{"id":"W4416840645","doi":"10.1136/bmjmed-2025-002024","title":"Reproducibility of meta-analytic results in systematic reviews of interventions: meta-research study","year":2025,"lang":"en","type":"article","venue":"BMJ Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Health and Medical Research Council; Australian Research Council","keywords":"Systematic review; Systematic error; Code (set theory); Reproducibility; Data extraction; Patient data","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8872292,0.0003842239,0.02703194,0.002739267,0.00006249244,0.00006532131,0.00298417,0.00007658802,0.007587507],"category_scores_gemma":[0.8503987,0.0001366381,0.007737778,0.009890272,0.0003202643,0.0001446361,0.0004124751,0.0003468989,0.0002418569],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000641693,"about_ca_system_score_gemma":0.0001322086,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004349507,"about_ca_topic_score_gemma":0.0007770042,"domain_scores_codex":[0.3357251,0.4272384,0.193153,0.01016179,0.03267015,0.001051646],"domain_scores_gemma":[0.7546444,0.1090394,0.04497908,0.07620425,0.01479617,0.0003367321],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0003656253,0.007346568,0.03599042,0.3055679,0.1724643,0.00006896741,0.01648271,0.0003348783,0.0003859627,0.01071864,0.4451638,0.005110179],"study_design_scores_gemma":[0.007664399,0.003703262,0.06881712,0.05367411,0.6979706,0.0000290267,0.07664676,0.004549372,0.0004322965,0.06860588,0.01683659,0.00107061],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.1756653,0.5527884,0.01160155,0.0424068,0.001953056,0.133861,0.0001097358,0.00002770979,0.08158644],"genre_scores_gemma":[0.9721137,0.0001359609,0.0006575354,0.00007008394,0.00003278405,0.001495232,0.000003920512,0.000007365725,0.0254834],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7964484,"threshold_uncertainty_score":0.9933197,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9716707446074663,"score_gpt":0.7243146200242608,"score_spread":0.2473561245832054,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}