{"id":"W4399772087","doi":"10.2196/60384","title":"Authors’ Response to Peer Reviews of “Performance Drift in Machine Learning Models for Cardiac Surgery Risk Prediction: Retrospective Analysis”","year":2024,"lang":"en","type":"article","venue":"JMIRx Med","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Machine learning; Medicine; Artificial intelligence","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007308423,0.0001899699,0.001735297,0.0007260518,0.0000530309,0.00002028905,0.00005643671,0.000133497,0.0000633391],"category_scores_gemma":[0.002783133,0.0001450226,0.001341889,0.001651813,0.00003392297,0.0001121385,0.00002286798,0.0003725374,0.00001528565],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001544367,"about_ca_system_score_gemma":0.0001013363,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004723105,"about_ca_topic_score_gemma":0.00001075241,"domain_scores_codex":[0.9975728,0.000555443,0.0006953936,0.0003972444,0.0005053014,0.0002738053],"domain_scores_gemma":[0.9975299,0.001676337,0.0001039583,0.0003263298,0.0001918345,0.0001716213],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003924842,0.00007212244,0.9738572,0.0004517201,0.00112772,0.00006084448,0.002465752,0.002517047,0.0003836113,0.0004055802,0.007032381,0.007701133],"study_design_scores_gemma":[0.0002946166,0.0001987988,0.3837187,0.0002442768,0.001098372,0.00000670149,0.0000572235,0.02625871,0.0001801665,0.00001840794,0.5877799,0.000144087],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9753788,0.00566723,0.004082511,0.00835493,0.000707522,0.002351402,0.0001721466,0.0001954937,0.003090041],"genre_scores_gemma":[0.9544673,0.001527226,0.0004667999,0.00007788311,0.000187362,0.0003149639,0.00009408122,0.00003184468,0.04283253],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5901385,"threshold_uncertainty_score":0.5913851,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03389387698986936,"score_gpt":0.3062406549357081,"score_spread":0.2723467779458387,"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."}}