{"id":"W4410528874","doi":"10.2196/71434","title":"Evaluating the Usability, Technical Performance, and Accuracy of Artificial Intelligence Scribes for Primary Care: Competitive Analysis","year":2025,"lang":"en","type":"article","venue":"JMIR Human Factors","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"Trillium Health Centre; Public Health Ontario; Women's College Hospital; University of Toronto","funders":"","keywords":"Preprint; Usability; Computer science; Artificial intelligence; Machine learning; Human–computer interaction; World Wide Web","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"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":[0.1978382,0.000962998,0.001322283,0.00941773,0.003305997,0.005357759,0.001816531,0.001100085,0.0009793494],"category_scores_gemma":[0.3428002,0.0004890274,0.002321491,0.005617269,0.003368208,0.002656506,0.004166701,0.001167659,0.0002492732],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008379903,"about_ca_system_score_gemma":0.01136152,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005210715,"about_ca_topic_score_gemma":0.009985446,"domain_scores_codex":[0.7426959,0.1776729,0.02283816,0.004749152,0.04978244,0.002261414],"domain_scores_gemma":[0.4231252,0.4187525,0.04722487,0.01308179,0.09465179,0.003163782],"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":[0.002560699,0.002300242,0.2696095,0.01355826,0.001563026,0.0004439343,0.1037028,0.003542328,0.005792267,0.004457369,0.003698188,0.5887715],"study_design_scores_gemma":[0.001194356,0.027328,0.7233291,0.01374638,0.003618634,0.001298659,0.1287185,0.03072982,0.02676415,0.01028945,0.0321038,0.000879206],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9447855,0.003190544,0.03529422,0.000858269,0.000115293,0.007304413,0.0003851411,0.0001320567,0.007934432],"genre_scores_gemma":[0.9085878,0.001225474,0.08301998,0.0003102511,0.00004883406,0.005979991,0.0003375017,0.00004864785,0.0004416149],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1978382,"threshold_uncertainty_score":0.9892083,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2285946247328254,"score_gpt":0.5077540762863881,"score_spread":0.2791594515535627,"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."}}