{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1347341,0.001054637,0.001222979,0.003582807,0.001096874,0.002135713,0.002106816,0.001697521,0.001029045],"category_scores_gemma":[0.2372612,0.001380068,0.004044561,0.002832245,0.003450422,0.001876653,0.004117483,0.001467788,0.0004291616],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001118641,"about_ca_system_score_gemma":0.0009868202,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001552386,"about_ca_topic_score_gemma":0.00291181,"domain_scores_codex":[0.8121532,0.1183957,0.01933048,0.01231119,0.03601455,0.001794844],"domain_scores_gemma":[0.5427336,0.3109254,0.04771017,0.04643126,0.05017478,0.002024791],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006192257,0.0006928025,0.8713155,0.001839876,0.01034018,0.0002327269,0.02032118,0.002763834,0.007670486,0.00182458,0.001132878,0.07567373],"study_design_scores_gemma":[0.0003884454,0.003245226,0.9766799,0.0003605823,0.00148431,0.0007313936,0.002167827,0.004396323,0.006564726,0.001270633,0.002479933,0.0002306424],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9572389,0.003525838,0.0317058,0.0001434795,0.0002345921,0.001287612,0.0006356661,0.00007597308,0.005152198],"genre_scores_gemma":[0.9881526,0.0003280002,0.009446355,0.00005939279,0.00008875303,0.0009710834,0.0005316036,0.00006073808,0.0003615464],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8652659,"threshold_uncertainty_score":0.7125505,"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."}}