3.2 Reduction in disorder‐specific anxiety symptom severity

The seven studies that compared therapist‐supported ICBT to face‐to‐face CBT assessed changes in symptom specific anxiety. Using these seven studies, including 228 treatment participants and 222 control participants, meta‐analysis resulted in a non‐significant SMD of 0.06 (95% CI ‐0.25 to 0.37; Analysis 3.3; see Figure 5), with substantial heterogeneity (I² = 60%). At 6 to 12 month follow‐up, data from six studies, including 184 treatment participants and 183 comparator participants, could be used to assess changes in symptom specific anxiety. Meta‐analysis resulted in a non‐significant SMD of ‐0.20 (95% CI ‐0.41 to 0.0; Analysis 3.4) with minimal heterogeneity that may not be important (I² = 0%). Results remained non‐significant following a sensitivity analysis excluding two studies that did not use ITT analysis and had high ROB (Andersson 2009; Andersson 2013).

Figure 5

Forest plot: therapist‐supported ICBT versus face‐to‐face CBT for anxiety symptom severity at post‐treatment.

Results for this outcome at post‐treatment were somewhat variable following subgroup analyses by anxiety disorder (social phobia: 2 trials; PD: 3 trials; specific phobia: 2 trials). A meta‐analysis of the studies investigating PD (Carlbring 2005; Kiropoulos 2008; Bergstrom 2010) found a significant SMD of 0.29 (95% CI 0.03 to 0.54) with minimal but difficult to estimate heterogeneity (I² = 0%) in favour of face‐to‐face CBT. In contrast, a meta‐analyses of two studies investigating social phobia (Tillfors 2008; Hedman 2011; SMD ‐0.18, 95% CI ‐0.92 to 0.5; I² = 76%) and two studies investigating specific phobia (Andersson 2009; Andersson 2013; SMD ‐0.02, 95% CI ‐0.56 to 0.52; I² = 0%) remained non‐significant (in line with the overall meta‐analysis).

Unexpectedly, at 6 to 12 month follow‐up it was only the meta‐analysis of social phobia studies (Tillfors 2008; Hedman 2011) that showed a significant difference between groups, with an SMD of ‐0.39 (95% CI ‐0.71 to ‐0.08) with minimal but difficult to estimate heterogeneity (I² = 0%) in favour of the experimental intervention, while the meta‐analysis of PD studies (Carlbring 2005; Bergstrom 2010) was non‐significant (SMD ‐0.04, 95% CI ‐0.36 to 0.28; I² = 0%), as was the meta‐analysis of specific phobia studies (Andersson 2009; Andersson 2013; SMD ‐0.09, 95% CI ‐0.64 to 0.46; I² = 0%).

Results for this outcome at post‐treatment were also somewhat variable following subgroup analyses by therapist contact (high: 2 trials; medium: 1 trial; low: 4 trials). A subgroup analysis of studies with high therapist contact (Kiropoulos 2008; Tillfors 2008) resulted in a significant SMD of 0.42 (95% CI 0.05 to 0.78), with minimal but difficult to estimate heterogeneity (I² = 0%), in favour of face‐to‐face CBT at post‐treatment. The meta‐analysis of studies with low therapist contact remained non‐significant (SMD ‐0.10, 95% CI ‐0.53 to 0.33) with substantial heterogeneity (I² = 64%) (Andersson 2009; Bergstrom 2010; Hedman 2011; Andersson 2013). At follow‐up, results for this outcome remained non‐significant following subgroup analyses by therapist contact.

Results for this outcome at post‐treatment and follow‐up remained non‐significant following subgroup analysis of the 6 trials by the Sweden research group (SMD=‐0.03, 95% CI ‐0.34 to 0.28, I² = 49%), however at post‐treatment the one Australia‐Klein trial (Kiropoulos 2008) resulted in a significant difference in favour of face‐to‐face CBT.

3.3 Reduction in general anxiety symptom severity

Six studies reported participants' levels of general anxiety post‐treatment. The six studies combined in the meta‐analysis included 176 treatment participants and 167 comparator participants and resulted in a non‐significant SMD of 0.06 (95% CI ‐0.42 to 0.55; Analysis 3.5), with substantial to considerable heterogeneity (I² = 78%). When the Kiropoulos 2008 study was removed from the analysis (because it presented transformed data, which we back‐transformed to include in the analysis), the resulting SMD remained non‐significant at ‐0.17 (95% CI ‐0.42 to 0.07) and heterogeneity was reduced (I² = 0%). At 6 to 12 month follow‐up, five studies reported participants' level of general anxiety. The five studies included 134 treatment participants and 129 comparator participants and yielded a non‐significant SMD of ‐0.13 (95% CI ‐0.37 to 0.11; Analysis 3.6) with minimal heterogeneity that may not be important (I² = 0%). Results remained non‐significant following a sensitivity analysis excluding two studies that did not use ITT analyses and had high ROB (Andersson 2009; Andersson 2013). Results for this outcome remained non‐significant following subgroup analyses at post‐treatment and follow‐up by anxiety disorder (social phobia: 2 trials; PD: 2 trials; specific phobia: 2 trials), therapist contact (high: 2 trials; medium: 1 trial; low: 3 trials), and research group (Sweden: 5 trials). The single Australia‐Klein trial (Kiropoulos 2008) did find a significant difference in favour of the face‐to‐face CBT condition at post‐treatment.

Secondary outcomes

3.4 Quality of life

Five studies reported on participants' quality of life following therapist‐supported ICBT (198 treatment participants) versus face‐to‐face CBT (194 comparator participants). Analysis resulted in a SMD of 0.26 (95% CI 0.06 to 0.45; Analysis 3.7) in favour of the experimental intervention, with minimal heterogeneity that may not be important (I² = 0%). This trend continued at 6 to 12 month follow‐up. Four studies comprising 158 treatment and 158 comparator participants resulted in a SMD of 0.33 (95% CI 0.11 to 0.55; Analysis 3.8) in favour of the experimental intervention, again with minimal heterogeneity (I² = 0%). No sensitivity analyses were required.

3.5 Participant satisfaction with the intervention

Two studies indexed participant satisfaction with the intervention. Overall, treatment satisfaction was high across both therapist‐supported ICBT and face‐to‐face CBT. In one study (Tillfors 2008), only one participant in the ICBT condition and two participants in the face‐to‐face condition reported being "neutral/somewhat dissatisfied with treatment" and no participants reported being "very dissatisfied" with treatment. Both studies found no significant difference between conditions in participants' overall satisfaction with the intervention or their perceptions of improvement as a result of treatment.

A notable significant difference between treatment conditions appeared in one instance: Kiropoulos 2008 found that participants receiving therapist‐supported ICBT reported significantly less enjoyment in communicating with their therapist as compared to participants receiving face‐to‐face CBT.

Sensitivity analysis

Sensitivity analyses are detailed in the results section above. Given the available studies for this review, some of the planned sensitivity analyses were not warranted. First, sensitivity analyses based on the blinding of participants or personnel, or both, in the included studies were not conducted because blinding of participants and personnel is not standard practice with this type of clinical trial. Second, as none of the included studies were cluster randomised trials and none of the included studies with multiple intervention arms had selective reporting of intervention comparisons, sensitivity analyses based on these characteristics were not conducted. Third, as we were not required to impute any standard deviations, we also eliminated that planned sensitivity analysis. Only one study included transformed data (Kiropoulos 2008) and it was discussed in section 3.3 above, and only two studies did not use ITT data (Andersson 2009; Andersson 2013) and are discussed in sections 3.2 and 3.3 above. Finally, as LOCF was the primary method of ITT analysis reported by authors, we did not exclude studies using LOCF.

Discussion

Summary of main results

Please refer to the summary of findings Table for the main comparison, summary of findings Table 2, and summary of findings Table 3 for a summary of the main results.

The present review investigated the efficacy of therapist‐supported ICBT in treating anxiety disorders in adults. We identified 38 studies to be included in the review, comparing the intervention of interest to a waiting list, attention, information, or online discussion group only control, unguided CBT, and face‐to‐face group or individual CBT.

The present findings suggest that therapist‐supported ICBT is more efficacious than a waiting list, attention, information, or online discussion group only control in leading to clinically important improvement in anxiety, reducing anxiety symptoms (both disorder‐specific and general), and improving quality of life. Results also generally showed no difference in outcomes following therapist‐supported ICBT versus unguided CBT at post‐treatment, though results are limited by very low quality evidence due to a limited number of studies (that is, imprecision). Moreover, results suggest that therapist‐supported ICBT may not be significantly different from face‐to‐face group and individual CBT in treating anxiety disorders. Meta‐analyses revealed no significant differences in clinically important improvement in anxiety or reduction in anxiety symptoms (both disorder‐specific and general) at post‐treatment or follow‐up for these two interventions.

At 6 to 12 month follow‐up, results generally mirror the post‐treatment findings but are limited by the small number of studies and the degree of variability in the interventions under investigation across studies. Thus, these findings should be interpreted with caution.

All findings largely remained robust following sensitivity analyses conducted to explore the impact of potential sources of bias or heterogeneity. Subgroup analyses suggest that there may be some differences in outcome based on the type of anxiety disorder being treated or the amount of therapist contact in the intervention; however, the small number of studies within each subgroup limits our ability to draw firm conclusions based on these analyses. More research is needed in these areas.

Overall completeness and applicability of evidence

Taken together, the studies included in the present review help us answer the question, is therapist‐supported ICBT an efficacious treatment for anxiety disorders in adults? In particular, the included studies are of sufficient number to comprehensively compare the efficacy of therapist‐supported ICBT to a waiting list, attention, information, or online discussion group only control. There are fewer, but still sufficient, studies to compare the efficacy of therapist‐supported ICBT to traditional face‐to‐face CBT. In comparison, the number of studies comparing therapist‐supported ICBT to unguided CBT (that is, self‐help) is limited and therefore findings with respect to this comparison must be interpreted with some caution.

