{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1108096,0.005444734,0.004237022,0.004235401,0.003780119,0.004474746,0.005090845,0.005058664,0.08677279],"category_scores_gemma":[0.1768254,0.003852566,0.006987803,0.00479501,0.002833835,0.002098862,0.00423408,0.006617763,0.01507915],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004404,"about_ca_system_score_gemma":0.01357023,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002959722,"about_ca_topic_score_gemma":0.003821297,"domain_scores_codex":[0.9106354,0.07041614,0.008046681,0.004343743,0.004945629,0.001612417],"domain_scores_gemma":[0.8358029,0.1137201,0.007080623,0.02437435,0.01747435,0.001547655],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0794628,0.02601847,0.004394651,0.0621908,0.004581382,0.002708631,0.01128343,0.05675353,0.01152892,0.05748615,0.1858203,0.4977709],"study_design_scores_gemma":[0.1581301,0.05351947,0.01454772,0.03077572,0.003042457,0.0006988685,0.005041878,0.08719241,0.02617571,0.07103554,0.547919,0.001921021],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"protocol","genre_scores_codex":[0.002447088,0.0001609329,0.05790197,0.0001986034,0.0004297852,0.9344799,0.002574,0.0006386896,0.001169053],"genre_scores_gemma":[0.0005761214,0.00003169318,0.02788353,0.00005072066,0.00001061091,0.9711043,0.0001368968,0.00002670417,0.0001793359],"genre_candidate":"protocol","genre_consensus":"protocol","teacher_disagreement_score":0.8891904,"threshold_uncertainty_score":0.5860239,"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."}}