{"id":"W2588203234","doi":"10.1186/s41512-016-0001-y","title":"Methods for Evaluating Medical Tests and Biomarkers","year":2017,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University; University of Calgary; McGill University; Jewish General Hospital; McMaster University; Population Health Research Institute","funders":"National Health and Medical Research Council; Haukeland Universitetssjukehus; Universiteit Leiden; Assistance publique-Hôpitaux de Paris; Leids Universitair Medisch Centrum; Universitetet i Bergen; Albert-Ludwigs-Universität Freiburg; Universiteit van Amsterdam; Abbott Diagnostics; University of Oxford; University of New South Wales; Medical Research Council; University of Notre Dame","keywords":"Equivalence (formal languages); Confidence interval; Statistics; Margin (machine learning); Sample (material); Medicine; Mathematics; Computer science; Chromatography; Machine learning; Chemistry","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":[0.008305135,0.0001342586,0.0002901317,0.0001267079,0.0008943085,0.0001838201,0.0001822238,0.0001690956,0.00005488493],"category_scores_gemma":[0.1333587,0.0001041704,0.00004099283,0.00006238474,0.0006658016,0.00006588643,0.0001726586,0.0003626939,0.000007694427],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002502587,"about_ca_system_score_gemma":0.0002965502,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001487737,"about_ca_topic_score_gemma":0.000004653888,"domain_scores_codex":[0.9979335,0.0003202262,0.0002289709,0.0004132905,0.00060219,0.0005017956],"domain_scores_gemma":[0.971777,0.02682052,0.00005913559,0.0004549255,0.0003493347,0.0005391407],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001133521,0.00006480738,0.2485907,0.0002330296,0.00008029071,0.00004417392,0.000154213,4.213979e-8,0.0003836642,0.0006481828,0.000254011,0.7494335],"study_design_scores_gemma":[0.004903272,0.002514328,0.937588,0.001633227,0.0001476475,0.0002542552,0.0002707211,0.0315311,0.0008619412,0.0123784,0.007629842,0.0002872247],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9754312,0.004820775,0.007484303,0.008029074,0.0003073857,0.001928467,0.00000828774,0.0000471958,0.00194333],"genre_scores_gemma":[0.9601881,0.001550631,0.0371006,0.0000752546,0.0003432933,0.0003929152,0.00001107732,0.00002975548,0.0003083602],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7491463,"threshold_uncertainty_score":0.8739414,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.269673097928087,"score_gpt":0.6196684686333537,"score_spread":0.3499953707052667,"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."}}