{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.006670528,0.0007542106,0.003420562,0.0002877312,0.00009189451,0.0002987584,0.0006880651,0.001097339,0.00002480922],"category_scores_gemma":[0.006599924,0.000495241,0.0004260133,0.0003353464,0.0001812257,0.00001034657,0.0003744558,0.0007293753,0.0001208977],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005940714,"about_ca_system_score_gemma":0.00008210944,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003576633,"about_ca_topic_score_gemma":0.00001973819,"domain_scores_codex":[0.9903124,0.00472158,0.002781152,0.001113696,0.00052814,0.0005430067],"domain_scores_gemma":[0.9961196,0.001325951,0.001130459,0.001221696,0.0001063383,0.00009598477],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000004969838,0.000009640573,0.00001106686,0.7706566,0.0002243029,0.0001650159,0.0001231734,0.000002865004,0.0002808068,0.000001120626,0.2278949,0.0006254707],"study_design_scores_gemma":[0.000162508,0.0001674101,0.0000283317,0.9391176,0.001214491,0.000581944,0.0001750585,0.004381028,0.00005124538,0.00001177299,0.05329049,0.0008181382],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"commentary","genre_scores_codex":[0.002262049,0.8395679,0.02835724,0.102831,0.003321792,0.02293298,0.00004990452,0.0003021498,0.0003749779],"genre_scores_gemma":[0.06541213,0.1164268,0.1831686,0.52672,0.02688785,0.05519746,0.001834753,0.002172693,0.02217977],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.7231411,"threshold_uncertainty_score":0.9997499,"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."}}