{"id":"W2770107865","doi":"10.1136/bmjopen-2017-018593","title":"UpToDate adherence to GRADE criteria for strong recommendations: an analytical survey","year":2017,"lang":"en","type":"article","venue":"BMJ Open","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":39,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Medicine; Family medicine; Epidemiology; Public health; MEDLINE; Medical education; Nursing; 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":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.02006152,0.0001796731,0.0004589971,0.00007222388,0.002795414,0.0004489635,0.001931092,0.0001945711,0.00144248],"category_scores_gemma":[0.01643708,0.0001699755,0.00004281833,0.00009739745,0.00006082909,0.001947529,0.0009700896,0.0004970137,0.0008654839],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001585224,"about_ca_system_score_gemma":0.001125257,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03423761,"about_ca_topic_score_gemma":0.05800968,"domain_scores_codex":[0.9923853,0.005133362,0.0009489266,0.0005768721,0.0002538148,0.0007017188],"domain_scores_gemma":[0.9936501,0.002500509,0.0007504616,0.001959052,0.0005643236,0.0005755889],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002149953,0.0004853613,0.4940298,0.0008331665,0.0001068167,0.00000988396,0.006817545,0.000001685287,0.0001143035,0.0208415,0.4367296,0.03788028],"study_design_scores_gemma":[0.001780839,0.0007969867,0.8883024,0.0008045345,0.00003520207,0.000001614304,0.004583321,0.001382547,0.00002912716,0.002310001,0.09945012,0.0005232687],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6758176,0.00001497911,0.007462675,0.2356096,0.006972013,0.05582897,0.002879194,0.0001866029,0.01522837],"genre_scores_gemma":[0.9460582,0.00001487303,0.02203819,0.01475101,0.0009941347,0.009315499,0.0005029038,0.00007923879,0.006245968],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3942726,"threshold_uncertainty_score":0.9999124,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9530837414453088,"score_gpt":0.7578939457473015,"score_spread":0.1951897956980073,"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."}}