{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":11,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":11,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"cbcfb6b7cdf0","filters":{"venue":"NEJM AI"}},"results":[{"id":"W4394845024","doi":"10.1056/aioa2300151","title":"Comparative Evaluation of LLMs in Clinical Oncology","year":2024,"lang":"en","type":"article","venue":"NEJM AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":97,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Sunnybrook Health Science Centre","funders":"National Cancer Institute; NIH Clinical Center; National Institutes of Health","keywords":"Clinical Oncology; Internal medicine; Oncology; Medicine; Cancer","authors":[{"name":"Nicholas R. Rydzewski","is_ca":false},{"name":"Deepak Dinakaran","is_ca":true},{"name":"Shuang G. Zhao","is_ca":false},{"name":"Eytan Ruppin","is_ca":false},{"name":"Barış Türkbey","is_ca":false},{"name":"Deborah E. Citrin","is_ca":false},{"name":"Krishnan R. Patel","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7465636406134145,"gpt":0.698507249034476,"spread":0.04805639157893848,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01255485,0.001593615,0.0007480932,0.002172422,0.0003997573,0.001851281,0.002630424,0.002035202,0.003774432],"category_scores_gemma":[0.05643551,0.000497699,0.001364866,0.001190238,0.0007873148,0.002963173,0.002304304,0.002149108,0.00139112],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003506851,"about_ca_system_score_gemma":0.001775065,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007098821,"about_ca_topic_score_gemma":0.007419812,"domain_scores_codex":[0.993604,0.003788892,0.0006048668,0.001060136,0.0007602062,0.0001818551],"domain_scores_gemma":[0.9526858,0.03948967,0.001313501,0.00280566,0.002743608,0.0009617656],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.006211509,0.002510506,0.07603617,0.002719878,0.001473498,0.0003260394,0.001269774,0.5099086,0.004945602,0.002977244,0.02066452,0.3709567],"study_design_scores_gemma":[0.0004409087,0.002541513,0.01604968,0.0003027366,0.0003255999,0.0003113084,0.0004181269,0.9592884,0.008375345,0.003914181,0.007926029,0.0001060441],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8820701,0.005648586,0.06475421,0.003494826,0.0004817096,0.0008965242,0.006717852,0.02305355,0.01288263],"genre_scores_gemma":[0.9567631,0.0007582796,0.03243149,0.0004624245,0.00006852769,0.0003201512,0.00767375,0.0003284362,0.001193814],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01255485,"threshold_uncertainty_score":0.06639719,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4404759704","doi":"10.1056/aics2400639","title":"Cognitive Biases and Artificial Intelligence","year":2024,"lang":"en","type":"article","venue":"NEJM AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Cognition; Psychology; Cognitive psychology; Artificial intelligence; Cognitive science; Computer science; Neuroscience","authors":[{"name":"Jonathan Wang","is_ca":true},{"name":"Donald A. Redelmeier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.329062206843413,"gpt":0.5014888764580899,"spread":0.1724266696146769,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004709251,0.0003784942,0.0004400354,0.002188253,0.001025285,0.005789343,0.0006059411,0.002490584,0.005922658],"category_scores_gemma":[0.02746803,0.0002480316,0.0002168285,0.001664747,0.009665363,0.006003902,0.001600982,0.00236043,0.0006504311],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001561317,"about_ca_system_score_gemma":0.001275012,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001829437,"about_ca_topic_score_gemma":0.001776357,"domain_scores_codex":[0.997256,0.001537237,0.0001092578,0.0002369261,0.0007259353,0.0001346963],"domain_scores_gemma":[0.9798644,0.01609928,0.0008145197,0.001275267,0.001446132,0.0005004766],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00004369598,0.00005456489,0.003150736,0.0001683301,0.00006688856,0.00008648339,0.00202593,0.001434877,0.0001283642,0.9309711,0.01065249,0.05121662],"study_design_scores_gemma":[0.000009160071,0.000005061934,0.0007843514,0.00005902127,0.000008706417,0.00004369705,0.0002827951,0.0005854224,0.00005410386,0.9850196,0.01314182,0.000006299793],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1114355,0.06236451,0.04087452,0.1882006,0.003289913,0.00006813306,0.0002929842,0.0001552824,0.5933186],"genre_scores_gemma":[0.946216,0.01525918,0.009367559,0.008137932,0.00216248,0.000102763,0.00009935984,0.00006596429,0.01858883],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005922658,"threshold_uncertainty_score":0.02490515,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4414499338","doi":"10.1056/aidbp2500120","title":"Assessment