{"id":"W2981744747","doi":"10.1503/cmaj.190085","title":"Testing group differences for confounder selection in nonrandomized studies: flawed practice","year":2019,"lang":"en","type":"article","venue":"Canadian Medical Association Journal","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Observational study; Confounding; Random assignment; Selection (genetic algorithm); Medicine; Key (lock); Selection bias; Intervention (counseling); Computer science; Family medicine; Pathology; Machine learning; Computer security; Nursing","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004960869,0.0001363086,0.0004348786,0.0001833886,0.000160263,0.0000790748,0.0001587,0.0002839669,0.0003417404],"category_scores_gemma":[0.07677388,0.0001182959,0.00006651554,0.0002884769,0.00002656782,0.0003701227,0.00001187745,0.0007502997,0.00001823662],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001813983,"about_ca_system_score_gemma":0.001070482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006109552,"about_ca_topic_score_gemma":0.01434943,"domain_scores_codex":[0.9977017,0.0003538546,0.0006077392,0.0001662809,0.0007267112,0.0004437034],"domain_scores_gemma":[0.9869367,0.01112246,0.000533185,0.00005864773,0.000969158,0.000379867],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001768171,0.001084288,0.3658244,0.0008169887,0.003347672,0.0004684981,0.01230948,0.00004208994,0.003317118,0.3126996,0.1470752,0.1512465],"study_design_scores_gemma":[0.03845836,0.001681644,0.02116205,0.002151683,0.0003667319,0.0009074535,0.01999763,0.01008927,0.0001921176,0.7958429,0.1074808,0.001669404],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8457897,0.0005921072,0.09075265,0.03478289,0.003440503,0.00474467,0.00003443639,0.0004567224,0.01940637],"genre_scores_gemma":[0.9264741,0.0001170971,0.06599211,0.005737761,0.0009310947,0.0001476824,0.000003881932,0.00004181492,0.0005544889],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4831433,"threshold_uncertainty_score":0.9310029,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1467618169955665,"score_gpt":0.4167907857041335,"score_spread":0.270028968708567,"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."}}