{"id":"W4415902225","doi":"10.1016/j.mlwa.2025.100789","title":"Beyond single-run metrics with CP-fuse: A rigorous multi-cohort evaluation of clinico-pathological fusion for improved survival prediction in TCGA","year":2025,"lang":"en","type":"article","venue":"Machine Learning with Applications","topic":"AI in cancer detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke; McGill University; Jewish General Hospital","funders":"Canada Research Chairs","keywords":"Benchmark (surveying); Fusion; Survival analysis; Calibration; Pattern recognition (psychology); Feature (linguistics)","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.01654193,0.002096378,0.001279062,0.001414884,0.0006490286,0.001555155,0.001436942,0.001880427,0.000812119],"category_scores_gemma":[0.02261305,0.0004048459,0.001557558,0.0007468826,0.001149127,0.001615118,0.00224727,0.001927517,0.0004412736],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001103004,"about_ca_system_score_gemma":0.001527418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008376206,"about_ca_topic_score_gemma":0.01159615,"domain_scores_codex":[0.9960119,0.001977328,0.0001915419,0.001039545,0.0005263282,0.0002534349],"domain_scores_gemma":[0.9928204,0.003907522,0.0004464788,0.00142017,0.001008005,0.000397376],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003518405,0.000917735,0.2276986,0.0008739315,0.005576475,0.0004238753,0.0007936372,0.4446764,0.01802569,0.002947527,0.01508341,0.2794644],"study_design_scores_gemma":[0.0001431518,0.002007552,0.03613874,0.0001881433,0.0007192311,0.000472898,0.0002568791,0.9361682,0.01514298,0.004627619,0.003993684,0.0001409126],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8608698,0.007530986,0.118252,0.001390174,0.0003762565,0.0002380199,0.003786546,0.004702601,0.002853643],"genre_scores_gemma":[0.972314,0.0003314417,0.02132807,0.0003903591,0.00006235854,0.0001025891,0.004402697,0.0002769674,0.0007914915],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01654193,"threshold_uncertainty_score":0.08748311,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0315292332960606,"score_gpt":0.3145519740706642,"score_spread":0.2830227407746035,"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."}}