of Large Language Models in Clinical Reasoning: A Novel Benchmarking Study","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":28,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Benchmarking; Language model; Quality (philosophy); Identification (biology); Measure (data warehouse)","authors":[{"name":"Liam G. McCoy","is_ca":true},{"name":"Rajiv Swamy","is_ca":false},{"name":"Nidhish Sagar","is_ca":false},{"name":"Minjia Wang","is_ca":false},{"name":"Stephen Bacchi","is_ca":false},{"name":"Jie Ming Nigel Fong","is_ca":false},{"name":"Nigel CK Tan","is_ca":false},{"name":"Kevin Tan","is_ca":false},{"name":"Thomas Buckley","is_ca":false},{"name":"Peter G. Brodeur","is_ca":false},{"name":"Leo Anthony Celi","is_ca":false},{"name":"Arjun K. Manrai","is_ca":false},{"name":"Aloysius J. Humbert","is_ca":false},{"name":"Adam Rodman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03709534502479742,"gpt":0.4589540775897755,"spread":0.4218587325649781,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03285042,0.001176179,0.000924357,0.002170961,0.00100646,0.003831927,0.002854826,0.002115275,0.004176596],"category_scores_gemma":[0.1691909,0.0004690127,0.001314042,0.002020314,0.001310194,0.006518242,0.004152421,0.002280743,0.0009169761],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002375114,"about_ca_system_score_gemma":0.002506743,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005561337,"about_ca_topic_score_gemma":0.005595571,"domain_scores_codex":[0.9691153,0.02262321,0.002062077,0.002534165,0.003160886,0.0005044734],"domain_scores_gemma":[0.7965477,0.1663193,0.004686168,0.01727273,0.01299838,0.002175751],"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":[0.01109319,0.01051299,0.1610761,0.002197625,0.002070023,0.0008973961,0.006523888,0.2370465,0.009432373,0.02393337,0.01963533,0.5155813],"study_design_scores_gemma":[0.001331806,0.004583814,0.03695701,0.0004134752,0.0007142548,0.0007398098,0.002705432,0.8999142,0.007938802,0.03420152,0.01032613,0.0001737381],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8368357,0.001664666,0.1446742,0.001337853,0.0001739121,0.0008842028,0.002644619,0.001717593,0.01006729],"genre_scores_gemma":[0.939348,0.0002142243,0.05490075,0.0002796171,0.00004803904,0.0003906329,0.003673323,0.0002578735,0.000887538],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03285042,"threshold_uncertainty_score":0.1737317,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4405533437","doi":"10.1056/aip2401088","title":"Tackling Algorithmic Bias and Promoting Transparency in Health Datasets: The STANDING Together Consensus Recommendations","year":2024,"lang":"en","type":"article","venue":"NEJM AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Google (Canada); SickKids Foundation; Hospital for Sick Children","funders":"Economic and Social Research Council; Medical Research Council","keywords":"Transparency (behavior); Data science; Computer science; Internet privacy; Computer security","authors":[{"name":"Joseph Alderman","is_ca":false},{"name":"Joanne Palmer","is_ca":false},{"name":"Elinor Laws","is_ca":false},{"name":"Melissa D. McCradden","is_ca":true},{"name":"Johan Ordish","is_ca":false},{"name":"Marzyeh Ghassemi","is_ca":false},{"name":"Stephen Pfohl","is_ca":false},{"name":"Negar Rostamzadeh","is_ca":true},{"name":"Heather Cole-Lewis","is_ca":false},{"name":"Ben Glocker","is_ca":false},{"name":"Melanie Calvert","is_ca":false},{"name":"Tom Pollard","is_ca":false},{"name":"Jaspret