{"id":"W4410629997","doi":"10.1136/bmjopen-2024-096107","title":"Completeness of reporting of simulation studies on responder analysis methods and simulation performance: a methodological survey","year":2025,"lang":"en","type":"review","venue":"BMJ Open","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Joseph’s Healthcare Hamilton; Impact; McMaster University; Ontario Clinical Oncology Group","funders":"","keywords":"Medicine; Binary data; Data extraction; Covariate; Statistics; Bayesian probability; Computer science; Binary number; MEDLINE; Mathematics","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":"reporting","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","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":["metaresearch"],"category_scores_codex":[0.6312907,0.001961874,0.009554079,0.02779094,0.002325673,0.01034205,0.006792048,0.005476682,0.005406721],"category_scores_gemma":[0.8891625,0.003773749,0.01369694,0.02775233,0.00573692,0.01152373,0.007795353,0.003437674,0.001311075],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01574327,"about_ca_system_score_gemma":0.03802335,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007357786,"about_ca_topic_score_gemma":0.005644303,"domain_scores_codex":[0.232604,0.3648556,0.3258077,0.01106967,0.06350943,0.00215375],"domain_scores_gemma":[0.0652179,0.66315,0.1431087,0.03375544,0.09328873,0.001479395],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.001375608,0.00009740274,0.02445579,0.7068067,0.01544785,0.00015038,0.005807817,0.001320057,0.0004615344,0.006841274,0.01608932,0.2211464],"study_design_scores_gemma":[0.0009659057,0.0005702788,0.01438113,0.894165,0.01843086,0.0005145764,0.002744609,0.001463479,0.001277506,0.008634133,0.05657182,0.0002806798],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.01868812,0.8188035,0.08283174,0.02376693,0.003640338,0.03445202,0.01145098,0.000464036,0.005902228],"genre_scores_gemma":[0.2250342,0.4727159,0.131772,0.01302475,0.001499183,0.1425669,0.01138681,0.0006389386,0.00136121],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3687093,"threshold_uncertainty_score":0.4546843,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9393145250890469,"score_gpt":0.7662681919119205,"score_spread":0.1730463331771264,"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."}}