Replication package for: Prenatal Cannabis Exposure and Adverse Neonatal Outcomes. A Systematic Review and Meta-Analysis with Meta-Regression Controlling for Tobacco Co-use

Everything needed to reproduce every number in the manuscript "Prenatal Cannabis Exposure and Adverse Neonatal Outcomes: A Systematic Review and Meta-Analysis with Meta-Regression Controlling for Tobacco Co-use", its supplement and its figures. The review searched eleven sources to 20 September 2026 and is registered as PROSPERO CRD420251065325. One hundred and three reports met the inclusion criteria; 79 studies from 75 independent source populations contribute pooled data across nine neonatal outcomes. The package contains the study-level analysis dataset with a provenance note for every non-obvious entry, the exposure-ascertainment coding, the full screening ledger, item-level Newcastle-Ottawa assessments for all 79 contributing studies, arm-level participant characteristics, and eight analysis scripts. It also contains the three result files the manuscript was written from, and a verification script that compares a fresh run against them value by value and reports any difference. The meta-analysis engine is self-contained. Restricted maximum likelihood estimation, the Hartung-Knapp-Sidik-Jonkman variance adjustment, Egger's regression and Duval-Tweedie trim-and-fill are implemented directly, so results do not depend on the version of any third-party meta-analysis package. Because of that, the whole analysis was also re-run in R with metafor on the same data. Eighty-one quantities were compared, covering every pooled estimate and interval, heterogeneity for all nine outcomes, Egger's p value and the trim-and-fill count wherever k is at least ten, and the slope and p value of every tobacco meta-regression including the size-adjusted models. All eighty-one agree to within restricted-maximum-likelihood convergence tolerance. Both implementations are included, along with the script that compares them. The central analysis regresses each study's effect on the difference in tobacco prevalence between its two arms rather than on tobacco prevalence in the exposed arm, on the reasoning that a study whose arms both smoke heavily is not confounded by tobacco while one where only the cannabis arm smokes is. No estimate that was itself adjusted for tobacco enters any pooled result. No individual participant data are included. Every value was extracted from published reports, all of which are cited in the manuscript. Requires Python 3.11 or later with numpy, scipy and pandas. The optional cross-check requires R with metafor and jsonlite. Version 1.0.1 corrects four things in version 1.0.0 and changes nothing in the dataset and nothing in any pooled estimate. The package's meta-regression script read the exposed-arm tobacco covariate from an arm-level file compiled for the original submission, which held no row for 44 of the 79 contributing studies, so that covariate was fitted on a subset; it now comes from the same table the between-arm differential is built from. Three data files, the arm-level characteristics, the participant characteristics and the study metadata, were also of that earlier vintage and are now derived from the published supplement and from the paper's own citation data. The sample-size file now carries analysed totals, which the participant counts in the GRADE profile depend on. And the within-study analysis no longer pools the four adjusted-versus-unadjusted pairs whose unadjusted estimate is not on the harmful side of the null, where the ratio of the two cannot measure attenuation; the pooled ratio moves from 0.786 to 0.782 and both are reported. A full list is in the README.