{"meta":{"query_hash":"dc36cf37c4ed","filters":{"venue":"JMIR AI"},"cohort_total":21,"direct_labels_cover":0,"predictions_cover":21,"exported":21,"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/dc36cf37c4ed","api":"https://metacan.xera.ac/api/v1/cohort?venue=JMIR+AI"},"results":[{"id":"W4317459808","doi":"10.2196/42046","title":"Introducing JMIR AI","year":2022,"lang":"en","type":"editorial","venue":"JMIR AI","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Scope (computer science); Focus (optics); Publishing; Data science; Computer science; Artificial intelligence; Political science","score_opus":0.012523897634655813,"score_gpt":0.35479406003269637,"score_spread":0.34227016239804053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4317459808","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00005685142,0.01896479,0.001000472,0.120193616,0.8428678,0.0002316062,0.00038028538,0.0003701164,0.015934512],"genre_scores_gemma":[0.001461093,0.023348846,0.0017331277,0.08758886,0.85844064,0.0004188817,0.0004217346,0.0003362657,0.02625061],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97055537,0.008138384,0.0043231,0.0016982874,0.014261565,0.0010233622],"domain_scores_gemma":[0.86858857,0.051412184,0.005306585,0.0033300382,0.06190583,0.009456705],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025765466,0.0020633582,0.0029672112,0.008380425,0.0034438842,0.017145922,0.0030981486,0.010392457,0.05830203],"category_scores_gemma":[0.16383207,0.0011491834,0.002601809,0.0031160717,0.0028814548,0.0070875436,0.0033763398,0.016211364,0.036623325],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020317013,0.0000061526407,0.000022291328,0.00044187516,0.000016394717,0.000020390364,0.000026425876,0.000009229401,0.000025897114,0.00056607637,0.9825965,0.016248364],"study_design_scores_gemma":[0.000058483456,0.000022991058,0.000112268914,0.0011952742,0.000035291287,0.00010498459,0.000044391294,0.000060351947,0.000043871358,0.001675286,0.99662447,0.000022296237],"about_ca_topic_score_codex":0.0026350403,"about_ca_topic_score_gemma":0.00518563,"teacher_disagreement_score":0.05830203,"about_ca_system_score_codex":0.0042700972,"about_ca_system_score_gemma":0.011694173,"threshold_uncertainty_score":0.19503975},"labels":[],"label_agreement":null},{"id":"W4322767845","doi":"10.2196/40973","title":"Preparing for an Artificial Intelligence–Enabled Future: Patient Perspectives on Engagement and Health Care Professional Training for Adopting Artificial Intelligence Technologies in Health Care Settings","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre for Addiction and Mental Health; Vector Institute; Michener Institute; Public Health Ontario; University of Toronto; University Health Network","funders":"Government of Canada","keywords":"Health care; Thematic analysis; Workflow; Population health; Population; Nursing; Psychology; Medical education; Medicine; Qualitative research; Public health; Computer science; Political science; Sociology","score_opus":0.2205622886818319,"score_gpt":0.48207317062381777,"score_spread":0.2615108819419859,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4322767845","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":[0.9484763,0.0011752598,0.001483618,0.039535515,0.00014640718,0.00012587312,0.000062249695,0.000033390705,0.0089613795],"genre_scores_gemma":[0.9947351,0.0006180666,0.0006517372,0.0033962862,0.000026505793,0.000054466494,0.00001922216,0.000008255891,0.0004903549],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9676462,0.024787052,0.000977061,0.00083220156,0.0019632438,0.0037942163],"domain_scores_gemma":[0.95286995,0.027404714,0.0047407695,0.00081347226,0.0032968665,0.010874203],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024521079,0.00040972308,0.0006144497,0.0013546938,0.013864132,0.008491115,0.0018069828,0.0040115104,0.0030289881],"category_scores_gemma":[0.044539098,0.00066252565,0.00069682207,0.0012131914,0.012585179,0.0061817304,0.009664404,0.008239698,0.00033635984],"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":[0.000035242534,0.00007655699,0.009535734,0.000117089236,0.000010216821,0.0007813073,0.9803114,0.000044784178,0.00032835547,0.0018113108,0.0014679251,0.005480106],"study_design_scores_gemma":[0.000009169903,0.000106134765,0.002789267,0.00016831943,0.00000742646,0.0006043787,0.9885748,0.00012663513,0.000117230105,0.0005775198,0.0068899863,0.000029013998],"about_ca_topic_score_codex":0.011760673,"about_ca_topic_score_gemma":0.013604367,"teacher_disagreement_score":0.024521079,"about_ca_system_score_codex":0.007888147,"about_ca_system_score_gemma":0.012613958,"threshold_uncertainty_score":0.12968141},"labels":[],"label_agreement":null},{"id":"W4323544013","doi":"10.2196/41264","title":"Developing