{"id":"W4416636213","doi":"10.1183/13993003.congress-2025.pa5076","title":"Content Validity of AI-Generated Medical Information on Idiopathic Pulmonary Fibrosis (IPF): A Comparative Analysis of ChatGPT-4 and Gemini 1.5 Pro","year":2025,"lang":"","type":"article","venue":"","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Paul's Hospital","funders":"","keywords":"Readability; Idiopathic pulmonary fibrosis; Intraclass correlation; Reliability (semiconductor); Content validity; Quality of Life Research; Benchmark (surveying)","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02609651,0.0004612733,0.0007838416,0.002912693,0.0004515467,0.001771606,0.0009750973,0.0007555334,0.002133343],"category_scores_gemma":[0.1399417,0.0002539045,0.001576256,0.001904239,0.001164105,0.001637517,0.002584283,0.0009214222,0.0005250858],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00207736,"about_ca_system_score_gemma":0.001410068,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003249477,"about_ca_topic_score_gemma":0.003719141,"domain_scores_codex":[0.9830915,0.009244855,0.002192752,0.001338463,0.003754106,0.0003782703],"domain_scores_gemma":[0.7970025,0.1589783,0.01862429,0.005623228,0.01648848,0.003283286],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004940749,0.001184789,0.8805686,0.001596609,0.0008207056,0.000342963,0.02905488,0.001982167,0.00199508,0.000505347,0.003280511,0.07372769],"study_design_scores_gemma":[0.0001838648,0.002044218,0.9761403,0.0004305431,0.0003458949,0.0003037284,0.006862324,0.008426112,0.001587403,0.0004624869,0.003123875,0.00008936592],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9957533,0.0001123397,0.0009755744,0.0001502702,0.00002628695,0.0003898919,0.0008232598,0.00006828852,0.001700941],"genre_scores_gemma":[0.993062,0.000140116,0.002905334,0.0001528353,0.00003537381,0.001059819,0.002069878,0.00004213812,0.0005324901],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9739035,"threshold_uncertainty_score":0.1380132,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2998085274056864,"score_gpt":0.4394533823106851,"score_spread":0.1396448549049987,"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."}}