{"id":"W4410397499","doi":"10.1016/j.sapharm.2025.05.008","title":"Consistency of Medical Subject Headings assignment: A test-retest reliability analysis","year":2025,"lang":"en","type":"article","venue":"Research in Social and Administrative Pharmacy","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Consistency (knowledge bases); Subject (documents); Reliability (semiconductor); Test (biology); Information retrieval; Computer science; Psychology; Reliability engineering; Artificial intelligence; World Wide Web; Engineering; Biology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001998088,0.0001053477,0.0002804976,0.0001215146,0.0001646468,0.00002182972,0.000234099,0.0001536043,0.00005344151],"category_scores_gemma":[0.002021315,0.00009019128,0.0001017245,0.001036449,0.001505846,0.000003226204,0.0002024841,0.000302979,7.730154e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003346201,"about_ca_system_score_gemma":0.0007654327,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002761653,"about_ca_topic_score_gemma":0.0001386038,"domain_scores_codex":[0.9981475,0.0003922768,0.0003087096,0.0003615756,0.0004858186,0.0003041343],"domain_scores_gemma":[0.9990845,0.0004678891,0.00005058981,0.0001179105,0.0001501878,0.0001288885],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004126797,0.0009096721,0.9095383,0.0002690293,0.00050592,0.00004353169,0.0004760042,4.502546e-7,0.02231796,0.003617639,0.002941121,0.05896774],"study_design_scores_gemma":[0.009464985,0.002956319,0.6673777,0.0004590729,0.000490844,0.00001680464,0.009035653,0.001268228,0.1769509,0.02261476,0.1083248,0.001039989],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9805516,0.001393874,0.0003131984,0.009954356,0.00003938626,0.0002089266,0.00008478495,0.00001003146,0.007443866],"genre_scores_gemma":[0.9989095,0.0004742433,0.0001421859,0.0001081716,0.00006591615,0.0000344869,0.00002128727,0.000002917258,0.0002413427],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2421606,"threshold_uncertainty_score":0.5548354,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1753936504615749,"score_gpt":0.5228577643158194,"score_spread":0.3474641138542446,"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."}}