Internet-delivered psychological treatment as an add-on to treatment as usual for common mental disorders: A systematic review with meta-analysis of randomized trials

Background: Psychological treatments for common mental disorders are increasingly being delivered remotely via the internet. Evidence suggests that internet-delivered cognitive behavioural therapy (iCBT) is superior to waitlist. However, the benefits are unclear of using this treatment modality as an add-on to treatment as usual (TAU) in regular healthcare. Methods: The literature was systematically searched up to August 2021 for randomized trials of internet-delivered psychological treatments using TAU as the comparator. Eligible participants were diagnosed with depressive, anxiety, obsessive-compulsive, or trauma-and stress-related disorders. Outcomes of interest were symptoms, functioning, quality of life, healthcare utilization, and negative effects. Results were synthesized using random-effects meta-analyses. Quality of evidence was assessed using GRADE. Results: The included studies evaluated iCBT for adults with depression (k = 9), depressive or anxiety disorders (k = 4), and post-traumatic stress disorder (k = 2) and were conducted in primary care or similar settings. For depression, low-certainty evidence suggested beneficial short-term effects on symptoms ( g = (cid:0) 0.23; 95 % CI: = (cid:0) 0.37, (cid:0) 0.09), response rate (OR = 2.46; 1.31, 4.64), and remission (OR = 1.70; 1.19, 2.42;). The certainty of evidence was very low for long-term effects, other outcomes, and other disorders. Limitations: TAU varied across studies and was often insufficiently described. Conclusions: iCBT as a complement to usual care for adult with depression may result in a small incremental effect, which potentially could be clinically important. Studies are lacking for several common disorders and for children, adolescents, and the elderly. More robust studies of long-term effects are also needed, to better inform clinical decision-making.


Introduction
Common mental disorders such as depressive disorders, anxiety disorders, obsessive-compulsive and related disorders, and trauma-and stress-related disorders constitute a global public health challenge (GBD 2019Mental Disorders Collaborators, 2022;Kessler et al., 2005).A large proportion of the general population will be affected during their life time, with a first onset typically occurring in childhood, adolescence, or young adulthood (Solmi et al., 2022).The long-term consequences can be dire, both for the individuals and society at large (Christensen et al., 2020;GBD 2019Mental Disorders Collaborators, 2022).To meet the increasing demand for treatment, mental health services in many countries are shifting from hospital-based specialty care to integrated mental health services in primary health care (Thornicroft et al., 2016).Still, a major proportion of those suffering from common mental disorders do not receive adequate treatment (Mekonen et al., 2021;Wang et al., 2007).Even if provision of treatment is increasing in some countries, the quality of care is still often suboptimal (Jorm et al., 2017).
Psychological treatment is frequently recommended as a first-line treatment (National Institute for Health and Care Excellence).In addition, the evidence also suggests that the effect of antidepressants can improve if they are combined with psychological treatment (Cuijpers et al., 2014a;Strawn et al., 2022).However, the traditional model of delivery of psychological treatment, including dependence on trained personnel that can deliver the treatment face-to-face, constitutes a key obstacle in terms of scalability (Kazdin, 2017(Kazdin, , 2019)).Delivery of treatment via the internet has been proposed as a possible avenue to improve accessibility (Andersson et al., 2019).This model of delivery typically includes a series of treatment modules made available on the internet.The modules are often text-based, but audio and video content may also be included.There is usually some level of support from a therapist, provided via text-message functions or telephone.Importantly, this format may require less therapist time per patient than traditional faceto-face treatment.The possible benefits of remote delivery of psychological treatment are further underscored by recent disruptions in mental health services observed in many countries during the COVID-19 pandemic.Patients not presenting and travel restrictions hindering access to health facilities have been pinpointed as some of the leading causes of an observed decrease in outpatient volumes (World Health Organization, 2020).
Previous systematic reviews suggest that internet-delivered cognitive behavioural therapy (iCBT) leads to greater symptom reduction than waitlist for depression and several anxiety disorders in adults (Andersson et al., 2019;Andrews et al., 2018;Arnberg et al., 2014;Pauley et al., 2021).In addition, a recent systematic review of studies conducted in routine care indicated large within-group effects on depression and anxiety over time (Etzelmueller et al., 2020).Based on the overall evidence there is reason to be optimistic about the usefulness of iCBT in primary care.Broad implementation of iCBT as a standardized add-on to usual care, with a view to increase the substandard remission and response rates among patients in real-world clinical practice, is a possible near-future scenario.On the other hand, the magnitude and clinical relevance of the added benefits, when introduced as an add-on to treatment as usual (TAU) are still unclear.Sub-analyses from a metaanalysis published in 2018 suggested that effect sizes compared to TAU are small to moderate, and significantly smaller than the effect size estimated from studies with waitlist comparators (Andrews et al., 2018).The number of studies on iCBT as an add-on to TAU has steadily increased during the last few years, suggesting that an updated metaanalysis could provide more precise estimates to guide clinical decision-making.
The present systematic review aimed to assess the benefits and risks of internet-delivered psychological treatments as an add-on to the usual care provided for depressive, anxiety, obsessive-compulsive, and trauma-related diagnoses across the life span, compared to usual care only.To increase the clinical relevance of the results, only studies including participants with a diagnosed disorder and treated in a regular healthcare setting were included.A wide range of outcomes were considered, including symptom reduction, remission, treatment response, functioning, quality of life, negative effects, and healthcare consumption.

