{"id":"W3164298370","doi":"10.21203/rs.2.11750/v1","title":"Feasibility and Evaluation of a Large-Scale External Validation Approach for Patient-Level Prediction in an International Data Network","year":2019,"lang":"en","type":"preprint","venue":"Research Square","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Scale (ratio); Computer science; Data mining; Geography; Cartography","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":[0.01829283,0.0001638807,0.0002650468,0.0003703813,0.0001232541,0.000235368,0.0019031,0.0002693019,0.00001981056],"category_scores_gemma":[0.001278142,0.0001691443,0.00004378102,0.0003274321,0.00005043536,0.0008202293,0.003776782,0.001026493,0.000001651993],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004728549,"about_ca_system_score_gemma":0.0007790203,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003053541,"about_ca_topic_score_gemma":0.0001796702,"domain_scores_codex":[0.9925845,0.002640253,0.0005803711,0.001346339,0.002412621,0.0004359545],"domain_scores_gemma":[0.9949791,0.0003666943,0.0002902959,0.002203089,0.002051395,0.0001094144],"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":[0.0001195522,0.0003491655,0.7674014,0.001176454,0.00001555078,3.692623e-7,0.002051481,0.1827872,0.00001263729,0.0007366168,0.0001911044,0.04515846],"study_design_scores_gemma":[0.0005355255,0.0001542733,0.3184527,0.000268843,0.000005030705,0.00000116992,0.0001004876,0.6772507,0.000008380408,0.003100247,0.00004379716,0.00007890877],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5310717,0.0003478347,0.4625708,0.0001772552,0.0005494786,0.004042538,0.001030547,0.00003978402,0.0001701684],"genre_scores_gemma":[0.9037182,0.00003834405,0.0920367,0.000008533851,0.0002693671,0.0003564016,0.003542212,0.0000164588,0.00001374296],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4944634,"threshold_uncertainty_score":0.6897506,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.343960059297652,"score_gpt":0.5026205554697459,"score_spread":0.1586604961720939,"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."}}