{"id":"W4412755825","doi":"10.1038/s44400-025-00015-1","title":"High inter-rater reliability in consensus diagnoses and overall assessment in the Asian Cohort for Alzheimer’s Disease Study","year":2025,"lang":"en","type":"article","venue":"npj Dementia","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canada Research Chairs; University of British Columbia; University of British Columbia Hospital","funders":"National Institute on Aging; University of California, Irvine; University of California, San Francisco; University of California, San Diego; National Eye Institute; University of Pennsylvania; National Institutes of Health; Alzheimer's Association; Korea Brain Research Institute; Brigham and Women's Hospital","keywords":"Inter-rater reliability; Medical diagnosis; Cohort; Disease; Medicine; Reliability (semiconductor); Psychology; Internal medicine; Pathology; Developmental psychology; Rating scale","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":[],"consensus_categories":[],"category_scores_codex":[0.001431721,0.0001612748,0.0002702571,0.0001868692,0.00007550605,0.00007325516,0.0001200084,0.00003650906,0.0001254568],"category_scores_gemma":[0.000195318,0.0001138004,0.00005783252,0.0001915504,0.0001002096,0.00005389473,0.000122888,0.0002267802,0.000003121443],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006506871,"about_ca_system_score_gemma":0.0001823078,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003830847,"about_ca_topic_score_gemma":0.0004778299,"domain_scores_codex":[0.9982144,0.0003224852,0.0003833456,0.0004303882,0.0003170599,0.0003323753],"domain_scores_gemma":[0.999137,0.000300257,0.00004143612,0.0003325538,0.0001024364,0.00008635502],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004489781,0.002242611,0.989621,0.00007524541,0.000283699,0.0001052404,0.0002098135,9.258462e-7,0.00002835159,0.0004061437,0.002088466,0.004489463],"study_design_scores_gemma":[0.005044534,0.0005843997,0.9891505,0.0000994399,0.000700562,9.462129e-7,0.00131553,0.00009650768,0.00007080931,0.001474604,0.001365053,0.00009708337],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9850019,0.0003519792,0.00008484802,0.007641249,0.0001153027,0.004842788,0.0000220175,0.00001545414,0.001924474],"genre_scores_gemma":[0.9970805,0.00003743585,0.000202097,0.001007646,0.00002454317,0.001398897,0.00004706386,0.000009480254,0.0001923154],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01207863,"threshold_uncertainty_score":0.4640647,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02182274915721513,"score_gpt":0.3700795027774596,"score_spread":0.3482567536202445,"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."}}