Gill","is_ca":false},{"name":"Jacqui Gath","is_ca":false},{"name":"Ade Adebajo","is_ca":false},{"name":"Jude Beng","is_ca":false},{"name":"Cheuk Wing Leung","is_ca":false},{"name":"Stephanie Kuku","is_ca":false},{"name":"Rubeta Matin","is_ca":false},{"name":"Bilal A. Mateen","is_ca":false},{"name":"Francis McKay","is_ca":false},{"name":"Katherine Heller","is_ca":false},{"name":"Alan Karthikesalingam","is_ca":false},{"name":"Darren Treanor","is_ca":false},{"name":"Maxine Mackintosh","is_ca":false},{"name":"Lauren Oakden‐Rayner","is_ca":false},{"name":"Russell Pearson","is_ca":false},{"name":"Arjun K. Manrai","is_ca":false},{"name":"Puja Myles","is_ca":false},{"name":"Judit Kumuthini","is_ca":false},{"name":"Zoher Kapacee","is_ca":false},{"name":"Neil J. Sebire","is_ca":false},{"name":"Lama Nazer","is_ca":false},{"name":"Jarrel Seah","is_ca":false},{"name":"Ashley Akbari","is_ca":false},{"name":"Lewis E. Berman","is_ca":false},{"name":"Judy Wawira Gichoya","is_ca":false},{"name":"Lorenzo Righetto","is_ca":false},{"name":"Diana Samuel","is_ca":false},{"name":"William Wasswa","is_ca":false},{"name":"Maria Charalambides","is_ca":false},{"name":"Anmol Arora","is_ca":false},{"name":"Sameer Pujari","is_ca":false},{"name":"Charlotte Summers","is_ca":false},{"name":"Elizabeth Sapey","is_ca":false},{"name":"Vishal Thakker","is_ca":false},{"name":"Alastair K. Denniston","is_ca":false},{"name":"Xiaoxuan Liu","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2216938960706069,"gpt":0.4714438908343587,"spread":0.2497499947637518,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2874652,0.002157713,0.005574612,0.006158065,0.003346671,0.01666158,0.01039319,0.02046445,0.01044849],"category_scores_gemma":[0.5649983,0.001792523,0.006674777,0.005231699,0.01372035,0.02952657,0.01124015,0.03767472,0.002795717],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005498623,"about_ca_system_score_gemma":0.02594288,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00703231,"about_ca_topic_score_gemma":0.01072698,"domain_scores_codex":[0.7968668,0.1449984,0.01860501,0.01361204,0.0228923,0.003025407],"domain_scores_gemma":[0.2728759,0.6070912,0.01442053,0.04106692,0.05708606,0.007459299],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001061624,0.0003403954,0.005164776,0.01194541,0.004989377,0.0002312858,0.001859249,0.005624267,0.0004271358,0.315464,0.2499965,0.4028959],"study_design_scores_gemma":[0.0006519587,0.0001270425,0.001020952,0.01392486,0.001564073,0.0001335426,0.000702631,0.007167771,0.0007633627,0.8514656,0.1223086,0.000169475],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.002131004,0.06685043,0.09208041,0.8208026,0.01169314,0.0004301226,0.0008843489,0.0003803044,0.004747533],"genre_scores_gemma":[0.1265932,0.07860144,0.4230026,0.3333219,0.02899892,0.002451336,0.002942228,0.0007061744,0.003382149],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7125348,"threshold_uncertainty_score":0.8786823,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4411672381","doi":"10.1056/aioa2401221","title":"Expert-Level Detection of Epilepsy Markers in EEG on Short and Long Timescales","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"EEG and Brain-Computer Interfaces","field":"Neuroscience","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"National Center for Advancing Translational Sciences; National Institute of Neurological Disorders and Stroke; National Institute of General Medical Sciences; National Heart, Lung, and Blood Institute; National Institute on Aging; U.S. Department of Veterans Affairs","keywords":"Epilepsy; Electroencephalography; Psychology; Audiology; Neuroscience; Pattern recognition (psychology); Computer science; Medicine; Cognitive psychology","authors":[{"name":"Jun Li","is_ca":false},{"name":"Daniel M. Goldenholz","is_ca":false},{"name":"Moritz Alkofer","is_ca":false},{"name":"Chenxi Sun","is_ca":false},{"name":"Fábio A. Nascimento","is_ca":false},{"name":"Jonathan J. Halford","is_ca":false},{"name":"Brian K. Dean","is_ca":false},{"name":"Mattia Galanti","is_ca":false},{"name":"Aaron F. Struck","is_ca":false},{"name":"Adam