{"id":"W4414193801","doi":"10.2139/ssrn.5489578","title":"Benchmarking Text Encoding Strategies in Multimodal Clinical Data for Surgical Case Duration Prediction","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta; Western University","funders":"","keywords":"Heuristics; Benchmarking; Benchmark (surveying); Unstructured data; Structured prediction; Perioperative; Encoding (memory); Set (abstract data type); Duration (music)","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":[],"consensus_categories":[],"category_scores_codex":[0.002966994,0.001118808,0.0008074525,0.004303399,0.0003088721,0.001587942,0.0008220162,0.001502248,0.003498024],"category_scores_gemma":[0.02455229,0.0001807982,0.0007755887,0.002555234,0.0002771814,0.001656589,0.001013416,0.001055095,0.00279233],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005308709,"about_ca_system_score_gemma":0.001487277,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003063991,"about_ca_topic_score_gemma":0.003080697,"domain_scores_codex":[0.9984457,0.0005535455,0.0003041321,0.0003729728,0.000211298,0.0001124715],"domain_scores_gemma":[0.9790644,0.01711242,0.0006070814,0.001028701,0.001686508,0.0005009011],"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":[0.004261704,0.001250365,0.0416012,0.001110521,0.000391199,0.0004392521,0.0003645638,0.03884165,0.01563669,0.001021054,0.02844129,0.8666405],"study_design_scores_gemma":[0.0005148783,0.00232776,0.03416714,0.0004133795,0.0006688454,0.001136357,0.001012133,0.9095167,0.030535,0.008247986,0.01131844,0.0001414041],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8129743,0.01019895,0.1238991,0.002683664,0.00139002,0.0006243978,0.03375665,0.008927319,0.005545525],"genre_scores_gemma":[0.8216896,0.001968344,0.1178981,0.0004355027,0.0005534739,0.0003361253,0.05400527,0.0004324104,0.002681168],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004303399,"threshold_uncertainty_score":0.01569116,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05811625148324234,"score_gpt":0.4016771221039853,"score_spread":0.343560870620743,"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."}}