{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.6640125,0.0002187752,0.002808703,0.0007333191,0.00007157615,0.001468099,0.001432578,0.0002483463,0.004514645],"category_scores_gemma":[0.8688986,0.0001172293,0.001242125,0.001593124,0.0001918273,0.004408023,0.0001126975,0.001126591,0.002302357],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001135882,"about_ca_system_score_gemma":0.000537024,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001115757,"about_ca_topic_score_gemma":0.000166619,"domain_scores_codex":[0.8506932,0.08778629,0.03938677,0.001101504,0.02061217,0.0004200505],"domain_scores_gemma":[0.4737909,0.4411206,0.05525518,0.004278196,0.02496275,0.0005923383],"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":[0.0002260844,0.002270007,0.2097798,0.00005705261,0.0006476611,0.0002233422,0.008922307,0.002082754,0.00009860855,0.004512444,0.5143753,0.2568046],"study_design_scores_gemma":[0.002228184,0.0007960357,0.108997,0.0006341267,0.0009052629,0.000296186,0.03365428,0.07141367,0.00003106022,0.09376761,0.6871356,0.0001410501],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7754228,0.01727859,0.02577194,0.1165737,0.0155853,0.002718456,0.00000329837,0.000005635093,0.04664025],"genre_scores_gemma":[0.9755923,0.002737883,0.01362814,0.004241237,0.001253155,0.0000139258,6.956479e-7,0.00001321022,0.002519511],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3769023,"threshold_uncertainty_score":0.9995685,"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."}}