{"id":"W4401641570","doi":"10.2196/57162","title":"Evaluation of AI-Driven LabTest Checker for Diagnostic Accuracy and Safety: Prospective Cohort Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical diagnosis; Medicine; Clinical decision support system; Test (biology); Gold standard (test); Patient safety; Medical physics; Health care; Medical emergency; Prospective cohort study; Diagnostic accuracy; Artificial intelligence; Computer science; Decision support system; Pathology; Internal medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002374154,0.0001039214,0.0002515489,0.0001105625,0.00006514299,0.00002829724,0.00005972765,0.0001279754,0.0002150062],"category_scores_gemma":[0.007956122,0.00007970713,0.0000426834,0.0002336509,0.0001071837,0.000231262,0.00002999041,0.0002735437,0.00003347165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001699048,"about_ca_system_score_gemma":0.001140174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005815634,"about_ca_topic_score_gemma":0.00004589497,"domain_scores_codex":[0.9977531,0.00004806732,0.0007729972,0.0001149304,0.001144457,0.0001664729],"domain_scores_gemma":[0.9976159,0.001085433,0.000096212,0.0001741569,0.0008391206,0.0001891291],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001256496,0.0007469188,0.5830284,0.001961443,0.0003017321,0.000004918086,0.05685014,0.00001724482,0.000006735169,0.001793551,0.004903858,0.3502594],"study_design_scores_gemma":[0.0008828274,0.002408519,0.4536738,0.001970454,0.001468016,0.00006294103,0.03217259,0.4976559,0.0008467812,0.003896166,0.004676711,0.0002852134],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9890484,0.0002740625,0.00183749,0.001628984,0.0004326368,0.005936068,0.00001238737,0.000056478,0.0007735466],"genre_scores_gemma":[0.9976959,0.0001244018,0.0003977977,0.0004003553,0.0002725675,0.001003879,0.00005363579,0.00001178297,0.0000397377],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4976387,"threshold_uncertainty_score":0.9524796,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1049382544730949,"score_gpt":0.481556908129329,"score_spread":0.376618653656234,"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."}}