{"id":"W1525076802","doi":"10.1111/j.1553-2712.2011.01214.x","title":"Consensus Conference Follow-up: Inter-rater Reliability Assessment of the Best Evidence in Emergency Medicine (BEEM) Rater Scale, a Medical Literature Rating Tool for Emergency Physicians","year":2011,"lang":"en","type":"article","venue":"Academic Emergency Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"National Center for Research Resources; National Cancer Institute; McMaster University","keywords":"Medicine; Inter-rater reliability; Rating scale; Reliability (semiconductor); Emergency department; Scale (ratio); Family medicine; Medical emergency; Emergency medicine; Psychiatry","routes":{"ca_aff":true,"ca_fund":true,"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":[0.2416077,0.001671365,0.005162084,0.01532838,0.003723803,0.004100591,0.004113059,0.003328085,0.004898767],"category_scores_gemma":[0.4618257,0.001337279,0.008326422,0.008598943,0.001813548,0.003760584,0.005545177,0.004647608,0.002419217],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003972751,"about_ca_system_score_gemma":0.01266186,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002171812,"about_ca_topic_score_gemma":0.004441119,"domain_scores_codex":[0.7318919,0.1235476,0.0995851,0.01176044,0.03038976,0.002825256],"domain_scores_gemma":[0.3731945,0.1857531,0.07986068,0.02344179,0.3332418,0.004508251],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01053428,0.002270571,0.132211,0.07465269,0.01237709,0.001449834,0.07599516,0.006576553,0.02056176,0.01035247,0.1472653,0.5057533],"study_design_scores_gemma":[0.01490218,0.008337502,0.3931966,0.05413881,0.01164019,0.003697412,0.04013309,0.07240862,0.03481871,0.03556709,0.3277027,0.003457118],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2898078,0.01844781,0.3259887,0.01233355,0.008595774,0.2745438,0.01654588,0.003630082,0.0501067],"genre_scores_gemma":[0.2905707,0.003133545,0.464847,0.002021244,0.0008537886,0.229616,0.005421674,0.0005074731,0.003028735],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7583923,"threshold_uncertainty_score":0.9352327,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6183455596864168,"score_gpt":0.5378222126301103,"score_spread":0.08052334705630648,"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."}}