{"id":"W4224236060","doi":"10.1371/journal.pone.0267110","title":"Delirium diagnosis without a gold standard: Evaluating diagnostic accuracy of combined delirium assessment tools","year":2022,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Intensive Care Unit Cognitive Disorders","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; Hotchkiss Brain Institute; Alberta Health Services","funders":"Networks of Centres of Excellence of Canada; Canadian Institutes of Health Research; Canadian Frailty Network; Alberta Health Services","keywords":"Delirium; Gold standard (test); Medicine; Intensive care medicine; MEDLINE; Emergency medicine; Internal medicine; Biology","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":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007540999,0.0002868777,0.0008341754,0.0002662399,0.0001793505,0.00003097888,0.000220161,0.00006611839,0.001627192],"category_scores_gemma":[0.106842,0.0003011986,0.0001911783,0.0004808071,0.000128932,0.0001413143,0.0003348558,0.000659354,0.00001909764],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004530861,"about_ca_system_score_gemma":0.0005988381,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008069426,"about_ca_topic_score_gemma":0.00001271684,"domain_scores_codex":[0.9962983,0.0003643336,0.0006552344,0.0004525158,0.001774738,0.0004549202],"domain_scores_gemma":[0.9729865,0.004447138,0.0004275223,0.0005838934,0.02136248,0.0001924796],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003980021,0.01394404,0.7920264,0.001910257,0.004561911,0.0003550161,0.002598194,0.0001989385,0.1197335,0.0006642098,0.04408067,0.01594679],"study_design_scores_gemma":[0.0544903,0.05513265,0.4710449,0.009541951,0.01846357,0.0001104335,0.07916364,0.01587424,0.2886634,0.001616838,0.002412973,0.003485112],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.983079,0.0004489453,0.0002802384,0.009699618,0.0001538616,0.002447412,0.0004385991,0.000107317,0.003345013],"genre_scores_gemma":[0.9716777,0.0001334848,0.003448573,0.02233174,0.0000764264,0.00163836,0.0003807024,0.00007907299,0.0002340083],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3209816,"threshold_uncertainty_score":0.999944,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08096985438925987,"score_gpt":0.3443086700890731,"score_spread":0.2633388156998133,"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."}}