{"id":"W2028270592","doi":"10.1046/j.1365-2923.2003.01713.x","title":"Evaluating surgeons’ informed decision making skills: pilot test using a videoconferenced standardised patient","year":2003,"lang":"en","type":"article","venue":"Medical Education","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Test (biology); Medical education; Educational measurement; Medicine; Psychology; MEDLINE; Medical physics; Curriculum; Pedagogy; 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","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001701827,0.0002430578,0.0003642842,0.0002785544,0.001422373,0.00002977355,0.0003741731,0.0002861276,0.00306007],"category_scores_gemma":[0.1091351,0.0002228829,0.00006771403,0.0006257452,0.0001092822,0.0002836971,0.0001756316,0.0009727213,0.0001404334],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00141543,"about_ca_system_score_gemma":0.02487581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004699474,"about_ca_topic_score_gemma":0.0004179847,"domain_scores_codex":[0.9939339,0.001763427,0.00161449,0.0004095794,0.001591076,0.0006874898],"domain_scores_gemma":[0.9884101,0.007859048,0.0008686649,0.0010303,0.00139024,0.0004416625],"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":[0.0002532929,0.002000298,0.3982533,0.0009061152,0.00003586919,0.000002396814,0.03983404,0.00009629779,0.0001761566,0.002696294,0.01347372,0.5422723],"study_design_scores_gemma":[0.01963071,0.004950954,0.2856962,0.0681441,0.0004090291,0.0001013888,0.2152925,0.08260957,0.0004709514,0.04293263,0.2754091,0.004352903],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9784225,0.0009144846,0.003574391,0.00199618,0.00448051,0.002951035,0.00002332591,0.0002057553,0.007431786],"genre_scores_gemma":[0.9697936,0.00006012708,0.02374222,0.005445366,0.0001049828,0.000690974,0.00006395647,0.00003951249,0.00005923116],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5379193,"threshold_uncertainty_score":0.9998776,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2375408128621948,"score_gpt":0.5363466417264319,"score_spread":0.298805828864237,"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."}}