{"id":"W3027437879","doi":"10.1016/j.jclinepi.2020.04.026","title":"Quality and risk of bias appraisals of systematic reviews are inconsistent across reviewers and centers","year":2020,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":67,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Interquartile range; Confidence interval; Systematic review; Publication bias; Reliability (semiconductor); MEDLINE; Surgery; Internal medicine; 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":[0.7748757,0.003295154,0.01765913,0.03468872,0.005360575,0.01785568,0.007540134,0.005617269,0.002543693],"category_scores_gemma":[0.9366687,0.006137219,0.02321176,0.03220148,0.01021717,0.01378544,0.01289832,0.005795851,0.001048406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01446951,"about_ca_system_score_gemma":0.03296211,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006240461,"about_ca_topic_score_gemma":0.01196635,"domain_scores_codex":[0.0642508,0.5368539,0.2861561,0.02116832,0.08958592,0.001984985],"domain_scores_gemma":[0.02377306,0.7101812,0.1438539,0.02946041,0.09152941,0.001202113],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.003088982,0.0001830965,0.1031801,0.4850965,0.1341502,0.0003546984,0.02578398,0.003139077,0.001500304,0.007755261,0.01687338,0.2188944],"study_design_scores_gemma":[0.005923626,0.002402699,0.1650999,0.5573618,0.1400803,0.002246232,0.01213325,0.01049897,0.004469825,0.03701211,0.06079732,0.001973917],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.09172861,0.6221808,0.1704755,0.03249114,0.0124984,0.04568382,0.009820006,0.001617166,0.01350457],"genre_scores_gemma":[0.6531912,0.07582067,0.2013582,0.008150155,0.004025361,0.05152286,0.003995563,0.001000819,0.0009351704],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2251243,"threshold_uncertainty_score":0.2776184,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9721662925162776,"score_gpt":0.7176930539635952,"score_spread":0.2544732385526824,"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."}}