{"id":"W3121451510","doi":"10.1200/op.20.00810","title":"Readiness of Healthcare Systems to Generate Real-World Evidence: Reliability of CT Radiographic End Points for Evaluation of First-Line Systemic Treatment","year":2021,"lang":"en","type":"article","venue":"JCO Oncology Practice","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Reliability (semiconductor); Health care; Healthcare system; Baseline (sea); Radiography; Resource (disambiguation); Clinical Practice","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"],"consensus_categories":[],"category_scores_codex":[0.007409962,0.0001694432,0.001039122,0.0002787496,0.00006653566,0.000006614541,0.0001078086,0.0001051455,0.00002162603],"category_scores_gemma":[0.01649395,0.0001413258,0.0001732429,0.0007159111,0.0001205198,0.0001360899,0.00003866281,0.0002325235,0.000001299143],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008222941,"about_ca_system_score_gemma":0.001533171,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005768731,"about_ca_topic_score_gemma":0.0005658211,"domain_scores_codex":[0.9960289,0.001579929,0.00111977,0.000432053,0.0006054366,0.0002338819],"domain_scores_gemma":[0.9906964,0.004378399,0.001110691,0.0006244066,0.003006757,0.0001833652],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01440415,0.008668859,0.6804613,0.03496944,0.003280615,0.0005207642,0.006151518,0.04834535,0.06488925,0.0048939,0.003620876,0.129794],"study_design_scores_gemma":[0.06026929,0.03701685,0.1500829,0.03102183,0.02931702,0.00736248,0.01020842,0.4972117,0.05825092,0.0009363563,0.1167387,0.001583548],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9558812,0.00417574,0.001157187,0.03486886,0.00091794,0.002321219,0.00003631849,0.00002229018,0.0006191998],"genre_scores_gemma":[0.9902932,0.00276648,0.005916981,0.0002767241,0.0002128941,0.0002621587,0.00006017044,0.00002370623,0.0001877067],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5303784,"threshold_uncertainty_score":0.9917905,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08495731440286869,"score_gpt":0.4554308684702615,"score_spread":0.3704735540673929,"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."}}