{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.1468213,0.0006058374,0.005351176,0.0006690356,0.0002185927,0.00009208074,0.001720461,0.002641811,0.00274929],"category_scores_gemma":[0.7325845,0.0004735393,0.0008399835,0.0009936922,0.0006404538,0.0001845057,0.002142818,0.002441234,0.0001695162],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001227702,"about_ca_system_score_gemma":0.002598795,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002541579,"about_ca_topic_score_gemma":0.0001560037,"domain_scores_codex":[0.9182609,0.06328966,0.01289608,0.002414907,0.001824931,0.001313484],"domain_scores_gemma":[0.4394219,0.5555419,0.002251476,0.001891236,0.0004783746,0.0004151213],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.006677102,0.002650171,0.001857132,0.0001015506,0.0001457274,0.0001972253,0.0001417714,0.000006085631,0.0003839883,0.07452884,0.03391885,0.8793916],"study_design_scores_gemma":[0.01050792,0.003352854,0.008498046,0.0006407902,0.0002406754,0.00001613291,0.0001017481,0.000638385,0.0004899161,0.8808959,0.09421265,0.0004050496],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06025844,0.0008938694,0.6064929,0.01429989,0.04285269,0.01194842,0.0002146689,0.0006290539,0.2624101],"genre_scores_gemma":[0.009862271,0.0008937158,0.9832104,0.002095479,0.001778531,0.0006495455,0.00002385368,0.00008222294,0.001403911],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8789865,"threshold_uncertainty_score":0.9998602,"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."}}