{"id":"W4410340882","doi":"10.1007/s43441-025-00785-z","title":"Using Large Language Models for Advanced and Flexible Labelling of Protocol Deviations in Clinical Development","year":2025,"lang":"en","type":"article","venue":"Therapeutic Innovation & Regulatory Science","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Roche (Canada)","funders":"","keywords":"Context (archaeology); Protocol (science); Computer science; Artificial intelligence; Natural language processing; Medicine; Pathology; Geography; Alternative medicine","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003967373,0.00009284006,0.0001669924,0.0005932892,0.0002747285,0.00004831798,0.0004657835,0.00005368091,0.000001365397],"category_scores_gemma":[0.000140759,0.00009056485,0.00001471307,0.003183507,0.0001499516,0.000577123,0.0001432912,0.0001345759,2.742413e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001099679,"about_ca_system_score_gemma":0.001127197,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005727638,"about_ca_topic_score_gemma":0.000006258533,"domain_scores_codex":[0.998275,0.00007479864,0.0007370281,0.0003798825,0.0002712024,0.0002620179],"domain_scores_gemma":[0.9987401,0.000136645,0.0002754095,0.0003531586,0.0004698189,0.0000248979],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002174114,0.00006658531,0.01231547,0.00016384,0.000005281669,1.752666e-7,0.001663341,0.009147605,0.009269219,0.8326381,0.000001432312,0.1347072],"study_design_scores_gemma":[0.0009880508,0.00004780697,0.04510678,0.000183765,0.000001948907,8.911025e-7,0.0001290549,0.9229448,0.01050438,0.01900008,0.0009538931,0.0001385351],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2646405,0.00003438213,0.7278178,0.0002115065,0.00009207462,0.007032428,6.20245e-7,0.00004531422,0.0001254631],"genre_scores_gemma":[0.8102096,3.798788e-7,0.1875887,0.0005016701,0.000007805103,0.001625142,8.59383e-7,0.000004335135,0.00006156196],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9137972,"threshold_uncertainty_score":0.3693127,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09952621716643238,"score_gpt":0.4547115080606196,"score_spread":0.3551852908941872,"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."}}