{"id":"W3129347641","doi":"10.1016/j.spinee.2021.02.007","title":"Using a national surgical database to predict complications following posterior lumbar surgery and comparing the area under the curve and F1-score for the assessment of prognostic capability","year":2021,"lang":"en","type":"article","venue":"The Spine Journal","topic":"Spine and Intervertebral Disc Pathology","field":"Medicine","cited_by":99,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Medicine; Receiver operating characteristic; Logistic regression; Lumbar; Retrospective cohort study; Area under the curve; Surgery; Cohort; Internal medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.001615675,0.0001054348,0.0002639615,0.00003520217,0.0004363863,0.00006246942,0.0001267977,0.00002537998,0.00002406652],"category_scores_gemma":[0.0005579127,0.0000442955,0.0001562916,0.0001470118,0.0001924568,0.000074249,0.0001691147,0.0002835928,2.176173e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004515518,"about_ca_system_score_gemma":0.0002827877,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003162428,"about_ca_topic_score_gemma":0.0000593558,"domain_scores_codex":[0.9988416,0.0001929361,0.0003756063,0.0001521735,0.0002696816,0.0001679557],"domain_scores_gemma":[0.9991012,0.0001596963,0.0001393067,0.0002685352,0.0002596671,0.00007163712],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007684364,0.0009649012,0.8859918,0.0003961104,0.001911431,0.0001175958,0.002017609,0.001443817,0.09148013,0.007018093,0.005151667,0.00343007],"study_design_scores_gemma":[0.0005475119,0.00008999946,0.9688397,0.0002398366,0.0002841837,0.002621795,0.0007206078,0.026267,0.00006242289,0.0001914324,0.00008437748,0.0000511352],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9693616,0.0008800761,0.009430287,0.01951003,0.0001446622,0.0005480004,0.00003146518,0.000005520568,0.00008834856],"genre_scores_gemma":[0.997987,0.00005176505,0.0009144213,0.0007927471,0.0001692767,0.00003334162,0.00001986892,0.000009470054,0.00002211058],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0914177,"threshold_uncertainty_score":0.3356375,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1725904093113591,"score_gpt":0.4051822759248301,"score_spread":0.2325918666134711,"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."}}