In terms of the applicability of the evidence to ICBT interventions and particular patient populations, several factors warrant consideration when interpreting the present findings. First, the included interventions are quite heterogeneous. While all studies investigated therapist‐supported ICBT, the nuances of each intervention (for example, length, number of online modules, nature of therapist support) varied widely. It seems prudent to note that while these interventions seem efficacious as a whole, the optimal characteristics of these interventions have yet to be identified and their heterogeneity reduces the quality of the body of evidence investigated here.

Second, the included studies investigated a number of different anxiety disorders with a particular focus on PD, social phobia, and GAD, either separately or as part of a transdiagnostic treatment package. As such, we can be most confident that the present findings apply to the treatment of these disorders. More research is needed into ICBT for other anxiety disorders, such as OCD, PTSD, and specific phobia.

Third, researchers have previously raised some concerns about the participants included in investigations of ICBT, as many of these studies recruit participants from the community via media advertisements (for example, Cuijpers 2009). There is some question as to whether these participants are similar enough to participants recruited for face‐to‐face CBT RCTs, who tend to be recruited via clinic referrals. Despite this concern, research by Titov and colleagues (Titov 2010b) found that ICBT participants are as severe in terms of symptom severity, distress, and disability as individuals attending a face‐to‐face clinic and more severe than individuals identified via an epidemiological survey. We also attempted to account for this possible difference in participant characteristics by including only individuals with an anxiety disorder diagnosed using a standardized instrument.

Despite the heterogeneity of the interventions and populations across studies, the robustness of findings following sensitivity analyses lends credence to the efficacy of therapist‐supported ICBT as an alternative method of delivering CBT to those with anxiety disorders who are in need of intervention.

Quality of the evidence

We considered the quality of the evidence of the included studies using the GRADE tool (Higgins 2011b). Most of the evidence contributing to the comparisons in this review was rated as low quality; however, simply dismissing findings based on this determination is hasty. A careful look at reasons for the downgrading of quality of evidence is warranted to best interpret the findings.

With respect to risk of bias, the included evidence is of moderate quality as there were only a few concerns with the internal validity of the included studies. There were some difficulties with (a) blinding of outcome assessors, and (b) incomplete outcome data in several of the included studies. Sensitivity analyses excluding these studies suggest that any potential bias introduced by these studies did not affect the meta‐analytic outcomes. Risk of bias ratings were inflated across comparisons in this review due to the nature of clinical psychotherapy trials, which precludes keeping the treatment condition concealed from the participant or the therapist delivering the treatment. It is important to note that there is no way to blind participants or study therapists, suggesting that reaching a higher quality study design would be difficult. Neverthless, downgrading evidence based on this characteristic of risk of bias gives an opportunity to make recommendations for future studies to limit (or at least explore potential impact of) this source of bias, perhaps by recruiting clients or therapists who do not favour interventions, by ensuring intervention providers in both groups believe they are delivering the ‘best’ treatment, and by measuring the potential impact by asking patients in all groups if they believe they received the ‘effective’ treatment. In addition to these risk of bias domains, it should be noted that there may also be some concerns with selective outcome reporting, but these remain unclear. Selective outcome reporting has been found to be an important concern in non‐pharmaceutical trials (Milette 2011), such as those included here. Approximately one third of the studies included in this review were not prospectively registered on a trial database. As such, it is impossible to discern if these studies are biased by selective reporting. It may be that with the advent of trial registration becoming more common (and expected), updates to this review will be able to provide a more clear estimate of the risk of selective outcome reporting.

There is a large degree of heterogeneity in a number of the meta‐analyses in this review, reducing the quality of some of the evidence. Subgroup and sensitivity analyses provide some indication of what may account for the heterogeneity, but there is by no means a clear answer. Some degree of heterogeneity may have emerged because we included studies of a range of anxiety disorders, including PD, social phobia, GAD, specific phobia, OCD, and PTSD, in the meta‐analyses. It seems possible there are nuances unique to each of these disorders and their treatment that might facilitate or hamper the efficacy of their treatment via therapist‐supported ICBT. Some subgroup analyses by anxiety disorder resulted in an important decrease in heterogeneity, however, in other cases heterogeneity did not decrease at all. This may have been in part because even within studies of the same disorder, researchers employed different outcome measures to assess treatment outcomes. The variability in outcome measures within and across studies may account for some important heterogeneity. Support for this hypothesis may be found in the fact that the quality of life outcome tends to show the least heterogeneity across comparisons as well as the least variability in assessment measures used. Also of importance, the nature of the ICBT interventions included in this review is quite diverse in terms of length, number of online modules, and nature of therapist contact. It may be that the nuances of these treatments led to nuanced differences in treatment outcome. However, subgroup analyses based on amount of therapist contact, for example, did not sufficiently and consistently reduce heterogeneity. Similarly, subgroup analyses by research group did not consistently lead to decreases in heterogeneity. This is surprising given the assumption that studies conducted within the same research laboratory would have some degree of consistency in methods, outcome measures, participants, etc. Nevertheless, these studies did vary over the years in terms of the anxiety disorder investigated and amount and nature of therapist contact with participants.

All subgroup analyses are complicated by the fact that only a small number of trials tended to be included in each analysis, making it difficult to estimate heterogeneity. Thus, heterogeneity does remain somewhat of a concern in the present review. While this was unexpected, it may be that there is simply too much variability in study methods, populations, outcome measures, etc. across studies and not enough studies to support meaningful subgroup analyses at this time. Importantly, our speculation is that this heterogeneity might be explained by the expected factors discussed here as opposed to any bias in the included studies. An increase in the number of studies in this area in the future may allow us to explore heterogeneity more robustly and meaningfully. In the future, we will also explore further subgroup analyses on other factors that may contribute to varying treatment outcomes, including length of the intervention in question. We are limited somewhat in the subgroup analyses we can conduct given that some potentially relevant factors (e.g., history of disorder) are variable within studies and thus not conducive to a study‐by‐study comparison.

In considering the quality of the evidence, we also examined indirectness of the included studies (that is, the degree to which the included studies address the review objective) and the imprecision of each study's findings. Across included studies, we had no concerns with indirectness. As far as imprecision, some of the included studies are limited by small sample size. The meta‐analyses attempt to address such small samples by combining studies, where appropriate. Precision of findings may also be affected by rates of dropout across interventions, particularly if there is the chance that one of the two interventions being compared is likely to lead to greater dropout. Given some of the characteristics of ICBT (for example, engaging from a distance, no requirement to commit to appointment times or be accountable), one might expect there to be greater dropout rates with this type of treatment. However, the present findings suggest this may not be the case. There were generally quite similar rates of dropout across the interventions investigated (experimental and comparator). Almost all studies used a rigorous and somewhat conservative method to account for missing data. Sensitivity analyses on dichotomous outcomes, assuming dropouts were treatment responders, did not significantly change the meta‐analytic outcomes. These details suggest that the precision of findings are not significantly threatened by treatment dropout rates.

Finally, we considered whether publication bias might have affected the evidence. The number of studies within each comparison in the present review only permitted the analysis of funnel plots for several outcomes for the comparison of the intervention and a waiting list control (Figure 6; Figure 7; Figure 8; Figure 9). A visual inspection of these funnel plots suggested that there may have been a small study effect (that is, the potential for some publication bias). Because there were less than 10 studies in the other meta‐analyses in this review (in accordance with the guidelines for the use of funnel plots in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a)), we did not analyze publication bias using funnel plots for the remaining comparisons. To complement the use of funnel plots, we looked to match trials recorded in clinical trial registries with published manuscripts. Accounting for the fact that many of the most recent registered trials are still ongoing or may be in the process of being published, we only observed a handful of registries that could not be matched with a published trial. This would suggest that, at least recently (since trial registration has been strongly encouraged), publication bias may not be a significant concern for this review. However, our findings with respect to the consistency between trial registries and published studies do not rule out earlier publication bias or the possibility of bias due to smaller‐scale, unfunded studies that may not have been registered. In a further effort to assess for publication bias, we contacted authors in the field to inquire about any unpublished findings and were only informed of two studies that were unfinished and unpublished, one due to difficulties with funding, the other due to difficulties with recruitment. With these factors in mind, we cannot make a conclusive statement about publication bias. Publication bias may not limit the quality of the included evidence but readers should keep the possibility of this bias in mind when interpreting the review findings.

Figure 6

Funnel plot of the outcome clinically important improvement in anxiety at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 7

Funnel plot the outcome anxiety symptom severity at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 8

Funnel plot of the outcome general anxiety symptom severity at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 9

Funnel plot of the outcome quality of life at post‐treatment for therapist‐supported ICBT versus waiting list control.

Overall, the included evidence, across studies and comparisons, is of low quality. In many cases reductions in quality tend to be due to heterogeneity, which may be explained by meta‐analytic methods rather than the evidence itself.

Potential biases in the review process

Given the variability within ICBT interventions, it is possible that there are several biases inherent in the present review. First, we elected to include only those interventions that did not include face‐to‐face therapist contact during active treatment. This may have excluded studies that were simply conducted within therapists' offices for practical purposes but could in fact have been followed online by a client at home as well. In this way, the included studies may not comprehensively include all possible ICBT treatments. Second, we included only interventions with active therapist involvement. This decision was made because (a) there seems to be an important distinction between guided and unguided treatment, and (b) some prior research has suggested that therapist involvement may be an important part of distance treatment (Spek 2007; Andersson 2009b). Nevertheless, this decision impacted the types of trials included in the present review and led to the exclusion of some Internet‐based studies that did not directly involve therapists but included interactive voice response software (Greist 2002). While preliminary work has been done, further research will need to investigate the importance of active versus automated versus no therapist involvement in ICBT.

