{"id":"W2160968678","doi":"10.1080/01421590802155597","title":"Quality assurance of item writing: During the introduction of multiple choice questions in medicine for high stakes examinations","year":2008,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Medical Education and Admissions","field":"Medicine","cited_by":52,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Alberta","keywords":"Quality assurance; Medical education; Quality (philosophy); MEDLINE; Psychology; Medicine; Political science; Epistemology; Pathology","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":[],"consensus_categories":[],"category_scores_codex":[0.0597975,0.0006926021,0.0007154875,0.001746462,0.0006420528,0.001276213,0.001352141,0.0008326212,0.003364122],"category_scores_gemma":[0.215348,0.0003467514,0.000578617,0.001433895,0.0008818574,0.001409166,0.001626863,0.0007507574,0.00120628],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001348135,"about_ca_system_score_gemma":0.00282912,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00185115,"about_ca_topic_score_gemma":0.002932306,"domain_scores_codex":[0.9579545,0.02626012,0.005924703,0.002485689,0.006329159,0.00104593],"domain_scores_gemma":[0.7790578,0.145068,0.03017309,0.01476394,0.02669294,0.004244077],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002181699,0.001740206,0.4754513,0.001127507,0.00009930263,0.0002265125,0.006432676,0.000845694,0.01221348,0.0004195082,0.00488133,0.4943807],"study_design_scores_gemma":[0.0001795253,0.003418822,0.9634664,0.0004879122,0.00007045993,0.0007816065,0.001304653,0.003646109,0.01695188,0.0003647103,0.009260193,0.00006777487],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.946291,0.0006711272,0.03659023,0.001136602,0.0002131032,0.006413436,0.0007801333,0.0009453775,0.006958921],"genre_scores_gemma":[0.9284793,0.0002011531,0.06346069,0.0003007254,0.0001167943,0.00359897,0.0007734239,0.00007930305,0.002989695],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0597975,"threshold_uncertainty_score":0.3162432,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08315160042075002,"score_gpt":0.3920798300413271,"score_spread":0.308928229620577,"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."}}