{"id":"W3191847692","doi":"10.1097/phm.0000000000001859","title":"Interrater Reliability in the American Board of Physical Medicine and Rehabilitation Part II Certification Examination","year":2021,"lang":"en","type":"article","venue":"American Journal of Physical Medicine & Rehabilitation","topic":"Medical Education and Admissions","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Inter-rater reliability; Certification; Medicine; Reliability (semiconductor); Physical examination; Physical therapy; Intra-rater reliability; Rehabilitation; Oral examination; Medical physics; Physical medicine and rehabilitation; Family medicine; Rating scale; Psychology; Surgery; Confidence interval","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","sts"],"consensus_categories":[],"category_scores_codex":[0.002019515,0.0002273203,0.001095749,0.0003708601,0.000085741,0.000006905133,0.0001551136,0.00003620456,0.00009605337],"category_scores_gemma":[0.04059273,0.000135014,0.0002050634,0.001529886,0.00328259,0.0002088655,0.00003450231,0.0005537996,0.000001839213],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001527816,"about_ca_system_score_gemma":0.0003610712,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001518459,"about_ca_topic_score_gemma":0.00001139834,"domain_scores_codex":[0.995815,0.001270214,0.001201394,0.0003782843,0.001113697,0.0002214058],"domain_scores_gemma":[0.9893419,0.00737501,0.001126239,0.0004801576,0.001248086,0.0004286411],"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":[0.0007570988,0.00370607,0.01542457,0.000453071,0.00007361182,0.00001171921,0.1454481,0.00003691678,0.02757447,0.003489229,0.003117767,0.7999073],"study_design_scores_gemma":[0.002938137,0.1110136,0.7307768,0.002829463,0.0004565015,0.00008660147,0.1354889,0.00192755,0.0003886777,0.00644193,0.007407801,0.0002441116],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.875912,0.00007448416,0.0001385442,0.1230839,0.0001506131,0.0003344173,0.000002047713,0.00001080176,0.0002932314],"genre_scores_gemma":[0.9958031,0.00001514271,0.001717033,0.001540367,0.0008060897,0.00003129913,0.0000242531,0.00001827146,0.00004440768],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7996632,"threshold_uncertainty_score":0.9994299,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02120119612402987,"score_gpt":0.3597731413697832,"score_spread":0.3385719452457533,"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."}}