Another potential bias in the review process may have been introduced as we elected to include only those studies in which participants were identified as meeting diagnostic criteria for an anxiety disorder, as determined by a validated measure. While this is good practice for the empirical validity of the present review, it may not accurately reflect clinical practice. It is likely that as a part of regular clinical practice, clients with subclinical diagnoses might be assigned to pursue ICBT. We might assume that these treatments would be as effective for individuals with subclinical symptom patterns as they are for those with diagnosed disorders (for example, Spek 2007); however, our exclusion of these populations prevents any firm statements in this regard.

The present review is also potentially biased in the way that we have measured one of our primary outcomes, clinically important improvement in anxiety. This outcome would possibly be more clinically useful had it been narrowed to assess diagnostic remission, in particular, or divided into two outcomes assessing remission and recovery separately. This issue is larger than the field of ICBT and is a result of the lack of consensus in clinical psychology research in general regarding the most robust way to assess clinically significant improvement resulting from treatment. Consequently, there are a variety of ways to assess treatment outcome, including measures of remission, recovery, clinically significant improvement, and high end state functioning. Because this issue is quite prevalent in ICBT for anxiety studies as well, we elected to combine each of these unique ways of determining notable changes in symptomatology and functioning post‐treatment by creating the clinically important improvement outcome. While relevant and useful, the nuances of remission and recovery may be lost by being subsumed within this category. As the field expands, and consistency in reporting treatment outcomes increases, it may be useful to subdivide this outcome to more clearly capture remission and recovery.

It is also worth mentioning that given the conceptual and operational overlap between quality of life and disability measures in the anxiety disorder literature (Mogotsi 2000), we included both outcomes within the same meta‐analysis. However, given research suggesting that these concepts are overlapping but also distinct (Hambrick 2003), it may be that some variability in the impact of treatment on these measures was missed through their amalgamation. Future studies on ICBT should consider assessing both quality of life and disability as separate treatment outcomes.

Finally, it is necessary to note that our method of statistical analysis may have introduced some bias into the results. In combining multiple measures within one study that assessed the same outcome (for example, combining several measures of panic symptoms into one mean and standard deviation) we made use of a method described by Borenstein 2009 that requires the availability of bivariate correlations between the study measures in order for them to be combined. In four studies in the present review, these correlations were not available. In these situations we simply excluded the measure in question from the overall mean and measure of variance. The general concordance between each of the symptom measures within each study (that is, a series of symptom measures tended to show similar direction and degree of change from pre‐ to post‐treatment) means that the exclusion of one measure should not significantly impact the results. However, this process may have introduced some small degree of bias into the findings.

Agreements and disagreements with other studies or reviews

A number of prior meta‐analyses have investigated the efficacy of ICBT. These meta‐analyses have ranged from a quite broad scope investigating the efficacy of Internet interventions for any health problems (d = 0.53; Barak 2008) to a more focused scope investigating the efficacy of ICBT for clinical and subclinical anxiety and mood symptoms (d = 0.96; Spek 2007) or anxiety and mood disorders in Sweden (d = 0.91; Andersson 2007). Taken together, these reviews support the present findings that ICBT is efficacious in reducing anxiety symptoms as compared to control interventions (for example, waiting list control). Within their broad meta‐analysis, Barak 2008 found that interventions designed to treat PTSD and those targeting PD showed the largest effect sizes (g = 0.88 and 0.80, respectively). Spek 2007 found that those interventions that included therapist contact, as opposed to those that did not, showed a particularly large effect size (d = 1.00). It should be noted that some concerns were raised about the methodological quality of the studies included in these types of reviews given their small sample sizes, the absence of details about randomisation and treatment allocation methods, and lack of adequate information about treatment compliance and credibility (Postel 2008).

Recently, Mayo‐Wilson 2013 completed a review of media‐delivered self‐help BT and CBT for anxiety disorders. Within their review they included ICBT studies delivered both with and without therapist contact. In line with the present findings, their review suggested that media‐delivered self‐help BT and CBT were more efficacious than no treatment (that is, a waiting list control). In contrast to the present findings, their review resulted in some suggestion that media‐delivered self‐help BT and CBT were somewhat inferior to face‐to‐face CBT with the conclusion that for those who can access it, face‐to‐face CBT is probably superior. The differences between these findings and those in the present review may be due in part to the differences in therapist involvement between included studies across the two reviews. Therapist involvement in media‐delivered treatments, such as ICBT, may lead the treatments to be more similar in efficacy to face‐to‐face CBT than those interventions without therapist support.

In addition, several meta‐analyses have investigated the effects of computer‐based psychotherapy for mental health problems more broadly, the results of which are also in accordance with the present findings. In a meta‐analysis of computer‐aided psychotherapy (including treatment delivered via stand‐alone or Internet‐linked computers, smart phones, palm pilots, interactive voice response, and CDs or DVDs) for anxiety disorders, Cuijpers 2009 found that computer‐aided psychotherapy was more effective than control conditions (d = 1.08) in reducing anxiety symptoms, and computer‐aided psychotherapy outcomes did not differ significantly from those outcomes achieved through face‐to‐face treatment. Similarly, Reger 2009 found medium to large effects sizes when comparing computer‐based CBT and ICBT to waiting list, placebo, or treatment as usual comparators in treating anxiety. When they investigated the effects of therapist involvement on their findings, no significant differences were identified based on amount of therapist contact. Most recently, Andrews 2010 investigated the effects of computer therapy for anxiety and depression (including both computer‐ and Internet‐aided treatments) as compared to control conditions and face‐to‐face treatment. They found computer‐based therapy to be superior to control for the treatment of social phobia (g = 0.92), PD (g = 0.83), and GAD (g = 1.12). They also found a non‐significant difference in outcome between computer‐based and face‐to‐face CBT.

It is important to note that these latter meta‐analyses looked more broadly at methods of administering treatment via computer technology, including but not limited to the Internet. Moreover, they included research into non‐therapist supported interventions, interventions administered using interactive voice response, as well as those that included substantial face‐to‐face contact. Nevertheless, despite the differences between these meta‐analyses and our own, the overall body of research serves to add further evidence for the efficacy of therapist‐supported ICBT in treating anxiety disorders.

Figure 1

PRISMA diagram of the search process.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 4

Forest plot: therapist‐supported ICBT versus waiting list control for anxiety symptom severity at post‐treatment.

Figure 5

Forest plot: therapist‐supported ICBT versus face‐to‐face CBT for anxiety symptom severity at post‐treatment.

Figure 6

Funnel plot of the outcome clinically important improvement in anxiety at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 7

Funnel plot the outcome anxiety symptom severity at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 8

Funnel plot of the outcome general anxiety symptom severity at post‐treatment for therapist‐supported ICBT versus waiting list control.

Figure 9

Funnel plot of the outcome quality of life at post‐treatment for therapist‐supported ICBT versus waiting list control.

Analysis 1.1

Comparison 1 Therapist‐supported ICBT versus waiting list control, Outcome 1 Clinically Important Improvement in Anxiety at Post‐Treatment.

Analysis 1.2

Comparison 1 Therapist‐supported ICBT versus waiting list control, Outcome 2 Anxiety Symptom Severity at Post‐Treatment.

Analysis 1.3

Comparison 1 Therapist‐supported ICBT versus waiting list control, Outcome 3 General Anxiety Symptom Severity at Post‐Treatment.

Analysis 1.4

Comparison 1 Therapist‐supported ICBT versus waiting list control, Outcome 4 Quality of Life at Post‐Treatment.

Analysis 2.1

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 1 Clinically Important Improvement in Anxiety at Post‐Treatment.

Analysis 2.2

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 2 Anxiety Symptom Severity at Post‐Treatment.

Analysis 2.3

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 3 Anxiety Symptom Severity at Follow‐up.

Analysis 2.4

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 4 General Anxiety Symptom Severity at Post‐Treatment.

Analysis 2.5

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 5 General Anxiety Symptom Severity at Follow‐up.

Analysis 2.6

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 6 Quality of Life at Post‐Treatment.

Analysis 2.7

Comparison 2 Therapist‐supported ICBT versus unguided CBT, Outcome 7 Quality of Life at Follow‐up.

Analysis 3.1

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 1 Clinically Important Improvement in Anxiety at Post‐Treatment.

Analysis 3.2

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 2 Clinically Important Improvement in Anxiety at Follow‐up.

Analysis 3.3

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 3 Anxiety Symptom Severity at Post‐Treatment.

Analysis 3.4

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 4 Anxiety Symptom Severity at Follow‐Up.

Analysis 3.5

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 5 General Anxiety Symptom Severity at Post‐Treatment.

Analysis 3.6

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 6 General Anxiety Symptom Severity at Follow‐up.

Analysis 3.7

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 7 Quality of Life at Post‐Treatment.

Analysis 3.8

Comparison 3 Therapist‐supported ICBT versus face‐to‐face CBT, Outcome 8 Quality of Life at Follow‐up.

