{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07266221,0.0007963072,0.0007178208,0.001588548,0.001319391,0.00183311,0.001028469,0.001110599,0.002978786],"category_scores_gemma":[0.2239287,0.0005491616,0.001603186,0.0008369883,0.001513564,0.001777804,0.00285202,0.001174979,0.000764579],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001216681,"about_ca_system_score_gemma":0.001746992,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001731187,"about_ca_topic_score_gemma":0.001803154,"domain_scores_codex":[0.9428654,0.03526757,0.004499162,0.005394885,0.01100536,0.000967591],"domain_scores_gemma":[0.7697601,0.1392006,0.02116107,0.01830877,0.04969045,0.001878959],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.009550233,0.006109438,0.6447442,0.003254274,0.002157483,0.000299844,0.04308582,0.007999748,0.01240631,0.003307389,0.006805985,0.2602793],"study_design_scores_gemma":[0.00282419,0.01859506,0.843767,0.00146113,0.001448905,0.0006200072,0.0134859,0.07844566,0.01774955,0.007130492,0.0140595,0.0004125251],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9023578,0.0002509357,0.07424492,0.000379715,0.000248047,0.01236347,0.0006886397,0.0002906553,0.009175828],"genre_scores_gemma":[0.9397377,0.0001364512,0.03983049,0.0003747644,0.0001093954,0.01833997,0.0005053876,0.00005831524,0.000907562],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07266221,"threshold_uncertainty_score":0.3842791,"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."}}