{"id":"W3032677176","doi":"10.3760/cma.j.issn.1007-9742.2003.03.130","title":"Evidence base of clinical diagnosis: The architecture of diagnostic research","year":2002,"lang":"en","type":"article","venue":"","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":143,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Test (biology); Diagnostic test; Medicine; Diagnostic accuracy; Suspect; Clinical trial; Intensive care medicine; Medical physics; Pathology; Psychology; Internal medicine; Pediatrics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.3385951,0.0001497707,0.002762297,0.0002930203,0.0001015705,0.0001669078,0.003418681,0.00008389133,0.07876624],"category_scores_gemma":[0.7565809,0.00004812328,0.001978064,0.002424822,0.0004921568,0.0001138203,0.0003181626,0.0003887115,0.003429936],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000005534775,"about_ca_system_score_gemma":0.00004363164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004451259,"about_ca_topic_score_gemma":0.00007122719,"domain_scores_codex":[0.909979,0.05701481,0.01820411,0.001096696,0.01332576,0.0003796479],"domain_scores_gemma":[0.4628092,0.5221081,0.004280615,0.00747168,0.003123141,0.0002073348],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.000005995747,0.0002461146,0.3486788,0.00008065447,0.0001307402,0.00000580511,0.0007036047,0.0001311385,0.00001760098,0.002606773,0.4469332,0.2004595],"study_design_scores_gemma":[0.0007514055,0.001234626,0.5545687,0.002275004,0.0009539993,0.00002833209,0.00480373,0.02972005,0.001271862,0.05354489,0.3502635,0.0005839339],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8674337,0.03862255,0.01047577,0.03035632,0.0005259904,0.002804076,0.0000233847,0.000006423461,0.04975182],"genre_scores_gemma":[0.9877346,0.001598878,0.001879909,0.0002342814,0.00009120814,0.0000797017,1.469791e-7,0.000005277683,0.008376056],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4650933,"threshold_uncertainty_score":0.997346,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9730843423361705,"score_gpt":0.6845970074891213,"score_spread":0.2884873348470491,"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."}}