Summary of findings for the main comparison. Therapist‐supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults

Therapist‐Supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults
Patient or population: patients with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: waiting list, attention, information, or online discussion group only control
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Waiting list, attention, information, or online discussion group only control	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	Study population		RR 3.75 (2.51 to 5.60)	866 (12 studies)	⊕⊕⊝⊝ low²
	14 per 100	53 per 100 (35 to 79)
	Moderate
	10 per 100	39 per 100 (26 to 58)
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean anxiety symptom severity at post‐treatment in the intervention groups was 1.06 standard deviations lower (1.29 to 0.82 lower)		2147 (28 studies)	⊕⊕⊝⊝ low^3,4	A standard deviation of 0.80 or greater represents a large difference between groups⁵
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.75 standard deviations lower (0.98 to 0.52 lower)		1496 (19 studies)	⊕⊕⊝⊝ low^4,6	A standard deviation of 0.80 or greater represents a large difference between groups⁵
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.47 standard deviations higher (0.38 to 0.57 higher)		1639 (23 studies)	⊕⊕⊕⊝ moderate⁶	A standard deviation of 0.50 represents a moderate difference between groups⁵
Adverse events at post‐treatment not reported	Study population		Not estimable	0 (0)	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
	See comment	See comment
	Moderate

Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	Study population		Not estimable	0 (13)	See comment	Studies reported high overall treatment satisfaction for therapist‐supported ICBT
	See comment	See comment
	Moderate

The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement. ² Downgraded for risk of bias (‐1) primarily because four of the included studies did not blind their outcome assessors to participants' group assignment and due to lack of blinding of participants and study therapists. Downgraded for publication bias (‐1) because only 12 studies reported this outcome. Not downgraded for inconsistency (0) because heterogeneity was reduced following subgroup analysis by anxiety disorder. ³ Downgraded for risk of bias (‐1) primarily due to minor concerns with selective outcome reporting, incomplete outcome data, baseline imbalances in a few studies, and lack of blinding of participants and study therapists. ⁴ Downgraded for inconsistency (‐1) because the heterogeneity amongst the included studies was quite high. This may be explained by the variety of anxiety disorders investigated and differences in the treatment details; however, the number of studies that could be included in subgroup analyses was not sufficient to provide useful reasons for this heterogeneity. ⁵ According to Cohen's (1969) interpretation of effect sizes. ⁶ Downgraded for risk of bias (‐1) primarily because two studies included baseline imbalances in participant severity across study groups and due to lack of blinding of participants and study therapists.

Summary of findings for the main comparison. Therapist‐supported ICBT compared to waiting list, attention, information, or online discussion group only control for anxiety disorders in adults

Summary of findings 2. Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults

Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults
Patient or population: patients with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: unguided ICBT
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Unguided ICBT	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	See comment	See comment	Not estimable	54 (1 study)	⊕⊝⊝⊝ very low^2,3	Not pooled because only one study in this comparison for this outcome
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean disorder‐specific anxiety symptom severity at post‐treatment in the intervention groups was 0.22 standard deviations lower (0.56 lower to 0.13 higher)		312 (5 studies)	⊕⊝⊝⊝ very low^4,5,6	A standard deviation of 0.20 represents a small difference between groups⁷
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.28 higher (2.21 lower to 2.78 higher)		138 (2 studies)	⊕⊝⊝⊝ very low^3,4
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.07 standard deviations higher (0.37 lower to 0.5 higher)		199 (3 studies)	⊕⊝⊝⊝ very low^4,5,6	A standard deviation of 0.10 represents a small difference between groups⁷
Adverse events at post‐treatment not reported	Study population		Not estimable	0 (0)	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
	See comment	See comment
	Moderate

Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	See comment	See comment	Not estimable	0 (2 studies)	See comment	Studies generally reported higher satisfaction with therapist‐supported ICBT
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement. ² Downgraded for risk of bias (‐1) primarily because of lack of blinding of outcome assessors, participants, and study therapists. ³ Downgraded for imprecision (‐2) as there is only one or two studies within the comparison for this outcome. ⁴ Downgraded for risk of bias (‐1) due to lack of blinding of participants and study therapists. ⁵ Downgraded for inconsistency (‐1) as the heterogeneity amongst the included studies was quite high. ⁶ Downgraded for imprecision (‐1) as there is a limited number of studies included in the comparison for this outcome. ⁷ According to Cohen's (1969) interpretation of effect sizes.

Summary of findings 2. Therapist‐supported ICBT compared to unguided CBT for anxiety disorders in adults

Summary of findings 3. Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults

Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults
Patient or population: adults with anxiety disorders Settings: outpatient care via Internet with e‐mail or telephone support, or both Intervention: therapist‐supported ICBT Comparison: face‐to‐face CBT
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Face‐to‐face CBT	Therapist‐supported ICBT
Clinically important improvement in anxiety at post‐treatment Indexed by a standardized interview or clinically accepted measure cut‐off score¹	Study population		RR 1.09 (0.89 to 1.34)	365 (4 studies)	⊕⊕⊝⊝ low^2,3
	41 per 100	44 per 100 (36 to 54)
	Moderate
	45 per 100	49 per 100 (40 to 61)
Disorder‐specific anxiety symptom severity at post‐treatment Indexed by a range of disorder‐specific self‐report measures		The mean anxiety symptom severity at post‐treatment in the intervention groups was 0.06 standard deviations higher (0.25 lower to 0.37 higher)		450 (7 studies)	⊕⊕⊝⊝ low^4,5	There was no significant difference between groups
General anxiety symptom severity at post‐treatment Indexed by a range of measures of anxiety symptoms in general		The mean general anxiety symptom severity at post‐treatment in the intervention groups was 0.06 standard deviations higher (0.42 lower to 0.55 higher)		343 (6 studies)	⊕⊕⊝⊝ low^4,5	There was no significant difference between groups
Quality of life at post‐treatment Indexed by self‐report measures of quality of life or functional disability		The mean quality of life at post‐treatment in the intervention groups was 0.26 standard deviations higher (0.06 to 0.45 higher)		392 (5 studies)	⊕⊕⊝⊝ low^2,3	A standard deviation of 0.20 represents a small difference between groups⁶
Adverse events at post‐treatment ‐ not reported	See comment	See comment	Not estimable	‐	See comment	Because adverse events were so rarely reported, they could not be meaningfully reported by comparison and are instead described in the review text
Participant satisfaction Indexed by a mix of qualitative and quantitative self‐report measures	Study population		Not estimable	0 (2)	See comment	Studies reported high overall treatment satisfaction across both conditions
	See comment	See comment
	Moderate

The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ For clinically important improvement in anxiety, an event is indicative of a participant achieving clinically important improvement at post‐treatment. ² Downgraded for imprecision (‐1) primarily due to small sample size. ³ Downgraded for risk of bias (‐1) due to lack of blinding of participants and study therapists. ⁴ Downgraded for risk of bias (‐1) primarily because two included studies provided incomplete outcome data (though sensitivity analyses suggest no difference in findings when these studies are excluded) and due to lack of blinding of participants and study therapists. ⁵ Downgraded for inconsistency (‐1) primarily due to unexplained heterogeneity. ⁶ According to Cohen's (1969) interpretation of effect sizes.

Summary of findings 3. Therapist‐supported ICBT compared to face‐to‐face CBT for anxiety disorders in adults