Study design
This systematic review was conducted in accordance with the Cochrane handbook (Higgins et al., 2022) and is reported in accordance with the PRISMA statement (Page et al., 2021).The work presented here was part of a broader review of the efficacy, safety, and costeffectiveness of internet-delivered psychological treatment versus other available treatment options for common mental disorders, conducted at the Swedish Agency for Health Technology Assessment and Assessment of Social Services (SBU).A protocol was preregistered in PROSPERO (CRD42020214686).
The full report is available in Swedish on the agency's website (SBU, 2021).It should be noted that the Swedish report includes aspect that were omitted from the present work.Studies comparing internetdelivered treatments to other specified treatments (e.g., face-to-face treatment or physical exercise) were outside the scope of present work.Cost-effectiveness and ethical aspects, assessed specifically from a Swedish perspective, were also omitted here, since the present work was intended for an international readership.

Population
People of all ages, gender, ethnicity, and cultural background with one or more of the following diagnoses, as defined by DSM-5 (American Psychiatric Association, 2013), ICD-11 (World Health Organization, 2018) or previous versions of these classifications: -Depressive disorders (major depressive disorder; persistent depressive disorder; disruptive mood dysregulation disorder; premenstrual dysphoric disorder; mixed depressive and anxiety disorder) -Anxiety disorders (generalized anxiety disorder, social anxiety disorder, panic disorder, agoraphobia, selective mutism; separation anxiety disorder, specific phobias) -Obsessive-compulsive and related disorders (obsessive-compulsive disorder, body dysmorphic disorder; olfactory reference disorder; hoarding disorder; trichotillomania/hair-pulling disorder; excoriation disorder/skin-picking disorder, hypochondriasis/health anxiety/illness anxiety) -Trauma-and stress-related disorders (reactive attachment disorder; disinhibited social engagement disorder; post-traumatic stress disorder (PTSD); acute stress disorder; adjustment disorders; prolonged grief disorder) Studies were excluded if the participants were selected primarily based on a general medical condition (e.g., diabetes or cardiovascular disease).

Intervention
Internet-delivered cognitive behavioural therapy (including mindfulness-and acceptance-based treatments), psychodynamic psychotherapy, or interpersonal psychotherapy.The treatment should be delivered via computer, mobile phone, or other relevant devices.The intervention should be an add on to treatment as usual, in a format where the patients complete the intervention themselves and any support was remotely delivered (e.g., text messages or telephone).
Blended interventions combining face-to-face and internet-based treatment, and treatments delivered face-to-face via video conference were excluded.

Control
TAU in a regular healthcare setting, with at least some information about the setting and specific content of the care provided.
U. Jonsson et al.

Outcome
Main outcome (short-term: <6 months after the intervention; longterm: ≥6 months): Symptoms of each specific disorder (including rates of remission or response) as measured by validated scales.
Additional outcomes (short-term: <6 months after treatment; longterm: ≥6 months): Quality of lifemeasured by validated scales.Functioningany information available.Adverse effectsany information available.Healthcare consumptionany information available from posttreatment onward.Adherencedefined as the extent to which a person's behavior corresponds with agreed recommendations from a health care provider (World Health Organization, 2003).

Study design
Only randomized controlled trials were considered for inclusion.

Language
English, Swedish, Norwegian, and Danish.

Publication type
Peer-reviewed original reports.

