This review and meta-analysis is registered with the International Prospective Register of Systematic Reviews (PROSPERO, CRD42019130314) and will be reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines . The PRISMA-P checklist can be found in Additional file 1.
A systematic literature search in the electronic databases PsycINFO, Medline/PubMed, Embase, and CENTRAL (Cochrane Central Register of Controlled Trials) with no restriction on dates of coverage and language will identify relevant studies. The sensitive search strategy employs a combination of search terms including the target diagnosis PTSD and the type of treatment delivery (i.e., Internet- or mobile-based). The predefined set of search strings can be found in Appendix 1.
A review of reference lists from identified studies (i.e., backward searches), a search of the WHO International Clinical Trials Registry Platform (ICTRP; Appendix 2) for ongoing studies with preliminary/unpublished results, and a citation search using Web of Science will complement the search. If necessary, study authors will be contacted to obtain further information regarding study characteristics. In case study protocols are identified without subsequent publication of results, protocol authors will be contacted to obtain missing or unpublished data and determine eligibility for inclusion in this review and meta-analysis.
Studies are eligible for inclusion if they focus on an adult target population (≥ 18 years; no upper limit) with PTSD or subthreshold PTSD assessed by standardized diagnostic interviews, observer-rated instruments with normed cutoff points, or validated self-reports (see Types of Outcomes). Studies of participants with different mental disorders (e.g., generalized anxiety disorder) will be included only if PTSD symptom severity is defined as primary outcome or main diagnosis of interest and results are reported specific to the respective disorders.
Studies will be included if at least one trial arm constitutes of an Internet- and mobile-based intervention (IMI) targeting PTSD symptoms. IMIs are defined as self-guided/unguided or guided interventions delivered via Internet- and mobile-based communication technology with a psychological/psychotherapeutic focus (e.g., trauma-focused CBT, CPT-based interventions [49, 50]). Interventions may vary concerning to the amount of guidance provided to participants (i.e., pure self-help interventions as well as guided interventions with different amounts and types of human support will be included). The following variations of IMIs are all eligible for inclusion: (a) guided interventions (i.e., with regular therapist contact), (b) mostly unguided interventions (i.e., predominantly self-help, e.g., with additional technical guidance on demand), and (c) completely unguided interventions (i.e., self-help, with no therapeutic support but assessment at most ). Therefore, studies are eligible for inclusion if IMIs involved an initial face-to-face interview or an initial face-to-face session.
Eligible comparisons differ by each research question. First, to investigate the efficacy of IMIs for PTSD, studies must include an active control group. Second, to investigate intervention components, studies must classify as additive or dismantling design study (see the “Study type” section). Third, to investigate mediators and mechanisms of change, studies are eligible for inclusion if they compare an IMI group with an active (e.g., IMI, face-to-face, treatment as usual, or placebo) or non-active control group (e.g., wait-list or no intervention).
To be classified as bonafide  in this systematic review and meta-analysis, the treatment for the control groups has to aim to reduce symptom severity (non-inferiority trials). If the authors postulate a superiority of the IMI group, we are classifying the control group as non-bonafide. In case authors do not specify their hypotheses clearly, we will decide the allocation depending on the existence of empirical evidence for the efficacy of the control treatment.
Change in PTSD symptom severity has to be measured by scores on a standardized, observer-rated instrument, for example, the Clinician-Administered PTSD Scale for DSM-IV (CAPS ), or a validated self-report measure of PTSD symptoms, for example, the Posttraumatic Stress Disorder Checklist - DSM-5 Version (PCL-5 ) or precursor version. Negative values characterize effect sizes favoring the intervention group.
To depict the influence of the participants’ engagement with the online intervention, adherence will be operationalized as (a) the mean number of main intervention units completed (e.g., lectures, diaries) and (b) the percentage of participants that completed the whole treatment [54, 55].
Possible mediators (e.g., functional cognitions, mindfulness, or problem-solving skills) are to be measured by validated psychometric instruments (e.g., Posttraumatic Cognitions Inventory ). Hereby, mediators are defined as intervening variables that statistically account for the relationship between an independent (i.e., the treatment) and a dependent variable (i.e., the outcome ).
