{"id":"W1518867746","doi":"10.1108/09526861311297325","title":"A comparison of two diagnostic performance measures","year":2013,"lang":"en","type":"article","venue":"International Journal of Health Care Quality Assurance","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Measure (data warehouse); Point (geometry); Receiver operating characteristic; Signal-to-noise ratio (imaging); Taguchi methods; Noise (video); Diagnostic accuracy; Computer science; Index (typography); Statistics; Mathematics; Artificial intelligence; Data mining; Medicine; Radiology","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"],"consensus_categories":[],"category_scores_codex":[0.01048177,0.0001379124,0.0006830692,0.0002383406,0.0001033636,0.0001375835,0.001469939,0.00004333722,0.0002748537],"category_scores_gemma":[0.01072864,0.00009899811,0.0002359185,0.0002504845,0.00012368,0.0006251631,0.00006949605,0.000295106,0.00008485084],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002086247,"about_ca_system_score_gemma":0.0004959633,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004724762,"about_ca_topic_score_gemma":0.0001074756,"domain_scores_codex":[0.9895951,0.001143876,0.003401807,0.0002270771,0.005400829,0.0002312675],"domain_scores_gemma":[0.9859018,0.003734626,0.003023934,0.0003401646,0.006857799,0.0001417078],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007291079,0.0002713182,0.6615962,0.00007696701,0.0000722754,0.000001462278,0.004610167,0.007097602,0.0002876767,0.0004978767,0.008256896,0.3171587],"study_design_scores_gemma":[0.001814145,0.0006891796,0.9466435,0.0007583743,0.00001237744,0.00002016044,0.00887457,0.0005214539,0.003271389,0.005010105,0.03213774,0.0002470354],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9696966,0.004093476,0.01041245,0.01239235,0.00246676,0.0002708259,0.00001805926,0.000007327369,0.0006421251],"genre_scores_gemma":[0.9967335,0.0001992734,0.002141722,0.000667165,0.0002120139,0.000007694159,0.00000180929,0.000005445973,0.00003140538],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3169116,"threshold_uncertainty_score":0.9976044,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2317686654379214,"score_gpt":0.5212233067730002,"score_spread":0.2894546413350788,"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."}}