Greenblatt","is_ca":false},{"name":"Alice Lam","is_ca":false},{"name":"Aline Herlopian","is_ca":false},{"name":"Chinasa Nwankwo","is_ca":false},{"name":"Dan Weber","is_ca":false},{"name":"Douglas Maus","is_ca":false},{"name":"Hiba A. Haider","is_ca":false},{"name":"Ioannis Karakis","is_ca":false},{"name":"Ji Yeoun Yoo","is_ca":false},{"name":"Marcus Ng","is_ca":true},{"name":"Olga Selioutski","is_ca":false},{"name":"Olga Taraschenko","is_ca":false},{"name":"Gamaleldin Osman","is_ca":false},{"name":"Roohi Katyal","is_ca":false},{"name":"Sarah E. Schmitt","is_ca":false},{"name":"Selim R. Benbadis","is_ca":false},{"name":"Sydney S. Cash","is_ca":false},{"name":"William O. Tatum","is_ca":false},{"name":"Zubeda Sheikh","is_ca":false},{"name":"Wan Yee Kong","is_ca":false},{"name":"Grace Bayas","is_ca":false},{"name":"Niels Turley","is_ca":false},{"name":"Shenda Hong","is_ca":false},{"name":"M. Brandon Westover","is_ca":false},{"name":"Jin Jing","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0267090744471451,"gpt":0.2920409229830009,"spread":0.2653318485358558,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002186498,0.0005948824,0.0003577455,0.001094888,0.0001132906,0.0007124508,0.0004225278,0.0005606618,0.001078899],"category_scores_gemma":[0.009577801,0.0001158122,0.000311879,0.0005567476,0.0001972907,0.0006415261,0.0005618462,0.0005479805,0.0004775328],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003713642,"about_ca_system_score_gemma":0.0004829142,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003450263,"about_ca_topic_score_gemma":0.006468299,"domain_scores_codex":[0.9992084,0.0002417716,0.00007360086,0.0002589755,0.0001526681,0.0000645969],"domain_scores_gemma":[0.9967144,0.001697893,0.0005351274,0.0002681444,0.0006563553,0.0001281756],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00129348,0.000418708,0.4817745,0.0003882711,0.0005226678,0.0005624369,0.0003885753,0.1084379,0.01926387,0.0008358746,0.008184797,0.3779289],"study_design_scores_gemma":[0.00006657916,0.0004466175,0.2368882,0.000100045,0.0001577305,0.000957469,0.000226396,0.7369502,0.01798171,0.002767056,0.003404366,0.00005354911],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.888405,0.001267064,0.1027821,0.0006046576,0.00006529387,0.0001192915,0.002409116,0.001020922,0.003326367],"genre_scores_gemma":[0.9852844,0.0001762619,0.01233399,0.00008072713,0.00003429369,0.00002433782,0.001433199,0.00001828674,0.0006145219],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003450263,"threshold_uncertainty_score":0.01156342,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4408023203","doi":"10.1056/aioa2400948","title":"Opportunistic Screening of Chronic Liver Disease with Deep-Learning–Enhanced Echocardiography","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Liver Disease and Transplantation","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"National Center for Advancing Translational Sciences; National Heart, Lung, and Blood Institute","keywords":"Medicine; Cardiology; Internal medicine; Chronic liver disease; Cirrhosis","authors":[{"name":"Yuki Sahashi","is_ca":false},{"name":"Miloš Vukadinovic","is_ca":false},{"name":"Fatemeh Amrollahi","is_ca":false},{"name":"Hirsh D. Trivedi","is_ca":false},{"name":"Justin Rhee","is_ca":false},{"name":"Jonathan H. Chen","is_ca":false},{"name":"Susan Cheng","is_ca":false},{"name":"David Ouyang","is_ca":true},{"name":"Alan C. Kwan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.007730231190715632,"gpt":0.2464394915518351,"spread":0.2387092603611194,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002047913,0.0006049849,0.00044088,0.001042488,0.0001625302,0.0005565844,0.0006670713,0.0003855345,0.0008472557],"category_scores_gemma":[0.006577026,0.0003019903,0.0004750046,0.0004322949,0.0002555051,0.0005826845,0.0009793446,0.0005767272,0.0003039956],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005619922,"about_ca_system_score_gemma":0.0006427636,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004613067,"about_ca_topic_score_gemma":0.01084866,"domain_scores_codex":[0.998901,0.0005303865,0.00007852315,0.0002