{"id":"W4399276598","doi":"10.2196/59617","title":"Viability of Open Large Language Models for Clinical Documentation in German Health Care: Real-World Model Evaluation Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":39,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Documentation; German; Computer science; Usability; Variety (cybernetics); USable; Health care; Language model; Artificial intelligence; Premise; Best practice; Natural language processing; Data science; World Wide Web; Programming language; Human–computer interaction; Linguistics; Political science","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02186841,0.002728842,0.001053675,0.002028286,0.0007311458,0.002270919,0.002485635,0.001891967,0.001885946],"category_scores_gemma":[0.04847363,0.0007169509,0.001607304,0.001443016,0.0009839085,0.00213531,0.001929375,0.002475617,0.0009694198],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004745857,"about_ca_system_score_gemma":0.002399873,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02623796,"about_ca_topic_score_gemma":0.02115298,"domain_scores_codex":[0.9893078,0.007708352,0.0007504485,0.001220505,0.0007410815,0.0002718113],"domain_scores_gemma":[0.9493264,0.04179826,0.001424406,0.003422633,0.003287207,0.0007409863],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.004757612,0.002860541,0.03435988,0.002001234,0.001840135,0.0008782171,0.002260096,0.6876683,0.003927939,0.00289748,0.02304338,0.2335051],"study_design_scores_gemma":[0.0004312366,0.0008778441,0.006922492,0.000162443,0.0003052331,0.0001811887,0.000410542,0.9824613,0.004226353,0.001495845,0.002445763,0.00007988274],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9339382,0.003181761,0.04661962,0.001568446,0.0002879158,0.001077905,0.004944149,0.004173207,0.004208744],"genre_scores_gemma":[0.9495193,0.0006175856,0.03846174,0.0002958915,0.00006607265,0.0005733377,0.008966401,0.0002740326,0.001225658],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9781316,"threshold_uncertainty_score":0.1156526,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2960965592878348,"score_gpt":0.6314565857016097,"score_spread":0.3353600264137749,"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."}}