{"id":"W6977082170","doi":"10.6084/m9.figshare.25223503.v1","title":"Additional file 1 of Characterizing the limitations of using diagnosis codes in the context of machine learning for healthcare","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Artificial Intelligence in Healthcare","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"SickKids Foundation","funders":"","keywords":"Context (archaeology); Health care; Key (lock); Training set","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002511813,0.00009329592,0.0002235371,0.0001031863,0.0003514665,0.000006421545,0.000220821,0.0001112181,0.5801389],"category_scores_gemma":[0.01474774,0.00006252612,0.0001031491,0.0003602619,0.00003212242,0.00009052437,0.00005864556,0.0005148289,0.00008615937],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005653333,"about_ca_system_score_gemma":0.0004920703,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007156231,"about_ca_topic_score_gemma":0.002502535,"domain_scores_codex":[0.9982446,0.0004812664,0.0006972624,0.0001468338,0.0001993216,0.0002306778],"domain_scores_gemma":[0.9535251,0.04545628,0.0003737227,0.0001660471,0.0004507829,0.0000281296],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001809966,0.00003384572,0.001361404,0.003028174,0.00001823045,0.000001365319,0.01573686,0.00004121395,0.00004098612,0.000393141,0.9660152,0.01331149],"study_design_scores_gemma":[0.00004607451,0.0001105954,0.004319672,0.02468405,0.00001087137,0.000001261274,0.02527994,0.02038391,0.0002903804,0.0002382583,0.9245454,0.00008959643],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"empirical","genre_scores_codex":[0.002270635,0.0007050681,0.000001625891,0.002267355,0.00006901037,0.001036114,0.9935225,0.00001821875,0.0001094247],"genre_scores_gemma":[0.5005845,0.00003449863,0.0001668647,0.0005356011,0.000148494,0.004604142,0.4938209,0.00002251398,0.00008241229],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.5800527,"threshold_uncertainty_score":0.9935514,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.470253144846657,"score_gpt":0.4764973894160241,"score_spread":0.006244244569367063,"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."}}