{"id":"W4416035400","doi":"10.18653/v1/2025.emnlp-main.1637","title":"Not What the Doctor Ordered: Surveying LLM-based De-identification and Quantifying Clinical Information Loss","year":2025,"lang":"en","type":"article","venue":"","topic":"Imbalanced Data Classification Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Arthritis Society; Canadian Institute for Advanced Research; Alberta Health Services","keywords":"Information system; MEDLINE; Data loss; Information loss; Risk assessment","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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":[0.1975322,0.001084539,0.001547034,0.01447685,0.001275615,0.006547305,0.003867162,0.001993834,0.001173798],"category_scores_gemma":[0.5718076,0.0005306319,0.001721947,0.01259277,0.002448781,0.007574828,0.004403705,0.002760744,0.0006082385],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004808814,"about_ca_system_score_gemma":0.004752026,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004995586,"about_ca_topic_score_gemma":0.003869738,"domain_scores_codex":[0.7906538,0.1349629,0.03032344,0.01130719,0.03122354,0.001529073],"domain_scores_gemma":[0.2340826,0.6274784,0.05649322,0.03484362,0.04555652,0.001545667],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001441161,0.000317059,0.3153483,0.005724624,0.001578695,0.0002727635,0.005819572,0.01935702,0.00173136,0.01438187,0.01947888,0.6145487],"study_design_scores_gemma":[0.0004979003,0.002338429,0.2775188,0.0137833,0.002564362,0.003311056,0.01374285,0.3977365,0.0401944,0.1179341,0.1294549,0.0009232786],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4045298,0.05683853,0.4451149,0.04987936,0.001480346,0.002263752,0.01867864,0.002809583,0.01840516],"genre_scores_gemma":[0.7964253,0.007746844,0.176898,0.005475228,0.0007111907,0.001318023,0.009957793,0.000347031,0.001120499],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1975322,"threshold_uncertainty_score":0.9895857,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06983332277345129,"score_gpt":0.3749101502068776,"score_spread":0.3050768274334263,"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."}}