{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08567204,0.004335656,0.00418057,0.02022165,0.001792903,0.01335387,0.003912753,0.005242841,0.01600394],"category_scores_gemma":[0.3359243,0.002476553,0.01088019,0.01213794,0.002015256,0.008097873,0.01048142,0.007025155,0.007926519],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003979574,"about_ca_system_score_gemma":0.01930251,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004831526,"about_ca_topic_score_gemma":0.009573287,"domain_scores_codex":[0.9200979,0.0437894,0.0252325,0.004883313,0.005445382,0.0005514451],"domain_scores_gemma":[0.5638815,0.3891133,0.01547176,0.01564003,0.01374889,0.002144576],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002933278,0.0005354601,0.006998425,0.06893115,0.005915695,0.001710222,0.005997416,0.029554,0.006589487,0.09277602,0.1010983,0.6769604],"study_design_scores_gemma":[0.002105461,0.0008327312,0.004396976,0.0268322,0.00602899,0.00193654,0.002181159,0.1718239,0.01987272,0.4411743,0.3220208,0.000794189],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002807026,0.004846542,0.904359,0.004518566,0.0003410932,0.00691762,0.06079452,0.01341257,0.002003094],"genre_scores_gemma":[0.0216142,0.001526673,0.9410142,0.0009276773,0.0001239516,0.00591663,0.02773078,0.0006956159,0.0004501932],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.914328,"threshold_uncertainty_score":0.4530824,"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."}}