{"id":"W4417311632","doi":"10.1038/s41698-025-01231-x","title":"Benchmarking progression-free survival ratio as primary endpoint in precision oncology clinical trials","year":2025,"lang":"en","type":"article","venue":"npj Precision Oncology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canada's Michael Smith Genome Sciences Centre","funders":"National Cancer Institute","keywords":"Censoring (clinical trials); Sample size determination; Clinical trial; Benchmarking; Clinical endpoint; Survival analysis; Progression-free survival; Correlation","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2594666,0.001269508,0.002565075,0.002958512,0.0004658585,0.004062683,0.001564423,0.001296567,0.004116138],"category_scores_gemma":[0.4493569,0.0004987196,0.003040414,0.003648199,0.001367397,0.001734883,0.003126233,0.002142678,0.001069231],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001384851,"about_ca_system_score_gemma":0.003663561,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004842285,"about_ca_topic_score_gemma":0.0003750002,"domain_scores_codex":[0.6052151,0.3545675,0.01423927,0.005765825,0.01911188,0.001100338],"domain_scores_gemma":[0.6315128,0.2905045,0.0352705,0.02423415,0.01715329,0.001324827],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.02253397,0.000762175,0.06522569,0.01534549,0.01754482,0.000296235,0.001611137,0.0881481,0.005506522,0.0516833,0.02724692,0.7040956],"study_design_scores_gemma":[0.01371606,0.03440651,0.1247604,0.009761071,0.0166686,0.001613476,0.0009601664,0.3870243,0.04371134,0.2145609,0.152099,0.0007183891],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09886632,0.02722615,0.8361421,0.002978298,0.001369316,0.009705431,0.004764146,0.003219823,0.01572846],"genre_scores_gemma":[0.7358984,0.002278394,0.2428422,0.00124844,0.0003429427,0.01277051,0.002735096,0.0007893458,0.001094687],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7405334,"threshold_uncertainty_score":0.9132095,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6552909208156281,"score_gpt":0.6898935354422623,"score_spread":0.03460261462663416,"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."}}