608346,0.0001335593,0.00009568682],"domain_scores_gemma":[0.9968635,0.001317969,0.0007197293,0.0002846236,0.0004736437,0.0003405342],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008095126,0.0004128798,0.941024,0.00007658558,0.0002219408,0.0001784367,0.00004970361,0.01076683,0.001173752,0.0001110179,0.001543668,0.04363167],"study_design_scores_gemma":[0.0002100623,0.00149607,0.4636086,0.0001266977,0.0002750028,0.001122306,0.0001298994,0.5248906,0.004518201,0.001195129,0.002357523,0.00006990785],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.985201,0.0006580096,0.00996232,0.0004125779,0.00002662878,0.000147098,0.001777161,0.0003213955,0.00149382],"genre_scores_gemma":[0.9921787,0.0001315034,0.006111725,0.0001152257,0.00002456034,0.00003906106,0.001235393,0.000009330557,0.0001544674],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004613067,"threshold_uncertainty_score":0.01083052,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4409767654","doi":"10.1056/aioa2400703","title":"Longitudinal Risk Prediction for Pediatric Glioma with Temporal Deep Learning","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Glioma Diagnosis and Treatment","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"National Institute of Dental and Craniofacial Research; National Cancer Institute","keywords":"Deep learning; Glioma; Artificial intelligence; Computer science; Medicine; Cancer research","authors":[{"name":"Divyanshu Tak","is_ca":true},{"name":"Biniam A. Garomsa","is_ca":true},{"name":"Anna Zapaishchykova","is_ca":true},{"name":"Zezhong Ye","is_ca":true},{"name":"Sridhar Vajapeyam","is_ca":false},{"name":"Maryamalsadat Mahootiha","is_ca":false},{"name":"Juan Carlos Pardo","is_ca":false},{"name":"Ceilidh Smith","is_ca":false},{"name":"Ariana Familiar","is_ca":false},{"name":"Tafadzwa L. Chaunzwa","is_ca":true},{"name":"Kevin X. Liu","is_ca":false},{"name":"Sanjay P. Prabhu","is_ca":false},{"name":"Pratiti Bandopadhayay","is_ca":false},{"name":"Ali Nabavizadeh","is_ca":false},{"name":"Sabine Mueller","is_ca":false},{"name":"Hugo J.W.L. Aerts","is_ca":true},{"name":"Daphne A. Haas‐Kogan","is_ca":false},{"name":"Tina Young Poussaint","is_ca":false},{"name":"Benjamin H. Kann","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.009416397693304265,"gpt":0.2674186025855746,"spread":0.2580022048922703,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001066655,0.0006812753,0.0003794963,0.0008029588,0.0001729127,0.0004211415,0.0006079308,0.0004175834,0.0008487393],"category_scores_gemma":[0.003068855,0.0002400979,0.0006976604,0.0004828675,0.0001875582,0.0005251917,0.000633804,0.0008902412,0.0002734741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008989636,"about_ca_system_score_gemma":0.00116576,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01337511,"about_ca_topic_score_gemma":0.01842505,"domain_scores_codex":[0.9997488,0.00006522969,0.00002119389,0.00008542972,0.00004046606,0.00003882264],"domain_scores_gemma":[0.9991633,0.0003253841,0.0002059758,0.0000702274,0.0001650579,0.0000700762],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001029491,0.000548631,0.3595159,0.0001418114,0.0004410089,0.000740077,0.0001183414,0.3814404,0.003635476,0.001592567,0.01120871,0.2395877],"study_design_scores_gemma":[0.00002543333,0.0001108123,0.0108239,0.00002075499,0.00005476055,0.0001698051,0.00002479212,0.9844332,0.001393531,0.002171192,0.0007572528,0.0000147132],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8734344,0.002165647,0.1142667,0.001821469,0.0001000246,0.00007711456,0.004798781,0.001454999,0.001880713],"genre_scores_gemma":[0.974878,0.0003994175,0.01918459,0.0001968109,0.00004991393,0.0000565716,0.004322798,0.00003687526,0.0008749368],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01337511,"threshold_uncertainty_score":0.02659452,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4411415068","doi":"10.1056/aipc2500153","title":"Lessons from the Failure of Canada’s Artificial Intelligence and Data Act","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Princess Margaret Cancer Centre; University Health Network; University