{"id":"W4417515976","doi":"10.1016/j.ajoint.2025.100216","title":"Assessing demographic variation in large language model outputs for patient education materials in cataract surgery","year":2025,"lang":"en","type":"article","venue":"AJO International","topic":"Health Literacy and Information Accessibility","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Western University; Queen's University","funders":"","keywords":"Readability; Patient education; Indigenous; Cataract surgery; Reading (process); Grade level; Socioeconomic status; Variation (astronomy); MEDLINE","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.009410704,0.0002995632,0.0002527919,0.001306903,0.0004102908,0.001186191,0.000470927,0.0003819191,0.004627627],"category_scores_gemma":[0.07324524,0.0002400764,0.00056589,0.000931664,0.0004898816,0.001134351,0.001243738,0.0006114143,0.001171891],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002151662,"about_ca_system_score_gemma":0.001652579,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009623535,"about_ca_topic_score_gemma":0.01869592,"domain_scores_codex":[0.9935097,0.0028029,0.0009902542,0.000741028,0.001591165,0.0003648676],"domain_scores_gemma":[0.9396315,0.03562357,0.01429565,0.002126123,0.006972067,0.001351178],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008035973,0.0004250597,0.9549401,0.000206036,0.00005795893,0.0001220224,0.009778206,0.0004205241,0.001147543,0.00009213463,0.001997253,0.03000952],"study_design_scores_gemma":[0.00001997597,0.0005215536,0.9914136,0.00007247845,0.00002160762,0.00009460439,0.004208865,0.0009303339,0.001093409,0.00005681861,0.00154561,0.00002128476],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9969956,0.00004171426,0.0004550473,0.0001039208,0.000009252996,0.000145138,0.001106435,0.00003733555,0.001105495],"genre_scores_gemma":[0.9964371,0.00006502304,0.001336426,0.00009089107,0.00001114648,0.0003569259,0.000898924,0.00002990254,0.0007736361],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009623535,"threshold_uncertainty_score":0.04976916,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04278803316391143,"score_gpt":0.4684787450061979,"score_spread":0.4256907118422865,"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."}}