{"id":"W4212970901","doi":"10.1002/jmri.28116","title":"Evaluating the Impact of Peer Review on the Completeness of Reporting in Imaging Diagnostic Test Accuracy Research","year":2022,"lang":"en","type":"article","venue":"Journal of Magnetic Resonance Imaging","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Ottawa Hospital; Queen's University; University of Ottawa","funders":"Mitacs; University of Ottawa; Radiological Society of North America; RSNA Research and Education Foundation","keywords":"Diagnostic accuracy; Medicine; Medical physics; Confidence interval; Wilcoxon signed-rank test; Transparency (behavior); Peer review; Radiology; Computer science; Internal medicine","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":[0.8393507,0.001844783,0.00527471,0.02072533,0.005324677,0.01547998,0.007357602,0.00441165,0.003134552],"category_scores_gemma":[0.9536982,0.002584475,0.007075739,0.02318013,0.01211427,0.01577881,0.01326366,0.00362163,0.000843637],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01143641,"about_ca_system_score_gemma":0.03775955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004441834,"about_ca_topic_score_gemma":0.004106014,"domain_scores_codex":[0.06544214,0.5810275,0.2284163,0.01341118,0.1094093,0.002293624],"domain_scores_gemma":[0.008523824,0.687822,0.1560399,0.04760594,0.09807982,0.001928461],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004899896,0.0004264084,0.5282212,0.06505162,0.01586285,0.0006144158,0.03129268,0.002920647,0.001335943,0.008420814,0.01844866,0.3225048],"study_design_scores_gemma":[0.003090189,0.006507257,0.6660649,0.1052721,0.0190297,0.002389348,0.014936,0.01892073,0.009363058,0.02973244,0.1230519,0.001642498],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4141277,0.1858176,0.1973526,0.06434264,0.01382458,0.06661704,0.01362793,0.001726143,0.04256376],"genre_scores_gemma":[0.8490568,0.01385107,0.1011256,0.003618129,0.002982275,0.02517798,0.002936962,0.0004270667,0.0008241194],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1606493,"threshold_uncertainty_score":0.1981093,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8184578349451404,"score_gpt":0.6398863174884083,"score_spread":0.1785715174567321,"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."}}