an Inpatient Electronic Medical Record Phenotype for Hospital-Acquired Pressure Injuries: Case Study Using Natural Language Processing Models","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Pressure Ulcer Prevention and Management","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Foothills Medical Centre; Alberta Health Services; University of Alberta; University of Calgary","funders":"","keywords":"Medicine; Medical record; Artificial intelligence; Random forest; Electronic health record; Machine learning; Gradient boosting; Computer science; Natural language processing; Health care; Emergency medicine; Internal medicine","score_opus":0.06514726895339741,"score_gpt":0.46718918352303673,"score_spread":0.40204191456963934,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4323544013","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":[0.96769756,0.00027289783,0.02713239,0.0019782742,0.000030337425,0.0004202159,0.0011578458,0.000120233126,0.0011902399],"genre_scores_gemma":[0.96837664,0.00026809992,0.02937201,0.00024741096,0.00003409769,0.00013512059,0.0010966578,0.000022745478,0.00044736746],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9985468,0.00052869326,0.00015658383,0.00029230758,0.00033956135,0.000136083],"domain_scores_gemma":[0.99460983,0.003285364,0.0009188263,0.00027887837,0.00069145905,0.00021562757],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00306868,0.000518799,0.000324216,0.001664795,0.0005470351,0.0013247795,0.0009326831,0.0011193611,0.00088141795],"category_scores_gemma":[0.011681795,0.0002798551,0.0007434729,0.0009895661,0.00051543367,0.00087109377,0.00069367717,0.001111942,0.00029070248],"study_design_candidate":"simulation_or_modeling","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":[0.00026192225,0.001211881,0.8892958,0.00027083512,0.00012716722,0.026344312,0.0018864949,0.010285631,0.0022443656,0.0016260208,0.0039974153,0.06244818],"study_design_scores_gemma":[0.000106220774,0.0010570263,0.3544842,0.0005481686,0.00029836697,0.038358092,0.008814784,0.5731882,0.008683902,0.0062777516,0.0079980325,0.0001852539],"about_ca_topic_score_codex":0.016076945,"about_ca_topic_score_gemma":0.020779956,"teacher_disagreement_score":0.016076945,"about_ca_system_score_codex":0.0015114018,"about_ca_system_score_gemma":0.0014309237,"threshold_uncertainty_score":0.031966746},"labels":[],"label_agreement":null},{"id":"W4361975492","doi":"10.2196/44779","title":"A Scalable Radiomics- and Natural Language Processing–Based Machine Learning Pipeline to Distinguish Between Painful and Painless Thoracic Spinal Bone Metastases: Retrospective Algorithm Development and Validation Study","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé; McGill University Health Centre; Faculty of Medicine and Health, University of Sydney; McGill University","keywords":"Medicine; Artificial intelligence; Receiver operating characteristic; Radiomics; Radiology; Machine learning; Retrospective cohort study; Random forest; Computer science; Surgery; Internal medicine","score_opus":0.013661754956650488,"score_gpt":0.34180601000543387,"score_spread":0.32814425504878336,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4361975492","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6343309,0.0017631252,0.352307,0.00038755068,0.000087780165,0.001986093,0.0015985911,0.00616529,0.0013736995],"genre_scores_gemma":[0.7750285,0.00029542277,0.21986401,0.00012620472,0.000025708798,0.0011014235,0.0027609924,0.00009694315,0.00070077804],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.998178,0.0006334601,0.00018269803,0.00054258626,0.00033640923,0.00012691598],"domain_scores_gemma":[0.995468,0.0024089415,0.00032743328,0.00047917574,0.0012122972,0.00010415951],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0062452294,0.0011300746,0.00084098557,