{"id":"W4414085277","doi":"10.2196/67529","title":"Real-World Evaluation of AI-Driven Diabetic Retinopathy Screening in Public Health Settings: Validation and Implementation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Public health; Health care; Diabetic retinopathy; Public healthcare; MEDLINE; Telemedicine; Retinopathy","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":[],"consensus_categories":[],"category_scores_codex":[0.004485389,0.0001012782,0.000351971,0.0006468478,0.00006026568,0.00003597622,0.00007088793,0.00004496031,0.00006819234],"category_scores_gemma":[0.0004401978,0.00008761285,0.00003811305,0.0009019879,0.00006115902,0.0002211057,0.00006009093,0.0002667132,0.000001484753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001200622,"about_ca_system_score_gemma":0.0005349231,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002180283,"about_ca_topic_score_gemma":0.0001118352,"domain_scores_codex":[0.9972198,0.0002583858,0.001053301,0.0001072408,0.001152516,0.0002087636],"domain_scores_gemma":[0.9989599,0.00009979598,0.0003209998,0.000178791,0.0002913257,0.0001491806],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001591881,0.000234635,0.7868928,0.0005378096,0.00007627236,0.000001125312,0.01148305,0.0000178225,0.00001762426,0.0001442872,0.0007683395,0.1998103],"study_design_scores_gemma":[0.005199085,0.000388103,0.7114702,0.001186429,0.0002743235,0.000004159705,0.04101102,0.239683,0.000208546,0.0001657923,0.0002796362,0.0001296877],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.988568,0.00002518336,0.001172753,0.008267505,0.00003289962,0.0009553493,0.000002478477,0.00002831413,0.0009474937],"genre_scores_gemma":[0.9973407,0.00003959151,0.001268318,0.001083706,0.00002317592,0.000075414,0.0001328018,0.000005431952,0.00003083274],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2396652,"threshold_uncertainty_score":0.3572749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04360005069517863,"score_gpt":0.4312347729464273,"score_spread":0.3876347222512486,"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."}}