{"id":"W3099884563","doi":"10.18653/v1/2020.clinicalnlp-1.2","title":"Multiple Sclerosis Severity Classification From Clinical Text","year":2020,"lang":"en","type":"article","venue":"","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; Toronto Rehabilitation Institute; University of Toronto; Ontario Institute for Cancer Research","funders":"","keywords":"Natural history; Computer science; Natural language processing; Multiple sclerosis; Artificial intelligence; Medicine; Internal medicine; Psychiatry","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002823292,0.00009260143,0.0001576757,0.00001662617,0.00009801819,0.00009814334,0.0008018854,0.00009679177,0.0001569993],"category_scores_gemma":[0.0008075433,0.00008437607,0.00007087925,0.0002529647,0.00003142632,0.0002373034,0.0002789744,0.0003557492,0.0007901539],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002111513,"about_ca_system_score_gemma":0.00006409402,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006675572,"about_ca_topic_score_gemma":0.00004369851,"domain_scores_codex":[0.9984159,0.000285828,0.0003597305,0.0005422847,0.0002251471,0.0001711158],"domain_scores_gemma":[0.9985331,0.0005232773,0.0001010938,0.0005246545,0.00006802487,0.0002498837],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001206007,0.00005260365,0.7067399,0.00001189284,0.000009899304,0.000002140666,0.0006522492,0.00004869973,0.0002688811,0.005633433,0.008186183,0.2783821],"study_design_scores_gemma":[0.000147039,0.00003410164,0.5159962,0.000004081151,0.000001125919,1.670584e-7,0.00001570304,0.476257,0.00004701738,0.0001535842,0.007275676,0.00006832819],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1618047,0.00004634185,0.7657914,0.06802531,0.0004739002,0.0001990403,0.000007490138,0.0008112664,0.002840503],"genre_scores_gemma":[0.8601388,0.00001444457,0.1326721,0.006877715,0.0002254473,0.000006174528,0.000008331702,0.000006547469,0.00005041007],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.698334,"threshold_uncertainty_score":0.9999878,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2310380793134054,"score_gpt":0.3534443390747336,"score_spread":0.1224062597613282,"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."}}