Protocol and registration
The current study was registered with PROSPERO (CRD42019133534) [52] and reported per the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) [53].
Eligibility criteria
The population-intervention-comparison-outcome-study design (PICOS) model [54] was used to define review eligibility:
- Population: Adult (aged 18 and older) police, paramedics and emergency management technicians, correctional officers, dispatchers, 9-1-1 communication officers, fire and safety officers.
- Intervention: Any psychological interventions approved by the American Psychological Association to treat PTSD, delivered alone or in combination with medications.
- Comparison: Any comparator condition, such as waitlist controls or no intervention.
- Outcomes: Self-report measures (e.g., symptom scores on psychopathology measures), remission for PTSIs (e.g., PTSD), and objective indices of functioning (e.g., absenteeism, occupational performance ratings).
- Study design: English-language, peer-reviewed, randomized controlled trials (RCTs) of any duration from any geographic location published since 2008.
Information sources and search
A systematic search of Cochrane Central Register, EMBASE, MEDLINE, PsycINFO, PubMed was performed from January 1, 2008, to October 8, 2019 (Appendix 1). The search was supplemented by reviewing the reference lists of included studies and searching for ongoing RCTs from trial registries.
Study selection
There were two co-authors (AB and PD) who independently screened all articles by title/abstract and then full-texts, using Cochrane’s Covidence, a web-based systematic review manager [55]. After removing duplicate citations, the initial screening selection of papers was verified at the title/abstract stage by having multiple reviewers screen 200 reports, with 99% agreement. All discrepancies were removed by consensus and third-party input (GA).
Data collection process
There were two co-authors (AB and PD) who independently extracted relevant data from the published full-text reports of each included article using Covidence. Data extracted was verified by one other author (AB, PD, or GA).
Data items
The following data items were extracted: sample size, age, sex, comorbidity status, and years of employment in PSP profession; type and duration of psychological intervention; comparator group; all relevant outcome measures (e.g., symptom scores for PTSD and other PTSI on psychopathology measures); author, design, location, and study duration (i.e., timing of follow-up evaluations).
Risk of bias in individual studies
The Cochrane risk of bias tool for RCTs was used to assess study quality [56]. In brief, the Cochrane risk of bias tool appraises randomization, allocation concealment, blinding, selective reporting, attrition bias, and potential bias from funding. There were two co-authors (AB and PD) who independently appraised each included trial against the risk of bias tool, with discrepancies resolved by consensus.
Summary measures and synthesis of results
Random-effects meta-analysis models were created to pool effect sizes for each psychotherapy's effectiveness and acceptability, and the results were graphed using Forest plots. The specific methods have been previously described [57–62]. The I2 statistic was used to quantify heterogeneity; values greater than 50% indicated high variance [63]. Effect sizes were pooled across RCTs using standardized mean differences (SMD, Cohen’s d) or rate ratios (RRs) and their 95% confidence intervals (CI), depending on whether the data was continuous or categorical. The SMD was the difference in mean symptom severity scores at the end of intervention between groups divided by the standard deviation of the difference between groups. A negative SMD indicates an improvement in symptom severity relative to the control condition. All confidence intervals containing zero were nonsignificant, given that an SMD of zero is null. RRs greater than 1 indicated that the result favoured the experimental intervention over the control; 95% CIs containing an RR of 1 were nonsignificant.
Risk of bias across studies
Publication bias from the overrepresentation of studies with positive results was assessed using funnel plots [64] and Egger’s funnel plot symmetry test [66].
Additional analyses
Random-effects and fixed-effects models were compared to assess sensitivity to heterogeneity. The effect sizes determined from the end-of-intervention were also compared to the end-of-follow-up to determine intervention-time effects.