{"id":"W2911791824","doi":"10.2196/12471","title":"Live Usability Testing of Two Complex Clinical Decision Support Tools: Observational Study","year":2019,"lang":"en","type":"article","venue":"JMIR Human Factors","topic":"Electronic Health Records Systems","field":"Health Professions","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institute of Allergy and Infectious Diseases; National Heart, Lung, and Blood Institute","keywords":"Usability; Observational study; Complaint; Clinical decision support system; Medicine; Context (archaeology); Health care; Computer science; Decision support system; Human–computer interaction; Artificial intelligence; Pathology","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":["insufficient_payload"],"category_scores_codex":[0.0054678,0.0002624692,0.0009282569,0.0001282772,0.0005660777,0.00001471502,0.0004907071,0.0002578156,0.004714568],"category_scores_gemma":[0.002387039,0.0002193664,0.0001581233,0.0003610364,0.00008985859,0.0002643848,0.0002590574,0.001214414,0.0009160868],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005521639,"about_ca_system_score_gemma":0.001348726,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001587235,"about_ca_topic_score_gemma":0.001626503,"domain_scores_codex":[0.9928553,0.001908475,0.002958057,0.0006485565,0.0008326133,0.0007970573],"domain_scores_gemma":[0.9860233,0.01098677,0.001132641,0.0008918391,0.0006890798,0.0002763541],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004534306,0.0006496281,0.991574,0.0002485532,0.00002823386,0.000001026673,0.003288183,0.000004134627,0.000169028,0.0003285558,0.002075039,0.001588274],"study_design_scores_gemma":[0.002016451,0.001674392,0.9879892,0.0001758444,0.0000124719,3.985951e-7,0.00392435,0.0001212643,0.000002552891,0.0004241033,0.003473196,0.0001857386],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9903528,0.000005588872,0.00002409942,0.00004273008,0.001048761,0.005237577,0.00005210825,0.000134987,0.003101353],"genre_scores_gemma":[0.9976327,5.806315e-7,0.0005850328,0.0002207631,0.0003627143,0.0001832746,0.00009305782,0.00004110906,0.0008807082],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00907829,"threshold_uncertainty_score":0.9998618,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.63938050514005,"score_gpt":0.6038143645491201,"score_spread":0.03556614059092988,"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."}}