{"id":"W4320062020","doi":"10.2196/43960","title":"Performance of a Web-Based Reference Database With Natural Language Searching Capabilities: Usability Evaluation of DynaMed and Micromedex With Watson","year":2023,"lang":"en","type":"article","venue":"JMIR Human Factors","topic":"Electronic Health Records Systems","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institutes of Health Research; Brigham and Women's Hospital","keywords":"Medicine; Usability; Likert scale; Watson; Health care; Think aloud protocol; Medical education; Human–computer interaction; Computer science; Artificial intelligence; Psychology","routes":{"ca_aff":false,"ca_fund":true,"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":[0.01790236,0.000875973,0.001204074,0.0023734,0.0006442656,0.00205668,0.001673835,0.0009220172,0.001998734],"category_scores_gemma":[0.0392132,0.0005549372,0.0008664044,0.001465968,0.0004993177,0.002445645,0.00240233,0.0005099331,0.000463197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008064059,"about_ca_system_score_gemma":0.001418866,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003980986,"about_ca_topic_score_gemma":0.005280657,"domain_scores_codex":[0.9912803,0.004348823,0.001689539,0.001055782,0.001403844,0.0002217286],"domain_scores_gemma":[0.9544436,0.0357403,0.00137631,0.002268174,0.004950091,0.001221535],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02356913,0.009640271,0.08425576,0.009394757,0.001325479,0.002490412,0.03020675,0.00466197,0.04765449,0.0009432139,0.009165264,0.7766926],"study_design_scores_gemma":[0.01345814,0.08812984,0.5699457,0.003326155,0.00434198,0.01182391,0.03098716,0.1006601,0.0809642,0.001923078,0.09278096,0.001658883],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9854166,0.000860801,0.008278014,0.0002119423,0.00004372358,0.001454219,0.0008387254,0.0009591776,0.001936713],"genre_scores_gemma":[0.9256026,0.0006576995,0.0670969,0.0002856213,0.00004216689,0.001982676,0.002296369,0.0002468078,0.001789043],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01790236,"threshold_uncertainty_score":0.09467787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08945395266214314,"score_gpt":0.4480171033544907,"score_spread":0.3585631506923476,"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."}}