of Toronto","funders":"","keywords":"Artificial intelligence; Computer science","authors":[{"name":"Abdullah Ishaque","is_ca":true},{"name":"Abdi Aidid","is_ca":true},{"name":"Gelareh Zadeh","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5619306336519689,"gpt":0.5951833525381953,"spread":0.03325271888622638,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.0740294,0.0008267554,0.001974508,0.002877807,0.02551053,0.02450356,0.005742964,0.04660949,0.009878671],"category_scores_gemma":[0.1644345,0.001170494,0.001406782,0.0033331,0.0551257,0.009665139,0.007953755,0.06194771,0.001406838],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.1366204,"about_ca_system_score_gemma":0.3374426,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9542326,"about_ca_topic_score_gemma":0.9595861,"domain_scores_codex":[0.9376349,0.02093427,0.00205655,0.004081818,0.02177893,0.01351343],"domain_scores_gemma":[0.7311477,0.1467457,0.006358236,0.007769039,0.05925326,0.04872616],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001145937,0.00005968465,0.003724491,0.00009475767,0.00009526981,0.0006420455,0.005120452,0.001151986,0.000113311,0.5147103,0.4544373,0.0197356],"study_design_scores_gemma":[0.0002847675,0.00005212126,0.006530741,0.001209883,0.00009995354,0.0003322554,0.006061274,0.001425881,0.0003209479,0.2925288,0.6908819,0.000271425],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.001873026,0.002180053,0.0005555931,0.9784782,0.001056637,0.000009955692,0.00007516617,0.00001925358,0.0157521],"genre_scores_gemma":[0.1716686,0.003823865,0.001896768,0.8003349,0.003419271,0.00008301396,0.0001169751,0.000180552,0.01847618],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9744895,"threshold_uncertainty_score":0.9912552,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7117104805","doi":"10.1056/aioa2500522","title":"International Retrospective Observational Study of Continual Learning for AI on Endotracheal Tube Placement from Chest Radiographs","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; Western University; University of Toronto","funders":"","keywords":"Observational study; Radiography; Retrospective cohort study; Endotracheal tube; MEDLINE","authors":[{"name":"Emma Chen","is_ca":false},{"name":"Agustina Saenz","is_ca":false},{"name":"Oishi Banerjee","is_ca":false},{"name":"Henrik Marklund","is_ca":false},{"name":"Xiaoman Zhang","is_ca":false},{"name":"Shreya Johri","is_ca":false},{"name":"Luyang Luo","is_ca":false},{"name":"Subathra Adithan","is_ca":false},{"name":"Kay Wu","is_ca":true},{"name":"Siddhant Dogra","is_ca":false},{"name":"Vijay Janapa Reddi","is_ca":false},{"name":"Dominic Buensalido","is_ca":false},{"name":"Helen Kavnoudias","is_ca":false},{"name":"Roman Kloeckner","is_ca":false},{"name":"Lukas Müller","is_ca":false},{"name":"Emmanuel Salinas-Miranda","is_ca":false},{"name":"Maria José Veloza Vega","is_ca":false},{"name":"Johannes Kolck","is_ca":false},{"name":"Tobias Penzkofer","is_ca":false},{"name":"Daiju Ueda","is_ca":false},{"name":"Shannon L. Walston","is_ca":false},{"name":"Armin Quispe Cornejo","is_ca":false},{"name":"Michele Salvagno","is_ca":false},{"name":"Christopher Lee","is_ca":false},{"name":"Jonathan Fournier","is_ca":false},{"name":"Rosa Castillo","is_ca":false},{"name":"Cibele Luna","is_ca":false},{"name":"Tara Bahramipour","is_ca":false},{"name":"Amanda R. Zuback","is_ca":false},{"name":"Rickmer Braren","is_ca":false},{"name":"Yutthaphan Wannasopha","is_ca":false},{"name":"Piyapong Khumrin","is_ca":false},{"name":"Desmond Lim Shi Wei Wei","is_ca":false},{"name":"James Thomas Patrick Decourcy Hallinan","is_ca":false},{"name":"Zhicheng Jiao","is_ca":false},{"name":"Thomas Yi","is_ca":false},{"name":"Juana María Plasencia Martinez","is_ca":false},{"name":"Nuria Isabel Casado Alarcon","is_ca":false},{"name":"Franz A. Fellner","is_ca":false},{"name":"Julian F. Niedermair","is_ca":false},{"name":"Derek Wu","is_ca":true},{"name":"Dongkeun Kim","is_ca":true},{"name":"Johannes Haubold","is_ca":false},{"name":"Lars Heiliger","is_ca":false},{"name":"Daniel Pérez-Chada","is_ca":false},{"name":"Pablo