Table 1. Summary of included studies table

Study	Diagnosis and Co‐morbidity	Participant Characteristics *(M age, age range, sex, country of residence)*	Co‐Use of Medication	N	Intervention Type & Therapist Duration Contact		Comparison	Assessment Points	Outcomes
Andersson et al (2009)	Specific Phobia, Spider Type co‐morbidity not reported	M age=25.6 (4.1) 18‐65 years 84.8% women Sweden	Not reported	27	IBT with email: 4 wks; 5 online modules	M total time spent per participant = 25 min	Orientation and 1 3‐hour live exposure session	post‐treatment 12‐month follow‐up	specific phobia sx; general anxiety sx
Andersson et al (2012a)	Social Phobia co‐morbidity included but not reported	ICBT M age=38.1 (11.3) WLC M age=38.4 (10.9) 19‐71 years 61% women Sweden	13.7% using medication	204	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	Online Discussion Group	post‐treatment	diagnostic status; social phobia sx; QOL; general anxiety sx;
Andersson et al (2012b)	GAD 22.2% Social Phobia, 19.8% PD, 3.7% OCD, 23.5% MDD	ICBT M age=44.4 (12.8) IPDTM age=36.4 (9.7) WLC M age=39.6 (13.7) 19‐66 years 76.5% women Sweden	32.1% using medication	81	ICBT with email: 8 wks; 8 online modules	M total time spent per participant = 92 min (SD=61)	(1) Waiting List Control (2) IPDT: 8 wks; 8 online modules	post‐treatment	diagnostic status, GAD sx; general anxiety sx; QOL
Andersson et al. (2013)	Specific Phobia, Snake Type comorbidity not reported	M age=27.2 (8.1) 19‐54 years 84.6% women Sweden	Not reported	30	IBT with email: 4 wks; 4 online modules	M total time spent per participant = 25 min	Orientation and 1 3‐hour live exposure session	post‐treatment 12‐month follow‐up	specific phobia sx; general anxiety sx
Berger et al (2009)	Social Phobia 26.9% co‐morbid Axis I disorder	M age=28.9 (5.3) 19‐43 years 44.2% women Switzerland, France, Belgium	Excluded	52	ICBT with email: 10 wks; 5 online modules	M=5.5 emails from participant weekly emails from therapist	Waiting List Control	post‐treatment	social phobia sx; treatment satisfaction
Berger et al (2011)	Social Phobia 38% co‐morbid Axis I disorder; 12% PD, 10% Specific Phobia, 2% GAD, 22% MDD/Dysthymia, 2% ED	M age=37.2 (11.2) 19‐62 years 53.1% women Switzerland	7.4% using medication	81	ICBT with email: 10 wks; 5 online modules	M=6.16 (SD=4.56; range=1‐17) emails from participant M=12.44 (SD=2.85; range=6‐17) emails from therapist	(1) Unguided ICBT 10 weeks; 5 online modules (2) Step‐up on demand ICBT	post‐treatment 6‐month follow‐up	diagnostic status; social phobia sx; treatment satisfaction
Berger et al. (2014)	33.3% PD with or without Agoraphobia 85.6% Social Phobia 25% GAD 37.1% PD with or without Agoraphobia, Social Phobia, or GAD, 13.6% MDD, 15.9% Specific Phobia, 5.3% OCD, 12.1% other Axis I disorder	M age=35.1 (11.4) 18‐65 years 56.1% women Switzerland, Germany, Austria	14.4% using medication	132	ICBT with email: 8 wks; 8 online modules	M=6.53 (SD=7.2; range=0‐36) emails from participant M=12.6 (SD=4.6; range=8‐35) emails from therapist	(1) Waiting List Control (2) Tailored ICBT: 8 wks; 8 online modules	post‐treatment 6‐month follow‐up	diagnostic status; anxiety sx; general anxiety sx; treatment satisfaction
Bergstrom et al (2010)	15.4% PD 84.6% PD with Agoraphobia co‐morbidity not reported	ICBT M age=33.8 (9.7) GCBT M age=34.6 (9.2) 18 years or older 61.5% women Sweden	45% using medication; 34% SSRI/SNRIs, 13% BZ, 24% BZ derivatives or neuroleptics; 5% TCAs	104	ICBT with email: 10 wks; 10 online modules	M=11.3 (SD=4.3) emails from therapist M total time spent per participant = 35.4 min (SD=19)	10 weekly 2‐hour sessions of GCBT	post‐treatment 6 month follow‐up	diagnostic status; PD sx; QOL
Carlbring et al (2001)	PD co‐morbidity included but not reported	M age=34 (7.5) 21‐51 years 71% women Sweden	64% using medication; 44% SSRIs, 10% BZ, 5% TCAs, 5% beta‐blockers	41	ICBT with email: 7‐12 wks; 6 online modules	M reciprocal emails = 7.5 (SD=1.2; range=6‐15) M total time spent per participant = 90 min	Waiting List Control	post‐treatment	diagnostic status; PD sx; QOL; general anxiety sx; treatment satisfaction
Carlbring et al (2005)	49% PD 51% PD with Agoraphobia 49% Anxiety Disorder; 6% MDD	M age=35.0 (7.7) 18‐60 years old 71% women Sweden	30.6% SSRIs, 8.2% BZ, 6.1% TCAs, 6.1% beta blockers	49	ICBT with email: 10 wks; 10 online modules	M reciprocal emails =15.4 (SD=5.5; range=4‐31) M total time spent per participant =150 min	10 weekly 45‐60 min sessions of individual CBT	post‐treatment 12‐month follow‐up	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL
Carlbring et al (2006)	PD co‐morbidity included but not reported	M age=36.7 (10) 18‐60 years 60% women Sweden	54% using medication	60	ICBT with email & phone: 10 wks; 10 online modules	M reciprocal contacts = 13.5 (SD =4.4; range=7‐29) M time spent per participant per week = 12 min M length phone call = 11.8 min (range= 9.6‐15.6)	Waiting List Control	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL; treatment satisfaction
Carlbring et al (2007)	Social Phobia co‐morbidity included but not reported	ICBT M age=32.4 (9.1) WLC M age=32.9 (9.2) 18‐60 years 64.9% women Sweden	Included but not reported	60	ICBT with email & phone: 9 wks; 9 online modules	M time spent per participant per week = 22 min M length phone call = 10.5 min (SD= 3.6)	Waiting List Control	post‐treatment	social phobia sx; general anxiety sx; QOL
Carlbring et al (2011)	9% PD 22% PD with Agoraphobia 39% Social Phobia 20% GAD 13% ADNOS 2% OCD, 2% PTSD, 20% MDD, 7% mild depression; 15% Dysthymia	M age=38.8 (10.7) 22‐63 years 76% women Sweden	26% using an antidepressant or anxiolytic	54	ICBT with email: 10 wks; 6‐10 online modules	M time spent per participant per week = 15 min	Attention Control 10 wks of posts in an online support forum	post‐treatment	diagnostic status; anxiety sx (broadly); QOL; general anxiety sx
Furmark et al (2009a)	Social Phobia co‐morbidity not reported	ICBT M age=35 (10.2) WLC M age=35.7 (10.9) Bib M age=37.7 (10.3) 18 years or older 67.5% women Sweden	13.9% using medication	120	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	(1) Biblio‐therapy: 9 wks; 9 lessons (2) Waiting List Control	post‐treatment	social phobia sx; general anxiety sx; QOL
Furmark et al (2009b)	Social Phobia co‐morbidity not reported	ICBT M age=34.9 (8.4) Bib M age=32.5 (8.5) Applied Relaxation M age=36.4 (9.8) 18 years or older 67.8% women Sweden	6.7% using medication	115	ICBT with email: 9 wks; 9 online modules	M time spent per participant per week = 15 min	(1) Biblio‐therapy: 9 wks; 9 lessons (2) Biblio‐therapy and discussion group: 9 wks; 9 lessons (2) Internet‐based applied relaxation: 9 wks; 9 online modules	post‐treatment	social phobia sx; general anxiety sx; QOL
Greist et al. (2012)	OCD 32% Anxiety Disorder, 31% Mood Disorder, 7% Substance Disorder, 7% ADHD, 2% ED	M age=38.34 (13.93) 18 years or older 63% women USA	19.5% using medication	87	ICBT with phone: 12 wks; 9 online modules	Weekly phone calls	(1) Unguided ICBT: 12 wks; 9 online modules (2) Lay Coaching ICBT: 12 wks; 9 online modules	post‐treatment	OCD sx; QOL; treatment satisfaction
Hedman et al (2011)	Social Phobia 47.5% Anxiety Disorder, 15.1% MDD	ICBT M age=35.2 (11.1) GCBT M age=35.5 (11.6) 18‐64 years 35.7% women Sweden	19.8% SSRIs, 4.8% SNRIs	126	ICBT with email: 15 wks; 15 online modules	M=17.4 emails per participant M time spent per participant per week = 5.5 min (SD=3.6)	15 weekly 2.5‐hour sessions of GCBT	post‐treatment 6 month follow‐up	diagnostic status; social phobia sx; QOL; general anxiety sx
Ivarsson et al. (2014)	PTSD comorbidity not reported	M age=46 (11.7) 21‐67 years 82.3% women Sweden	Included but not reported	62	ICBT with email: 8 wks; 8 online modules	M time spent per participant per week = 28 min (SD=19.8)	Attention Control: sent weekly question on wellbeing, stress, sleep	post‐treatment	diagnostic status; PTSD sx; general anxiety sx; QOL
Johnston et al (2011)	20.6% PD with or without Agoraphobia 34.4% Social Phobia 45% GAD 29% Anxiety Disorder, 9.2% Affective Disorder, 32.1% both disorders	M age=41.62 (12.83) 19‐79 years 58.8% women Australia	29% using medication	139	ICBT with email & phone: 10 wks; 8 online modules	M=8.83 (SD=3.29) emails per participant M=7.54 (SD=2.43) phone calls per participant M total time spent per participant = 69.09 min (SD=32.29)	Waiting List Control	post‐treatment	disorder‐specific sx; general anxiety sx; QOL; treatment satisfaction
Kiropoulos et al (2008)	41.9% PD 58.1% PD with Agoraphobia 72.1% co‐morbid Mood, Anxiety, Somatoform, or Substance Disorder	M age=38.96 (11.13) 20‐64 years 72.1% women Australia	47.7% using medication	86	ICBT with email: 6 wks, 6 required & 2 optional online modules	M=18.24 (SD=9.82) emails from therapist M=10.64 (SD=8.21) emails from participant M total time spent per participant = 352 min (SD=240)	12 weekly 1‐hour sessions of individual CBT	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL; treatment satisfaction
Kok et al. (2012)	41% PD with Agoraphobia 17% Agoraphobia 53.3% Social Phobia 83.5% Specific Phobia Comorbidity included but not reported	M age=34.6 (11.7) 18 years or older 61% women Netherlands	43% using medication	212	IBT with email: 5wks, 8 online modules	weekly contact by therapist	Waiting List Control: also sent self‐help book without instructions	post‐treatment	phobia sx; general anxiety sx; treatment satisfaction
Newby et al. (2013)	84% GAD Comorbidity included but not fully reported; 56% MDD	M age=44.3 (12.2) 21‐80 years 77.8% women Australia	40.4% using medication	100	ICBT with email and phone: 10 wks; 6 online modules	M total time spent per participant = 23.37 min (SD=12.15)	Waiting List Control	post‐treatment	GAD sx; QOL; treatment satisfaction
Nordgren et al. (2012)	31% PD with or without Agoraphobia 8% Agoraphobia 32% Social Phobia 10% GAD 19% ADNOS 21% Anxiety Disorder, 43% Mood Disorder, 1% Hypochon‐driasis	ICBT M age=35 (13) WLC M age=36 (12) 19‐68 years 63% women Sweden	26% using medication	100	ICBT with email: 10 wks; 7‐10 online modules	M time spent per participant = 15 min/week	Attention Control: sent weekly questions on wellbeing	post‐treatment	anxiety sx; general anxiety sx; QOL
Paxling et al (2012)	GAD co‐morbidity included but not fully reported; 22.5% MDD	M age=39.3 (10.8) 18‐66 years 79.8% women Sweden	37.1% using medication	89	ICBT with email: 8 wks; 8 online modules	M total time spent per participant = 97 min (SD=52)	Waiting List Control	post‐treatment	GAD sx; general anxiety sx; QOL
Richards et al (2006)	21.9% PD 78.1% PD with Agoraphobia 22% Social Phobia, 13% GAD, 9% Specific Phobia, 6% PTSD, 9% MDD, 6% Hypochondriasis, 3% Somatization	M age=36.59 (9.9) 18‐70 years 68.8% women Australia	15.6% anti‐depressants, 12.5% BZ, 9.4% antidepressants and BZ	23	ICBT with email: 8 wks, 6 online modules	M=18 (SD=6.5) emails from therapist M=15.3 (SD=12.8) emails from participant M total time spent per participant =376.3 min (SD=156.8)	Information Only Control Weekly status updates to clinician and access to online non‐CBT info	post‐treatment	diagnostic status; PD and agoraphobia sx; general anxiety sx; QOL
Robinson et al (2010)	GAD co‐morbidity included but not reported	M age=46.96 (12.7) 18‐80 years 68.3% women Australia	Included but not reported	101	ICBT with email and phone: 10 wks; 6 online modules	M =33.2 (SD=4) emails/calls per participant M total time spent per participant = 80.8 min (SD=22.6)	Waiting List Control	post‐treatment	diagnostic status; GAD sx; QOL; treatment satisfaction
Silfvernagel et al (2012)	7% PD 83% PD with Agoraphobia 16% Social Phobia 19% GAD 2% ADNOS 32% co‐morbid disorder	M age=32.4 (6.9) 20‐45 years 65% women Sweden	47% using medication	57	ICBT with email: 8 wks; 6‐8 online modules	M time spent per participant = 15 min/week	Waiting List Control	post‐treatment	diagnostic status; PD sx; general anxiety sx; QOL
Spence et al (2011)	PTSD 62% MDD, 33% Social Phobia, 31% PD with or without Agoraphobia, 26% GAD; 17% OCD	M age=42.6 (13.1) 21‐68 years 81% women Australia	60% using medication	44	ICBT with email & phone: 8 wks; 7 online modules	M=5.39 (SD=3.54) emails per participant M=7.87 (SD=2.56) phone calls per participant M total time spent per participant = 103.91 min (SD=96.53)	Waiting List Control	post‐treatment	diagnostic remission; PTSD sx; QOL; general anxiety sx; treatment satisfaction
Tillfors et al (2008)	Social Phobia co‐morbidity included but not reported	ICBT M age=32.3 (9.7) ICBT+exposure M age= 30.4 (6.3) 19‐53 years 78.9% women Sweden	Included but not reported	38	ICBT with email: 9 wks; 9 online modules	M=35 min per participant per week	ICBT with email (9 online modules) + 5 live 2.25‐hour exposure sessions; 9 wks	post‐treatment 12‐month follow‐up	social phobia sx; general anxiety sx; QOL; treatment satisfaction
Titov et al (2008a)	Social Phobia co‐morbidity included but not reported	M age=38.13 (12.24) 18‐72 years 59% women Australia	29% using medication	105	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 125 min (SD=25)	Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2008b)	Social Phobia co‐morbidity included but not reported	M age=36.79 (10.93) 20‐61 years 62.96% women Australia	25.9% using medication	88	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 126.76 min (SD=30.89)	Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2008c)	Social Phobia co‐morbidity included but not reported	M age=37.97 (11.29) 18‐64 years 61.05% women Australia	25.9% using medication	98	ICBT with email: 10 wks; 6 online modules	M total time spent per participant = 168 min (SD=40)	(1) Unguided ICBT 10 wks; 6 online modules (2) Waiting List Control	post‐treatment	social phobia sx; QOL; treatment satisfaction
Titov et al (2009)	GAD co‐morbidity included but not reported	M age=44 (12.98) 18 years or older 76% women Australia	29% using medication	48	ICBT with email & phone: 9 wks, 6 online modules	M =23.7 emails, 5.5 instant messages, and 4.1 calls per participant M total time spent per participant = 130 min	Waiting List Control	post‐treatment	diagnostic status; GAD sx; QOL; treatment satisfaction
Titov et al (2010)	26.9% PD with Agoraphobia 29.5% Social Phobia 43.6% GAD 28.2% Anxiety Disorder, 20.5% Affective Disorder, 26.9% both disorders	M age=39.5 (13) 18 years or older 67.9% women Australia	47.4% using medication	86	ICBT with email & phone: 8 wks; 6 online modules	M =23.6emails from therapist M total time spent per participant = 46 min (SD=16)	Waiting List Control	post‐treatment	diagnostic status; disorder‐specific anxiety sx; QOL; treatment satisfaction
Titov et al (2011)	10% PD with or without Agoraphobia 11% Social Phobia 28% GAD 51% MDD 81% had a co‐morbidity	M age=43.9 (14.6) 18‐79 years 73% women Australia	54% using medication	74	ICBT with email & phone: 10 wks; 8 online modules	M=5.45 (SD=3.57) emails per participant M=9.35 (SD=2.96) phone calls per participant M total time spent per participant = 84.76 min (SD=50.37)	Waiting List Control	post‐treatment	disorder‐specific sx; general anxiety sx; QOL; treatment satisfaction
Van Ballegooijen et al (2013)	78% PD with or without Agoraphobia 14% Agoraphobia co‐morbidity included but not reported	M age=36.6 (11.4) 18‐67 years 67.5% women Netherlands	Included but not reported	126	ICBT with email: 12 wks; 6 online modules	M total time spent per participant = 1 to 2 hours	Waiting List Control	post‐treatment	PD sx; general anxiety sx
Wims et al (2010)	PD with or without Agoraphobia 21% Social Phobia, 31% GAD, 10% OCD, 7% PTSD, 21% MDD	M age=42.08 (12.29) 18 years or older 76% women Australia	31% using medication	59	ICBT with email: 8 wks; 6 online modules	M =7.5 emails from therapist M total time spent per participant = 75 min	Waiting List Control	post‐treatment	diagnostic status; PD & agoraphobia sx; QOL
Wootton et al. (2013)	OCD 26.9% Social Phobia, 40.4% GAD, 15.4% PD, 11.5% PTSD, 38.5% MDD	ICBT M age=39.93 (12.57) Bib M age=35.55 (9.68) WLC M age=38.58 (10.51) 18‐64 years 75% women Australia	61.5% using SSRIs	56	ICBT with phone: 8 wks, 5 online modules	M=15.05 (SD=3.93) phone calls per participant M total time spent per participant = 88.63 min (SD=46.41)	1) Guided bibliotherapy: 8 wks, 5 lessons 2) Waiting List Control	post‐treatment	diagnostic status; OCD sx; general anxiety sx; treatment satisfaction
Notes: All data in the above table represent only that included in/relevant to the present review. ADNOS = anxiety disorder, not otherwise specified; Bib = Bibliotherapy; BZ = benzodiazepine; ED = eating disorder; GAD = generalized anxiety disorder; GCBT = group cognitive behavioural therapy; IBT = internet‐based behavioural therapy; ICBT = internet‐based cognitive behavioural therapy; IPDT = internet‐based psychodynamic therapy; MDD = major depressive disorder; PD = panic disorder; QOL = quality of life; SNRI = serotonin‐norepinephrine re‐uptake inhibitor; SSRI = selective serotonin re‐uptake inhibitor; sx = symptoms; TCA = tricyclic antidepressant; VCBT = videoconferencing cognitive‐behavioural therapy; WLC = waiting list control.

