{"meta":{"query_hash":"5570edbb5698","filters":{"venue":"BMJ Digital Health & AI"},"cohort_total":4,"direct_labels_cover":0,"predictions_cover":4,"exported":4,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/5570edbb5698","api":"https://metacan.xera.ac/api/v1/cohort?venue=BMJ+Digital+Health+%26+AI"},"results":[{"id":"W4413252984","doi":"10.1136/bmjdhai-2025-000114","title":"Development and prospective evaluation of a machine learning model to predict serious cardiac outcomes among paediatric cardiac inpatients","year":2025,"lang":"en","type":"article","venue":"BMJ Digital Health & AI","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Patient Safety Institute; SickKids Foundation; Toronto General Hospital; Casey House; Institute for Clinical Evaluative Sciences; Hospital for Sick Children","funders":"","keywords":"Retrospective cohort study; Medicine; Receiver operating characteristic; Logistic regression; Prospective cohort study; Emergency medicine; Internal medicine","score_opus":0.02054444801677871,"score_gpt":0.3433876705748698,"score_spread":0.32284322255809106,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413252984","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88265777,0.00083297933,0.1077011,0.0021613445,0.00017446364,0.00049043994,0.0032608432,0.0010964616,0.0016246438],"genre_scores_gemma":[0.94197893,0.0002306388,0.051263694,0.00031106517,0.00007511943,0.00041848517,0.0044373143,0.00004736919,0.001237408],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9988403,0.000583039,0.00008973804,0.00027170999,0.00011797463,0.00009717683],"domain_scores_gemma":[0.99485105,0.003668464,0.0002917177,0.00019044885,0.0007874389,0.00021086558],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006446878,0.0010757784,0.0008507327,0.0010583985,0.00032869205,0.0011345455,0.0014246969,0.0009125407,0.0016443321],"category_scores_gemma":[0.010020866,0.0004196037,0.0010130603,0.0005514685,0.00022164873,0.0006442266,0.0008560414,0.0015123455,0.0006081031],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011485246,0.0013914388,0.25895065,0.0001799007,0.00063467,0.00031055696,0.00019918094,0.6210732,0.0013770247,0.0007614069,0.006400057,0.107573554],"study_design_scores_gemma":[0.00004080916,0.00025378584,0.0071181227,0.000026652273,0.00004320463,0.000039693143,0.000034855126,0.99135876,0.0004305403,0.00033445974,0.00030914805,0.000010124805],"about_ca_topic_score_codex":0.009499257,"about_ca_topic_score_gemma":0.007196816,"teacher_disagreement_score":0.009499257,"about_ca_system_score_codex":0.001092394,"about_ca_system_score_gemma":0.0017784754,"threshold_uncertainty_score":0.03409481},"labels":[],"label_agreement":null},{"id":"W4414189579","doi":"10.1136/bmjdhai-2025-000014","title":"Optimising large language models for clinical information extraction: a benchmarking study in the context of ulcerative colitis research","year":2025,"lang":"en","type":"article","venue":"BMJ Digital Health & AI","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Clinical and Translational Science Institute, University of California, Los Angeles; Clinical and Translational Science Institute, University of California, San Francisco","keywords":"Benchmarking; Context (archaeology); Adaptation (eye); Oracle; Language model; Set (abstract data type); Predictive modelling; Colonoscopy","score_opus":0.11497063511652149,"score_gpt":0.5230203048782074,"score_spread":0.4080496697616859,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414189579","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6816127,0.0058194315,0.27555737,0.0033223245,0.000408588,0.0011934413,0.0073755123,0.019818954,0.0048918],"genre_scores_gemma":[0.771923,0.00095693406,0.20784768,0.00059015217,0.00010644674,0.00067125104,0.015409132,0.0009125505,0.0015828643],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9874959,0.009346471,0.00077206915,0.0015501648,0.0006260784,0.00020926427],"domain_scores_gemma":[0.9216779,0.069966815,0.0011412716,0.0037952878,0.0029835757,0.0004351848],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01933746,0.0016412617,0.0008698837,0.0017141904,0.00050755456,0.0020250229,0.0022567336,0.0016492546,0.0020712023],"category_scores_gemma":[0.06170137,0.0007287963,0.0016757963,0.0020004078,0.0008421748,0.0032311894,0.0019405187,0.002243378,0.0011071782],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0028230567,0.0012269986,0.024489934,0.0024195223,0.0011242471,0.0007399291,0.0017280657,0.5579401,0.0059963586,0.0029923748,0.016833343,0.38168603],"study_design_scores_gemma":[0.00036909254,0.00063503097,0.0053952523,0.0001491724,0.00030797484,0.00025384035,0.00044095886,0.97656095,0.00591034,0.0043341755,0.0055528916,0.000090264664],"about_ca_topic_score_codex":0.012079808,"about_ca_topic_score_gemma":0.012775487,"teacher_disagreement_score":0.9806625,"about_ca_system_score_codex":0.0023218587,"about_ca_system_score_gemma":0.0020259665,"threshold_uncertainty_score":0.102267504},"labels":[],"label_agreement":null},{"id":"W7105645746","doi":"10.1136/bmjdhai-2025-000078","title":"“Trying