Information sources and search strategy
An initial literature search for relevant systematic reviews was conducted in the CADTH publication database, CINAHL (EBSCO), Cochrane Library (Wiley), Embase (Elsevier), Epistemonikos, Evidence search (NICE), the Norwegian Institute of Public Health publication database, International HTA-database, KSR Evidence, Ovid MEDLINE, PROS-PERO, PsycINFO (EBSCO), Psychology and Behavioral Sciences Collection (EBSCO) and Scopus (Elsevier).This initial search was conducted in May 2020 and updated in October 2020.
A total of 1918 records were screened, and 416 full-text documents were retrieved.No systematic review covering the full range of the commissioned report was identified.It was therefore decided that the review would build on relevant parts of a previous review by SBU from 2013 (Arnberg et al., 2014;SBU, 2013).
Literature search of original articles was conducted in CINAHL (EBSCO), Cochrane Library (Wiley), Embase (Elsevier), Ovid MEDLINE, PsycINFO (EBSCO), Psychology and Behavioral Sciences Collection (EBSCO) and Scopus (Elsevier).The search was conducted in October 2020, with an updated search in April 2021 and final update on 26 August 2021 (or a few days later for some databases).The search for articles was supplemented by citation searching of key articles in Scopus as well as literature that we identify in reference lists from published articles and overviews.The search was restricted to records in English, Swedish, Norwegian, and Danish.For diagnoses or diagnostic terms included in the previous review, the databases were searched from 2012, while the search for new diagnoses and terms appearing in DSM-5 or ICD-11 was limited from 1990.The reference management program EndNote was used to deduplicate records from the literature search (Bramer et al., 2016;The EndNote Team, 2013).The full search strategy for each database is provided in Appendix A.
Two review authors independently screened all abstracts/titles.All studies selected by at least one reviewer were checked in full text by two reviewers independently.Discrepancies were resolved by discussion and ultimately by a third reviewer.The Rayyan tool was used for the selection process.

Data collection process and data items
One author extracted the data, while a second author checked the integrity of the extracted information.The data extraction form included the following information: author and year; country of origin; design; number of participants; assessment points; participant characteristics (age, gender, education, diagnosis, and method of diagnostic assessment); setting; treatment (including treatment paradigm, level of support, duration, content of TAU); primary and secondary outcome measures (including relevant outcome data); adverse events or deterioration; attrition; and information related to treatment adherence.In one of the publications (Rollman et al., 2018), the outcome data were extracted from figs.

Study risk of bias assessment
Risk of bias was assessed independently by two review authors, using The Cochrane risk-of-bias tool for randomized trials, RoB 2. Disagreements between the review authors over the risk of bias in particular studies were resolved by discussion, with involvement of a third review author when necessary.

Effect measures
If the included studies used different scales, standardized mean differences (SMDs) were calculated, in the form of Hedge's g.If the same scale was used, the results were presented as both standardized mean difference and mean difference in the scale's original metric.To facilitate interpretation of clinical relevance, transformation of the SMD to a well-established scale was considered for key results.Ultimately, this was only done for the meta-analysis focusing on depressive symptoms at posttreatment, where the scores were transformed to the Beck Depression Inventory II (BDI-II).As a basis for the transformation, a standard deviation of 10 was selected based on the standard deviations at baseline for the included studies using BDI-II and other studies on similar populations (Wang and Gorenstein, 2013).If minimally important differences (MIDs) had been established for the scales used, these were considered when interpreting the results.Binary outcomes were reported as both relative (odds ratios) and absolute (risk difference) effects.

Synthesis methods
Each included study was summarized and described according to the characteristics of the participants, interventions, follow up and outcomes measured.Meta-analysis was employed if the included studies were sufficiently homogenous.
Results were synthesized for each specific diagnosis, or, when deemed appropriate, diagnostic groups (e.g., depressive disorders or anxiety disorders).When the included populations were mixed in terms of primary diagnosis, studies were included in a synthesis if the majority were diagnosed with a condition relevant for the specific analysis.When appropriate, sensitivity analyses were considered based on the following: age (children <18 years, adults, and elderly); type of psychological treatment (cognitive behavioural therapy, psychodynamic psychotherapy, interpersonal psychotherapy); models of delivery; degree of support (pure self-help vs coach or therapist-guided); design; and risk of bias of the individual studies.All meta-analyses were randomeffects models calculated using Review Manager 5.4.1.

Reporting bias assessment
Reporting bias was assessed using funnel plots for outcomes available from more than five studies.
U. Jonsson et al.

Certainty assessment
The certainty of evidence was assessed using Grading of Recommendations, Assessment, Development and Evaluations, GRADE (Balshem et al., 2011).

