{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02070823,0.000179486,0.00211888,0.0001511641,0.00004355268,0.000007645142,0.0001722998,0.0003636535,0.0001647392],"category_scores_gemma":[0.1859695,0.0001323823,0.0006239748,0.0002270119,0.0001616374,0.00008895542,0.000113561,0.001464144,0.00004332418],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003333795,"about_ca_system_score_gemma":0.0001179278,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004500994,"about_ca_topic_score_gemma":0.00001165726,"domain_scores_codex":[0.9946821,0.001025597,0.003319549,0.00041386,0.0001846574,0.0003742729],"domain_scores_gemma":[0.9657168,0.03125222,0.001340816,0.0004193553,0.0005150173,0.0007558093],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001916185,0.0006541464,0.7554196,0.00005466569,0.00009910311,0.00006540339,0.0003172671,0.000009058662,0.0001994683,0.0002426066,0.01426468,0.2267578],"study_design_scores_gemma":[0.0006517207,0.002875086,0.9543945,0.0002429437,0.00009297487,0.0005301864,0.0001205993,0.0001850319,0.00003618945,0.001008051,0.03973293,0.000129755],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9698079,0.00007367769,0.009343267,0.0152268,0.002901169,0.0004632686,0.00001212331,0.00001388782,0.002157878],"genre_scores_gemma":[0.8880516,0.00008292601,0.09108386,0.01666253,0.003992192,0.0000112852,0.00000352316,0.00001820327,0.00009382595],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2266281,"threshold_uncertainty_score":0.8208874,"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."}}