{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4009609,0.002675691,0.007518878,0.01419218,0.005185362,0.007905691,0.004351264,0.004388341,0.004108429],"category_scores_gemma":[0.7640869,0.003699939,0.01347675,0.009111588,0.0072965,0.007338601,0.009423066,0.005509128,0.002192172],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004934418,"about_ca_system_score_gemma":0.008837329,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008201389,"about_ca_topic_score_gemma":0.01893686,"domain_scores_codex":[0.4181233,0.2609306,0.1993606,0.04053154,0.07586472,0.005189237],"domain_scores_gemma":[0.1066177,0.688329,0.04321095,0.05464771,0.1050191,0.002175562],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005258096,0.0009347632,0.7117068,0.02913933,0.04472923,0.000815817,0.03356622,0.006289247,0.004067342,0.007354912,0.02130092,0.1348372],"study_design_scores_gemma":[0.003267799,0.004433087,0.714798,0.01517747,0.03505852,0.002375102,0.01728936,0.05933008,0.01422249,0.0457221,0.0851653,0.003160641],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6130728,0.03206343,0.2386462,0.008275764,0.01201452,0.03476002,0.01296462,0.003656039,0.0445466],"genre_scores_gemma":[0.8735378,0.001499542,0.1031091,0.001476389,0.0005932339,0.01305978,0.003463403,0.0009886954,0.002272125],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5990391,"threshold_uncertainty_score":0.7387218,"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."}}