{"id":"W4395031821","doi":"10.2196/56165","title":"Clinical Accuracy, Relevance, Clarity, and Emotional Sensitivity of Large Language Models to Surgical Patient Questions: Cross-Sectional Study","year":2024,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"CLARITY; Cross-sectional study; Relevance (law); Clinical significance; Psychology; Medicine; Pathology; Political science","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":[0.02432784,0.0003447291,0.0005315816,0.001491278,0.0004698141,0.00173336,0.0004181209,0.0008012967,0.001557145],"category_scores_gemma":[0.09489702,0.0004757795,0.0008596519,0.0007965189,0.001207653,0.001932745,0.001555261,0.001229914,0.0005872068],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004575952,"about_ca_system_score_gemma":0.0005102589,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001329883,"about_ca_topic_score_gemma":0.001349662,"domain_scores_codex":[0.987775,0.006837752,0.001671895,0.001326539,0.001910786,0.0004781366],"domain_scores_gemma":[0.8855031,0.07162273,0.02004723,0.007466068,0.01331776,0.002043104],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001853905,0.0001946006,0.9927821,0.00004358019,0.0001282629,0.00004868212,0.002731753,0.00008956088,0.0002347738,0.00004065201,0.000122302,0.003398431],"study_design_scores_gemma":[0.00001638757,0.001426851,0.9903501,0.00006616559,0.0001437051,0.0005734049,0.004380302,0.001090909,0.0007514899,0.0001656311,0.0009949441,0.00004007054],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9984542,0.000142358,0.0006059006,0.00007259433,0.00001234403,0.00004707353,0.0001624092,0.000005747174,0.0004972736],"genre_scores_gemma":[0.9991739,0.00006497352,0.0003677103,0.00006698897,0.00001404102,0.00004582532,0.0001444588,0.000005407115,0.0001165577],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02432784,"threshold_uncertainty_score":0.1286595,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3125165973552284,"score_gpt":0.6058303830983638,"score_spread":0.2933137857431354,"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."}}