{"id":"W4366241702","doi":"10.1097/lbr.0000000000000920","title":"Reporting Standards for Diagnostic Testing","year":2023,"lang":"en","type":"article","venue":"Journal of Bronchology & Interventional Pulmonology","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; McGill University","funders":"","keywords":"Medicine; Checklist; Test (biology); Diagnostic test; Contingency table; MEDLINE; Medical physics; Statistics; Pediatrics; Psychology","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":[0.001703855,0.0001265709,0.0005601033,0.0003280242,0.00007723686,0.00001186263,0.0001006363,0.0001318502,0.0001721226],"category_scores_gemma":[0.01345324,0.0001031741,0.0005307806,0.0002195199,0.00009865116,0.00007638318,0.00005097302,0.0002539325,0.0000207684],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004106911,"about_ca_system_score_gemma":0.0003627089,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009227159,"about_ca_topic_score_gemma":0.000007855017,"domain_scores_codex":[0.9975424,0.00005739186,0.001645032,0.0002065187,0.0002466328,0.000301982],"domain_scores_gemma":[0.9954381,0.001412142,0.00194577,0.0001464919,0.000951863,0.0001056225],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009739463,0.001256665,0.8131301,0.0004614346,0.001802796,0.004465142,0.0001143725,0.0002330915,0.002034904,0.001680603,0.1063802,0.06746674],"study_design_scores_gemma":[0.004937804,0.007862223,0.936864,0.001128195,0.0006410714,0.009681756,0.0001206588,0.0001437567,0.001070358,0.008242892,0.02915705,0.0001502633],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9842,0.001383364,0.001985164,0.009933494,0.001461066,0.0003169033,0.00004099758,0.00004823702,0.000630778],"genre_scores_gemma":[0.9955217,0.0000843631,0.002815098,0.0003362157,0.0006111776,0.00006342785,0.00002996693,0.00002340206,0.0005147137],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1237338,"threshold_uncertainty_score":0.9948568,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07585108831056069,"score_gpt":0.4080725586681805,"score_spread":0.3322214703576198,"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."}}