Timing of outcome assessment
To be able to meaningfully comment on any post-intervention symptom reduction, separate analyses based on different periods of assessment will be performed (e.g., immediately post-treatment and after the follow-up period). The resulting follow-up periods will be clustered in three time frames: short-term (1–3 months after post-treatment), medium-term (> 3 to ≤ 12 months after post-treatment), and long-term (> 12 months after post-treatment) effects. If studies report more than one follow-up assessment point, the longest follow-up period will be used to provide the best estimate of the crucial long-term outcomes of the PTSD intervention.
In order to investigate the efficacy of IMIs for PTSD and active intervention components, only randomized controlled trials (RCTs ) with active control groups or comparing IMIs to dismantled variations of the same intervention will be included. The latter allows to investigate specific effects of single components of IMIs by adding or subtracting specific elements . To study potential mediator variables, both original RCTs and secondary analyses on previous RCTs will be eligible. They have to include repeated measures and use well-established mediation analyses (e.g., ) or include a quantitative assessment of changes in investigated psychological mediators. Studies have to be written in English and either be published in peer-reviewed journals or classified as ongoing trials in ICTRP with already existing results.
Study selection process
Two independent reviewers (LS, AS) will conduct the selection of articles. First, one reviewer (LS) will screen all titles and abstracts of articles yielded by the database search. Second, full texts of the selected articles will be retrieved and screened by both reviewers (LS, AS) in terms of the aforementioned eligibility criteria. Third, reference lists of finally included articles will be screened in the same way. Disagreement will be resolved by discussion among reviewers or, if necessary, by consultation of a third reviewer (MD). To illustrate the study selection process and reasons for exclusion, a PRISMA-P flow chart  will be provided. Records will be managed using the literature management program CITAVI 6.
The following data items will be extracted by two independent reviewers (LS, AS) for each study: (a) study identification items (year of publication, first author, country), (b) study and intervention characteristics (inclusion and exclusion criteria, screening instruments, therapeutic background, sample size, intervention design/type, level of human support/guidance, control group, duration of intervention/number of sessions, follow-up assessments, recruitment strategy, communication mode, prompts, standardization), (c) target population items (gender, age, specific population groups if applicable, comorbidities, trauma type, type of PTSD), (d) setting (nationality, environment, recruitment strategy), (e) drop-out rate, (f) human support characteristics (e.g., qualification of e-coaches if any), (g) platform characteristics, (h) frequency of adverse events if assessed and reported in primary studies, and (i) dimensional clinical outcomes.
In case of overlapping studies or multiple studies on the same data set, all information will be extracted with a note, highlighting the shared data set. Should there be any data missing or the reported data unclear, study authors will be contacted and asked for further clarification.
In case of multiple outcome measures for the assessment of PTSD symptom severity, the primary outcome measure of the study will be selected. If outcomes are assessed by several instruments, data will be extracted as follows: (a) the primary outcome measure of the study will be prioritized; (b) in case of multiple outcome measures of the same hierarchical level, the most used outcome measure will be chosen for the meta-analysis; and (c) should the aforementioned steps be not be possible, we will randomly select one outcome measure.
Two independent reviewers (LS, AS) will assess the risk of bias using the Cochrane Collaboration’s tool for assessing risk of bias in RCTs  in order to evaluate the quality of included studies. As recommended, each study will be rated in the following domains: (a) random sequence generation, (b) allocation concealment, (c) blinding of participants and personnel, (d) blinding of outcome assessment, (e) incomplete outcome data, and (f) other bias (e.g., study has been claimed to be deceitful, baseline differences between intervention and control group). Studies will be rated as showing an “unclear”, “low”, or “high” risk of bias on each domain. Inter-rater reliability will be calculated by means of Cohen’s kappa, whereby a value between .60 and .80 can be considered as substantial and a value > .80 as perfect . It is important to note that the third domain on blinding of participants and personnel is not warranted in (guided) IMIs, which would result in a high risk of bias rating. To prevent a distorted rating, we will rate this domain as “unclear” if this is the case.
