{"id":"W4411025380","doi":"10.1001/jamanetworkopen.2025.13685","title":"Detecting and Remediating Harmful Data Shifts for the Responsible Deployment of Clinical AI Models","year":2025,"lang":"en","type":"article","venue":"JAMA Network Open","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":43,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Michael's Hospital; University Health Network; University of Toronto; Vector Institute; York University; Hospital for Sick Children","funders":"Canadian Institutes of Health Research","keywords":"Receiver operating characteristic; Medicine; Demographics; Harm; Pipeline (software); Emergency medicine; Artificial intelligence; Machine learning; Medical emergency; Computer science; Psychology; Demography; Internal medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006338271,0.00007117464,0.0002649812,0.00002869163,0.0002436175,0.0000742204,0.0003996032,0.0001007318,0.000009179638],"category_scores_gemma":[0.003474852,0.00005094007,0.00003370046,0.0002217734,0.00005574538,0.0002294911,0.0004921413,0.0002583688,0.000001336093],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002091166,"about_ca_system_score_gemma":0.0005903796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008800681,"about_ca_topic_score_gemma":0.0006080475,"domain_scores_codex":[0.9985043,0.0001594804,0.0007296221,0.000288193,0.0001114698,0.0002069503],"domain_scores_gemma":[0.9944987,0.004374223,0.0001929879,0.0006655458,0.0001989016,0.00006967304],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00185413,0.0000632011,0.08737324,0.0001331588,0.0001003893,9.444935e-7,0.0007740718,0.001214707,0.00002145822,0.002148587,0.03737348,0.8689426],"study_design_scores_gemma":[0.000518356,0.000411885,0.03321555,0.001466371,0.0002967866,0.00000341297,0.003231107,0.8932786,0.001390946,0.03825895,0.02778247,0.0001455369],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7773778,0.004644485,0.06436697,0.142561,0.00281551,0.005417713,0.0000223037,0.00005123467,0.002743041],"genre_scores_gemma":[0.9822963,0.0004881135,0.01108298,0.004564675,0.001076091,0.00008517181,0.00002322051,0.00001189712,0.0003715158],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8920639,"threshold_uncertainty_score":0.4159974,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5452111985725556,"score_gpt":0.5722105086674241,"score_spread":0.02699931009486845,"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."}}