{"id":"W2965485927","doi":"10.48550/arxiv.1908.00690","title":"Feature Robustness in Non-stationary Health Records: Caveats to Deployable Model Performance in Common Clinical Machine Learning Tasks","year":2019,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":55,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Robustness (evolution); Computer science; Health records; Artificial intelligence; Machine learning; Feature (linguistics); Health care","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":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.001921327,0.0004673149,0.0009336328,0.0008420881,0.0002006449,0.00008665004,0.002178445,0.0005588479,0.000006356149],"category_scores_gemma":[0.0001073671,0.0005677013,0.0001520899,0.001304577,0.00004836236,0.0004682906,0.002420716,0.004613354,0.00005587782],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009660299,"about_ca_system_score_gemma":0.00102754,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003235811,"about_ca_topic_score_gemma":0.004503991,"domain_scores_codex":[0.9955426,0.00100722,0.0006361547,0.001807529,0.0002142701,0.000792241],"domain_scores_gemma":[0.9971504,0.0003629134,0.0005470946,0.001425806,0.0001565192,0.0003573266],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00005510112,0.00006132187,0.4303428,0.0002674783,0.00000668,0.00007199741,0.0003498905,0.5656265,1.230013e-7,0.0007001669,0.0001107028,0.002407235],"study_design_scores_gemma":[0.000637503,0.0002076445,0.1477519,0.0006038955,0.00000469757,0.000004594382,0.00003915694,0.8496742,4.435643e-7,0.000398388,0.0002414524,0.0004361071],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7057377,0.0001385235,0.2902304,0.001855883,0.0006178802,0.0008835422,0.00001496933,0.00017844,0.0003427082],"genre_scores_gemma":[0.9798437,0.0005570564,0.01584556,0.0005866945,0.00005644569,0.000006110432,0.00009037341,0.00004901608,0.002965092],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2840477,"threshold_uncertainty_score":0.9996774,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08851988355023072,"score_gpt":0.2733285824830718,"score_spread":0.1848086989328411,"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."}}