{"id":"W2163839878","doi":"10.1067/mem.2001.115881","title":"Evaluating the quality of systematic reviews in the emergency medicine literature","year":2001,"lang":"en","type":"article","venue":"Annals of Emergency Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":81,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"University of Alberta","keywords":"Systematic review; Medicine; MEDLINE; Meta-analysis; Quality (philosophy); Confidence interval; Family medicine; Internal medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4330122,0.006602986,0.02973046,0.03538393,0.002408613,0.01043665,0.005098144,0.008053145,0.001677904],"category_scores_gemma":[0.8151273,0.006180365,0.04911495,0.01838954,0.005808502,0.007903129,0.00751929,0.00550802,0.0002352264],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01181256,"about_ca_system_score_gemma":0.0147377,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009216921,"about_ca_topic_score_gemma":0.01667352,"domain_scores_codex":[0.3343356,0.4192974,0.1962846,0.01395775,0.03434934,0.001775331],"domain_scores_gemma":[0.0731097,0.8460733,0.05296504,0.01137694,0.01529964,0.001175355],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"observational","study_design_scores_codex":[0.004436587,0.00008624388,0.01967913,0.1883719,0.7637886,0.0002513082,0.0007250595,0.001562111,0.0003518594,0.0004669017,0.0006669952,0.01961326],"study_design_scores_gemma":[0.003372064,0.001213785,0.01297082,0.03865968,0.9357095,0.0003638871,0.0003529004,0.001877235,0.0007243363,0.002839309,0.001732598,0.0001839346],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.03113965,0.94945,0.0100329,0.003487972,0.00157576,0.002401167,0.001191673,0.0001143292,0.0006066088],"genre_scores_gemma":[0.6687151,0.2768027,0.03948388,0.003401851,0.002904354,0.006368809,0.001810946,0.0001944861,0.0003179284],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9702696,"threshold_uncertainty_score":0.6991969,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9580746477098889,"score_gpt":0.7052984042000917,"score_spread":0.2527762435097972,"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."}}