0.0011555953,0.00046198946,0.00089551194,0.0014800049,0.0008756646,0.0011186411],"category_scores_gemma":[0.0098566525,0.00046991106,0.001026234,0.0007900647,0.0005060256,0.0009533855,0.0010182273,0.0009911989,0.0005509673],"study_design_candidate":"bench_or_experimental","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":[0.0018950257,0.0017624705,0.096032724,0.0005084061,0.00074802584,0.00051384623,0.00026251553,0.29049602,0.020517193,0.00092077354,0.004649746,0.5816933],"study_design_scores_gemma":[0.00013623005,0.00051358726,0.0108928345,0.00002323965,0.000113298556,0.00018023349,0.00006155162,0.9806138,0.0061047436,0.0004341989,0.00090432924,0.000021844147],"about_ca_topic_score_codex":0.0125350235,"about_ca_topic_score_gemma":0.008914345,"teacher_disagreement_score":0.0125350235,"about_ca_system_score_codex":0.0013881875,"about_ca_system_score_gemma":0.0027189497,"threshold_uncertainty_score":0.033028305},"labels":[],"label_agreement":null},{"id":"W4366307907","doi":"10.2196/44835","title":"Natural Language Processing for Clinical Laboratory Data Repository Systems: Implementation and Evaluation for Respiratory Viruses","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; Sinai Health System; Vector Institute; Public Health Ontario; York University; University Health Network; University of Toronto","funders":"Canadian Institutes of Health Research; Hospital for Sick Children","keywords":"Computer science; Artificial intelligence; Natural language processing; Generalizability theory; Parsing; Classifier (UML); Machine learning; Information extraction; Task (project management)","score_opus":0.1563436978837656,"score_gpt":0.5307190196558628,"score_spread":0.3743753217720972,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366307907","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":[0.7272162,0.0017781182,0.17627814,0.0020564888,0.00040037508,0.0047657625,0.005966243,0.07561905,0.005919729],"genre_scores_gemma":[0.6406522,0.0008408894,0.34147027,0.00058683864,0.00006123612,0.0017706483,0.010968541,0.0008535244,0.0027958401],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.996345,0.0012883132,0.000448976,0.00090819417,0.0008010579,0.00020857622],"domain_scores_gemma":[0.98994815,0.0064112023,0.00043971807,0.0008638227,0.0018690601,0.00046806358],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007174647,0.0009995921,0.00060461776,0.0010818596,0.00067974895,0.0014313578,0.0035055636,0.0014608268,0.0033778625],"category_scores_gemma":[0.018562445,0.0005005164,0.00071076525,0.0010493202,0.00077755126,0.0027499998,0.001896624,0.0018534457,0.0016871273],"study_design_candidate":"bench_or_experimental","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":[0.004484861,0.005051716,0.026202483,0.0035657983,0.00069832517,0.0018038785,0.0027572438,0.11880503,0.038437672,0.0023635675,0.04794212,0.7478873],"study_design_scores_gemma":[0.00085423706,0.0011771715,0.009609471,0.0001646344,0.00019142817,0.00044375833,0.00080525974,0.93651235,0.0351666,0.0022501845,0.012712404,0.00011248182],"about_ca_topic_score_codex":0.018411856,"about_ca_topic_score_gemma":0.014442605,"teacher_disagreement_score":0.018411856,"about_ca_system_score_codex":0.0021035993,"about_ca_system_score_gemma":0.0030359006,"threshold_uncertainty_score":0.0379436},"labels":[],"label_agreement":null},{"id":"W4367319158","doi":"10.2196/47995","title":"Reporting and Methodological Observations on Prognostic and Diagnostic Machine Learning Studies","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Agricultural Research Institute of Ontario; University of Ottawa","funders":"","keywords":"Key (lock); Data science; Artificial intelligence; Computer science; Psychology; Machine