{"id":"W2911395613","doi":"10.1016/j.pec.2019.01.018","title":"Reliability of narrative assessment data on communication skills in a summative OSCE","year":2019,"lang":"en","type":"article","venue":"Patient Education and Counseling","topic":"Empathy and Medical Education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Summative assessment; Reliability (semiconductor); Narrative; Objective structured clinical examination; Psychology; Computer science; Medical education; Formative assessment; Medicine; Mathematics education; Art; Literature","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04158667,0.0005820176,0.0008295414,0.003258258,0.0009990161,0.001511784,0.0008260695,0.001073466,0.002191057],"category_scores_gemma":[0.1765242,0.0004244122,0.001442008,0.001359139,0.001551322,0.001812144,0.002747081,0.0008702214,0.001372049],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006960207,"about_ca_system_score_gemma":0.00084995,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001023875,"about_ca_topic_score_gemma":0.001524443,"domain_scores_codex":[0.95811,0.0154856,0.009138338,0.003343917,0.01249048,0.001431564],"domain_scores_gemma":[0.7419453,0.1530765,0.01883243,0.01725416,0.06665917,0.002232498],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002496054,0.0007361779,0.9085861,0.0005199844,0.0006453715,0.0002144541,0.01302747,0.001261152,0.00877616,0.0006741154,0.00186101,0.06120193],"study_design_scores_gemma":[0.0001134278,0.002107415,0.9728159,0.0002423913,0.0002658568,0.0005597762,0.00378748,0.005348096,0.009085467,0.0006901814,0.004881219,0.0001026767],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9724457,0.0004909624,0.01333679,0.0001318658,0.00031334,0.0007069545,0.0006798602,0.0001676243,0.01172692],"genre_scores_gemma":[0.992459,0.0001392091,0.004682363,0.00007380464,0.00005973566,0.0006247274,0.0007755386,0.00005706945,0.001128601],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9584134,"threshold_uncertainty_score":0.2199339,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01681552603304047,"score_gpt":0.3528935463075387,"score_spread":0.3360780202744982,"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."}}