to Do No Harm”: exploring clinician concerns towards the use of AI for risk prediction in psychiatry","year":2025,"lang":"en","type":"article","venue":"BMJ Digital Health & AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Waypoint Centre for Mental Health Care; University of Toronto","funders":"","keywords":"Mental health; Thematic analysis; Focus group; Health care; Reflexivity; Quality (philosophy); MEDLINE; Mental health care","score_opus":0.4641667768071848,"score_gpt":0.5330326401816575,"score_spread":0.0688658633744727,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7105645746","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9402299,0.0020856336,0.007143344,0.044659737,0.00031522464,0.00033036314,0.00010680323,0.000050217608,0.0050787926],"genre_scores_gemma":[0.9847452,0.0014488397,0.0030362513,0.009245651,0.00006449726,0.00028644284,0.000030686508,0.000035976173,0.0011065025],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9334252,0.055670407,0.0019070193,0.0016704248,0.004586824,0.0027402213],"domain_scores_gemma":[0.8202993,0.14530678,0.010988149,0.002851838,0.011673934,0.00887993],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.054604866,0.0006056715,0.00084192323,0.0013564297,0.009292783,0.0064892257,0.0027329624,0.0042524077,0.0023428637],"category_scores_gemma":[0.12700169,0.0011062605,0.0006410702,0.0009372825,0.017781174,0.006798801,0.008234088,0.0069455323,0.00047229673],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000030350862,0.000027755108,0.0033157386,0.00026614213,0.000009165697,0.0005326624,0.98968303,0.00003571959,0.00043236348,0.000900215,0.0009031727,0.0038636308],"study_design_scores_gemma":[0.00000974033,0.00007514999,0.0012438812,0.00037679964,0.000008902458,0.00053462427,0.9904951,0.00013861129,0.0002232334,0.0006505576,0.0062180804,0.000025389032],"about_ca_topic_score_codex":0.011078569,"about_ca_topic_score_gemma":0.011968589,"teacher_disagreement_score":0.054604866,"about_ca_system_score_codex":0.009662748,"about_ca_system_score_gemma":0.013664392,"threshold_uncertainty_score":0.28878164},"labels":[],"label_agreement":null},{"id":"W7105695419","doi":"10.1136/bmjdhai-2025-000134","title":"Retrospective evaluation of a machine learning model to facilitate pharmacogenetic testing","year":2025,"lang":"en","type":"article","venue":"BMJ Digital Health & AI","topic":"Pharmacogenetics and Drug Metabolism","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto General Hospital; Institute for Clinical Evaluative Sciences; Hospital for Sick Children","funders":"Hospital for Sick Children","keywords":"Pharmacogenetics; Logistic regression; Medical prescription; Epidemiology; Retrospective cohort study; Cohort; Gradient boosting","score_opus":0.3195143293878861,"score_gpt":0.5259982311863415,"score_spread":0.20648390179845538,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7105695419","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3750523,0.0011998926,0.6145654,0.0015858025,0.0001745382,0.00035840934,0.0022934554,0.0018960342,0.0028742312],"genre_scores_gemma":[0.9052494,0.00031825688,0.09027218,0.00022114221,0.000088340676,0.00025437394,0.0023243404,0.00007009007,0.001201872],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.99746025,0.0014150244,0.00023142724,0.0004901859,0.00026865545,0.00013448147],"domain_scores_gemma":[0.9864031,0.0099278595,0.0009811298,0.0007720562,0.0017527331,0.00016299661],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011550466,0.0010853834,0.0008536279,0.0012138926,0.0003214807,0.0015696671,0.0011693247,0.00094296446,0.0017144906],"category_scores_gemma":[0.024807064,0.0005179086,0.000965496,0.00071512646,0.00039854957,0.0010787828,0.00090171525,0.0013073026,0.00067828974],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005919615,0.00022564588,0.109178655,0.00012619777,0.0003165953,0.00029424173,0.0001418954,0.8104613,0.0014560259,0.002941646,0.0025423565,0.07172356],"study_design_scores_gemma":[0.000010723592,0.00008206791,0.0031126097,0.00002039834,0.000024928439,0.00006352697,0.000014394544,0.9949805,0.0004657035,0.0007221044,0.0004936495,0.000009354327],"about_ca_topic_score_codex":0.008522661,"about_ca_topic_score_gemma":0.004628108,"teacher_disagreement_score":0.011550466,"about_ca_system_score_codex":0.001021221,"about_ca_system_score_gemma":0.0012539302,"threshold_uncertainty_score":0.061085463},"labels":[],"label_agreement":null}]}