{"id":"W1993349052","doi":"10.1016/j.jclinepi.2013.03.003","title":"Testing the Newcastle Ottawa Scale showed low reliability between individual reviewers","year":2013,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":446,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ottawa Hospital; University of Ottawa; University of Alberta","funders":"University of Alberta","keywords":"Confidence interval; Reliability (semiconductor); Medicine; Meta-analysis; Odds ratio; Kappa; Cohort; Scale (ratio); Statistics; Cohort study; Demography; Internal medicine; Mathematics; Geography; Cartography","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.790966,0.0003186971,0.01440829,0.0001735981,0.0001785495,0.0002184506,0.004106568,0.0003656679,0.007135311],"category_scores_gemma":[0.9386033,0.0001096894,0.006148599,0.001050001,0.0005696266,0.0003620388,0.0003105048,0.00154459,0.003421884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000246694,"about_ca_system_score_gemma":0.0002407139,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004578269,"about_ca_topic_score_gemma":0.000009388273,"domain_scores_codex":[0.6699322,0.2289099,0.09433719,0.001274062,0.0047784,0.0007682093],"domain_scores_gemma":[0.03041906,0.8724794,0.0825381,0.006027647,0.007437935,0.001097809],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000002752567,0.00003243337,0.6452075,0.00001557258,0.00012142,0.000001140911,0.0000237287,0.00007411699,0.000001495852,0.00006423008,0.2618393,0.09261633],"study_design_scores_gemma":[0.000205135,0.000209784,0.8413873,0.00008577811,0.0003188798,0.00003253178,0.0001038716,0.001095696,4.777834e-7,0.03978399,0.1166782,0.000098413],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9254113,0.001834396,0.01463071,0.05067459,0.001933529,0.0008618233,0.00001026645,0.000004983956,0.004638396],"genre_scores_gemma":[0.9074231,0.0001572245,0.07731826,0.009996674,0.003009048,0.00001782482,0.000002403647,0.00001890302,0.002056506],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6435696,"threshold_uncertainty_score":0.9973541,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9347665164969218,"score_gpt":0.6547938755714102,"score_spread":0.2799726409255116,"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."}}