Study characteristics
Of the included studies, 14 were parallel arms RCTs and one was a cluster-randomized trial (Löbner et al., 2018).Nine of the included studies included participants with depressive disorders exclusively, while four studies included patients with depressive or anxiety disorders (Table 1).In three of the latter studies (Hallgren et al., 2016;Hallgren et al., 2015;Proudfoot et al., 2003;Rollman et al., 2018) there was a substantial overlap between the diagnoses, with >80 % of the participants having a depressive disorder.In the fourth study, which explicitly evaluated a transdiagnostic treatment protocol, 30 % of the participants had a depressive disorder and 70 % an anxiety disorder (Gonzalez-Robles et al., 2020).The participants in two studies had a diagnosis of PTSD or unspecified trauma and stressor-related disorder (Acosta et al., 2017;Engel et al., 2015).
The sample size of the included studies ranged from 43 (Milgrom et al., 2016) to 647 (Löbner et al., 2018).With the exception of the studies focusing on PTSD, the included participants were predominantly females.In most studies, participants were, on average, in their mid-30s to mid-40s.The proportion of participants with higher education ranged from 30 % (Löbner et al., 2018) to 100 % (Raevuori et al., 2021) (Table 1).
The studies were mainly conducted in primary care or similar healthcare facilities.The evaluated interventions were CBT-based programs, some of which incorporated mindfulness practice.The duration of the interventions varied considerably, with the number of modules ranging from 5 (Löbner et al., 2018;Yeung et al., 2018) to 24 (Acosta et al., 2017;Ritvo et al., 2021) over a treatment perioded lasting from 5 weeks (Yeung et al., 2018) to 26 weeks (Rollman et al., 2018).In most studies, participants received either clinical support or help by nonclinicians to navigate the program.However, some of the evaluated programs were purely self-guided (Acosta et al., 2017;Löbner et al., 2018;Yeung et al., 2018).One study evaluated both guided and selfguided iCBT in a three-armed trial (Montero-Marin et al., 2016).Participants' adherence to iCBT, as indicated by completed modules, was reported in 11 of the studies.Results suggested that participants completed approximately half of the modules on average, while the proportion completing all modules ranged from 13 % (Löbner et al., 2018) to 86 % (Milgrom et al., 2016) (Table 2).
TAU, as reported by the authors, mainly comprised pharmacological and psychological treatment.In some studies, a clear majority of the participants in both groups received pharmacological treatment (Gonzalez-Robles et al., 2020;Hatcher et al., 2018;Montero-Marin et al., 2016;Rollman et al., 2018).Other studies reported that about half (Löbner et al., 2018;Pfeiffer et al., 2020;Raevuori et al., 2021), or about one third or less received such treatment (Hallgren et al., 2016;Hallgren  U. Jonsson et al. et al., 2015;Milgrom et al., 2016;Proudfoot et al., 2003).With a few exceptions (Kivi et al., 2014;Milgrom et al., 2016), the proportion receiving pharmacological treatment was similar in both groups.In several studies, other forms of psychological treatment were not allowed or not reported.When reported, a minority in both groups received such treatment.An exception was the study by Hallgren et al. (Hallgren et al., 2016;Hallgren et al., 2015), in which 73 % of the participants in TAU group received psychological treatment (Table 2).
All studies measured change in core symptoms with validated scales.Three studies reported treatment response, defined as at least 50 % reduction of the symptom scores.Six of the studies included a definition of remission, mainly defined as self-reported symptoms below a cut-off on the primary outcome measure.Ten studies included a measure of quality of life, while only two studies used a measure of functioning.There was some form of monitoring of adverse effects in seven of the studies, some of which measured deterioration on the primary outcome measure (Table 3).

Risk of bias in studies
There were concerns about risk of bias in all included studies.The overall risk of bias was assessed as high for 11 studies, while there were some concerns for the remaining 4 studies.Missing outcome data was the most frequent reason for rating a study as having a high risk of bias.A summary of the ratings can be found in Appendix C: Table S1, and the ratings for specific outcomes are included in the forest plots (Figs.2-5; Appendix C: Figs.S2-S26).

Improvement in core symptoms
A meta-analysis of the effects on depressive symptoms at posttreatment included the results from 12 studies that primarily included participants with depressive disorders (Fig. 2).The estimated effects of the individual studies at posttreatment varied considerably, with point estimates ranging from large effects sizes in favour of iCBT to small and non-significant effects.The largest effect sizes were observed in smaller trials, while the trials with larger samples suggested small or no effect.Reporting bias could not be ruled out (see Appendix C: Fig. S1).The meta-analysis, including a total of 2224 participants, suggested in a small overall effect size in favour of iCBT.Sensitivity analyses excluding trials of high risk of bias (Appendix C:  scale (BDI-II) indicated that the estimated effect corresponds to approximately 2.5 points (Fig. 3), which approximately matches the minimally important difference estimated for patients in the mild to moderate range on the BDI-II (Button et al., 2015).Five trials reporting results from short-term follow-ups, suggesting a similar effect size after up to 6 months (Appendix C: Fig. S5).Long-term follow-ups of 6 months or more were reported in five trials: while two relatively small trials indicated moderate effects in favour of iCBT, three trials (two of which were larger) suggested small or no effects.Meta-analysis suggested an effect size of a similar magnitude as the estimated effect at posttreatment (Appendix C: Fig. S6).The certainty of evidence was assessed as low (⊕⊕⊝⊝) for short-term outcome, and very low (⊕⊝⊝⊝) for long-term outcome (Table 4).
Of the studies that included patients with either a primary depressive disorder or anxiety disorder, three used anxiety symptoms as an outcome (Gonzalez-Robles et al., 2020;Proudfoot et al., 2003;Rollman et al., 2018).The majority of participants in these studies (62 to 86 %) had an anxiety disorder or mixed anxiety and depression.Meta-analysis of the studies indicated a small effect in favour of iCBT (Appendix C: Fig. S7).Data on short-term follow-up after 3 months presented in two studies indicated similarly small effects, while the long-term outcomes did not suggest a significant advantage of the group receiving iCBT (Appendix C: Figs.S8 and S9).The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for both the short-term and the long-term outcomes (Table 5).
The two studies evaluating iCBT as an add-on to TAU for patients with PTSD (Acosta et al., 2017;Engel et al., 2015) used the same primary outcome measure (PTSD Checklist) measured posttreatment and after approximately 1 and 3 months.Meta-analyses did not suggest any added effects of iCBT on PTSD-symptoms at any of the time points (Appendix C: Figs.S10 to S15).The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for this outcome (Table 5).