learning","score_opus":0.3717432222961256,"score_gpt":0.4492624438026238,"score_spread":0.0775192215064982,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4367319158","genre_codex":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023259398,0.48923075,0.04302958,0.33349642,0.06961022,0.0055902544,0.01427195,0.0007013882,0.020810068],"genre_scores_gemma":[0.37250727,0.23592524,0.08519226,0.17689149,0.086223364,0.024346543,0.012210151,0.0023589288,0.004344748],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.20506306,0.34501114,0.31097227,0.021381456,0.11330279,0.0042692907],"domain_scores_gemma":[0.03306918,0.73411673,0.089303456,0.029974647,0.11170257,0.0018335124],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.59159136,0.0015217483,0.002910862,0.025056895,0.00341943,0.014769365,0.0066569983,0.0049384,0.003440621],"category_scores_gemma":[0.9148299,0.0020444598,0.003928684,0.022895072,0.009187081,0.008531571,0.009486941,0.006616355,0.0010443168],"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":[0.0035734263,0.00012333816,0.038869753,0.17209306,0.01145478,0.0006717465,0.027329022,0.00091672014,0.002688388,0.038581815,0.32439792,0.37930003],"study_design_scores_gemma":[0.0019181616,0.0007067969,0.05884064,0.2306553,0.011510951,0.0017717981,0.014674676,0.0022944498,0.006928315,0.05864503,0.61097115,0.0010826148],"about_ca_topic_score_codex":0.0063098483,"about_ca_topic_score_gemma":0.00627398,"teacher_disagreement_score":0.40840864,"about_ca_system_score_codex":0.009715016,"about_ca_system_score_gemma":0.017916791,"threshold_uncertainty_score":0.50364065},"labels":[],"label_agreement":null},{"id":"W4385449689","doi":"10.2196/45257","title":"Machine Learning Models Versus the National Early Warning Score System for Predicting Deterioration: Retrospective Cohort Study in the United Arab Emirates","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Tamkeen; York University; New York University Abu Dhabi; Cleveland Clinic","keywords":"Early warning score; Medicine; Vital signs; Abu dhabi; Retrospective cohort study; Emergency medicine; Cohort; Emergency department; Warning system; Cohort study; Medical emergency; Internal medicine; Surgery","score_opus":0.145280695388654,"score_gpt":0.3718261755518861,"score_spread":0.22654548016323206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385449689","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":[0.9992449,0.00008910747,0.00020715952,0.000053586347,0.000011119027,0.000015197819,0.00031241932,0.0000023967275,0.00006410908],"genre_scores_gemma":[0.9986053,0.000091079724,0.0002607454,0.000037234313,0.000014006625,0.000026335583,0.0008567093,0.0000028556144,0.000105748375],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9985275,0.0006424455,0.00013670356,0.0003528374,0.00018141897,0.0001591408],"domain_scores_gemma":[0.9943082,0.0026802288,0.0012194075,0.0007042071,0.00076028117,0.0003277187],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006013985,0.00068611454,0.00052958494,0.00090992334,0.0004856935,0.00088125595,0.00069775543,0.00070708117,0.00097358035],"category_scores_gemma":[0.010889547,0.0004583627,0.0011660219,0.00059727475,0.0003111114,0.00052645674,0.0006619024,0.0010865616,0.0002491546],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002979087,0.00012089426,0.9976828,0.0000073906353,0.00016105443,0.00008868235,0.000080101316,0.00044790667,0.00007254787,0.00002394324,0.00018710653,0.0008297036],"study_design_scores_gemma":[0.00008882281,0.0010714859,0.9648839,0.000050072882,0.0003196757,0.00063581316,0.0012241462,0.030659303,0.00029389528,0.00010295102,0.0006410004,0.000028781844],"about_ca_topic_score_codex":0.013423177,"about_ca_topic_score_gemma":0.009496625,"teacher_disagreement_score":0.013423177,"about_ca_system_score_codex":0.00061114103,"about_ca_system_score_gemma":0.0006547343,"threshold_uncertainty_score":0.031805396},"labels":[],"label_agreement":null},{"id":"W4385465543","doi":"10.2196/47353","title":"Insights on the Current State and Future Outlook of AI in Health Care: Expert Interview Study","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vector Institute; University of Toronto; St. Michael's Hospital","funders":"Volkswagen Foundation","keywords":"Current (fluid); State (computer science); Health care; Political science; Psychology; Medicine; Engineering ethics; Data science; Computer science; Engineering; Electrical