{"id":"W4414957620","doi":"10.2196/85437","title":"Evaluation of latent health risk prediction models: A protocol for a mixed-methods study using clinical triage as a vehicle for comparison and discussion (Preprint)","year":2025,"lang":"en","type":"preprint","venue":"JMIR Research Protocols","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Triage; Protocol (science); Data collection; Qualitative research; Health care; Phase (matter); Emergency department; Resource (disambiguation)","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.1251095,0.000414906,0.00126269,0.0006816501,0.0007719204,0.0004130432,0.001594122,0.000533064,0.000005536679],"category_scores_gemma":[0.007360243,0.0003122074,0.0003239212,0.0006275557,0.0001531136,0.0003209455,0.004228871,0.002749283,7.367119e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009252846,"about_ca_system_score_gemma":0.006637738,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000853609,"about_ca_topic_score_gemma":0.00008699751,"domain_scores_codex":[0.9565996,0.03430945,0.00291674,0.002313325,0.002962853,0.0008980595],"domain_scores_gemma":[0.9888038,0.002968649,0.001870753,0.002205182,0.003786015,0.0003655777],"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":[0.002568882,0.004017816,0.02993701,0.01173521,0.0001356141,4.449307e-7,0.005906329,0.04696185,0.000013777,0.0006372519,0.0007270104,0.8973588],"study_design_scores_gemma":[0.009010841,0.008361076,0.009740077,0.006702897,0.00002191894,8.477797e-7,0.0002794565,0.9243212,0.00006526824,0.03917485,0.002100893,0.0002206853],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"protocol","genre_scores_codex":[0.001044873,0.000004369481,0.3922687,0.0004642409,0.00004360319,0.6059909,0.00006592413,0.0001016277,0.00001577939],"genre_scores_gemma":[0.002470291,6.247579e-7,0.1838724,0.00001153545,0.0001400052,0.8134212,0.00001203968,0.00003242452,0.00003941593],"genre_candidate":"protocol","genre_consensus":"protocol","teacher_disagreement_score":0.8971381,"threshold_uncertainty_score":0.999933,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6922968907199762,"score_gpt":0.7356693796455016,"score_spread":0.04337248892552537,"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."}}