{"id":"W4415569782","doi":"10.1016/j.exer.2025.110700","title":"Postoperative data leakage in AI-based IOL power prediction formulas: a hidden source of bias","year":2025,"lang":"en","type":"article","venue":"Experimental Eye Research","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Université de Montréal; McGill University","funders":"","keywords":"Biometrics; Mean squared prediction error; Refraction; Power (physics); Predictive modelling; Predictive power; Range (aeronautics); Biometric data; Set (abstract data type)","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001066502,0.0001120295,0.0002842555,0.0006343097,0.00009216482,0.00003490602,0.0002725441,0.00006388713,0.0002175767],"category_scores_gemma":[0.0003867724,0.00009263725,0.00007077339,0.0008797314,0.0002226659,0.0001528861,0.0002771784,0.0004257044,0.00002739029],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002069674,"about_ca_system_score_gemma":0.0002745949,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001009175,"about_ca_topic_score_gemma":0.00001495557,"domain_scores_codex":[0.9981027,0.0002649167,0.0003064609,0.0003902478,0.0006319466,0.0003037614],"domain_scores_gemma":[0.9988971,0.0001443745,0.00003183575,0.0006645019,0.0001784083,0.00008376825],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00130715,0.001725874,0.1508781,0.0001341824,0.0001852478,0.00008626844,0.002551003,0.00004194825,0.8292026,0.0002134139,0.009396289,0.00427789],"study_design_scores_gemma":[0.003399844,0.001032102,0.01606548,0.0008917395,0.00003312855,0.000004729445,0.01174681,0.04581416,0.9140651,0.00002649908,0.006780017,0.0001403411],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9851043,0.001469565,0.0004337751,0.00329573,0.00003825708,0.0004616144,0.00003023805,0.00003065635,0.009135853],"genre_scores_gemma":[0.9948014,0.00001522757,0.0003771262,0.0002449178,0.00002501387,0.00003458875,0.0001593451,0.00001393346,0.004328429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1348126,"threshold_uncertainty_score":0.3777637,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1176369063483444,"score_gpt":0.4819524744199014,"score_spread":0.3643155680715571,"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."}}