{"id":"W4414702335","doi":"10.1177/19160216251377349","title":"Evaluating the Efficacy of Large Language Models for Dizzy History Taking and Peripheral Vestibular Disorder Diagnosis","year":2025,"lang":"en","type":"article","venue":"Journal of Otolaryngology - Head and Neck Surgery","topic":"Vestibular and auditory disorders","field":"Neuroscience","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Medical diagnosis; Benign paroxysmal positional vertigo; Vestibular system; Diagnostic accuracy; Kappa; Differential diagnosis; Vestibular disorders; Labyrinthitis","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01444714,0.002551853,0.001095239,0.002626,0.0007718494,0.002079485,0.001939979,0.0019036,0.001812771],"category_scores_gemma":[0.03609335,0.0006693033,0.002036242,0.001019683,0.0007800742,0.001672088,0.001689785,0.002430078,0.000694235],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002408473,"about_ca_system_score_gemma":0.002641413,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02050856,"about_ca_topic_score_gemma":0.02068525,"domain_scores_codex":[0.9945609,0.003317617,0.000455082,0.00103651,0.0004128705,0.0002171548],"domain_scores_gemma":[0.950802,0.04381847,0.001387823,0.001468853,0.001725407,0.0007974544],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.005484377,0.001583017,0.1321594,0.001003882,0.001868382,0.0005369385,0.0004514521,0.6648412,0.002370261,0.001402583,0.0109968,0.1773017],"study_design_scores_gemma":[0.0001417803,0.0003205708,0.003756931,0.00008889205,0.0001757508,0.00009267786,0.00009818418,0.9921072,0.0009899974,0.001489446,0.0007094354,0.00002913942],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9019649,0.008306401,0.06447295,0.005596438,0.0005751374,0.0007440727,0.008962648,0.00470229,0.004675219],"genre_scores_gemma":[0.9530835,0.0004954289,0.03702679,0.0005523537,0.0001506166,0.0001832119,0.007863561,0.0000708666,0.0005737423],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02050856,"threshold_uncertainty_score":0.07640469,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04784512659981769,"score_gpt":0.3356961051145103,"score_spread":0.2878509785146925,"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."}}