{"id":"W4396912353","doi":"10.1016/j.neuchi.2024.101566","title":"Reliability of study endpoint adjudication in a pragmatic trial on brain arteriovenous malformations","year":2024,"lang":"en","type":"article","venue":"Neurochirurgie","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta Hospital; Centre Hospitalier de l’Université de Montréal; Health Sciences Centre","funders":"","keywords":"Medicine; Clinical endpoint; Adjudication; Inter-rater reliability; Adverse effect; Clinical trial; Reliability (semiconductor); Pediatrics; Surgery; Internal medicine; Psychology; Rating scale; Developmental psychology","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":[],"category_scores_codex":[0.00823251,0.0001548465,0.0003482253,0.0004510974,0.00007410065,0.0001748949,0.0005219942,0.00005158223,0.0003252786],"category_scores_gemma":[0.009701512,0.0001105514,0.0001513294,0.001087551,0.00006606372,0.0003411378,0.00008397814,0.0002775746,0.0002972691],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009276145,"about_ca_system_score_gemma":0.0001403428,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002822247,"about_ca_topic_score_gemma":0.00005769993,"domain_scores_codex":[0.9948919,0.0009081785,0.001501229,0.0005736546,0.001913331,0.0002117011],"domain_scores_gemma":[0.9950802,0.003516383,0.0002201959,0.0009792125,0.0001397888,0.000064225],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.07265651,0.06939993,0.02309579,0.001540362,0.0002657984,0.000378928,0.0707227,0.04301908,0.02170593,0.03234005,0.09963685,0.5652381],"study_design_scores_gemma":[0.03681269,0.02137331,0.6537602,0.0006108443,0.0001136832,0.00001093078,0.006141628,0.0300365,0.001042128,0.1818458,0.06730283,0.0009494289],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9888766,0.0000321323,0.0005513845,0.004329209,0.0008514078,0.002027114,0.00001856329,0.0000542573,0.003259316],"genre_scores_gemma":[0.9992356,0.000005855726,0.00009598917,0.0002805776,0.00006025013,0.0001494167,0.00000385309,0.000008459981,0.0001599813],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6306644,"threshold_uncertainty_score":0.9986402,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09841165656943207,"score_gpt":0.3866456987383119,"score_spread":0.2882340421688798,"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."}}