{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7628012,0.0002960405,0.0362667,0.0001116193,0.00004957039,0.0000394495,0.00111837,0.000232208,0.0001741833],"category_scores_gemma":[0.9572428,0.0001202548,0.006018223,0.0005033153,0.000584475,0.0001419644,0.0002500695,0.0005923333,0.00003794868],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000007882539,"about_ca_system_score_gemma":0.00006224335,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001331156,"about_ca_topic_score_gemma":0.0000122831,"domain_scores_codex":[0.3942958,0.4402826,0.1607586,0.001201554,0.003018223,0.0004431467],"domain_scores_gemma":[0.09767225,0.5852872,0.3088566,0.003286855,0.003721363,0.001175737],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006318164,0.00005612019,0.9635451,0.01598287,0.000721036,0.000002941291,0.000266881,0.00001854002,0.000004661596,0.0003595536,0.0105604,0.008418741],"study_design_scores_gemma":[0.001323986,0.0009092925,0.9360446,0.01136075,0.002571701,0.0001070446,0.002509441,0.001728644,0.000002530218,0.005822103,0.03734501,0.0002748387],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8568959,0.1145408,0.01564511,0.01051771,0.0007063505,0.001490665,0.00005635703,0.000001619389,0.0001455771],"genre_scores_gemma":[0.8689923,0.1064303,0.01942667,0.004793494,0.0002606994,0.00001087058,8.712857e-7,0.00001266353,0.00007208868],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2966236,"threshold_uncertainty_score":0.9780409,"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."}}