Treatment response
Three studies reported response rates for depression, defined as at least 50 % reduction in scores on symptom scales (Hallgren et al., 2016;Hallgren et al., 2015;Löbner et al., 2018;Pfeiffer et al., 2020).Metaanalysis of the two studies suggested an odds ratio of 2.46 (1.31 to   and S18).The study by Hallgren et al. (2015Hallgren et al. ( , 2016) ) reported that 47 % in both groups were responders 9 months after the treatment ended.The certainty of evidence was assessed as low (⊕⊕⊝⊝) for short-term outcome, and very low (⊕⊝⊝⊝) for long-term outcome (Table 4).Response rates were not reported for other outcomes.depression, with definitions based on symptom rating scales (Hatcher et al., 2018;Kivi et al., 2014;Löbner et al., 2018;Pfeiffer et al., 2020).Two of the studies also included short-term follow-up data (Löbner et al., 2018;Pfeiffer et al., 2020).The definitions of remission varied between studies (Table 3).Meta-analysis of the four studies suggested an odds ratio of 1.70 (1.19 to 2.42) at posttreatment in favour of iCBT (Fig. 5; Appendix C: Fig. S19), while the two studies with follow-up data suggested that the effects were sustained in the short-term (Appendix C: Figs.S20 and 21).One additional study (Milgrom et al., 2016) reported remission rates defined as the proportion no longer meeting the criteria for a major depressive episode posttreatment, with a significantly larger proportion reported in the group receiving iCBT compared to the control group (79 % vs. 18 %).The certainty of evidence was assessed as low (⊕⊕⊝⊝) for remission in the short term (Table 4).

Remission Four studies reported posttreatment remission in adults with
In a study that mainly included participants with mixed anxiety disorder (Gonzalez-Robles et al., 2020), recovery was defined as reliable change based on the Reliable Change Index (RCI) and a score located within the range of the functional distribution on the Beck Anxiety Inventory (BAI).This criterium was met by 56 % of the iCBT group and 37 % of the control group posttreatment, with the difference evened out after 3 months (49 % vs. 47 %).The certainty of evidence was assessed as very low (⊕⊕⊝⊝) for this outcome (Table 5).The studies focusing on PTSD did not report remission rates.
The effects at posttreatment and short-term follow-up were synthesized in meta-analyses.When more than one measure was available, the mental component score of SF-12 was used if available, with EQ-5D as the second choice.The meta-analyses suggested small effects favouring iCBT at posttreatment (Appendix C: Fig. S22), while the few studies with follow-up assessments in the short-term and long-term were inconsistent (Appendix C: Figs.S23 and S24).The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for short-term and long-term outcome (Table 4).
The transdiagnostic study included EQ-5D and reported small to moderate effect sizes posttreatment and 3 months after the treatment.One of the studies focusing on PTSD measured quality of life with the WHOQOL-BREF (Acosta et al., 2017), while the other included SF-36 (Engel et al., 2015).No significant effects were reported.The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for this outcome (Table 5).

Functioning
Two studies focusing on depression included information about effects on functioning.In one of the studies, where the participants were asked to rate their work capacity on a scale from 0 to 10, no significant difference was observed at the end of the treatment (Hallgren et al., 2015).The other study suggested a small and non-significant effect on the Work and Social Adjustment Scale (WSAS) favouring iCBT at posttreatment and six months after the treatment ended (Proudfoot et al., 2003).The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for this outcome (Table 4).

Healthcare consumption
All studies provided at least some information about the treatment received at baseline, during the study, or at the end of the treatment (Table 2).However, no information about how the healthcare consumption developed after the intervention period was available.

