Open materials: Network meta-analysis of psychosocial interventions for childhood maltreatment

Analysis code, input data, and outputs for a frequentist network and pairwise meta-analysis of psychological/psychosocial interventions on social functioning in people with childhood maltreatment (CM). Follows the published protocol (Fares-Otero et al., 2025, Eur J Psychotraumatol; PROSPERO CRD42022347034). All code is self-contained and uses relative paths. Download the folder and run from its root. Folder structure R/: analysis scripts (run in numeric order) 00-setup.R: data load, node/domain labels, RoB D1 exclusion, helpers 01-nma-main.R: primary per-domain NMA (7 outcome domains) 02-pairwise.R: conventional pairwise meta-analysis 03-moderators.R: subgroup / moderator NMAs (categorical + median-split) 04-sensitivity.R: sensitivity scenarios (RoB, blinding, ITT, post-hoc) 05-dropout.R: acceptability (all-cause dropout) NMA 06-smallstudy.R: comparison-adjusted funnels + Egger test (per domain) 07-report-tables.R: assemble publication tables into report-tables.xlsx + CSVs 08-build-word.py: render the CSV tables into a formatted Word document 09-followup.R: follow-up analyses (NMA where feasible, else pairwise) 10-transitivity-cinema.R: transitivity table + CINeMA input files 11-overall.R: pooled "overall social functioning" composite NMA (+ its small-study/Egger screen and CINeMA input) run-all.R: master runner (sources the R scripts in order) data/: input data (see "Data" below) results/: all generated outputs (see "Outputs") environment/: R version, package versions, sessionInfo, install helper Data Three input files, all read by the pipeline: File Used by Contents data/data_clean_mod_drop_corrected.xlsx all primary analyses Contrast-level SMDs (TE/seTE), 6 nodes, 7 outcome domains, moderators, dropout counts. Sheets nma* (collapsed nodes) and pairwise* (specific labels); names auto-detected. data/robvis_corrected.xlsx 00-setup.R RoB 2 domain-level and overall judgments (sheet robvis). data/complete-dataset_corrected.xlsx 09-followup.R Arm-level data used to reconstruct follow-up time bands. Nodes (6): trauma-focused CBT (TF-CBT), non-trauma-focused CBT (CBT), relational/insight/mentalization (REL), mindfulness/somatic/community (MIND), waitlist (WL, reference), treatment as usual (TAU). Outcome domains (7): global functioning, interpersonal/social relations, emotion regulation, emotional processing/social cognition, quality of life, internalizing behavior, externalizing behavior. Effect measure: SMD (Cohen's d), random-effects, graph-theoretical NMA (netmeta). Positive SMD = better functioning (directions pre-aligned). Key analysis choices (per protocol): Primary timepoint = post-treatment. §2.9 exclusion of studies at high risk of bias in randomization (RoB 2 domain 1) from primary analyses, retained as a sensitivity scenario. Ranking by P-scores plus mean ranks and rankograms. Inconsistency via design-by-treatment (global) and SIDE/node-splitting (local). Small-study effects via comparison-adjusted funnel + Egger-type test for networks with ≥10 studies. Outputs (results/) nma/ per-domain NMA: results.xlsx, forest/netgraph/rankogram.pdf, summary.txt + 00-nma-summary.xlsx (all domains) pairwise/ node-level & specific-intervention pairwise MA overall/ pooled "overall social functioning" composite NMA + age/clinical subgroups moderators/ subgroup NMAs + feasibility + moderator distribution sensitivity/ effect/ranking robustness across exclusion scenarios dropout/ acceptability (dropout) NMA smallstudy/ comparison-adjusted funnel plots + Egger tests (incl. overall-funnel.pdf) followup/ follow-up NMA/pairwise by time band cinema/ per-domain + overall CINeMA input files (L/M/H) + README (margins) report/ report-tables.xlsx + transitivity + NMA_results_tables.docx (publication tables) CINeMA and RoB-MEN are run outside R (in the CINeMA web app); the inputs are prepared in results/cinema/. Environment R 4.5.2 (see environment/R-version.txt, environment/sessionInfo.txt). CRAN packages (pinned versions in environment/package-versions.csv): dplyr 1.1.4, tidyr 1.3.2, stringr 1.6.0, readxl 1.4.5, openxlsx 4.2.8.1, netmeta 3.2.0, meta 8.2.1. Reproduce the library with Rscript environment/install-packages.R (uses remotes::install_version to match the pinned versions). Optional (Word report only): Python 3 with python-docx (pip install python-docx; see environment/requirements-python.txt). set.seed(2026) is set in 00-setup.R; rankogram simulations use fixed seeds, so results are reproducible run-to-run. Notes / deviations Meta-regression (netmetareg) was numerically unstable on these sparse networks; moderators are analysed as subgroup NMAs (continuous moderators split at the median, computed over unique trials). tol.multiarm = 0.1 is set on all netmeta() calls to accommodate pooled multi-arm trials whose independently computed contrasts carry ~0.02-0.03 SMD rounding inconsistency. Disconnected domain networks are reduced to their largest connected component.