{"id":"W2123707693","doi":"10.1080/17453670710013807","title":"Reviewer agreement in scoring 419 abstracts for scientific orthopedics meetings","year":2007,"lang":"en","type":"article","venue":"Acta Orthopaedica","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Hamilton General Hospital","funders":"","keywords":"Medicine; Intraclass correlation; Orthopedic surgery; Workload; Inter-rater reliability; MEDLINE; Physical therapy; Family medicine; Statistics; Surgery; Psychometrics; Mathematics; Clinical 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":[0.4007483,0.001343771,0.002862692,0.01060698,0.001837916,0.004112038,0.002535038,0.001307709,0.002220125],"category_scores_gemma":[0.5466381,0.0009990019,0.004055424,0.006851094,0.001850067,0.002330836,0.004945348,0.001140218,0.00134209],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002327273,"about_ca_system_score_gemma":0.00550925,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001072542,"about_ca_topic_score_gemma":0.001898787,"domain_scores_codex":[0.4206961,0.3413576,0.1551684,0.0139596,0.06669152,0.002126765],"domain_scores_gemma":[0.2858158,0.3920304,0.0760439,0.02893963,0.2138159,0.003354251],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.009413303,0.0003750454,0.5751486,0.01811053,0.008694685,0.0009530725,0.02857448,0.003502751,0.01097365,0.001755078,0.01421539,0.3282833],"study_design_scores_gemma":[0.001725382,0.006042674,0.8323565,0.0107584,0.009523502,0.005383058,0.01292221,0.02666729,0.02369195,0.008010176,0.06148313,0.00143578],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7846993,0.02555438,0.1498519,0.002399679,0.004471995,0.01185488,0.003204654,0.0009700456,0.0169932],"genre_scores_gemma":[0.8795542,0.003326878,0.1045316,0.0004351327,0.0006309832,0.007163419,0.002192484,0.0003672108,0.001798114],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5992516,"threshold_uncertainty_score":0.738984,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5551584336990506,"score_gpt":0.4904386912035031,"score_spread":0.06471974249554752,"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."}}