{"id":"W3004577559","doi":"10.1016/j.jclinepi.2019.12.021","title":"GRADE guidelines: 21 part 2. Test accuracy: inconsistency, imprecision, publication bias, and other domains for rating the certainty of evidence and presenting it in evidence profiles and summary of findings tables","year":2020,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":364,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ottawa Public Health; McGill University; McMaster University Medical Centre; McMaster University","funders":"European Commission","keywords":"Certainty; Operationalization; Test (biology); MEDLINE; Evidence-based medicine; Research design","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4311213,0.0001965624,0.00532224,0.0001795875,0.00009147789,0.0001118409,0.0009325577,0.0002205295,0.0000871291],"category_scores_gemma":[0.9745336,0.00007807885,0.0008044839,0.0005856739,0.0004796481,0.0004897067,0.0003037757,0.0004501938,0.000001255646],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000007496188,"about_ca_system_score_gemma":0.0002035653,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004434658,"about_ca_topic_score_gemma":0.00005718354,"domain_scores_codex":[0.9035581,0.03734572,0.05553922,0.001184176,0.001945172,0.0004275915],"domain_scores_gemma":[0.05130843,0.9051943,0.03818933,0.001086208,0.003902014,0.0003196893],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006285015,0.00002691508,0.937952,0.0002054007,0.00007418855,6.781735e-7,0.0003102075,0.00003954608,0.0001746125,0.000594653,0.04329006,0.0172689],"study_design_scores_gemma":[0.001731937,0.001301663,0.7372087,0.005836436,0.0007753865,0.0001245471,0.004315526,0.1103549,0.00013369,0.05161525,0.08623413,0.0003678523],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7917432,0.01446528,0.008254199,0.1843342,0.0001548523,0.0009867589,0.00002399255,0.000001537062,0.00003601353],"genre_scores_gemma":[0.9469834,0.006388499,0.03978589,0.006397442,0.0003162576,0.00002322581,9.981757e-7,0.00001003433,0.0000942324],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8678486,"threshold_uncertainty_score":0.5857803,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9592689633150591,"score_gpt":0.668763848994735,"score_spread":0.2905051143203241,"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."}}