{"id":"W2461165364","doi":"10.1186/s12885-016-2408-9","title":"Use and misuse of common terminology criteria for adverse events in cancer clinical trials","year":2016,"lang":"en","type":"article","venue":"BMC Cancer","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"ca_institutions":"Princess Margaret Cancer Centre; University of Toronto","funders":"Fudan University","keywords":"Common Terminology Criteria for Adverse Events; Medicine; Clinical trial; Adverse effect; Terminology; MEDLINE; Randomized controlled trial; Neutropenia; Intensive care medicine; Family medicine; Toxicity; 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":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08779406,0.0001543421,0.003791867,0.0001902718,0.00002481018,0.00002683324,0.0005234058,0.0001163241,0.004804257],"category_scores_gemma":[0.063673,0.000062042,0.001007848,0.0002326998,0.0000930481,0.0002032042,0.00008242262,0.00004533464,0.00004466165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003514457,"about_ca_system_score_gemma":0.0001379777,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003325657,"about_ca_topic_score_gemma":0.005265899,"domain_scores_codex":[0.9779184,0.01147452,0.008969027,0.0006332893,0.0008112898,0.0001934606],"domain_scores_gemma":[0.9738961,0.02040355,0.003951351,0.001241538,0.0004112002,0.00009625194],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001387128,0.00005182532,0.9285392,0.0000572051,0.00008784175,5.039233e-7,0.00007913038,0.000003094915,0.0004547921,0.0001057361,0.01985404,0.05062796],"study_design_scores_gemma":[0.002479499,0.00008016229,0.8888909,0.0003203594,0.0003441518,0.000001599194,0.00008496184,0.0008307838,0.0001961114,0.002572739,0.1040235,0.0001752808],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9945208,0.001068508,0.001890906,0.0005538597,0.0007219705,0.000992924,0.0002111112,0.000001587347,0.00003826389],"genre_scores_gemma":[0.9910656,0.0006652789,0.001803658,0.0002323175,0.0001637553,0.0003453292,0.000001221838,0.000009396671,0.005713495],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08416942,"threshold_uncertainty_score":0.9961055,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9482729871807495,"score_gpt":0.6979188045437343,"score_spread":0.2503541826370151,"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."}}