{"id":"W4414223584","doi":"10.1016/j.jvs.2025.06.062","title":"Evaluating Vascularis, A Modular Retrieval-augmented Generation-based Vascular Surgery Artificial Intelligence Chat Interface, Against General Purpose Large Language Models in Management of Complex Aortic Disease","year":2025,"lang":"en","type":"article","venue":"Journal of Vascular Surgery","topic":"Renal and Vascular Pathologies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Dalhousie University; University of Toronto; University Health Network; Sunnybrook Health Science Centre; St. Michael's Hospital","funders":"","keywords":"Modular design; Vascular surgery; Complex disease; Disease; Aortic surgery","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.00711479,0.0004812777,0.001745352,0.001640853,0.000128753,0.00007743025,0.0003085515,0.0001915412,0.00007496685],"category_scores_gemma":[0.0009469963,0.0004149961,0.002461632,0.001465663,0.0001345645,0.0003166473,0.0001700079,0.0005231738,0.000006020374],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002973846,"about_ca_system_score_gemma":0.0006751791,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002807133,"about_ca_topic_score_gemma":0.000007008879,"domain_scores_codex":[0.9936439,0.0008995092,0.002541811,0.0005873625,0.001664298,0.0006631654],"domain_scores_gemma":[0.9968331,0.0003719246,0.0008215694,0.000991308,0.0006873512,0.0002947521],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003524477,0.006728292,0.03442128,0.007682927,0.009061361,0.006310517,0.0009320316,0.6670728,0.1313597,0.0008788338,0.0008940903,0.1311336],"study_design_scores_gemma":[0.002441565,0.0001959358,0.04943028,0.006506518,0.002491827,0.0000348659,0.001025755,0.8945626,0.04128978,0.0009761829,0.0002298393,0.0008148582],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8333256,0.01383861,0.1512788,0.0002709316,0.0004945791,0.0007176751,0.00001555305,0.00003543679,0.00002288522],"genre_scores_gemma":[0.9910727,0.003728955,0.004433509,0.0002992556,0.0001854009,0.00003396762,0.0001448918,0.0000549413,0.00004642248],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2274898,"threshold_uncertainty_score":0.9998302,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09065741694089986,"score_gpt":0.3546862267299372,"score_spread":0.2640288097890373,"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."}}