{"id":"W2884485345","doi":"10.1111/jep.12993","title":"What confidence should we have in <scp>GRADE</scp>?","year":2018,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; University of Toronto","funders":"","keywords":"Bayes' theorem; Confidence interval; Bayesian probability; Frequentist inference; Set (abstract data type); Evidence-based medicine; Quality (philosophy); Hierarchy; Medicine; Computer science; Bayesian inference; Artificial intelligence; Epistemology; Alternative medicine; Internal 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":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2360692,0.001672761,0.005468105,0.01050458,0.004702702,0.01849931,0.01092637,0.01784883,0.007742477],"category_scores_gemma":[0.6466197,0.001634364,0.004353398,0.008869656,0.01863154,0.0186326,0.006787322,0.02278178,0.006863956],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02023331,"about_ca_system_score_gemma":0.02984324,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02025942,"about_ca_topic_score_gemma":0.01436257,"domain_scores_codex":[0.7807929,0.1079876,0.04658911,0.007623657,0.05403656,0.00297015],"domain_scores_gemma":[0.4571486,0.2919005,0.04513312,0.03153608,0.1576463,0.01663537],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001302991,0.0002052402,0.008894102,0.02282757,0.001592033,0.0004110044,0.003559982,0.002510373,0.00104108,0.1556158,0.4586628,0.3433771],"study_design_scores_gemma":[0.000883496,0.0007043696,0.008832908,0.06229696,0.002088312,0.001608336,0.003376674,0.006279604,0.00313307,0.3935227,0.5161545,0.001119181],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.004563333,0.06987002,0.04388703,0.8000004,0.03595751,0.001644226,0.002309966,0.0005680712,0.04119939],"genre_scores_gemma":[0.2882707,0.06203618,0.2247161,0.3731018,0.02792466,0.009437922,0.003267181,0.001136551,0.01010879],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7639308,"threshold_uncertainty_score":0.9420627,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9160063757252691,"score_gpt":0.7096496822924008,"score_spread":0.2063566934328683,"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."}}