{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5165368,0.001816139,0.00891775,0.00166803,0.002078559,0.004470903,0.002474091,0.005899836,0.001254551],"category_scores_gemma":[0.7384639,0.002802362,0.006465,0.001849572,0.00590747,0.004741868,0.005065285,0.006605204,0.0003337974],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00327436,"about_ca_system_score_gemma":0.005577867,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001402552,"about_ca_topic_score_gemma":0.001148712,"domain_scores_codex":[0.140735,0.8202752,0.02104611,0.006569521,0.01026028,0.001113824],"domain_scores_gemma":[0.09819961,0.813611,0.03835615,0.03469202,0.01344286,0.001698275],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.6814874,0.003629645,0.09965466,0.01063738,0.05487527,0.0004032232,0.009142526,0.01410094,0.001708769,0.009888196,0.006354529,0.1081175],"study_design_scores_gemma":[0.3880537,0.1615047,0.1908096,0.005222992,0.04564345,0.001249419,0.002310976,0.1426491,0.004483656,0.04444757,0.01185414,0.001770779],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8899367,0.01512203,0.06820606,0.005207185,0.003148527,0.0121092,0.000715402,0.0003212404,0.005233688],"genre_scores_gemma":[0.9811121,0.0003031167,0.01357298,0.0006212597,0.0003146956,0.003676621,0.0002220735,0.00004037641,0.0001368214],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4834632,"threshold_uncertainty_score":0.5961963,"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."}}