{"id":"W1995014100","doi":"10.1016/j.jclinepi.2012.02.012","title":"Treatment effects on patient-important outcomes can be small, even with large effects on surrogate markers","year":2012,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Surrogate endpoint; Medicine; Randomized controlled trial; Confidence interval; Intensive care medicine; Clinical trial; Outcome (game theory); Internal medicine","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":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.009873944,0.0004585995,0.003297107,0.0001496205,0.0000722771,0.000005805333,0.0002655744,0.0004340117,0.00002742644],"category_scores_gemma":[0.07730119,0.0002495947,0.0007077255,0.00009695333,0.0001283215,0.0001140127,0.00005857942,0.001013671,0.00001151768],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002398382,"about_ca_system_score_gemma":0.000127356,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001602425,"about_ca_topic_score_gemma":0.00005018841,"domain_scores_codex":[0.990575,0.004592419,0.003340962,0.0003152396,0.0002959189,0.0008804374],"domain_scores_gemma":[0.8757413,0.1194238,0.003636892,0.0004975449,0.0001314922,0.0005689478],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002137965,0.001873578,0.9199738,0.0001293507,0.000828158,0.0003056612,0.0001748128,0.000009301063,0.00003065573,0.03443595,0.002365598,0.03773516],"study_design_scores_gemma":[0.01253062,0.0938473,0.6748552,0.001776341,0.001138065,0.0003291355,0.0001453696,0.00008680954,0.006346589,0.1922554,0.01551436,0.001174906],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9890312,0.0001243806,0.003052229,0.00573211,0.0009501689,0.0006518249,0.00001485314,0.00008292989,0.0003603219],"genre_scores_gemma":[0.9553344,0.0003145167,0.03729922,0.006357492,0.0004888846,0.00004572829,0.000004744521,0.00005854512,0.00009644997],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2451187,"threshold_uncertainty_score":0.9999956,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3246215329788542,"score_gpt":0.5232699659370967,"score_spread":0.1986484329582426,"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."}}