Table 1. Summary of included studies table

Table 2. Subgroup analyses. Comparison 1: therapist‐supported ICBT versus waiting list control

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	2	42	39	RR, M‐H, Random	18.32 [2.50, 134.18]	3
		ii) Social Phobia	1	102	102	RR, M‐H, Random	6.00 [2.64, 13.62]	‐‐
		iii) GAD	3	91	94	RR, M‐H, Random	2.58 [1.48, 4.51]	36
		iv) PTSD	2	54	50	RR, M‐H, Random	3.48 [1.78, 6.80]	0
		v) OCD	1	15	14	RR, M‐H, Random	14.06 [0.88, 225.47]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	3	129	134	RR, M‐H, Random	3.76 [1.68, 8.43]	53
	b. By Therapist Contact
		i) High	1	12	9	RR, M‐H, Random	6.92 [0.42, 114.19]	‐‐
		ii) Medium	6	155	153	RR, M‐H, Random	4.34 [2.43, 7.76]	31
		iii) Low	4	204	202	RR, M‐H, Random	4.13 [1.73, 9.82]	70
	c. By Research Group
		i) Sweden	5	210	215	RR, M‐H, Random	5.12 [2.63, 9.98]	38
		ii) Australia‐Klein	1	47	48	RR, M‐H, Random	2.10 [1.44, 3.04]	‐‐
		iii) Australia‐Titov	5	149	140	RR, M‐H, Random	3.33 [1.88, 5.91]	44
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	62	69	RR, M‐H, Random	2.30 [1.41, 3.74]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	6	165	158	SMD, Random	‐1.52 [‐2.56, ‐0.48]	93
		ii) Social Phobia	8	336	325	SMD, Random	‐1.38 [‐1.63, ‐1.13]	48
		iii) GAD	6	194	200	SMD, Random	‐0.80 [‐1.19, ‐0.42]	69
		iv) PTSD	2	54	50	SMD, Random	‐0.78 [‐1.38, ‐0.17]	55
		v) OCD	1	15	17	SMD, Random	‐0.63 [‐1.35, 0.08]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	7	319	314	SMD, Random	‐0.81, [‐1.20, ‐0.41]	81
	b. By Therapist Contact
		i) High	1	12	9	SMD, Random	‐0.80 [‐1.70, 0.11]	‐‐
		ii) Medium	14	472	468	SMD, Random	‐1.34 [‐1.69, ‐0.99]	82
		iii) Low	10	420	416	SMD, Random	‐0.83 [‐1.15, ‐0.50]	78
	c. By Research Group
		i) Sweden	12	464	471	SMD, Random	‐1.50 [‐1.88, ‐1.11]	85
		ii) Australia‐Klein	1	12	9	SMD, Random	‐0.80 [‐1.70, 0.11]	‐‐
		iii) Australia‐Titov	12	365	350	SMD, Random	‐0.84 [‐1.06, ‐0.61]	50
		iv) Netherlands‐Kok	1	104	106	SMD, Random	‐0.16 [‐0.44, 0.11]	‐‐
		v) Netherlands‐van Ballegooijen	1	63	63	SMD, Random	‐0.30 [‐0.66, 0.05]	‐‐
		vi) Switzerland	2	75	65	SMD, Random	‐1.04 [‐1.63, ‐0.46]	59
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	4	126	122	SMD, Random	‐0.74 [‐1.35, ‐0.13]	78
		ii) Social Phobia	3	171	170	SMD, Random	‐0.64 [‐0.85, ‐0.42]	0
		iii) GAD	2	67	71	SMD, Random	‐1.91 [‐3.57, ‐0.26]	94
		iv) PTSD	2	54	50	SMD, Random	‐0.63 [‐1.02, ‐0.23]	0
		v) OCD	1	15	17	SMD, Random	‐1.05 [‐1.79, ‐0.30]	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	7	319	314	SMD, Random	‐0.49 [‐0.69, ‐0.29]	31
	b. By Therapist Contact
		i) High	1	12	9	SMD, Random	0.07 [‐0.79, 0.94]	‐‐
		ii) Medium	10	326	324	SMD, Random	‐0.92 [‐1.32, ‐0.52]	83
		iii) Low	6	266	261	SMD, Random	‐0.61 [‐0.82, ‐0.39]	25
	c. By Research Group
		i) Sweden	11	426	427	SMD, Random	‐0.94 [‐1.29, ‐0.59]	82
		ii) Australia‐Klein	1	12	9	SMD, Random	0.07 [‐0.79, 0.94]	‐‐
		iii) Australia‐Titov	4	103	95	SMD, Random	‐0.59 [‐0.88, ‐0.30]	0
		iv) Netherlands‐Kok	1	104	106	SMD, Random	‐0.21 [‐0.48, 0.07]	‐‐
		v) Netherlands‐van Ballegooijen	1	63	63	SMD, Random	‐0.39 [‐0.74, ‐0.04]	‐‐
		vi) Switzerland	1	44	44	SMD, Random	‐0.91 [‐1.23, ‐0.59]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Table 2. Subgroup analyses. Comparison 1: therapist‐supported ICBT versus waiting list control

