{"id":"W2891885683","doi":"10.1016/j.hpe.2018.09.001","title":"Guidelines for Creating Written Clinical Reasoning Exams: Insight from a Delphi Study","year":2018,"lang":"en","type":"article","venue":"Health Professions Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke","funders":"","keywords":"Likert scale; Delphi method; Guideline; Medical education; Scale (ratio); Delphi; Psychology; Snowball sampling; Relevance (law); Medicine; Computer science; Artificial intelligence; Pathology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.172491,0.001104286,0.0007430688,0.003874173,0.006252605,0.007178156,0.003013011,0.002842681,0.002690427],"category_scores_gemma":[0.2351078,0.001197389,0.0008217284,0.003134922,0.005747788,0.007870426,0.009573048,0.004422909,0.0008747315],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005986519,"about_ca_system_score_gemma":0.01465613,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005004229,"about_ca_topic_score_gemma":0.007533499,"domain_scores_codex":[0.7620175,0.2097615,0.007817981,0.002716142,0.01284609,0.004840823],"domain_scores_gemma":[0.7192427,0.2344832,0.005788946,0.005180401,0.03172849,0.003576402],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001380898,0.0003232664,0.008491183,0.001525888,0.00003927228,0.001459281,0.8674039,0.001349216,0.002245352,0.01853819,0.006754548,0.09173191],"study_design_scores_gemma":[0.00006121946,0.0004252165,0.005740392,0.002917846,0.00004252094,0.0009255746,0.9254653,0.00484115,0.001489726,0.01928534,0.03864015,0.0001655547],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.676689,0.002506167,0.2016189,0.05019277,0.0005858883,0.01457931,0.0005040974,0.0002620507,0.05306182],"genre_scores_gemma":[0.8302257,0.00213163,0.1520834,0.005651661,0.00007255049,0.006461387,0.0002278471,0.0001133318,0.003032592],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.172491,"threshold_uncertainty_score":0.9122306,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.282272881405195,"score_gpt":0.5888860901688526,"score_spread":0.3066132087636577,"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."}}