{"id":"W4407009455","doi":"10.1038/s41746-025-01457-w","title":"Language models for data extraction and risk of bias assessment in complementary medicine","year":2025,"lang":"en","type":"article","venue":"npj Digital Medicine","topic":"Complementary and Alternative Medicine Studies","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"Fundamental Research Funds for the Central Universities; China Academy of Chinese Medical Sciences; National Natural Science Foundation of China","keywords":"Computer science; Data extraction; Extraction (chemistry); Data science; Natural language processing; MEDLINE; Chemistry; Chromatography","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":[0.0008163649,0.0002159487,0.0007481982,0.0003845586,0.00006370925,0.000005177102,0.0001773369,0.00001833725,0.0001114329],"category_scores_gemma":[0.0008555615,0.0001476274,0.00003326596,0.0002980118,0.0004048998,0.0002858298,0.0002329644,0.000224905,3.169424e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006854334,"about_ca_system_score_gemma":0.00005863042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001301961,"about_ca_topic_score_gemma":0.000340208,"domain_scores_codex":[0.9982145,0.00004448679,0.0007181527,0.0004284694,0.0003653633,0.000228966],"domain_scores_gemma":[0.9979385,0.001163016,0.0002079847,0.000488931,0.000113178,0.00008838852],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008784794,0.0005243903,0.6169227,0.001191828,0.0009888607,0.00006088703,0.002869971,0.00001407453,0.001924519,0.006270338,0.01773771,0.3506162],"study_design_scores_gemma":[0.05589871,0.009952252,0.7138602,0.009982394,0.0028771,0.00008472643,0.07511319,0.05155041,0.0005845592,0.02113412,0.0583531,0.0006092389],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9241746,0.004499541,0.02719085,0.0126037,0.0005746676,0.001774003,0.0006253929,0.0000485618,0.02850869],"genre_scores_gemma":[0.9951743,0.001084248,0.001198624,0.0003828701,0.0003401184,0.00003464353,0.001318076,0.0000129158,0.0004542347],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.350007,"threshold_uncertainty_score":0.6020069,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1603237042729094,"score_gpt":0.4542686677577073,"score_spread":0.2939449634847979,"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."}}