Borrowing Strength: How Meta-Analytic Predictive Priors Can Accelerate Mental Health Treatment Research [Data & Materials]

This repository reproduces the three use cases and the within-group effective sample size (ESS) table reported in the paper. Robust Meta-Analytic-Predictive (MAP) priors are fit with a random-effects model in JAGS and a HalfNormal(0, 0.5) prior on the between-study heterogeneity tau, robustified with a 20% vague mixture component, and transformed into power curves with RBesT. Repository structure 📁 _zenodo/ 📄 00-analysis.R - main script: runs all analyses, writes figures/ and results/ 📄 01-sensitivity.R - sensitivity analysis of the operating characteristic for examples 1 and 2 📁 _utils/ 📄 prep-smc.R - run-first helper: builds standardized-change.csv from raw data 📁 data/ 🗃️ depression-data.csv - Metapsy depression database export (study/arm-level) 🗃️ sd-reference.json - reference-SD catalogue per instrument (for standardization) 🗃️ standardized-change.csv - arm-level standardized mean change (SMC); derived, shipped ready-made 📁 figures/ 📈 power.pdf - Figure 2: superiority (a) + non-inferiority (b) power curves 📈 update.pdf - Figure 3: single-study shrinkage ("meta-analysis with one study") 📈 sensitivity.pdf - Results of sensitivity analyses for use cases 1 and 2 (Type I error rate and power) 📁 results/ - CSV outputs written by 00-analysis.R 📄 README.md Requirements JAGS >= 4.3 (system install; https://mcmc-jags.sourceforge.io) R >= 4.1 with packages: tidyverse, RBesT, metafor, rjags, metapsyTools, jsonlite, and (optional) patchwork for the combined Figure 2. metapsyTools is installed from the Metapsy project: remotes::install_github("metapsy-project/metapsyTools"). How to run source("00-analysis.R") The script reads data/standardized-change.csv directly. If that file is missing it regenerates it from data/depression-data.csv and data/sd-reference.json by sourcing _utils/prep-smc.R. On completion it writes: figures/power.pdf, figures/update.pdf results/ess-summary.csv (use cases 1-2) results/single-study-update.csv (use case 3) results/within-group-ess.csv (Table 1 within-group block) A fixed seed (set.seed(123)) is set, but MAP fits use MCMC, so figures and the third/fourth decimal of some estimates may differ negligibly between runs and across JAGS versions. What each step does Step Paper element Description Use case 1 Fig. 2a Superiority trial of life-review therapy vs waitlist; the waitlist comparator arm borrows the robust MAP. Powered at the patient-derived smallest worthwhile difference (SWD = 0.47 SMD; Sahker et al. 2025). Use case 2 Fig. 2b Non-inferiority trial of a new therapy vs CBT in postpartum depression (margin 0.24 SMD; Cuijpers et al. 2014); the CBT comparator borrows the robust MAP. Use case 3 Fig. 3 "Meta-analysis with a single study" (van Zwet et al. 2025): an enthusiastic pilot (g = 1.10, n = 15/arm) is fused with a MAP prior pooled via the influence model, shrinking the inflated effect. Table 1 within-group block Within-group (pre-post) ESS ceiling 1/tau^2 per condition, tau from the influence model. The between-group block of Table 1 is sourced from published network meta-analyses (Cipriani 2018; Schneider-Thoma 2022; Barth 2013; Hoppen 2023; Papola 2024) and is not recomputed here. Notes Risk-of-bias filters. Use case 2 and use case 3 restrict to the lowest risk-of-bias rating (rob == 4), matching the paper's "lowest risk of bias" / "low risk of bias evidence only" wording. Remove the rob == 4 filter in 00-analysis.R to use all qualifying studies. Within-group conditions. within.conditions is set to c("cbt", "wl", "cau"); extend this vector to reproduce additional rows of Table 1. Data provenance. depression-data.csv is a snapshot of the Metapsy depression psychotherapy database (01-09-2025 update). sd-reference.json is the SD-reference catalogue (Fábián, Cuijpers & Harrer 2026).