{"id":"W7117367844","doi":"10.1016/j.imed.2025.12.008","title":"The performance of Covidence: An artificial intelligence-based tool for title and abstract screening in a breast cancer evidence-based clinical practice guideline","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hamilton Health Sciences; Population Health Research Institute; McMaster University; Impact","funders":"Hamilton Health Sciences Foundation","keywords":"Workload; Clinical Practice; Guideline; Clinical trial; Breast cancer; Stage (stratigraphy); Identification (biology)","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2576095,0.002397568,0.008291963,0.03296618,0.001187337,0.008884719,0.003212571,0.002683987,0.005943136],"category_scores_gemma":[0.6967176,0.001949212,0.01245067,0.01944467,0.001795182,0.009149944,0.006015779,0.002124287,0.00123959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003075589,"about_ca_system_score_gemma":0.01038423,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002814683,"about_ca_topic_score_gemma":0.006028994,"domain_scores_codex":[0.5476261,0.2928877,0.1054113,0.01112652,0.04166434,0.001284012],"domain_scores_gemma":[0.1727069,0.7049015,0.07180505,0.01516592,0.03359388,0.001826646],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02140273,0.0007878057,0.09233755,0.1636501,0.05595041,0.0002716078,0.004025036,0.004347207,0.002600444,0.002679391,0.02042901,0.6315187],"study_design_scores_gemma":[0.04501571,0.01807965,0.3113245,0.1689623,0.2181987,0.003054861,0.004756244,0.06404837,0.01514442,0.02587681,0.1223366,0.003201924],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4097265,0.2484207,0.1557116,0.01976325,0.003878289,0.0717438,0.04115687,0.005425378,0.04417367],"genre_scores_gemma":[0.5959985,0.02158269,0.3300968,0.002823829,0.001056306,0.04055312,0.006306697,0.000426319,0.00115584],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7423905,"threshold_uncertainty_score":0.9154997,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.361299736082622,"score_gpt":0.5694698143195533,"score_spread":0.2081700782369313,"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."}}