Internet-delivered transdiagnostic psychological treatments for individuals with depression, anxiety or both: a systematic review with meta-analysis of randomised controlled trials

Search strategy

This systematic review with meta-analysis adhered to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 statement20 and wasregistered on PROSPERO (CRD42021243172) on 16 April 2021 and at the Open Science Framework (https://osf.io/dtcey) on 18 May 2021, that is, before the initial database searches. We searched Medline (Ovid), Cochrane Library (Wiley), the Web of Science Core Collection (Clarivate) and PsycInfo (EBSCO) for RCTs of internet-delivered transdiagnostic psychological treatments for individuals with depression, anxiety or both. A search strategy was developed in collaboration with the Karolinska Institute library service. Candidate search strings were validated against a selection of articles that we were aware of prior to this systematic review, and after discussion within the research group, as well as with an external expert in psychodynamic online therapies. Four types of search terms were combined: terms for treatment (eg, ‘psychotherapy’), terms for online communication (eg, ‘internet-delivered’), terms for diagnostic focus (eg, ‘transdiagnostic’, ‘depression’ and ‘anxiety’) and terms for study design (eg, ‘RCT’). Only publications published in 1995 or later were considered because there were no RCTs of internet-delivered transdiagnostic psychological treatments before then. Full search documentation for all databases can be found in online supplemental file A. An initial search was conducted on 24 May 2021, with an update on 6 February 2023 in accordance with Bramer and Bain.21 We also reviewed the reference lists of previous meta-analyses that we found relevant.16 17

Supplemental material

Selection of studies

We employed a systematic deduplication algorithm in Endnote V.X9,22 after which unique search hits were imported into Rayyan.23 The publications were then assessed by two independent assessors in two phases. In phase I, all titles and abstracts were screened for eligibility by two authors (of KK, AHB, EH-L and EA), and non-relevant publications were excluded. In phase II, publications that had been included by at least one assessor in phase I were reviewed in full text. One assessor (KK) did this sequentially, that is, from criterion a to h (see below), for all full texts. Each full text was also reviewed by a second assessor (either AHB, EH-L, JH or EA) who suggested a decision of exclusion based on the first exclusion criterion identified, that is, in any order. Cases of disagreement were discussed in the research group and a final decision was reached either in consensus or, if necessary, by means of voting. If information about the eligibility criteria a–h was missing, authors were not contacted to provide additional information.

Data extraction

Prespecified variables were extracted by the first author (KK), including participant demographics (eg, age and gender). Studies were considered to concern children or adolescents if the mean age was below 18, and to concern older adults if the mean age was above 65. The following study and sample characteristics were extracted: study setting (primary care, not primary care or unclear), recruitment path (routine care or not routine care), main inclusion criterion (psychiatric diagnosis or ‘cut-off symptoms’), mean depression severity (mild, moderate or severe) and mean anxiety severity (mild, moderate or severe). Symptom severity ratings were based on published norm data.25–28 Extracted treatment characteristics were as follows: delivery format (website, mobile app, video, mixed or other), therapist support (guided or unguided), treatment type (CBT, mindfulness, mixed CBT and mindfulness or other), specific treatment protocol, treatment length (≥6 weeks or <6 weeks; intended to contrast what are usually considered brief therapies with therapies of a more conventional length), component flexibility (standardised or tailored) and therapist time (minutes per participant in treatment). In line with recent findings demonstrating that many treatment-as-usual control groups are virtually indistinguishable from waitlist controls,29 we did not lump all treatment-as-usual controls into the same category but rather made a distinction among (a) rudimentary passive controls that comprised waitlist controls and other controls without a structured intervention, (b) attention/engagement controls where the participants received some sort of standardised intervention that controlled for the attention from a caregiver or engagement in a study but without this being a conventional psychological treatment and (c) bona-fide treatments ‘delivered by trained therapists and […] based on psychological principles, […] offered to the psychotherapy community as viable treatments’.30 For each trial, the primary outcome for depression and anxiety was tabulated. If no outcome was defined as primary, or the primary outcome measure was deemed not to be valid, we tabulated the first valid measure that was mentioned. KK contacted the corresponding author when main outcome data were missing. All authors were also asked to provide pre–post correlations for the calculation of within-group effects. Intention-to-treat estimates were tabulated whenever available.

