{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03728166,0.0007822284,0.0008473348,0.0009534282,0.0004125157,0.001299601,0.001091519,0.0008217045,0.001139274],"category_scores_gemma":[0.07228839,0.0003769784,0.00114064,0.0004873033,0.0008712793,0.0009419198,0.001350795,0.001307537,0.0001787363],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001669681,"about_ca_system_score_gemma":0.001685648,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008415349,"about_ca_topic_score_gemma":0.003667073,"domain_scores_codex":[0.9919751,0.005602748,0.0005291902,0.001034816,0.0005556147,0.00030239],"domain_scores_gemma":[0.9332902,0.05462289,0.005100803,0.002836694,0.003399799,0.0007495682],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001529167,0.0002969637,0.5762857,0.0001009331,0.0008611868,0.0001210866,0.0001765414,0.3902938,0.000431486,0.001307853,0.0009581106,0.02763704],"study_design_scores_gemma":[0.00008138957,0.0005335025,0.02676389,0.00005287829,0.0001326104,0.00004975852,0.0000716382,0.969326,0.0008394818,0.001837806,0.0002921179,0.00001901109],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9640022,0.0002931234,0.03351358,0.0008208615,0.00006149487,0.0001374987,0.0003592214,0.0001612819,0.0006508323],"genre_scores_gemma":[0.9951321,0.00003303407,0.004303464,0.00009880688,0.00001963613,0.00003190591,0.0002683775,0.00000628508,0.0001063402],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9627184,"threshold_uncertainty_score":0.1971666,"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."}}