{"id":"W4409572808","doi":"10.1007/s00464-025-11723-3","title":"Large language model-generated clinical practice guideline for appendicitis","year":2025,"lang":"en","type":"article","venue":"Surgical Endoscopy","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Guideline; Appendicitis; Clinical Practice; Medicine; Computer science; General surgery; Family medicine; Pathology","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008884112,0.0005643403,0.0003200391,0.001027915,0.0002968107,0.0007940018,0.0008613946,0.001231474,0.00461247],"category_scores_gemma":[0.009579121,0.0002403629,0.0007568432,0.0005555122,0.0001860806,0.0003941238,0.0004330265,0.0009367673,0.001351649],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001345403,"about_ca_system_score_gemma":0.00280273,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03698634,"about_ca_topic_score_gemma":0.08526151,"domain_scores_codex":[0.9992393,0.0003595506,0.00007444349,0.0001412934,0.000139138,0.0000463102],"domain_scores_gemma":[0.9945617,0.004162557,0.0001409176,0.0001614986,0.0008588137,0.0001144101],"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":[0.002335997,0.001423304,0.03995615,0.002707089,0.0007700684,0.006736691,0.001132791,0.3857948,0.01700595,0.009154627,0.2484365,0.2845461],"study_design_scores_gemma":[0.0005636854,0.0002186456,0.005330971,0.0002179388,0.0002265872,0.0006954832,0.0002708469,0.955838,0.005389036,0.005909674,0.02527611,0.00006307582],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.502398,0.002393225,0.2767248,0.01474849,0.001671317,0.002615638,0.1501984,0.03026484,0.01898535],"genre_scores_gemma":[0.6975436,0.0005110328,0.2016981,0.001757931,0.0001078592,0.0007389865,0.09313467,0.0004612602,0.004046645],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03698634,"threshold_uncertainty_score":0.07354212,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1200831951731252,"score_gpt":0.5367283093942888,"score_spread":0.4166451142211637,"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."}}