{"id":"W4388180784","doi":"10.1016/j.jclinepi.2023.10.005","title":"GRADE guidance 37: rating imprecision in a body of evidence on test accuracy","year":2023,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Certainty; Test (biology); Grading (engineering); Sample size determination; Rating scale; Statistics; Sample (material); Computer science; Confidence interval; Mathematics; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.335896,0.0001747558,0.002997884,0.0008997353,0.00006277871,0.00002095412,0.001741835,0.0003309564,0.00006615427],"category_scores_gemma":[0.9949576,0.0000942895,0.0007430693,0.003552208,0.000279834,0.0003227582,0.0003206743,0.001326839,0.00007615453],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003071709,"about_ca_system_score_gemma":0.0002186993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002078828,"about_ca_topic_score_gemma":0.000004564929,"domain_scores_codex":[0.9614305,0.01977395,0.01601141,0.0007261938,0.001403886,0.0006540552],"domain_scores_gemma":[0.04198987,0.9459634,0.01038013,0.0007321291,0.0006877374,0.0002467577],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001937464,0.0001036915,0.7508121,0.000007097396,0.00001023018,0.00004654893,0.00002882379,0.001237062,0.0005124688,0.0002056372,0.008793155,0.2380495],"study_design_scores_gemma":[0.0006126172,0.001598725,0.8909871,0.0005615174,0.000007331255,0.00003080034,0.000117943,0.00811046,0.00007852942,0.09604509,0.001762038,0.00008783094],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9692255,0.001042065,0.01182436,0.01439313,0.002952668,0.0001309789,0.000003516508,0.00001893143,0.0004088917],"genre_scores_gemma":[0.9505088,0.002422502,0.04509067,0.001237463,0.0006294528,0.00000227096,1.732524e-7,0.000009462829,0.00009917351],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9261894,"threshold_uncertainty_score":0.6838349,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9563425643518924,"score_gpt":0.7288577227698326,"score_spread":0.2274848415820597,"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."}}