{"id":"W7117294252","doi":"10.1001/jamanetworkopen.2025.50454","title":"Diagnostic Codes in AI Prediction Models and Label Leakage of Same-Admission Clinical Outcomes","year":2025,"lang":"en","type":"article","venue":"JAMA Network Open","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"U.S. National Library of Medicine","keywords":"Predictive modelling; Trustworthiness; Multiple Models; Clinical Practice; Diagnosis code; Artificial neural network; Interpretability","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":[],"consensus_categories":[],"category_scores_codex":[0.0007684911,0.0001272247,0.0005748236,0.00006315205,0.00004533866,0.00004368172,0.0001135933,0.0001524153,0.00003777109],"category_scores_gemma":[0.0007956827,0.0000936028,0.0000632062,0.0001883142,0.00005253208,0.0001610886,0.0002188798,0.0001878793,0.000002782554],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000456083,"about_ca_system_score_gemma":0.00009106228,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003648507,"about_ca_topic_score_gemma":0.0001445697,"domain_scores_codex":[0.9987189,0.0001494928,0.0005384843,0.0002854318,0.0001317542,0.0001759448],"domain_scores_gemma":[0.9972475,0.002233328,0.00009404615,0.0002691766,0.000053875,0.0001020296],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002511447,0.0007692567,0.9383034,0.0000414814,0.0001477095,0.00002651771,0.00005714779,0.00051709,0.00000144705,0.001645282,0.02496394,0.03327559],"study_design_scores_gemma":[0.005524901,0.0003333843,0.9786297,0.001462873,0.0001891677,0.000002109696,0.00005450364,0.00864896,0.00002786429,0.003309002,0.00175315,0.00006436442],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9764742,0.002492596,0.0001937365,0.01549285,0.0002429304,0.001197164,0.00001652458,0.00002291206,0.003867039],"genre_scores_gemma":[0.9894764,0.004196226,0.001016395,0.004405808,0.00008260264,0.0001148023,0.0000362356,0.00001125739,0.0006602017],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04032632,"threshold_uncertainty_score":0.3817011,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1324866201201463,"score_gpt":0.4361251982803647,"score_spread":0.3036385781602184,"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."}}