{"id":"W4416969343","doi":"10.2196/78309","title":"Unsupervised Characterization of Temporal Dataset Shifts as an Early Indicator of AI Performance Variations: Evaluation Study Using the Medical Information Mart for Intensive Care-IV Dataset","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Healthcare Technology and Patient Monitoring","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Training set; Clinical Practice; Characterization (materials science); Health informatics; Change detection","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005582435,0.0005861416,0.0004065836,0.001538626,0.0003032245,0.0007106243,0.0008763035,0.0004433224,0.0003058419],"category_scores_gemma":[0.01341349,0.0001440556,0.0007769627,0.001324513,0.0004672713,0.001031244,0.0009501043,0.0009445637,0.0001456695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008003738,"about_ca_system_score_gemma":0.0009753053,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005166952,"about_ca_topic_score_gemma":0.00734711,"domain_scores_codex":[0.9981292,0.0006903079,0.0001756502,0.0005359013,0.0003394559,0.000129518],"domain_scores_gemma":[0.9893294,0.005285413,0.00159599,0.001742995,0.001609343,0.0004367767],"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":[0.001255446,0.001052634,0.7110999,0.0002601015,0.0007287018,0.0002466865,0.0005255493,0.1736366,0.00611686,0.001134994,0.005632663,0.09830987],"study_design_scores_gemma":[0.00004121001,0.0006217522,0.253867,0.00002778873,0.00009975566,0.0003032987,0.0003373119,0.7361179,0.005589929,0.001205535,0.001745443,0.0000431152],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9833068,0.0001906715,0.01313197,0.0001609925,0.00002912615,0.00009632335,0.0022557,0.0003130272,0.0005155682],"genre_scores_gemma":[0.9770615,0.00007471754,0.0147955,0.0000372479,0.00002725252,0.00006851616,0.007731823,0.00003238694,0.0001711026],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005582435,"threshold_uncertainty_score":0.02952307,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05172330938516912,"score_gpt":0.4106883063506143,"score_spread":0.3589649969654452,"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."}}