Pratesi","is_ca":false},{"name":"Ryan Cummings","is_ca":false},{"name":"Narges Razavian","is_ca":false},{"name":"Anastasia Oikonomou","is_ca":true},{"name":"William T. Tran","is_ca":true},{"name":"Thomas Küstner","is_ca":false},{"name":"Saif Afat","is_ca":false},{"name":"Adrienne N. Dula","is_ca":false},{"name":"Justin F. Rousseau","is_ca":false},{"name":"Franko Hržić","is_ca":false},{"name":"Michael Fuchsjäger","is_ca":false},{"name":"Pranav Rajpurkar","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1655640285438163,"gpt":0.4549574241101151,"spread":0.2893933955662988,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009471786,0.0004718079,0.0006454238,0.001299863,0.0008798305,0.001132797,0.0007181988,0.001049667,0.00254839],"category_scores_gemma":[0.004808052,0.0006172236,0.000748443,0.002174716,0.0006673161,0.0009360287,0.0008788058,0.001391276,0.0006363146],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006423729,"about_ca_system_score_gemma":0.001103849,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005998189,"about_ca_topic_score_gemma":0.005307574,"domain_scores_codex":[0.9986738,0.0002793601,0.0002832119,0.0003451753,0.0002079744,0.0002105251],"domain_scores_gemma":[0.9930917,0.001004726,0.003717089,0.0005473726,0.0006761025,0.0009630386],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001070732,0.0001044328,0.9986842,0.0000207266,0.00004161424,0.0002273385,0.0001290157,0.000007364322,0.00005624411,0.00001383447,0.00009992842,0.000508232],"study_design_scores_gemma":[0.0000162251,0.0004178767,0.9960809,0.00003497405,0.00006344113,0.001577998,0.001283227,0.00007114813,0.00004069152,0.0000225749,0.0003785623,0.0000122906],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9984207,0.0003210863,0.00008119919,0.00003873404,0.00001006695,0.00002963948,0.0004111105,0.00000311287,0.0006843598],"genre_scores_gemma":[0.9985643,0.0002147258,0.0001228511,0.0001065384,0.00003582759,0.00003429855,0.0007293904,0.0000051856,0.000186889],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005998189,"threshold_uncertainty_score":0.01192659,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4413815394","doi":"10.1056/aip2500154","title":"Energy Considerations for Scaling Artificial Intelligence Adoption in Medicine: First Do No Harm","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Climate Change and Health Impacts","field":"Environmental Science","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Toronto Public Health; Public Health Ontario; University of Toronto","funders":"","keywords":"Harm; Do no harm; Scaling; Energy (signal processing); Computer science; Artificial intelligence; Psychology; Social psychology; Mathematics; Psychiatry; Statistics","authors":[{"name":"Armaan K. Malhotra","is_ca":true},{"name":"Amol A. Verma","is_ca":true},{"name":"Christopher W Selby Smith","is_ca":true},{"name":"Husain Shakil","is_ca":true},{"name":"Muhammad Mamdani","is_ca":true},{"name":"Laura C. Rosella","is_ca":true},{"name":"Christopher D. Witiw","is_ca":true},{"name":"Jefferson R. Wilson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1083559187372145,"gpt":0.3702471005966559,"spread":0.2618911818594414,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01491316,0.001071643,0.001156254,0.001941233,0.002582835,0.008733656,0.002584645,0.00949845,0.02609767],"category_scores_gemma":[0.07807539,0.0006150113,0.001601682,0.001224454,0.006897179,0.01550922,0.006363115,0.009071028,0.00449016],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003435721,"about_ca_system_score_gemma":0.008553125,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005081927,"about_ca_topic_score_gemma":0.006315526,"domain_scores_codex":[0.9885439,0.005620471,0.0004971189,0.0007212961,0.003710031,0.0009071288],"domain_scores_gemma":[0.9579985,0.02435852,0.001630806,0.004465042,0.009258764,0.002288514],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004601833,0.0003465159,0.00274691,0.00101417,0.0003742745,0.0002956032,0.0005897448,0.01388452,0.002766794,0.6545333,0.1172541,0.2057337],"study_design_scores_gemma":[0.0001491167,0.0004894697,0.003572008,0.002097264,0.0001359946,0.0004484256,0.001614891,0.007342017,0.001727559,0.6736929,0.3085549,0.0001753735],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.01235541,0.02585774,0.06573028,0.6880307,0.01135118,0.0002606613,0.0005011839,0.0005422063,0.1953706],"genre_scores_gemma":[0.6784683,0.02459899,0.08702333,0.1607306,0.01184933,0.0008849556,0.0003952927,0.0008583855,0.03519066],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.02609767,"threshold_uncertainty_score":0.08730543,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4409990539","doi":"10.1056/aics2401008","title":"Clinical