To examine possible publication bias, the trim and fill procedure [63, 64], Egger’s test of funnel plot asymmetry , and visual inspection of funnel plots will be utilized.
Additionally, mediation studies will be qualitatively rated with the following criteria, originally proposed by Kazdin : (a) underlying theory or conceptual model, (b) RCT and inclusion of a control group, (c) sufficient sample size per condition (defined as n ≥ 40), (d) examination of multiple mediators in one study, (e) assessment of temporality (defined as ≥ 3 assessments in treatment phase), and (f) direct experimental manipulation of the mediator.
Data synthesis and presentation
Both text and tables will provide a detailed description of the results for all included studies. Characteristics of selected studies will be listed and qualitatively described. Characteristics include (a) study design and characteristics (sample size, duration, follow-up period) and patient population (age, gender, specific population group), (b) intervention characteristics (name, intervention content (e.g., CBT)), (c) technical implementation (e.g., Internet-only or Internet- and mobile-based), (d) duration, (e) level of human support/guidance, (f) study and intervention drop-out rate, (g) assessment tool used to determine presence of PTSD (clinical interview, questionnaire), (h) recruitment procedure, and (i) any covariates assessed (list of variables).
To assess the efficacy of IMIs compared to active control conditions and the efficacy of their intervention components, data analyses will be performed using the Review Manager 5.3 software developed by the Cochrane Collaboration . Meta-analytic pooling will be conducted when at least three studies report outcome parameters. If applicable, random-effects meta-analyses will be used to compute overall estimates of treatment outcomes. Standardized mean differences (SMD) and their 95% confidence intervals (CI) will be calculated for all continuous outcomes. As Hedges’ g is less biased than Cohen’s d in small samples , SMD will be given as values of the former measure. Dichotomous outcomes will be analyzed using risk ratios and corresponding 95% CIs and completer rate using odds ratios and their 95% CIs.
Heterogeneity will be evaluated with the I2 statistic, whereby an I2 value < 40% may not be important, a value between 30 and 60% may represent moderate heterogeneity, between 50 and 90% substantial heterogeneity, and > 75% considerable heterogeneity .
A forest plot will be created and used to visually investigate the presence and nature of statistical heterogeneity. In addition, statistical heterogeneity of the effect sizes will be evaluated using the Q statistic , with a significant Q indicating heterogeneity across studies that warrants further exploration.
For further comparisons of intervention and study characteristics (concerning, e.g., content, form, therapeutic background, guidance, trauma type, study quality, target population, and active intervention components), subgroup analysis will be performed if feasible. The possible influence of publication bias will be determined by inspection of funnel plots if feasible . Feasibility will be given with at least three trials per subgroup for subgroup analysis and at least ten trials to determine possible influence of publication bias .
Sensitivity analyses will be conducted to test the robustness of the results by comparing the pooled SMD of the different times of outcome assessment (see the “Timing of outcome assessment” section): (1) short term, (2) medium term, and (3) long term. Further sensitivity analyses will be conducted to examine the effect of including studies at high risk of bias. Should a quantitative synthesis not be appropriate, results will be summarized qualitatively.
To evaluate and synthesize evidence for possible mediators, a two-stage structural equation modeling (TSSEM) approach employing R will be used if appropriate [68, 69]. This approach combines meta-analytic techniques and structural equation modeling (SEM) and has been considered superior to conventional approaches to synthesize studies that use SEM (e.g., Pearson correlations, generalized least squares ). Means, SD, t statistics, F statistics, and effect sizes will be used to calculate bivariate correlations if studies do not report bivariate correlation between treatment, mediator, and outcome . Should a quantitative synthesis not be appropriate, results will be summarized qualitatively by summarizing the (significant) mediators found.
To measure the confidence in the cumulative evidence, the strength of the body of evidence will be assessed using the GRADE approach . This system classifies the quality of evidence in four levels ranging from “very low” and “low” to “moderate” and “high” (influenced by, e.g., study limitations, inconsistency of results) and offers “strong” and “weak” as grades of recommendation (influenced by, e.g., quality of evidence, uncertainty, or variability in values and preferences ).