{"id":"W4416435674","doi":"10.1016/j.jcrc.2025.155358","title":"Can large language models approximate the results of meta-analyses in critical care? A meta-research study","year":2025,"lang":"en","type":"article","venue":"Journal of Critical Care","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa; Western University; St. Michael's Hospital","funders":"","keywords":"MEDLINE; Critical illness; Evidence-based medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3122171,0.006667251,0.0156504,0.004397016,0.001244552,0.00862256,0.007625589,0.005937511,0.005210675],"category_scores_gemma":[0.6953067,0.003970235,0.02786611,0.0063123,0.003402726,0.01625754,0.004011916,0.008882876,0.001010628],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002928507,"about_ca_system_score_gemma":0.004001536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007759626,"about_ca_topic_score_gemma":0.006284513,"domain_scores_codex":[0.587977,0.3820901,0.01179026,0.01097549,0.005884627,0.001282616],"domain_scores_gemma":[0.1438495,0.8181874,0.01015204,0.02416546,0.003219635,0.0004259876],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.02358292,0.0006350983,0.02826898,0.03566499,0.6090519,0.001257065,0.001806606,0.1242043,0.0007559231,0.04024979,0.008544922,0.1259774],"study_design_scores_gemma":[0.01114446,0.002859843,0.007760366,0.006524301,0.3373274,0.00133497,0.000594561,0.259904,0.0008916335,0.3631875,0.007951274,0.0005197526],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05188985,0.1821478,0.7434888,0.01121434,0.002330701,0.001579958,0.001473652,0.002174186,0.003700594],"genre_scores_gemma":[0.7905801,0.01737275,0.1766088,0.007944167,0.001169795,0.002898836,0.001076223,0.0009746067,0.001374847],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6877829,"threshold_uncertainty_score":0.8481588,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9054250625592039,"score_gpt":0.6908616910681487,"score_spread":0.2145633714910552,"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."}}