{"id":"W1980472079","doi":"10.1136/bmj.c117","title":"Is a subgroup effect believable? Updating criteria to evaluate the credibility of subgroup analyses","year":2010,"lang":"en","type":"article","venue":"BMJ","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":812,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Subgroup analysis; Credibility; Psychology; Computer science; Information retrieval; Mathematics; Statistics; Political science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05210993,0.0001989426,0.0005820959,0.0003848359,0.0002585465,0.0002472036,0.001533039,0.0001034805,0.001950391],"category_scores_gemma":[0.3680751,0.0001033338,0.0002576032,0.003468795,0.0001644727,0.000178368,0.0004634486,0.0003171815,0.0001281999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001466392,"about_ca_system_score_gemma":0.00005704839,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003766624,"about_ca_topic_score_gemma":0.00007445804,"domain_scores_codex":[0.9941485,0.001821451,0.001201906,0.0007075477,0.001716189,0.0004043984],"domain_scores_gemma":[0.9256036,0.07074651,0.0005233983,0.002279896,0.0006732249,0.0001733789],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000197372,0.00007285809,0.6027064,0.00002913739,0.00006816762,0.000005413349,0.001032575,0.0000525382,0.1972111,0.0002506351,0.06077076,0.137603],"study_design_scores_gemma":[0.0004405424,0.0003618735,0.950068,0.00002947949,0.00006166435,0.00001965875,0.0006044412,0.006016744,0.02582115,0.01304283,0.003298929,0.0002346762],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9839177,0.00007732087,0.003968733,0.002394823,0.001428874,0.0005750382,0.00002126142,0.00004360427,0.007572626],"genre_scores_gemma":[0.9683033,0.000001679801,0.03063382,0.0005131312,0.0003049655,0.00003712335,0.000001282904,0.00001086592,0.0001938963],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3473616,"threshold_uncertainty_score":0.998962,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5243444902286908,"score_gpt":0.5884504748589897,"score_spread":0.06410598463029893,"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."}}