{"id":"W4384155640","doi":"10.1101/2023.07.13.23292601","title":"Design differences explain variation in results between randomized trials and their non-randomized emulations","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"U.S. Food and Drug Administration; Hamilton Health Sciences Foundation; National Institutes of Health; Brigham and Women's Hospital","keywords":"Randomized controlled trial; Medicine; Confounding; Discontinuation; Randomized experiment; Research design; Sample size determination; Completely randomized design; Randomization; Observational study; Statistics; Surgery; Internal medicine","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5913207,0.003957021,0.007437991,0.006132916,0.001142287,0.006798507,0.006227009,0.005398124,0.005865934],"category_scores_gemma":[0.7772731,0.002746262,0.02391922,0.006135905,0.007713663,0.005542611,0.006001106,0.005483584,0.001189074],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004348691,"about_ca_system_score_gemma":0.003121343,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001791788,"about_ca_topic_score_gemma":0.001524884,"domain_scores_codex":[0.1891778,0.6632736,0.07300937,0.04535799,0.0266077,0.002573503],"domain_scores_gemma":[0.1265685,0.7520938,0.04528217,0.06626955,0.009159,0.0006270081],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01890487,0.001146418,0.1730767,0.05656978,0.3173149,0.002324482,0.007184472,0.0450724,0.003855634,0.1068996,0.01274993,0.2549009],"study_design_scores_gemma":[0.0157882,0.008544485,0.1476403,0.02783497,0.1492907,0.002619659,0.002471127,0.1717614,0.01540312,0.4067048,0.05057264,0.001368486],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08223014,0.03393291,0.83982,0.006160224,0.00456428,0.01915414,0.003763497,0.001727354,0.008647361],"genre_scores_gemma":[0.684975,0.002900236,0.2758,0.004229551,0.0006440364,0.02762608,0.001784056,0.0005402812,0.001500811],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4086793,"threshold_uncertainty_score":0.5039744,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4261275238457517,"score_gpt":0.4427547650123362,"score_spread":0.01662724116658454,"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."}}