{"id":"W2750029871","doi":"10.1016/j.jclinepi.2017.06.001","title":"The PRECIS-2 tool has good interrater reliability and modest discriminant validity","year":2017,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":69,"is_retracted":false,"has_abstract":false,"ca_institutions":"North York General Hospital; Centre for Family Medicine; Western University","funders":"Medical Research Council; Chief Scientist Office; University of Aberdeen; Scottish Government; University of Dundee","keywords":"Inter-rater reliability; Discriminant validity; External validity; Intraclass correlation; Reliability (semiconductor); Face validity; Medicine; Linear discriminant analysis; Applied psychology; Validity; Clinical psychology; Psychology; Computer science; Psychometrics; Artificial intelligence; Social psychology; Developmental psychology; Rating scale","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","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":"methods","study_design":"observational","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7460754,0.00023617,0.009188504,0.0000627697,0.0007170996,0.000790519,0.003660971,0.0002609631,0.0005458306],"category_scores_gemma":[0.9041359,0.00007303007,0.003994357,0.00006801746,0.00132486,0.0004016048,0.0005366948,0.0009100093,0.000222253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001639625,"about_ca_system_score_gemma":0.000150045,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001845478,"about_ca_topic_score_gemma":0.00006826854,"domain_scores_codex":[0.7756268,0.156944,0.06307024,0.001245519,0.002534032,0.0005793487],"domain_scores_gemma":[0.333975,0.5556789,0.0946098,0.01157616,0.003394175,0.0007659879],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001166259,0.00008287453,0.8427022,0.00001384051,0.000164653,0.00000790437,0.00006926218,0.00004030404,0.000002288124,0.004093522,0.07826993,0.07443667],"study_design_scores_gemma":[0.0002380977,0.0002669882,0.6458027,0.00003339141,0.0001727641,0.00005740821,0.00005882819,0.006550061,0.000001173603,0.1823428,0.1644044,0.00007135155],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8805658,0.0006110346,0.04148401,0.07116447,0.003472031,0.0003017784,0.00000599161,0.000001368115,0.00239355],"genre_scores_gemma":[0.9891549,0.0005456184,0.006109705,0.0008309113,0.000627193,0.000004309934,2.357515e-7,0.000005884696,0.002721207],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4416519,"threshold_uncertainty_score":0.7622993,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9632173445817618,"score_gpt":0.6924030426664284,"score_spread":0.2708143019153334,"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."}}