{"id":"W2170538328","doi":"10.1016/j.jclinepi.2010.07.012","title":"Minimal change is sensitive, less specific to recovery: a diagnostic testing approach to interpretability","year":2010,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Sports injuries and prevention","field":"Medicine","cited_by":59,"is_retracted":false,"has_abstract":false,"ca_institutions":"Institute for Work & Health; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; University of Toronto","keywords":"Interpretability; Medicine; Statistics; Computer science; Mathematics; Machine learning","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08520252,0.002158803,0.003514817,0.007529187,0.00167384,0.005595704,0.004153801,0.003728977,0.001871691],"category_scores_gemma":[0.3199507,0.001069692,0.002744594,0.001888428,0.008274258,0.006293467,0.004455669,0.00406285,0.0002725314],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001781109,"about_ca_system_score_gemma":0.001695531,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002172218,"about_ca_topic_score_gemma":0.002199352,"domain_scores_codex":[0.9249175,0.04372571,0.008567076,0.007160948,0.01399664,0.001632153],"domain_scores_gemma":[0.6655046,0.2835622,0.01916484,0.0154005,0.01469967,0.001668255],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004041212,0.0007184113,0.7993418,0.001200095,0.002148469,0.00198062,0.005187364,0.004360068,0.008028617,0.03258742,0.004302426,0.1361036],"study_design_scores_gemma":[0.0005180298,0.004399097,0.5014523,0.00114526,0.003451684,0.01322584,0.006131488,0.1773628,0.01909999,0.2631195,0.009499303,0.0005947312],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.676742,0.004631738,0.2808783,0.01345177,0.001185568,0.001375214,0.0008995305,0.000591793,0.02024396],"genre_scores_gemma":[0.9657893,0.0001959435,0.03193191,0.0009698345,0.0003315915,0.0002345054,0.0001161844,0.00005597934,0.0003747752],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08520252,"threshold_uncertainty_score":0.4505994,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3751571079880765,"score_gpt":0.4901804318391478,"score_spread":0.1150233238510713,"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."}}