{"id":"W7117668415","doi":"10.1145/3761712.3761788","title":"The \"Silent\" Failures of Scoring CDSS: Challenge in Detecting Malfunctions","year":2025,"lang":"","type":"article","venue":"","topic":"Electronic Health Records Systems","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Missing data; Clinical decision support system; Decision support system; Focus (optics); Poisson distribution; Time point; Change detection; Point (geometry); Outcome (game theory)","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":[0.01358244,0.001121135,0.001234692,0.00198761,0.000792768,0.002450845,0.002177174,0.00146007,0.0009711712],"category_scores_gemma":[0.07340942,0.0005194076,0.0006198605,0.001506153,0.001211487,0.002674608,0.001535805,0.001787998,0.000772302],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008956019,"about_ca_system_score_gemma":0.001908249,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005766935,"about_ca_topic_score_gemma":0.005899545,"domain_scores_codex":[0.98812,0.004646824,0.001424476,0.00231704,0.002987679,0.0005039982],"domain_scores_gemma":[0.9246466,0.04252985,0.01229333,0.008940571,0.01055657,0.001033038],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005357423,0.0004011769,0.6252545,0.0009128097,0.0003718468,0.001086048,0.002956813,0.02858546,0.0101186,0.003904355,0.01061709,0.3152556],"study_design_scores_gemma":[0.00009945875,0.001054389,0.3773256,0.000681308,0.0002416522,0.003692989,0.005040383,0.5483523,0.02048574,0.02891328,0.0138824,0.0002306178],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6259705,0.002127593,0.3561847,0.006537758,0.0004628777,0.0003592015,0.001638291,0.002655467,0.004063552],"genre_scores_gemma":[0.9540979,0.0002489567,0.04390958,0.0006159698,0.00012063,0.00008782136,0.0003784532,0.00009323924,0.0004472841],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01358244,"threshold_uncertainty_score":0.07183164,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06397986370725198,"score_gpt":0.4245135789585615,"score_spread":0.3605337152513095,"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."}}