{"id":"W6904814505","doi":"10.1371/journal.pone.0311772.s003","title":"Evaluation of articles by the Newcastle-Ottawa Quality Assessment Scale.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Risk assessment; Quality assurance; Data collection","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2172605,0.003215136,0.01904902,0.04143264,0.003072634,0.007498552,0.005747393,0.002382282,0.01419292],"category_scores_gemma":[0.5226187,0.001797957,0.03135214,0.0240422,0.00322914,0.00626782,0.005661095,0.002653716,0.001695303],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006694953,"about_ca_system_score_gemma":0.01777616,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009987869,"about_ca_topic_score_gemma":0.01905816,"domain_scores_codex":[0.655697,0.07464469,0.2120622,0.01067903,0.04536816,0.001548945],"domain_scores_gemma":[0.402196,0.3350256,0.1187404,0.03283468,0.1079104,0.003292968],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.004130111,0.0001345922,0.02154106,0.7615348,0.09606884,0.0002723913,0.002463494,0.0009440723,0.001331833,0.002032319,0.02946986,0.08007664],"study_design_scores_gemma":[0.01106772,0.002052068,0.1080568,0.3664975,0.3366691,0.001173359,0.003204741,0.005289303,0.004178023,0.01563302,0.1448459,0.001332548],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.0431363,0.5014356,0.0772155,0.009674991,0.00667264,0.1446459,0.192154,0.002810004,0.02225521],"genre_scores_gemma":[0.349897,0.1048914,0.1666124,0.003345213,0.001737806,0.3027648,0.06339143,0.001169469,0.006190412],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7827395,"threshold_uncertainty_score":0.9652572,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9372789554589371,"score_gpt":0.6474359821440813,"score_spread":0.2898429733148559,"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."}}