Deployment of Real-Time Left Ventricular Ejection Fraction Estimation from Coronary Angiography","year":2025,"lang":"en","type":"article","venue":"NEJM AI","topic":"Cardiac Imaging and Diagnostics","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary; Université de Montréal; Montreal Heart Institute; University of Ottawa","funders":"","keywords":"Ejection fraction; Cardiology; Coronary angiography; Internal medicine; Software deployment; Angiography; Fraction (chemistry); Estimation; Medicine; Computer science; Heart failure; Myocardial infarction; Economics; Chemistry","authors":[{"name":"Pascal Thériault-Lauzier","is_ca":true},{"name":"Stanley Ly","is_ca":true},{"name":"Michael Knafo","is_ca":true},{"name":"A Grandchamp","is_ca":true},{"name":"Marie-Gabrielle Lessard","is_ca":true},{"name":"Denis Corbin","is_ca":true},{"name":"Bahareh Taji","is_ca":true},{"name":"Olivier Tastet","is_ca":true},{"name":"Aun‐Yeong Chong","is_ca":true},{"name":"Michael Froeschl","is_ca":true},{"name":"Alexander Dick","is_ca":true},{"name":"Marino Labinaz","is_ca":true},{"name":"Chris Glover","is_ca":true},{"name":"Michel Le May","is_ca":true},{"name":"Zeeshan Ahmed","is_ca":true},{"name":"Omar Abdel‐Razek","is_ca":true},{"name":"Pietro Di Santo","is_ca":true},{"name":"David Messika–Zeitoun","is_ca":true},{"name":"Geoffrey H. Tison","is_ca":false},{"name":"Juan Russo","is_ca":true},{"name":"Guillaume Marquis‐Gravel","is_ca":true},{"name":"Jean‐François Tanguay","is_ca":true},{"name":"Richard Gallo","is_ca":true},{"name":"Quoc Hung Ly","is_ca":true},{"name":"Anita Asgar","is_ca":true},{"name":"Serge Doucet","is_ca":true},{"name":"Gilbert Gosselin","is_ca":true},{"name":"Jean Grégoire","is_ca":true},{"name":"Réda Ibrahim","is_ca":true},{"name":"Philippe L. L’Allier","is_ca":true},{"name":"Mohamed Nosair","is_ca":true},{"name":"Derek So","is_ca":true},{"name":"Robert Avram","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01010705609729034,"gpt":0.3206113315904624,"spread":0.3105042754931721,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001833932,0.0006009841,0.0006268346,0.0006807395,0.0002104767,0.001590879,0.0006125662,0.001198881,0.003336349],"category_scores_gemma":[0.007421003,0.0003953413,0.0001989271,0.0002529754,0.0002948722,0.0007572096,0.0006120457,0.0008663978,0.001889311],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001664168,"about_ca_system_score_gemma":0.000359819,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005492768,"about_ca_topic_score_gemma":0.0009039871,"domain_scores_codex":[0.9990135,0.0004798526,0.00007030533,0.0001698728,0.0001991539,0.00006734247],"domain_scores_gemma":[0.9975904,0.001242501,0.0001845445,0.0002415353,0.000472333,0.0002686362],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01080086,0.001037188,0.1482211,0.0006004189,0.0001845077,0.00720769,0.0005145642,0.004007797,0.1341835,0.001592379,0.01706825,0.6745818],"study_design_scores_gemma":[0.002517615,0.01474608,0.4486544,0.001276548,0.001147016,0.04847499,0.00141594,0.2348367,0.1584166,0.00741374,0.08051253,0.0005879888],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7115248,0.01619482,0.2223758,0.00723746,0.001747215,0.0007541213,0.001211607,0.004338225,0.03461608],"genre_scores_gemma":[0.9553339,0.002155875,0.03784834,0.001463919,0.0009352352,0.0001451526,0.0003382066,0.0001849559,0.001594435],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003336349,"threshold_uncertainty_score":0.01116115,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}