{"id":"W4408937960","doi":"10.2196/58366","title":"The AI Reviewer: Evaluating AI’s Role in Citation Screening for Streamlined Systematic Reviews","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; Montfort Hospital; University of Ottawa; University of Toronto","funders":"","keywords":"Citation; Systematic review; Medicine; Data science; Information retrieval; Computer science; Library science; MEDLINE; Chemistry","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.7665437,0.003219709,0.009485959,0.02587725,0.008704009,0.01976492,0.005922846,0.01399873,0.01334133],"category_scores_gemma":[0.9549509,0.003578372,0.008960449,0.02332144,0.01038355,0.02094177,0.009554737,0.009829503,0.004538109],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01040483,"about_ca_system_score_gemma":0.05997236,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005019825,"about_ca_topic_score_gemma":0.01255575,"domain_scores_codex":[0.1405871,0.5701164,0.2052788,0.02155813,0.0597888,0.00267081],"domain_scores_gemma":[0.017721,0.6997028,0.06702022,0.03456688,0.1761653,0.0048238],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0113241,0.0005002554,0.05726643,0.1625205,0.02050334,0.0008614794,0.02263834,0.00172375,0.003504853,0.01517603,0.3625827,0.3413982],"study_design_scores_gemma":[0.02283257,0.00434204,0.04644021,0.2691434,0.0651081,0.003184931,0.01178105,0.04979596,0.01376611,0.1126644,0.3973794,0.003561879],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.04924495,0.0777831,0.163689,0.4860041,0.1452483,0.03549346,0.009537474,0.006276893,0.0267227],"genre_scores_gemma":[0.3755692,0.01171409,0.4311944,0.09309657,0.02137002,0.05615414,0.002302514,0.002104705,0.006494372],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2334563,"threshold_uncertainty_score":0.2878932,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5049512980112671,"score_gpt":0.6412632887552181,"score_spread":0.136311990743951,"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."}}