{"id":"W4417082287","doi":"10.2196/76674","title":"Leveraging AI Large Language Models for Writing Clinical Trial Proposals in Dermatology: Instrument Validation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Dermatology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Process (computing); Expert system; Clinical trial; Subject-matter expert; Model validation","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007229997,0.0001399196,0.0005382203,0.0003222751,0.0001344089,0.00002361009,0.0001028694,0.0002642247,0.00001856888],"category_scores_gemma":[0.0003934,0.000135752,0.00009477358,0.0002626119,0.00004946598,0.0001402529,0.00005182679,0.0003689715,0.00001628248],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001044425,"about_ca_system_score_gemma":0.0004681178,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001238771,"about_ca_topic_score_gemma":0.0001428158,"domain_scores_codex":[0.9974506,0.0003637286,0.001271304,0.0003831233,0.0001168917,0.000414383],"domain_scores_gemma":[0.9988725,0.0004822132,0.0001450607,0.0002766916,0.0001513613,0.00007217145],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.009419555,0.004262262,0.9073148,0.0006968468,0.0001163254,0.0001724402,0.01370358,0.00003112225,0.00007589196,0.004920417,0.003328531,0.05595822],"study_design_scores_gemma":[0.1520292,0.006136603,0.07111339,0.002478324,0.001089781,0.002048344,0.3193247,0.2844986,0.05997172,0.08160126,0.01733436,0.002373662],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9730322,0.0001161161,0.005964943,0.01501333,0.0009156622,0.004613906,0.000002687844,0.00007239767,0.0002687669],"genre_scores_gemma":[0.9954847,0.00001369762,0.0004937642,0.002501846,0.0001667944,0.001193793,0.00005936006,0.00001421619,0.00007190649],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8362014,"threshold_uncertainty_score":0.5535807,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2268048425356374,"score_gpt":0.5355604158180052,"score_spread":0.3087555732823678,"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."}}