Table 3. Subgroup analyses. Comparison 2: therapist‐supported ICBT versus unguided CBT

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Low	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
	c. By Research Group
		i) Sweden	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	RR, M‐H, Random	1.07 [0.67, 1.69]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	4	127	126	SMD, Random	‐0.24 [‐0.69, 0.21]	68
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	1	31	28	SMD, Random	‐0.14 [‐0.65, 0.37]	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	3	100	99	SMD, Random	‐0.18 [‐0.78, 0.41]	77
		iii) Low	1	27	27	SMD, Random	‐0.43 [‐0.97, 0.11]	‐‐
	c. By Research Group
		i) Sweden	2	69	69	SMD, Random	0.12 [‐0.22, 0.45]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	1	31	30	SMD, Random	‐0.82 [‐1.34, ‐0.29]	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	SMD, Random	‐0.43 [‐0.97, 0.11]	‐‐
		vii) USA	1	31	28	SMD, Random	‐0.14 [‐0.65, 0.37]	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	MD, Random	0.28 [‐2.21, 2.78]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Clinically Important Improvement in Anxiety at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	3	96	96	SMD, Random	‐0.30 [‐0.58, ‐0.01]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact at Follow‐up
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	SMD, Random	‐0.31 [‐0.65, 0.03]	3
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	SMD, Random	‐0.31 [‐0.65, 0.03]	3
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	1	27	27	SMD, Random	‐0.26 [‐0.80, 0.27]	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Social Phobia	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		iii) Low	0	‐‐	‐‐	‐‐	‐‐	‐‐
	c. By Research Group
		i) Sweden	2	69	69	MD, Random	0.72 [‐2.12, 3.57]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Table 3. Subgroup analyses. Comparison 2: therapist‐supported ICBT versus unguided CBT

Table 4. Subgroup analyses. Comparison 3: therapist‐supported ICBT versus face‐to‐face CBT

Outcome and Subgroup			No. of Studies	No. of Participants ICBT Comparator		Statistical Method	Effect Size	I²
Clinically Important Improvement in Anxiety at Post‐Treatment
	a. By Disorder
		i) Panic	3	121	118	RR, M‐H, Random	1.06 [0.85, 1.32]	0
		ii) Social Phobia	1	64	62	RR, M‐H, Random	1.45 [0.77, 2.76]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) OCD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	46	40	RR, M‐H, Random	1.11 [0.57, 2.15]	‐‐
		ii) Medium	1	25	24	RR, M‐H, Random	1.20 [0.85, 1.69]	‐‐
		iii) Low	2	114	116	RR, M‐H, Random	1.08 [0.72, 1.60]	34
	c. By Research Group
		i) Sweden	3	139	140	RR, M‐H, Random	1.09 [0.88, 1.36]	0
		ii) Australia‐Klein	1	46	40	RR, M‐H, Random	1.11 [0.57, 2.15]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	3	119	115	SMD, Random	0.29 [0.03, 0.54]	0
		ii) Social Phobia	2	83	80	SMD, Random	‐0.18 [‐0.92, 0.57]	76
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	27	SMD, Random	‐0.02 [‐0.56, 0.52]	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	2	63	55	SMD, Random	0.42 [0.05, 0.78]	0
		ii) Medium	1	25	24	SMD, Random	0.05 [‐0.51, 0.61]	‐‐
		iii) Low	4	140	143	SMD, Random	‐0.10 [‐0.53, 0.33]	64
	c. By Research Group
		i) Sweden	6	184	185	SMD, Random	‐0.03 [‐0.34, 0.28]	49
		ii) Australia‐Klein	1	44	37	SMD, Random	0.49 [0.05, 0.93]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Post‐Treatment
	a. By Disorder
		i) Panic	2	67	62	SMD, Random	0.42 [‐0.75, 1.60]	90
		ii) Social Phobia	2	83	80	SMD, Random	‐0.18 [‐0.49, 0.13]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.14 [‐1.07, 0.79]	64
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	2	61	56	SMD, Random	0.49 [‐0.57, 1.56]	86
		ii) Medium	1	25	24	SMD, Random	‐0.19 [‐0.75, 0.38]	‐‐
		iii) Low	3	90	87	SMD, Random	‐0.18 [‐0.59, 0.22]	29
	c. By Research Group
		i) Sweden	5	134	129	SMD, Random	‐0.17 [‐0.42, 0.07]	0
		ii) Australia‐Klein	1	42	38	SMD, Random	1.01 [0.54, 1.48]	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Clinically Important Improvement in Anxiety at Follow‐up
	a. By Disorder
		i) Panic	2	75	78	RR, M‐H, Random	1.09 [0.93, 1.28]	0
		ii) Social Phobia	1	64	62	RR, M‐H, Random	1.15 [0.73, 1.83]	‐‐
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	0	‐‐	‐‐	‐‐	‐‐	‐‐
		ii) Medium	1	25	24	RR, M‐H, Random	1.05 [0.87, 1.27]	‐‐
		iii) Low	2	114	116	RR, M‐H, Random	1.17 [0.92, 1.50]	0
	c. By Research Group
		i) Sweden	3	139	140	RR, M‐H, Random	1.10 [0.94, 1.27]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	2	75	78	SMD, Random	‐0.04 [‐0.36, 0.28]	0
		ii) Social Phobia	2	83	80	SMD, Random	‐0.39 [‐0.71, ‐0.08]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.09 [‐0.64, 0.46]	0
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	19	18	SMD, Random	‐0.11 [‐0.76, 0.53]	‐‐
		ii) Medium	1	25	24	SMD, Random	‐0.04 [‐0.60, 0.52]	‐‐
		iii) Low	4	140	141	SMD, Random	‐0.24 [‐0.48, 0.00]	3
	c. By Research Group
		i) Sweden	6	184	183	SMD, Random	‐0.20 [‐0.41, 0.00]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐
General Anxiety Symptom Severity at Follow‐up
	a. By Disorder
		i) Panic	1	25	24	SMD, Random	‐0.17 [‐0.74, 0.39]	‐‐
		ii) Social Phobia	2	83	80	SMD, Random	‐0.14 [‐0.45, 0.17]	0
		iii) GAD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) PTSD	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Specific Phobia	2	26	25	SMD, Random	‐0.06 [‐0.61, 0.49]	0
		vi) Trans‐diagnostic	0	‐‐	‐‐	‐‐	‐‐	‐‐
	b. By Therapist Contact
		i) High	1	19	18	SMD, Random	‐0.19 [‐0.83, 0.46]	‐‐
		ii) Medium	1	25	24	SMD, Random	‐0.17 [‐0.74, 0.39]	‐‐
		iii) Low	3	90	87	SMD, Random	‐0.10 [‐0.40, 0.19]	0
	c. By Research Group
		i) Sweden	5	134	129	SMD, Random	‐0.13 [‐0.37, 0.11]	0
		ii) Australia‐Klein	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iii) Australia‐Titov	0	‐‐	‐‐	‐‐	‐‐	‐‐
		iv) Netherlands‐Kok	0	‐‐	‐‐	‐‐	‐‐	‐‐
		v) Netherlands‐van Ballegooijen	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vi) Switzerland	0	‐‐	‐‐	‐‐	‐‐	‐‐
		vii) USA	0	‐‐	‐‐	‐‐	‐‐	‐‐

Table 4. Subgroup analyses. Comparison 3: therapist‐supported ICBT versus face‐to‐face CBT

