{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5034562,0.0004578794,0.007214568,0.0004954636,0.0001436705,0.00001420638,0.004031315,0.0001062653,0.04590312],"category_scores_gemma":[0.3878448,0.0001295851,0.001955821,0.007258487,0.0001928375,0.0002034898,0.0001029222,0.0003412475,0.0001754163],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000007143753,"about_ca_system_score_gemma":0.00003358796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002124504,"about_ca_topic_score_gemma":0.0001857228,"domain_scores_codex":[0.8535004,0.0737299,0.05617231,0.001225659,0.01477131,0.0006004773],"domain_scores_gemma":[0.9450225,0.01307743,0.02708325,0.008939433,0.005708525,0.0001689335],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005212132,0.0002534169,0.03533024,0.02422912,0.0004588712,0.000003928852,0.03830414,0.00005511902,0.001817328,0.01025447,0.8849903,0.00425091],"study_design_scores_gemma":[0.002984619,0.0053041,0.2629137,0.1040986,0.005398181,0.00008358977,0.07793192,0.01152751,0.00008426357,0.2148642,0.3128296,0.001979734],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.3163191,0.5532575,0.001546445,0.06086043,0.004757687,0.008581677,0.00001752767,0.000007777895,0.05465178],"genre_scores_gemma":[0.8875086,0.08553447,0.0001632473,0.001162254,0.001192212,0.0004165233,0.00002404334,0.0000263982,0.02397221],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5721608,"threshold_uncertainty_score":0.954969,"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."}}