Assessment of risk of bias and quality of the evidence base

Trial risk of bias was assessed using a modified version of the Cochrane risk-of-bias tool for randomised trials V.2 (RoB 2).31 The RoB 2 covers five domains: (1) the randomisation process, (2) deviations from intended interventions, (3) missing outcome data, (4) measurement of the outcome and (5) selection of the reported results. Because of the difficulty in blinding therapists to the psychological treatment they are delivering, we omitted items related to blinding. Consistent with previous similar research, non-blinding did not therefore lead to an elevated risk of bias.15 32 Each trial received one rating on each risk of bias domain, and this rating concerned both the depression and the anxiety outcome. Reporting one score was sufficient because all trials administered the depression and anxiety measures in parallel, which resulted in identical ratings. Authors KK and EL independently assessed the trials and discussed cases of disagreement to reach consensus. The decision was made that if five articles in a row were rated identically by both assessors, it would no longer be necessary with two assessments per article. This target was achieved after 25 articles, and KK assessed the remaining trials. Based partially on the risk of bias ratings, we assessed the overall quality of the evidence base according to Grading of Recommendations Assessment, Development and Evaluation (GRADE).33 We assigned separate ratings for the evidence base pertaining to (1) effects versus rudimentary passive controls for adults, (2) effects versus attention/engagement controls for adults, (3) effects versus other bona-fide treatments for adults, (4) effects for children and adolescents and (5) effects in primary care. Similar to the risk of bias assessment, one common score was given for depression and anxiety.

Statistical analysis

We report studies focusing on children and adolescents, and studies focusing on older adults, separately due to their specific focus. Effects versus other bona-fide treatments were too few to be included in meta-analysis. At the end of the tabulation phase, we also decided to report studies with a treatment length of 2 weeks or shorter separately, a duration which would not typically be regarded as bona fide in routine clinical care. Nine randomised factorial trials were also included in the meta-analysis of within-group effects, but their main results were summarised separately.

We proceeded with meta-analysis using random-effects models fitted with the restricted maximum likelihood estimator using the inverse variance method, in R V.4.2.034 with the metafor 3.8–1 package.35 Effects were quantified in terms of Hedges’ g, that is, the bias-corrected standardised mean difference.36 Although not our primary focus, we first conducted a meta-analysis of the within-group change in treatment. The point of this was to ensure that change had been observed per se, to determine if heterogeneity was much reduced without the influence of variance due to different control groups, and to facilitate the interpretation of within-group effects in future studies. For this analysis, we standardised g on the pretreatment SD, and used the pre–post correlation (imputed pooled estimate if missing) to determine the sampling variance:37 38

Embedded Image

Embedded Image

We then conducted a meta-analysis of between-group effects versus rudimentary passive controls, and a meta-analysis of between-group effects versus attention/engagement controls. For these analyses, we standardised g on on the pooled post-treatment SD, and employed the large-sample approximation of the sampling variance:36 39

Embedded Image

Embedded Image

Absolute values for g of 0.2 are usually regarded as small, 0.5 as moderate and 0.8 as large.40 We quantified absolute heterogeneity in terms of the τ2, and also report the I2 which stands for the proportion of the between-sample variance that is due to true study differences as opposed to sampling error. An I2 of 25% is usually regarded as low, 50% as moderate and 75% as high,41 though the I2 also increases with the precision of the original trials.42 Pertaining to the meta-analysis of trials of transdiagnostic treatment versus rudimentary passive controls, a series of planned moderator analyses were conducted on the basis of trial, sample and treatment characteristics whenever strata included at least four studies. One exception was made to this rule where we analysed primary care setting as a potential moderator based on four arms from three trials because we found this test important. In the analysis of trials with rudimentary controls, we also investigated publication bias based on the visual inspection of funnel plots in combination with Egger’s test43 and the Duval and Tweedie trim and fill procedure.44 In line with recommendations,45 the latter was repeated using the R0, L0 and Q0 estimators.

Patient and public involvement

None

This post was originally published on https://bmjopen.bmj.com