{"id":"W4225120362","doi":"10.2196/30988","title":"Video-Based Communication Assessment of Physician Error Disclosure Skills by Crowdsourced Laypeople and Patient Advocates Who Experienced Medical Harm: Reliability Assessment With Generalizability Theory","year":2022,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Patient Safety and Medication Errors","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Washington","keywords":"Generalizability theory; Harm; Reliability (semiconductor); Medical education; Psychology; Scale (ratio); Medicine; Applied psychology; Family medicine; Social psychology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003129568,0.000245885,0.0004652899,0.0001021411,0.001084313,0.000003893378,0.0005115028,0.0002310436,0.003422919],"category_scores_gemma":[0.00128771,0.0001964423,0.00007408008,0.0004678639,0.0006727059,0.000169822,0.0002964821,0.001094569,0.00000227951],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000600201,"about_ca_system_score_gemma":0.006564371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002340742,"about_ca_topic_score_gemma":0.00004982485,"domain_scores_codex":[0.9918191,0.003899873,0.001056389,0.0005624641,0.002261995,0.0004001297],"domain_scores_gemma":[0.9960399,0.001202782,0.000853697,0.0009380264,0.0003386829,0.0006269725],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009457607,0.02760218,0.6510164,0.001856441,0.0001764609,0.000002610979,0.1309251,0.000733136,0.0008113357,0.01238289,0.1024521,0.07109568],"study_design_scores_gemma":[0.006073771,0.00168319,0.7455888,0.001408985,0.0001296192,0.000003946626,0.1123542,0.01652004,0.0002817573,0.001768738,0.1133344,0.0008524462],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.959914,0.0001638654,0.00419582,0.03191459,0.0004470302,0.002082097,0.00006978661,0.00008167872,0.001131137],"genre_scores_gemma":[0.9820482,0.0000391694,0.001192169,0.009897818,0.00006172102,0.005636843,0.0009316468,0.00002877288,0.0001636468],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09457241,"threshold_uncertainty_score":0.9990675,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01306510995129056,"score_gpt":0.4168844679443189,"score_spread":0.4038193579930284,"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."}}