{"id":"W4411515984","doi":"10.1002/sim.70176","title":"Precision of Treatment Hierarchy: A Metric for Quantifying Certainty in Treatment Hierarchies From Network Meta‐Analysis","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Actua; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Deutsche Forschungsgemeinschaft; Hellenic Foundation for Research and Innovation; European Commission","keywords":"Metric (unit); Ranking (information retrieval); Hierarchy; Pairwise comparison; Statistics; Mathematics; Frequentist inference; Rank (graph theory); Certainty; Variance (accounting); Similarity (geometry); Computer science; Bayesian probability; Econometrics; Artificial intelligence; Bayesian inference","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2088231,0.00363374,0.008324383,0.02025588,0.001941453,0.007126093,0.003993788,0.004323836,0.003773572],"category_scores_gemma":[0.5735497,0.00178557,0.01293634,0.01510957,0.003814173,0.007576298,0.006467558,0.006650037,0.0004835526],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004137916,"about_ca_system_score_gemma":0.003961599,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003484739,"about_ca_topic_score_gemma":0.003468778,"domain_scores_codex":[0.7194709,0.2171203,0.02325918,0.01842176,0.02090543,0.0008224128],"domain_scores_gemma":[0.2847533,0.6404408,0.03224007,0.03294134,0.008797179,0.0008274579],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002743427,0.0001748314,0.06694666,0.04221006,0.1344106,0.0007324441,0.002390021,0.2121443,0.002510931,0.105776,0.01597301,0.4139877],"study_design_scores_gemma":[0.001538194,0.001520281,0.02423696,0.008256963,0.0489657,0.00123631,0.0004144035,0.2400609,0.005365844,0.6333563,0.03416729,0.0008808535],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01480797,0.02820026,0.9419364,0.002644197,0.0004028456,0.001419927,0.006270031,0.001346162,0.002972102],"genre_scores_gemma":[0.4304458,0.007601338,0.5481261,0.002029583,0.0009283687,0.004379377,0.005322501,0.0006672862,0.0004996928],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7911769,"threshold_uncertainty_score":0.9756619,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7096301178491243,"score_gpt":0.5685468593222872,"score_spread":0.1410832585268371,"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."}}