{"id":"W4404904568","doi":"10.2196/63109","title":"Diagnostic Decision-Making Variability Between Novice and Expert Optometrists for Glaucoma: Comparative Analysis to Inform AI System Design","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Concordance; Preprint; Clinical decision support system; Glaucoma; Computer science; Clinical decision making; Decision support system; Data science; Medicine; Artificial intelligence; Optometry; Family medicine; Ophthalmology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002185343,0.0002645298,0.001035037,0.00101358,0.0001611794,0.000179698,0.0002150864,0.0001940205,0.00005973191],"category_scores_gemma":[0.00373739,0.0001901674,0.0002812761,0.003385421,0.000122173,0.0003115213,0.0001348547,0.0003848847,0.00006232341],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002314485,"about_ca_system_score_gemma":0.0002210357,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001766405,"about_ca_topic_score_gemma":0.000002375297,"domain_scores_codex":[0.9970194,0.00006308869,0.001187583,0.0002421607,0.001110046,0.0003777787],"domain_scores_gemma":[0.9884115,0.01009165,0.0001278438,0.000382198,0.000326404,0.0006603805],"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":[0.0009644708,0.0006201745,0.2149467,0.01326179,0.01629816,0.0003123634,0.07703753,0.00046286,0.0000202971,0.00323074,0.06013546,0.6127095],"study_design_scores_gemma":[0.001103475,0.0005037854,0.02891718,0.003986593,0.003614879,0.00009479807,0.005342018,0.9401338,0.00007469249,0.0001816208,0.01555426,0.0004929292],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2189395,0.0002230839,0.7781529,0.0008579691,0.0000856662,0.0008792186,0.00003118,0.0001770975,0.0006533667],"genre_scores_gemma":[0.9534668,0.00002369307,0.04495059,0.001052472,0.0002072559,0.0002034586,0.00005547352,0.00001252448,0.00002771791],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9396709,"threshold_uncertainty_score":0.77548,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0297020711444331,"score_gpt":0.4054785838180814,"score_spread":0.3757765126736483,"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."}}