{"id":"W4404620563","doi":"10.1186/s13643-024-02699-7","title":"Computer-assisted screening in systematic evidence synthesis requires robust and well-evaluated stopping criteria","year":2024,"lang":"en","type":"letter","venue":"Systematic Reviews","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"European Commission; Wellcome Trust","keywords":"Medicine; Systematic review; Evidence-based medicine; Medical physics; MEDLINE; Alternative 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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.509275,0.001540417,0.007117704,0.005864132,0.002653648,0.01202995,0.005052128,0.03013536,0.008487554],"category_scores_gemma":[0.878509,0.002680857,0.007697479,0.006468593,0.006335808,0.01025784,0.005472149,0.02716868,0.004570891],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0076768,"about_ca_system_score_gemma":0.02074794,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002120275,"about_ca_topic_score_gemma":0.003662593,"domain_scores_codex":[0.3285965,0.5348394,0.09300765,0.009817093,0.03171685,0.0020225],"domain_scores_gemma":[0.02306869,0.9373601,0.01420686,0.009007933,0.01472787,0.00162851],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.007082865,0.0004225318,0.005702714,0.02719645,0.006227027,0.001997231,0.002840132,0.006955229,0.0009675983,0.09430168,0.4571385,0.389168],"study_design_scores_gemma":[0.008702395,0.002122125,0.005382009,0.043373,0.004277384,0.002007605,0.0006154128,0.05346973,0.002791204,0.6646739,0.2115024,0.001082881],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.004244949,0.01887874,0.1691885,0.7639266,0.02307932,0.006054566,0.001664576,0.001234048,0.01172874],"genre_scores_gemma":[0.0885044,0.008276119,0.4645177,0.3931369,0.02018035,0.02093528,0.0007292249,0.0004908151,0.003229157],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.490725,"threshold_uncertainty_score":0.6051513,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.128503148657038,"score_gpt":0.3552626784208391,"score_spread":0.2267595297638011,"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."}}