{"id":"W4416935767","doi":"10.1093/jamiaopen/ooaf158","title":"Evaluating sociodemographic bias in a deployed machine-learned patient deterioration model","year":2025,"lang":"en","type":"article","venue":"JAMIA Open","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Standardization; Risk assessment; Decision support system; MEDLINE; Data collection","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":[],"consensus_categories":[],"category_scores_codex":[0.001415192,0.0001320025,0.0004089233,0.0001620014,0.0006786172,0.00006178154,0.0003175079,0.0002176711,0.00008427645],"category_scores_gemma":[0.0003576484,0.0001184949,0.00005350019,0.000488012,0.0000272722,0.0002283342,0.0004068751,0.0004690306,0.00004030061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001835153,"about_ca_system_score_gemma":0.0004959235,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005803593,"about_ca_topic_score_gemma":0.02809688,"domain_scores_codex":[0.9975567,0.001045166,0.0005709908,0.0002867435,0.0001978004,0.0003425307],"domain_scores_gemma":[0.9991169,0.0002674187,0.0001858457,0.0002459415,0.0001322042,0.00005166705],"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":[0.0003377523,0.0002152488,0.7103398,0.0003247897,0.00005281191,0.00000666842,0.111678,0.001934958,0.001315175,0.009702218,0.0003741263,0.1637184],"study_design_scores_gemma":[0.006260514,0.0002489815,0.2050798,0.001579022,0.00007486548,4.466195e-8,0.1043854,0.6239113,0.0001250858,0.05661817,0.001062729,0.0006541308],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9898154,0.0001585547,0.0004405197,0.0008698108,0.0002064112,0.001666981,0.000009644304,0.0000605202,0.006772152],"genre_scores_gemma":[0.9962272,0.00005621661,0.001621573,0.0006796569,0.00002934776,0.0006694858,0.00002572334,0.00002116435,0.0006696452],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6219764,"threshold_uncertainty_score":0.9896378,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.288285744714407,"score_gpt":0.5431989938870504,"score_spread":0.2549132491726434,"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."}}