{"id":"W2000477303","doi":"10.1097/acm.0b013e3181637925","title":"The Reported Validity and Reliability of Methods for Evaluating Continuing Medical Education: A Systematic Review","year":2008,"lang":"en","type":"review","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":69,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Johns Hopkins University","keywords":"Reliability (semiconductor); Medical education; MEDLINE; Validity; Continuing medical education; Psychology; Continuing education; Medicine; Clinical psychology; Psychometrics; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0331861,0.0004793893,0.005108916,0.0002316826,0.0002125796,0.000004169063,0.0005114562,0.0009206451,0.00008633632],"category_scores_gemma":[0.4503038,0.0002535103,0.0003321325,0.001173776,0.001037707,0.00005634914,0.0001113802,0.002110413,0.000002590549],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002927302,"about_ca_system_score_gemma":0.003507096,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001186839,"about_ca_topic_score_gemma":1.471133e-7,"domain_scores_codex":[0.9898832,0.001927849,0.005814861,0.0006819689,0.001320927,0.000371147],"domain_scores_gemma":[0.9818696,0.009695758,0.004502195,0.00157867,0.002050919,0.0003028809],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000004601812,0.00001413496,0.000003253568,0.5389181,0.000080274,6.349362e-7,0.000199608,1.568145e-9,2.180271e-7,0.000157702,0.02147067,0.4391508],"study_design_scores_gemma":[0.0002779376,0.0001422195,0.000002119272,0.6005931,0.00665143,0.001560648,0.0001699942,0.00009436564,3.128399e-7,0.0002582342,0.3901227,0.000126892],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001692128,0.9801853,0.002826379,0.005844231,0.001341608,0.009560312,0.000002229333,0.00004821771,0.0001747768],"genre_scores_gemma":[0.00000266678,0.9844615,0.008644837,0.002017693,0.0009472821,0.002860861,0.0002244997,0.0000589736,0.0007817215],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4390239,"threshold_uncertainty_score":0.9999917,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2238340500536985,"score_gpt":0.5939144964063777,"score_spread":0.3700804463526792,"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."}}