{"id":"W4412514246","doi":"10.2196/77893","title":"Magnitude and Impact of Hallucinations in Tabular Synthetic Health Data on Prognostic Machine Learning Models: Validation Study","year":2025,"lang":"en","type":"article","venue":"Journal of Medical Internet Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Children's Hospital of Eastern Ontario; University of Ottawa","funders":"","keywords":"Preprint; Computer science; Artificial intelligence; Psychology; Data science; Machine learning; Medicine; World Wide Web","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01901789,0.0009945907,0.0005854758,0.00108206,0.0003664604,0.001111998,0.001116201,0.0009685613,0.001201859],"category_scores_gemma":[0.06714715,0.0003028481,0.001410642,0.0007600923,0.001536922,0.001122482,0.001793854,0.00128918,0.0002371532],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0012861,"about_ca_system_score_gemma":0.0007173803,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003554074,"about_ca_topic_score_gemma":0.002209402,"domain_scores_codex":[0.994128,0.004201625,0.0004433598,0.0006274048,0.0004542273,0.0001454013],"domain_scores_gemma":[0.8703309,0.1126247,0.004162672,0.008097943,0.004218602,0.0005650451],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003443293,0.00120369,0.2027552,0.001004114,0.001157793,0.000330356,0.0009169995,0.7128553,0.003407752,0.002180046,0.002908747,0.06783676],"study_design_scores_gemma":[0.0002033824,0.002008986,0.04447329,0.0001949193,0.0001974125,0.0002642981,0.0003603471,0.9430782,0.005799921,0.002054337,0.00128983,0.00007496749],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9805585,0.0005379443,0.01630246,0.0001849111,0.00004124324,0.0002303777,0.001279689,0.000157426,0.000707395],"genre_scores_gemma":[0.9895133,0.0001336338,0.007168679,0.00004978651,0.00001632392,0.00008291032,0.002858394,0.00001685389,0.000160049],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9809821,"threshold_uncertainty_score":0.1005774,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4719190187181834,"score_gpt":0.6019612767654643,"score_spread":0.1300422580472809,"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."}}