{"id":"W2953849708","doi":"10.1186/s12888-019-2171-y","title":"The reproducibility of psychiatric evaluations of work disability: two reliability and agreement studies","year":2019,"lang":"en","type":"article","venue":"BMC Psychiatry","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Hamilton Health Sciences","funders":"Bundesamt für Sozialversicherungen; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; McMaster University; Northwestern University","keywords":"Intraclass correlation; Reproducibility; Reliability (semiconductor); Inter-rater reliability; Standard error; Work (physics); Medicine; Psychology; Physical therapy; Statistics; Mathematics; Rating scale","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.005982419,0.0001457698,0.0004240957,0.00004326119,0.0000999206,0.000004723689,0.0001008354,0.00005781575,0.00004274323],"category_scores_gemma":[0.002686571,0.00008747412,0.0002340686,0.0004794611,0.0006590855,0.00004917353,0.00007196722,0.0001461469,0.000008821902],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006077815,"about_ca_system_score_gemma":0.0001829793,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003988464,"about_ca_topic_score_gemma":0.0001124102,"domain_scores_codex":[0.9974025,0.0003897703,0.0008404087,0.0007558218,0.0004438065,0.0001676222],"domain_scores_gemma":[0.9961413,0.0006454397,0.000289742,0.002517018,0.0003422799,0.00006426148],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003739924,0.0005645702,0.9902614,0.002868379,0.00007932843,2.370937e-9,0.0003407356,0.00004533073,0.0002059252,0.001546729,0.0004610507,0.003252522],"study_design_scores_gemma":[0.001326676,0.0005004191,0.9817197,0.0001963244,0.0001375745,5.046347e-7,0.001716111,0.00007843839,0.00003636549,0.01404078,0.0001628627,0.00008422988],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9833497,0.006432144,0.00003816908,0.006026914,0.001312226,0.001839827,0.000004783687,0.00001992932,0.0009763507],"genre_scores_gemma":[0.9899728,0.0001410511,0.009383677,0.00004772978,0.0001297248,0.0000422612,0.000004410791,0.000009724405,0.0002686491],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01249405,"threshold_uncertainty_score":0.3567091,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02781797813394696,"score_gpt":0.3632240959628039,"score_spread":0.3354061178288569,"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."}}