Adverse effects
Information about adverse effects was either missing or only briefly mentioned in most studies.Five of the studies focusing on depression include at least some information.One study report that no adverse events that could be linked to the intervention were recorded by the participating GPs or by the contact persons for the study during the trial (Löbner et al., 2018).Two additional studies briefly stated that no adverse events occurred (Raevuori et al., 2021;Yeung et al., 2018), while two studies reported that safety had been monitored during the trial (Milgrom et al., 2016;Rollman et al., 2018).
Deterioration rates, defined as reliable change on the BDI-II using the Fig. 3. Meta-analysis of 12 trials reporting depressive symptoms at posttreatment in adults with depressive disorders (mean differences transformed to the Beck Depression Inventory-II scale; lower scores indicate lower level of depression).
U. Jonsson et al.RCI, was reported in two of the studies focusing on depression.Results were available at posttreatment in one of the studies (Kivi et al., 2014), and follow-up after approximately 4.5 months in the other (Löbner et al., 2018).A meta-analysis of the two results did not suggest a significant difference between the groups (Appendix C: Figs.S25 and S26).One study reported deterioration rates regarding anxiety symptoms at posttreatment and after 3 months, defined using the RCI on the BIA (Gonzalez-Robles et al., 2020).There was a statistically significant difference only at posttreatment, favouring the group receiving iCBT.The certainty of evidence was assessed as very low (⊕⊝⊝⊝) for this outcome (Tables 4  and 5).

Discussion
This systematic review suggests that iCBT as an add-on to usual care for mild to moderate depression in adults may have a small incremental effect, at least in the short-term perspective.The estimated effect size was at par with a tentative clinical relevance cutoff for depression proposed by Cuijpers and colleagues (Cuijpers et al., 2014b).The add-on effect should be interpreted in light of the improvements seen in the comparison groups, where the average posttreatment symptom severity generally was in the mild range.From this modest level of symptom severity, a small additional effect might be sufficient to help patients cross the threshold to full or partial remission.Indeed, such a pattern was suggested by the few studies that reported data on remission rates.The effects were unclear for other important outcomes, including quality of life and functioning.The same was true for long-term outcomes, partly due to a lack of information about treatments that patients might have received after iCBT.For other common mental disorders, and for all disorders in children and adolescents, there were few or no studies available, and the certainty of evidence was assessed as very low.
The estimated effect on depressive symptoms was comparable to the overall effects reported for psychological treatment of adult depression, at least when comparing to usual care (Cuijpers et al., 2019;Munder et al., 2019).While effectiveness studies have shown large within-group effects in clinical settings (Etzelmueller et al., 2020), previous estimates of the effect of iCBT compared to TAU are similar to the effects estimated here (Andrews et al., 2018).Importantly, a recent systematic review suggests that the effect size might vary as a function of the intensity of TAU (Munder et al., 2019).It might not be a coincidence, then, that the two trials with the largest estimated effect sizes in our review also reported unusually low rates of pharmacological or psychological treatment in the control group.Indeed, it is quite possible that the value of broad implementation of iCBT would be small or negligible if the overall quality of usual care were improved.On the other hand, the main reason for implementing iCBT in primary care would arguably be to raise the base level of quality of the care provision.Further, the evidence on benefits of combined pharmacotherapy and psychotherapy in adults (Cuijpers et al., 2014a) may extrapolate to internet-delivered treatments.If so, there would potentially be an additional positive effect of receiving iCBT for adults who receive antidepressants as TAU, particularly for some disorders, whereas it is unclear if iCBT as an add-on would be beneficial, or even counterproductive, for patients who are receiving some other form of psychosocial intervention as TAU.It is striking that the included studies mainly concerned depressive disorders.Anxiety disorders were only included in transdiagnostic studies, together with depressive disorders.No studies targeted obsessive-compulsive and related disorders, while the only two studies on trauma-and stress-related disorders explicitly focused on veterans in the US.This is surprising, given that iCBT has shown favourable results compared to waitlist for anxiety disorders in particular (Andrews et al., 2018;Arnberg et al., 2014;Pauley et al., 2021) and that CBT in general has been reported to have positive effects compared to TAU for anxiety disorders (Cuijpers et al., 2016).Promising results have also been reported for iCBT compared to waitlist for PTSD in adults, although the certainty of the current evidence recently might be very low (Simon et al., 2021).
There was also a lack of studies on children, adolescents, and the elderly.Effects of both psychological and pharmacological treatments may differ between age groups (Cuijpers et al., 2020;Hetrick et al., 2021;Tham et al., 2016), underscoring the importance of age-specific adaptations and evaluations of interventions.The lack of studies on children and adolescents might be particularly important, as iCBT for these age groups is already being offered within healthcare settings in some countries (e.g., Sweden and Australia).It is also noticeable, although not surprising, that the included studies exclusively evaluated the effect of iCBT.While other forms of psychotherapy have been adapted to an internet format, only few studies on these programs have been published to date (Johansson et al., 2013;Johansson et al., 2012;Lindqvist et al., 2020).Finally, it is unclear if the addition of iCBT to usual care improves other outcomes besides core symptoms.The World Health Organization's (WHO) definition of health as "a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity" (World Health Organization, 1948) points to the value of including a broad array of outcomes, including quality of life and functioning.
There is reason to be optimistic that some of these gaps in the literature will be filled in the near future.The number of published trials with TAU as the comparator is steadily growing, indicating that this is one of the major questions the field is currently grappling with.Some additional studies on this topic have already been published since the last search date for the present systematic review.One trial compared iCBT or only TAU for primary care patients with depression (Wright et al., 2022).Not all participants had a confirmed diagnosis; however, the overall results are in line with our findings on posttreatment effects on depressive symptoms, response, and remission rates.Further, a recent trial comparing iCBT, face-to-face treatment, and TAU for postnatal depression (Milgrom et al., 2021), suggested that iCBT was superior to TAU.