Comparison 1. Therapist‐supported ICBT versus waiting list control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Clinically Important Improvement in Anxiety at Post‐Treatment Show forest plot	12	866	Risk Ratio (M‐H, Random, 95% CI)	3.75 [2.51, 5.60]

1.1 Social Phobia	1	204	Risk Ratio (M‐H, Random, 95% CI)	6.0 [2.64, 13.62]
1.2 GAD	3	185	Risk Ratio (M‐H, Random, 95% CI)	2.58 [1.48, 4.51]
1.3 Panic Disorder	2	81	Risk Ratio (M‐H, Random, 95% CI)	18.32 [2.50, 134.18]
1.4 PTSD	2	104	Risk Ratio (M‐H, Random, 95% CI)	3.48 [1.78, 6.80]
1.5 OCD	1	29	Risk Ratio (M‐H, Random, 95% CI)	14.06 [0.88, 225.47]
1.6 Mixed Anxiety	3	263	Risk Ratio (M‐H, Random, 95% CI)	3.76 [1.68, 8.43]
2 Anxiety Symptom Severity at Post‐Treatment Show forest plot	30	2147	Std. Mean Difference (IV, Random, 95% CI)	‐1.06 [‐1.29, ‐0.82]

2.1 Social Phobia	8	661	Std. Mean Difference (IV, Random, 95% CI)	‐1.38 [‐1.63, ‐1.13]
2.2 GAD	6	394	Std. Mean Difference (IV, Random, 95% CI)	‐0.80 [‐1.19, ‐0.42]
2.3 Panic Disorder	6	323	Std. Mean Difference (IV, Random, 95% CI)	‐1.52 [‐2.56, ‐0.48]
2.4 PTSD	2	104	Std. Mean Difference (IV, Random, 95% CI)	‐0.78 [‐1.38, ‐0.17]
2.5 OCD	1	32	Std. Mean Difference (IV, Random, 95% CI)	‐0.63 [‐1.35, 0.08]
2.6 Mixed Anxiety	7	633	Std. Mean Difference (IV, Random, 95% CI)	‐0.81 [‐1.20, ‐0.41]
3 General Anxiety Symptom Severity at Post‐Treatment Show forest plot	19	1496	Std. Mean Difference (IV, Random, 95% CI)	‐0.75 [‐0.98, ‐0.52]

3.1 Social Phobia	3	341	Std. Mean Difference (IV, Random, 95% CI)	‐0.64 [‐0.85, ‐0.42]
3.2 GAD	2	138	Std. Mean Difference (IV, Random, 95% CI)	‐1.91 [‐3.57, ‐0.26]
3.3 Panic Disorder	4	248	Std. Mean Difference (IV, Random, 95% CI)	‐0.74 [‐1.35, ‐0.13]
3.4 PTSD	2	104	Std. Mean Difference (IV, Random, 95% CI)	‐0.63 [‐1.02, ‐0.23]
3.5 OCD	1	32	Std. Mean Difference (IV, Random, 95% CI)	‐1.05 [‐1.79, ‐0.30]
3.6 Mixed Anxiety	7	633	Std. Mean Difference (IV, Random, 95% CI)	‐0.49 [‐0.69, ‐0.29]
4 Quality of Life at Post‐Treatment Show forest plot	23	1639	Std. Mean Difference (IV, Random, 95% CI)	0.47 [0.38, 0.57]

4.1 Social Phobia	6	586	Std. Mean Difference (IV, Random, 95% CI)	0.44 [0.28, 0.61]
4.2 GAD	5	360	Std. Mean Difference (IV, Random, 95% CI)	0.57 [0.35, 0.78]
4.3 Panic Disorder	4	176	Std. Mean Difference (IV, Random, 95% CI)	0.45 [0.15, 0.75]
4.4 PTSD	2	104	Std. Mean Difference (IV, Random, 95% CI)	0.41 [0.02, 0.80]
4.5 OCD	0	0	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
4.6 Mixed Anxiety	6	413	Std. Mean Difference (IV, Random, 95% CI)	0.47 [0.27, 0.67]

Comparison 1. Therapist‐supported ICBT versus waiting list control

Comparison 2. Therapist‐supported ICBT versus unguided CBT

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Clinically Important Improvement in Anxiety at Post‐Treatment Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

1.1 Social Phobia	1		Risk Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]
2 Anxiety Symptom Severity at Post‐Treatment Show forest plot	5	312	Std. Mean Difference (IV, Random, 95% CI)	‐0.22 [‐0.56, 0.13]

2.1 Social Phobia	4	253	Std. Mean Difference (IV, Random, 95% CI)	‐0.24 [‐0.69, 0.21]
2.2 OCD	1	59	Std. Mean Difference (IV, Random, 95% CI)	‐0.14 [‐0.65, 0.37]
3 Anxiety Symptom Severity at Follow‐up Show forest plot	3	192	Std. Mean Difference (IV, Random, 95% CI)	‐0.30 [‐0.58, ‐0.01]

3.1 Social Phobia	3	192	Std. Mean Difference (IV, Random, 95% CI)	‐0.30 [‐0.58, ‐0.01]
4 General Anxiety Symptom Severity at Post‐Treatment Show forest plot	2	138	Mean Difference (IV, Random, 95% CI)	0.28 [‐2.21, 2.78]

4.1 Social Phobia	2	138	Mean Difference (IV, Random, 95% CI)	0.28 [‐2.21, 2.78]
5 General Anxiety Symptom Severity at Follow‐up Show forest plot	2	138	Mean Difference (IV, Random, 95% CI)	0.72 [‐2.12, 3.57]

5.1 Social Phobia	2	138	Mean Difference (IV, Random, 95% CI)	0.72 [‐2.12, 3.57]
6 Quality of Life at Post‐Treatment Show forest plot	3	199	Std. Mean Difference (IV, Random, 95% CI)	0.07 [‐0.37, 0.50]

6.1 Social Phobia	3	199	Std. Mean Difference (IV, Random, 95% CI)	0.07 [‐0.37, 0.50]
7 Quality of Life at Follow‐up Show forest plot	2	138	Std. Mean Difference (IV, Random, 95% CI)	‐0.19 [‐0.53, 0.14]

7.1 Social Phobia	2	138	Std. Mean Difference (IV, Random, 95% CI)	‐0.19 [‐0.53, 0.14]

Comparison 2. Therapist‐supported ICBT versus unguided CBT

Comparison 3. Therapist‐supported ICBT versus face‐to‐face CBT

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Clinically Important Improvement in Anxiety at Post‐Treatment Show forest plot	4	365	Risk Ratio (M‐H, Random, 95% CI)	1.09 [0.89, 1.34]

1.1 Social Phobia	1	126	Risk Ratio (M‐H, Random, 95% CI)	1.45 [0.77, 2.76]
1.2 Panic Disorder	3	239	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.85, 1.32]
2 Clinically Important Improvement in Anxiety at Follow‐up Show forest plot	3	279	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.94, 1.27]

2.1 Social Phobia	1	126	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.73, 1.83]
2.2 Panic Disorder	2	153	Risk Ratio (M‐H, Random, 95% CI)	1.09 [0.93, 1.28]
3 Anxiety Symptom Severity at Post‐Treatment Show forest plot	7	450	Std. Mean Difference (IV, Random, 95% CI)	0.06 [‐0.25, 0.37]

3.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.92, 0.57]
3.2 Panic Disorder	3	234	Std. Mean Difference (IV, Random, 95% CI)	0.29 [0.03, 0.54]
3.3 Specific Phobia	2	53	Std. Mean Difference (IV, Random, 95% CI)	‐0.02 [‐0.56, 0.52]
4 Anxiety Symptom Severity at Follow‐Up Show forest plot	6	367	Std. Mean Difference (IV, Random, 95% CI)	‐0.20 [‐0.41, 0.00]

4.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	‐0.39 [‐0.71, ‐0.08]
4.2 Panic Disorder	2	153	Std. Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.36, 0.28]
4.3 Specific Phobia	2	51	Std. Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.64, 0.46]
5 General Anxiety Symptom Severity at Post‐Treatment Show forest plot	6	343	Std. Mean Difference (IV, Random, 95% CI)	0.06 [‐0.42, 0.55]

5.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.49, 0.13]
5.2 Panic Disorder	2	129	Std. Mean Difference (IV, Random, 95% CI)	0.42 [‐0.75, 1.60]
5.3 Specific Phobia	2	51	Std. Mean Difference (IV, Random, 95% CI)	‐0.14 [‐1.07, 0.79]
6 General Anxiety Symptom Severity at Follow‐up Show forest plot	5	263	Std. Mean Difference (IV, Random, 95% CI)	‐0.13 [‐0.37, 0.11]

6.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	‐0.14 [‐0.45, 0.17]
6.2 Panic Disorder	1	49	Std. Mean Difference (IV, Random, 95% CI)	‐0.17 [‐0.74, 0.39]
6.3 Specific Phobia	2	51	Std. Mean Difference (IV, Random, 95% CI)	‐0.06 [‐0.61, 0.49]
7 Quality of Life at Post‐Treatment Show forest plot	5	392	Std. Mean Difference (IV, Random, 95% CI)	0.26 [0.06, 0.45]

7.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	0.22 [‐0.09, 0.53]
7.2 Panic Disorder	3	229	Std. Mean Difference (IV, Random, 95% CI)	0.28 [0.02, 0.54]
8 Quality of Life at Follow‐up Show forest plot	4	316	Std. Mean Difference (IV, Random, 95% CI)	0.33 [0.11, 0.55]

8.1 Social Phobia	2	163	Std. Mean Difference (IV, Random, 95% CI)	0.36 [0.01, 0.70]
8.2 Panic Disorder	2	153	Std. Mean Difference (IV, Random, 95% CI)	0.28 [‐0.04, 0.60]

Comparison 3. Therapist‐supported ICBT versus face‐to‐face CBT