{"id":"W1989953251","doi":"10.1097/acm.0000000000000074","title":"Am I Right When I Am Sure? Data Consistency Influences the Relationship Between Diagnostic Accuracy and Certainty","year":2013,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Western Hospital","funders":"","keywords":"Certainty; Medical diagnosis; Diagnostic accuracy; Consistency (knowledge bases); Context (archaeology); Medicine; Statistics; Psychology; Computer science; Mathematics; Artificial intelligence; Internal medicine; Pathology","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09246857,0.0005783733,0.0009098939,0.002157626,0.0009107622,0.003628545,0.001111772,0.002070169,0.002043595],"category_scores_gemma":[0.6332655,0.0009202198,0.001409564,0.001057999,0.00357481,0.005842354,0.003506935,0.002075552,0.0002511954],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001372415,"about_ca_system_score_gemma":0.001502419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001829567,"about_ca_topic_score_gemma":0.001150795,"domain_scores_codex":[0.8771653,0.08521719,0.01093659,0.007700855,0.01708839,0.001891702],"domain_scores_gemma":[0.1398844,0.7894452,0.04477602,0.01516288,0.008918721,0.001812753],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005836516,0.0005845959,0.8939934,0.001342174,0.002051529,0.0002767174,0.009306957,0.005560109,0.005169997,0.002626925,0.0005120206,0.07273901],"study_design_scores_gemma":[0.0005787314,0.006146861,0.9206015,0.001803413,0.001900901,0.001522694,0.005200074,0.02929574,0.01123895,0.01818623,0.003037176,0.0004877017],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9786049,0.001247329,0.01424058,0.001143383,0.00005605823,0.0002186869,0.0001461341,0.00006810296,0.004274725],"genre_scores_gemma":[0.9940608,0.0001592014,0.005387223,0.000165548,0.00002799455,0.00006957234,0.00004750444,0.00001115382,0.00007097053],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09246857,"threshold_uncertainty_score":0.4890264,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1343602694132445,"score_gpt":0.3999045361787397,"score_spread":0.2655442667654952,"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."}}