{"id":"W7117158505","doi":"10.1093/jamia/ocaf223","title":"AutoReporter: development of an artificial intelligence tool for automated assessment of research reporting guideline adherence","year":2025,"lang":"en","type":"article","venue":"Journal of the American Medical Informatics Association","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Princess Margaret Cancer Centre; BC Cancer Agency; McMaster University; University of Toronto","funders":"","keywords":"Guideline; Quality (philosophy); Control (management); MEDLINE; Scalability","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":[0.3835296,0.0001179523,0.002182666,0.0004918512,0.000179687,0.0002182169,0.001658979,0.00007630287,0.0001722694],"category_scores_gemma":[0.3395412,0.00005518744,0.0007648629,0.002180646,0.0001649618,0.0002537804,0.0001885379,0.000350386,0.000003120181],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003369433,"about_ca_system_score_gemma":0.002931589,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001277988,"about_ca_topic_score_gemma":0.00001497754,"domain_scores_codex":[0.9415314,0.003242834,0.04234017,0.0001541667,0.01248837,0.0002430457],"domain_scores_gemma":[0.8564183,0.007307616,0.1224697,0.001225841,0.01243936,0.0001392554],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000515003,0.0005240189,0.07548559,0.0003862385,0.00142951,0.000003532603,0.006520748,0.004495254,0.0008232875,0.003748445,0.04371,0.8628219],"study_design_scores_gemma":[0.0001069881,0.0002236044,0.02756406,0.0004032122,0.0001581721,0.00001023569,0.007642193,0.9521688,0.00145524,0.006763876,0.003415037,0.00008862056],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6197767,0.00001035611,0.3783488,0.0009052645,0.0003150199,0.0003344069,0.000003437603,0.000003127968,0.0003028533],"genre_scores_gemma":[0.7288002,0.000004779155,0.270802,0.0001850647,0.00005129275,0.000008708184,0.000002811891,0.000003256527,0.0001418504],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9476735,"threshold_uncertainty_score":0.6660221,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.642779943123821,"score_gpt":0.6432471026700052,"score_spread":0.0004671595461841482,"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."}}