{"id":"W7085459742","doi":"10.2139/ssrn.5574679","title":"From Text to Tools: A Clinician-Validated Framework for Structured Evidence Extraction from Clinical Trials Using Language Models","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Lanthanide and Transition Metal Complexes","field":"Materials Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Pipeline (software); Automation; Data extraction; Information extraction; Clinical trial; Domain (mathematical analysis); Evidence-based medicine; Statistical model; Statistical power","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":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01524233,0.0005441771,0.001829437,0.0002299349,0.0003391469,0.0008633848,0.001144432,0.0009360791,0.0004684774],"category_scores_gemma":[0.006505207,0.000462203,0.001047804,0.0002138275,0.0000651997,0.0006577692,0.0002985547,0.004412481,0.00004010949],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007462303,"about_ca_system_score_gemma":0.004643999,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002474196,"about_ca_topic_score_gemma":0.001335912,"domain_scores_codex":[0.9908993,0.002774061,0.002745057,0.001174898,0.0006370553,0.001769615],"domain_scores_gemma":[0.9898779,0.006994421,0.001685426,0.0008302254,0.0003667992,0.0002451858],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.01672305,0.0005838331,0.0001276257,0.0002464402,0.003650964,0.00006329029,0.003785684,0.05180082,0.7501422,0.03811112,0.0004293356,0.1343356],"study_design_scores_gemma":[0.000935695,0.0002339814,0.00006980279,0.001702918,0.001119996,0.00001941934,0.001068402,0.005381556,0.0221426,0.9666765,0.00009596978,0.0005531721],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4329929,0.004273382,0.556673,0.0004004583,0.003275458,0.0007715107,0.001532581,0.00007212045,0.000008601513],"genre_scores_gemma":[0.8666815,0.002829077,0.1238266,0.0006335767,0.005565225,0.00005861756,0.0002669807,0.00005956923,0.00007890537],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9285654,"threshold_uncertainty_score":0.999783,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2932872230679449,"score_gpt":0.4957688938339777,"score_spread":0.2024816707660327,"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."}}