{"id":"W4411040734","doi":"10.1016/j.jdent.2025.105877","title":"Large language models for the screening step in systematic reviews in dentistry","year":2025,"lang":"en","type":"article","venue":"Journal of Dentistry","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Dentistry; Orthodontics; Computer science; 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":[0.002103085,0.00007156775,0.0003952727,0.0001709744,0.00004705294,0.00002897944,0.0001349764,0.00007430634,0.0000162146],"category_scores_gemma":[0.00111467,0.00004768204,0.000131497,0.0002625437,0.00001341381,0.0001184513,0.00001685026,0.0002945973,0.000004483669],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001176581,"about_ca_system_score_gemma":0.0001457936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008755725,"about_ca_topic_score_gemma":0.0004741975,"domain_scores_codex":[0.9984368,0.0001068406,0.001054246,0.00007944319,0.0001558378,0.0001668182],"domain_scores_gemma":[0.9988155,0.0005136129,0.0003409191,0.0001694201,0.0001189883,0.00004155069],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0020767,0.003233707,0.3713951,0.3550979,0.0008928296,0.003472316,0.03159659,0.002850382,0.004943452,0.00837484,0.05242769,0.1636385],"study_design_scores_gemma":[0.004608347,0.0007977686,0.08535601,0.4079653,0.00220638,0.006340939,0.2873496,0.1753123,0.01023469,0.01087081,0.007994846,0.0009630234],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7473937,0.05044835,0.1966826,0.001817694,0.00188991,0.001598949,0.00000435823,0.00000551727,0.0001589047],"genre_scores_gemma":[0.9964785,0.0007146165,0.00131136,0.0003629287,0.0001778926,0.00004437884,0.000002220174,0.000006241268,0.0009018372],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2860391,"threshold_uncertainty_score":0.1944417,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1994065509622011,"score_gpt":0.4799455652121281,"score_spread":0.280539014249927,"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."}}