{"id":"W4407394923","doi":"10.1101/2025.02.10.25322041","title":"Performance of DeepSeek-R1 in Ophthalmology: An Evaluation of Clinical Decision-Making and Cost-Effectiveness","year":2025,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University; University of Waterloo; University of Toronto","funders":"","keywords":"Clinical decision making; Optometry; Ophthalmology; Psychology; Medicine; Family medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01811179,0.0010221,0.00130588,0.001852285,0.0002884802,0.001806229,0.001554776,0.0007533083,0.004347362],"category_scores_gemma":[0.0493633,0.0003529618,0.001843498,0.001164753,0.0006783091,0.00191942,0.002678856,0.0008900206,0.0008026742],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002531971,"about_ca_system_score_gemma":0.002449507,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002659049,"about_ca_topic_score_gemma":0.002329358,"domain_scores_codex":[0.98457,0.008846893,0.001400793,0.001642616,0.002880719,0.0006589888],"domain_scores_gemma":[0.9351453,0.04772209,0.00780867,0.002527161,0.003708151,0.003088555],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.02298333,0.007579977,0.4609616,0.002015034,0.00130056,0.0002640127,0.001619655,0.03974332,0.001800518,0.0008612826,0.003654554,0.4572162],"study_design_scores_gemma":[0.003070673,0.05152925,0.610763,0.0009557497,0.001436376,0.001363123,0.003681182,0.3085546,0.009038378,0.002025449,0.007145317,0.0004368409],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.99151,0.0007084552,0.003428085,0.0003852263,0.00003116302,0.0005094219,0.0006292556,0.0002280023,0.002570353],"genre_scores_gemma":[0.9893701,0.0003266906,0.008812366,0.0001037977,0.00002514518,0.0002723004,0.0006538216,0.00002815652,0.0004077846],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01811179,"threshold_uncertainty_score":0.0957855,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5560535159959448,"score_gpt":0.5644865920010473,"score_spread":0.008433076005102569,"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."}}