Data and code for: Oral decoctions, patent medicines, and injections as adjunctive therapy for coronary microvascular disease: a systematic review and Bayesian indirect comparison

Companion data and code deposit for a systematic review and Bayesian indirect comparison of three delivery formats of Chinese medicine (oral decoctions, oral patent medicines, intravenous injections) as adjunctive therapy for coronary microvascular disease. PROSPERO CRD420261506455. Contents:- et1_corrected.csv — the five endothelin-1 trials exactly as entered into the model: trial, node, contrast d, standard error se, and a note recording where each value came from.- et1_corrected_results.json — posterior output of both fitted models (primary, and sensitivity excluding Zhao ZK 2025): estimate, 95% credible interval, R-hat and effective sample size for every node and for tau.- fit_et1.py — the Bayesian random-effects sampler that produced those two models.- make_fig3.py — draws Fig. 3 (forest plots for all four outcomes).- make_figures.py — draws Fig. 1 (PRISMA 2020 flow diagram), Fig. 2 (network plot) and Fig. 6 (graphical abstract).- README.md — reproduction instructions and data notes. Reproducing the endothelin-1 models: python3 fit_et1.py --selftest python3 fit_et1.py et1_corrected.csv python3 fit_et1.py et1_corrected.csv --drop "Zhao ZK 2025" python3 fit_et1.py et1_corrected.csv --tau-scale 7.61 The sampler is a Gibbs step for the treatment effects (conjugate normal) with a random-walk Metropolis step for log(tau), run as four chains of 120,000 iterations with 30,000 discarded as burn-in and thinned by 15 — 6,000 retained draws per chain, 24,000 in total. The model is a star-shaped random-effects network, d_i ~ Normal(theta_node(i), se_i^2 + tau^2), with theta ~ Normal(0, 10^4) and tau ~ HalfNormal(scale), the scale set to the empirical standard deviation of the observed contrasts (7.61 for endothelin-1). Data notes:Zhu W 2019 — an earlier extraction carried only the experimental arm's change (−25.80 ng/L) and omitted the control arm's improvement (−7.00 ng/L), which overstated that trial's effect by 7.00 ng/L. The value used here is the difference in changes, −18.80 ng/L.Zhao ZK 2025 — reports endothelin-1 in μmol/L where the other trials report ng/L; 16.99 μmol/L would be about seven orders of magnitude above the physiological range, so the unit is treated as a labelling error and the trial is retained in the primary model, with its exclusion reported as a sensitivity analysis.Yang J 2015 — no control-arm value was recorded, so it entered as an experimental-arm change. It is the only contributor to the injection node on this outcome. If its control arm did report endothelin-1 and improved at all during the trial, the injection estimate is correspondingly overstated and would move towards the null. The source report has not been re-read; this is flagged as a residual uncertainty in the paper.