{"id":"W4402146094","doi":"10.1093/aje/kwae332","title":"Assessing the certainty of the evidence in systematic reviews: importance, process, and use","year":2024,"lang":"en","type":"article","venue":"American Journal of Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Certainty; Systematic review; Process (computing); Medicine; Systematic process; MEDLINE; Risk analysis (engineering); Computer science; Epistemology; Engineering; Political science; Philosophy; Work in process; Operations management; Law","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4019574,0.0001936528,0.006505883,0.0002509656,0.00006353785,0.0002321423,0.001582453,0.00004961435,0.0001545239],"category_scores_gemma":[0.5722752,0.00005297631,0.001336433,0.002066001,0.0005906631,0.0004910111,0.00008804556,0.0005244748,0.00001949592],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003853648,"about_ca_system_score_gemma":0.0001734523,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003784111,"about_ca_topic_score_gemma":0.00003299459,"domain_scores_codex":[0.8875456,0.0890429,0.02115081,0.0005165001,0.001464156,0.0002800133],"domain_scores_gemma":[0.706802,0.2507699,0.03786622,0.003148209,0.001251085,0.0001626068],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006978858,0.00001577924,0.9611196,0.004346604,0.0002614393,0.00001591717,0.001237334,0.0003569048,0.00003562876,0.003262145,0.01256267,0.01677902],"study_design_scores_gemma":[0.0002696904,0.0006474283,0.6773314,0.1042109,0.003051353,0.00315284,0.02204421,0.04973927,0.0000102889,0.102206,0.03669693,0.0006396774],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8447108,0.1275982,0.01007302,0.01577842,0.0004953801,0.001177812,0.00000169379,0.000001729848,0.0001629346],"genre_scores_gemma":[0.9939331,0.002390479,0.001895085,0.001580609,0.0000547764,0.00001753516,6.179433e-8,0.000005974101,0.0001223719],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2837882,"threshold_uncertainty_score":0.6158108,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7904981829431675,"score_gpt":0.6035752628648117,"score_spread":0.1869229200783559,"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."}}