Generalizability
The included studies were conducted in a range of different countries with diverse healthcare systems.On the other hand, interpretation is hampered by shortcomings related to the sparce descriptions of the content and quality of TAU.However, the samples in the trials on depression appear to be representative of adults treated in primary care.
Most studies predominantly included females, reflecting the higher prevalence of depressive disorders and anxiety disorders in females (Boyd et al., 2015;Seedat et al., 2009).Some studies included quite specific patient groups (e.g., students, or mothers with postpartum depression).Still, the overall results cannot be readily generalized to some of the major groups of patients treated in primary care (e.g., children, adolescents, young adults, and older adults).In addition, cultural and socioeconomic aspect were not explored in the included studies.Finally, we did not include studies evaluating blended approaches or videoconference therapies.Thus, our results should not be generalized across all digital therapies.

Limitations
There are several limitations to the extant literature, which makes the overall evidence uncertain.First, as noted above, there are major gaps in the literature.Second, information was scant about the specific content of TAU.In addition, it was unclear if the TAU provided in the studies was comparable to the treatment that patients at the specific facilities typically receive.As noted above, the true added effect is likely to be contingent on the intensity and quality of the treatment already provided (Munder et al., 2019).To make informed decisions about implementation, this must clearly be factored in.There has been a longstanding discussion about insufficient reporting of TAU comparators in the field of psychiatry (Burns and Priebe, 1996;Watts et al., 2015), but it is evident that the level of details provided can still be improved.Indeed, there is a need for methodological guidelines standards of design and reporting of control conditions in the mental health field in general (Faltinsen et al., 2022).Third, the mean number of completed modules was in many studies substantially lower than previously reported for iCBT and for face-to-face CBT (van Ballegooijen et al., 2014).While the results might still reflect the true effect of the intervention as provided, the poor completion rate in some studies raises questions as to whether the interventions were implemented in an optimal way.The full scope is unclear of the participants' adherence to the treatment, including to what extent participants followed the recommendations provided by the programs (World Health Organization, 2003).Fourth, the measurement and reporting of potential harms were suboptimal in most studies.However, a clear improvement has occurred over the last decade (Arnberg et al., 2014;Jonsson et al., 2014).For instance, recent individual participant data meta-analyses indicate that participants in both guided and self-guided iCBT are less likely to deteriorate than waitlist controls (Ebert et al., 2016;Karyotaki et al., 2018;Rozental et al., 2017).Fifth, there were at least some concerns related to risk of bias for all included studies.High risk of bias was

Table 5
Summary of findings for internet-delivered cognitive behavioural therapy (iCBT) as an add-on to treatment as usual (TAU) for post-traumatic stress disorder (PTSD) in adults and for anxiety in adult populations with mixed anxiety disorders and/or depressive disorders.U. Jonsson et al. assessed for most studies, often related to attrition.On the other hand, the attrition rate is arguably a true reflection of the clinical reality in primary care.There were also some limitations related to our review process.The meta-analyses were not based on change scores and consequently did not account for potential baseline differences between the groups.It should also be noted that some of the included studies did not view iCBT as an add-on to TAU, but rather as an alternative.However, the participants receiving iCBT also received at least a base level of usual care in those studies.Finally, we included only studies with participants who fulfilled criteria for a diagnosis in order to increase the generalizability to clinical settings and to lower heterogeneity across studies.Therefore, some studies were excluded due to using only symptom severity as the inclusion criterion.It is possible that a majority of participants in some of these studies would have fulfilled diagnostic criteria if they had been subjected to a diagnostic interview.Given the unspecific nature of symptoms of depression and anxiety, however, such assumptions are very uncertain.

