{"id":"W4409970142","doi":"10.7717/peerj-cs.2822","title":"Validation of automated paper screening for esophagectomy systematic review using large language models","year":2025,"lang":"en","type":"article","venue":"PeerJ Computer Science","topic":"Esophageal Cancer Research and Treatment","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of Calgary; University of Toronto; Dalhousie University; University of Ottawa","funders":"","keywords":"Esophagectomy; Computer science; Natural language processing; Medicine; Esophageal cancer; 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":[0.001072753,0.00009656422,0.0004032814,0.0001843652,0.0001257675,0.00003617831,0.0001999228,0.00001954914,0.000008550158],"category_scores_gemma":[0.0001457262,0.0000714228,0.0000868054,0.0007477233,0.00006882332,0.0002661411,0.0001054336,0.00005104984,0.000001755791],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001084268,"about_ca_system_score_gemma":0.0002723816,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000388058,"about_ca_topic_score_gemma":9.804663e-7,"domain_scores_codex":[0.9986577,0.00004386183,0.0002950376,0.0002647241,0.0004709697,0.0002676822],"domain_scores_gemma":[0.9990276,0.00008786745,0.00009602005,0.0003486214,0.0003547135,0.00008521382],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001701505,0.0008738221,0.002261236,0.9592295,0.000599974,0.0001773426,0.001793355,0.007270792,0.01253221,0.007655553,0.001754292,0.005681713],"study_design_scores_gemma":[0.001003188,0.0001353626,0.0004005856,0.04101337,0.0002365772,0.00001662177,0.00001414024,0.9462611,0.01074947,0.0001026933,0.000001829953,0.00006504011],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03124472,0.04968125,0.9145746,0.0005440406,0.0001408747,0.003423221,0.00002391446,0.0002111739,0.0001562381],"genre_scores_gemma":[0.9377853,0.00006831554,0.0612493,0.0006808147,0.0000229843,0.00007622802,0.00001402364,0.000006582972,0.00009640517],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9389904,"threshold_uncertainty_score":0.2912537,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04105738923212442,"score_gpt":0.3866673114043312,"score_spread":0.3456099221722068,"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."}}