{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1240627,0.002385621,0.005539012,0.01018257,0.002558925,0.00544758,0.00182464,0.002173288,0.005964065],"category_scores_gemma":[0.3692974,0.002070738,0.01024804,0.007667714,0.002673508,0.00389429,0.005567506,0.002361529,0.002443234],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001548493,"about_ca_system_score_gemma":0.005172408,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001633063,"about_ca_topic_score_gemma":0.008550759,"domain_scores_codex":[0.7908817,0.09107798,0.06422864,0.01237065,0.03944113,0.001999927],"domain_scores_gemma":[0.4262778,0.4450502,0.03897494,0.02473091,0.06341513,0.001551033],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003529352,0.0006557094,0.613945,0.02019529,0.03257153,0.0005346836,0.006075936,0.00330556,0.002876226,0.005011367,0.02757684,0.2837225],"study_design_scores_gemma":[0.003536385,0.002445187,0.8197258,0.008172087,0.02465714,0.004392035,0.005478556,0.02442171,0.01022778,0.02633336,0.06898772,0.001622287],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5957344,0.04029192,0.2432219,0.005339772,0.00441125,0.02765165,0.02452065,0.004377389,0.05445105],"genre_scores_gemma":[0.7370869,0.005423681,0.2090443,0.002019537,0.001207011,0.02642617,0.01000828,0.001341398,0.007442797],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8759373,"threshold_uncertainty_score":0.6561141,"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."}}