Clinical implications and future directions
It can be expected that full remission from depression and recovery from other common mental disorders will benefit not only the patient and their extended family, but also the healthcare system and the society at large.To the extent usual care at primary care facilities fails to provide effective treatments, it is urgent to find cost-effective ways to improve the outcome.The estimated add-on effect for depression in adults could prove to be clinically important, especially if an additional proportion of patients would regain their previous level of quality of life and functioning.However, the uncertainty of the current findings, including long-term effects, underscores the need for more research.
Information about long-term effects and potential harms is also important for the assessment of cost-effectiveness of iCBT.A costeffectiveness study based on one of the trials included in our review provide some indications, with a small reported gain in quality-adjusted life-years (QALYs) and an estimated cost per QALY gained of 10,166 euro over 12 months (price year 2012) from a health-care perspective (Kraepelien et al., 2018).Though iCBT as an add-on to TAU clearly has the potential of being cost-effective due to the low cost of the intervention, a full assessment is nonetheless precluded by the uncertainty surrounding the long-term effects.
In addition to these gaps in the literature, our review suggests some future research directions that might help inform clinical decisionmaking.First, better reporting of the TAU provided and the extent to which it reflects the real-world treatments provided for common mental disorders in divers healthcare systems would improve the usefulness of the research.Second, the estimates from the meta-analyses included in the present review might also guide researchers in conducting realistic power analyses for future trials.This is important, not the least since the funnel plots suggest possible publication bias with larger effects observed in smaller trials.Finally, methods to improve participants' adherence when iCBT is implemented in regular care seems warranted.On a similar note, the variability between participants in terms of completed number of modules might point to the possibility that iCBT is more well-suited for some patients than for others.From an implementation perspective, this is an important issue to address in future research.

Conclusion
iCBT as an addition to usual care may result in a small additive effect for adult with mild to moderate depression, but the certainty of evidence is still low.A small effect could potentially be clinically important.However, it is unclear if the added benefit is lasting, and whether any reduction in symptom severity will be accompanied by improvements in quality of life and functioning.To better inform clinical decision-making, more robust studies are needed on long-term effects.Studies focusing on anxiety disorders and other mental disorders commonly treated in primary care are also needed, as well as studies into potential benefits for children, adolescents, and the elderly.
Fig. S2), trials evaluating selfguided interventions (Appendix C: Fig. S3), and the clusterrandomized trial by Löbner et al. (Appendix C: Fig. S4) suggested similar effects.Transformation of the results to the most frequently used

Fig. 4 .
Fig. 4. Meta-analysis of 2 trials reporting posttreatment response rates in adults with depressive disorders.

Fig. 5 .
Fig. 5. Meta-analysis of 4 trials reporting posttreatment remission rates in adults with depressive disorders.

Table 1
Sample characteristics of the included trials.
Abbreviations: iCBT, internet-delivered Cognitive Behavioural Therapy; MDD, Major Depressive Disorder; PTSD, Post-Traumatic Stress Disorder; SD, Standard Deviation; TAU, treatment as usual.aNumber of participants randomized to interventions or comparators relevant to the present review.U.Jonsson et al.

Table 2
Treatment received by the intervention group and the control group.

Table 3
Outcome assessment in the included trials.

Table 4
Summary of findings for internet-delivered cognitive behavioural therapy (iCBT) as an add-on to treatment as usual for depression in adults.Downgraded one level: serious risk of bias due to lack of blinding and unclarities related to the content of treatment as usual.cDowngradedone level: imprecision for confidence interval overlapping with a possibly unimportant effect.dDowngraded two levels: very serious risk of bias due large attrition in combination with a lack of blinding and unclarities related to the content of treatment as usual.eDowngraded one level: inconsistency of the estimated effects.f Downgraded one level: imprecision for wide confidence interval.
a Transformed to scores on Beck Depression Inventory II.b gU.Jonsson et al.
Civilian version; PCL-M, PTSD Checklist -Military version; RD, Risk Difference; SMD, Standardized Mean Difference; WHOQOL-BREF, WHO Quality of Life assessment, short form.a Downgraded two levels: very serious risk of bias due to ambiguity regarding the randomization, large attrition in combination with a lack of blinding and ambiguity regarding the content of treatment as usual.b Downgraded one level: imprecision due to the confidence interval overlapping with possibly trivial effects.c Downgraded one level: serious risk of bias due to lack of blinding and ambiguity regarding the content of treatment as usual.d Downgraded two levels: imprecision due to a wide confidence interval overlapping with no effect.e Downgraded one level: imprecision, few observations.f Downgraded one level: indirectness, due to the uncertainty of generalizing information from only one study.g Downgraded two levels: imprecision for confidence interval overlapping with no effect and a relatively low number of participants.h Downgraded two levels: indirectness due to the specific population included (Veterans in the US).