{"id":"W2978087496","doi":"10.2217/nmt-2019-0017","title":"What is Test Accuracy? Comparing Unitary Accuracy Metrics for Cognitive Screening Instruments","year":2019,"lang":"en","type":"article","venue":"Neurodegenerative Disease Management","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Receiver operating characteristic; Metric (unit); Dementia; Test (biology); Cognition; Cognitive impairment; Raw score; Statistics; Measure (data warehouse); Psychology; Computer science; Medicine; Mathematics; Data mining; Raw data; Psychiatry; Engineering","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0003277801,0.0004296118,0.0004534871,0.0005587413,0.0002955696,0.0003969767,0.0002653415,0.00004630293,0.0005369231],"category_scores_gemma":[0.000629341,0.0004049619,0.0002759123,0.0007702392,0.00008794361,0.001017033,0.0004919178,0.0002601701,0.0002598877],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001017685,"about_ca_system_score_gemma":0.0000918904,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000633217,"about_ca_topic_score_gemma":0.000001568336,"domain_scores_codex":[0.9967007,0.0001305246,0.0004485635,0.0009807427,0.001011883,0.0007275843],"domain_scores_gemma":[0.9974117,0.0007996982,0.0001766338,0.0004786206,0.000585789,0.0005475494],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002784327,0.002241356,0.7938418,0.001413612,0.001273836,0.0002599849,0.0003495676,0.0000543405,0.00107873,0.000841678,0.004543142,0.1913176],"study_design_scores_gemma":[0.01781835,0.001830165,0.9013622,0.001509528,0.00197546,0.000006930112,0.005185568,0.03067322,0.01893813,0.0002508628,0.01947942,0.0009701344],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9805527,0.0005755581,0.004182224,0.001611587,0.0004952426,0.006149656,0.0001745561,0.000106686,0.006151797],"genre_scores_gemma":[0.9840777,0.0009814578,0.0007363696,0.006332384,0.0001380529,0.0006496885,0.0007510997,0.00007125726,0.006261995],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1903474,"threshold_uncertainty_score":0.9998402,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04453971085852782,"score_gpt":0.3395502077307218,"score_spread":0.295010496872194,"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."}}