{"id":"W2073059260","doi":"10.1016/j.jclinepi.2014.01.006","title":"Applying Grading of Recommendations Assessment, Development and Evaluation (GRADE) to diagnostic tests was challenging but doable","year":2014,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":164,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"European Commission","keywords":"Grading (engineering); Operationalization; Quality of evidence; Test (biology); Systematic review; Publication bias; Perspective (graphical); Evidence-based medicine; MEDLINE; Medicine; Medical physics; Psychology; Meta-analysis; Computer science; Alternative medicine; Artificial intelligence; Pathology","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":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5984932,0.0001806069,0.006379001,0.0004385746,0.0001357817,0.00006029952,0.0007210885,0.0001409984,0.0006153567],"category_scores_gemma":[0.7603567,0.00009435984,0.001007214,0.0003794931,0.00006359112,0.0001772206,0.0001276879,0.0004167313,0.00007001351],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004328226,"about_ca_system_score_gemma":0.0001819253,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004705259,"about_ca_topic_score_gemma":0.00001395485,"domain_scores_codex":[0.8713415,0.0781678,0.04630759,0.0008604177,0.00289345,0.0004292727],"domain_scores_gemma":[0.3907606,0.5723894,0.03252717,0.001371946,0.00235013,0.0006008493],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000006316573,0.00009812089,0.442621,0.00003585567,0.0002398157,9.768706e-7,0.0001357074,0.002070402,0.00002220852,0.004568914,0.0121854,0.5380153],"study_design_scores_gemma":[0.0008288519,0.0004024941,0.680961,0.0003872849,0.0004913827,0.0000509102,0.0006324674,0.07565992,0.00002101847,0.0710448,0.1693057,0.0002141799],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6105593,0.0005623284,0.3694519,0.01647481,0.001029777,0.0006800773,0.00000144733,0.000002003748,0.001238406],"genre_scores_gemma":[0.8300625,0.0002243753,0.1682806,0.001045096,0.0002345829,0.00003891084,0.000001722509,0.000006211711,0.0001059724],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5378011,"threshold_uncertainty_score":0.6737724,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9248872393294887,"score_gpt":0.7005075891732893,"score_spread":0.2243796501561994,"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."}}