engineering","score_opus":0.17019160021695126,"score_gpt":0.49377730085421834,"score_spread":0.3235857006372671,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385465543","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":[0.98288584,0.00038776445,0.007483036,0.003082797,0.00002758755,0.0006650712,0.0003171989,0.000016270571,0.005134596],"genre_scores_gemma":[0.9892762,0.0008442997,0.0056117782,0.001387842,0.00002547938,0.0009851754,0.00022245999,0.000011404052,0.0016353041],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9790495,0.015141876,0.0013248363,0.0008955416,0.0010607527,0.0025274106],"domain_scores_gemma":[0.9388975,0.050039522,0.0025260923,0.00084253,0.0058697346,0.0018246978],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033749502,0.00040818314,0.00047165027,0.003484684,0.005322841,0.0038203204,0.0011236479,0.0021761432,0.0024917994],"category_scores_gemma":[0.048631765,0.0006145692,0.00039552862,0.0033050962,0.0029499778,0.0042750165,0.003918911,0.002173674,0.00041337046],"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":[0.000046817626,0.000063397674,0.009756694,0.00020277662,0.0000073962174,0.000567145,0.9763871,0.00018532366,0.0007540088,0.0019098988,0.0009250502,0.009194451],"study_design_scores_gemma":[0.0000056660692,0.00003985253,0.0027351219,0.00018185235,0.0000047421945,0.00013296789,0.99097323,0.00057773216,0.00030795808,0.0005810147,0.0044435677,0.000016289428],"about_ca_topic_score_codex":0.01044831,"about_ca_topic_score_gemma":0.013144466,"teacher_disagreement_score":0.033749502,"about_ca_system_score_codex":0.007881389,"about_ca_system_score_gemma":0.0076900744,"threshold_uncertainty_score":0.17848659},"labels":[],"label_agreement":null},{"id":"W4388192518","doi":"10.2196/49531","title":"Momentary Depressive Feeling Detection Using X (Formerly Twitter) Data: Contextual Language Approach","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Mental Health via Writing","field":"Psychology","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan","funders":"Mitacs","keywords":"Feeling; Lexicon; Transfer of learning; Psychology; Artificial intelligence; Context (archaeology); Computer science; Machine learning; Distress; Categorization; Binary classification; Convolutional neural network; Social psychology; Clinical psychology; Support vector machine","score_opus":0.12400983285868158,"score_gpt":0.4383199983535391,"score_spread":0.31431016549485746,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388192518","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":[0.8899679,0.0012927687,0.07722086,0.0011889288,0.0002838109,0.0003605676,0.01824418,0.003104846,0.008336135],"genre_scores_gemma":[0.92840767,0.00040116708,0.057911836,0.00020226135,0.00014256401,0.00019054407,0.01052445,0.00006634333,0.0021532066],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9997002,0.00010000662,0.000029874785,0.00008094629,0.000043796448,0.000045119094],"domain_scores_gemma":[0.99937785,0.0002686736,0.000102726364,0.00007654095,0.0001391572,0.000035047346],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00045343375,0.00052244216,0.00023679258,0.001019701,0.00028801573,0.0005185157,0.00031303932,0.0003641682,0.0012478372],"category_scores_gemma":[0.0018760659,0.00009514099,0.00028756933,0.0006496524,0.0001987849,0.00065653614,0.00079864584,0.00041804963,0.00080048887],"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":[0.002138807,0.0005949048,0.22232056,0.0015697178,0.00021833106,0.0023746272,0.0021675257,0.022674104,0.095202535,0.0059259036,0.03951565,0.6052974],"study_design_scores_gemma":[0.00010783328,0.000851078,0.2519019,0.00029400227,0.00026499663,0.0016482128,0.006330091,0.6225766,0.0501326,0.0098853735,0.055820096,0.00018726886],"about_ca_topic_score_codex":0.0050637904,"about_ca_topic_score_gemma":0.01015257,"teacher_disagreement_score":0.0050637904,"about_ca_system_score_codex":0.0002741079,"about_ca_system_score_gemma":0.0003848629,"threshold_uncertainty_score":0.010068655},"labels":[],"label_agreement":null},{"id":"W4388812715","doi":"10.2196/50442","title":"Assessment of ChatGPT-3.5's Knowledge in Oncology: Comparative Study with ASCO-SEP Benchmarks","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Preprint; Oncology; Clinical Oncology; Medicine; Internal medicine; Medical physics; Computer science; Cancer; World Wide Web","score_opus":0.21428723043483547,"score_gpt":0.5726386980995014,"score_spread":0.35835146766466597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388812715","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":[0.95071274,0.016692115,0.0059163417,0.0015244614,0.00018163634,0.0063076145,0.009427303,0.00018014293,0.009057577],"genre_scores_gemma":[0.9693182,0.0033072978,0.013036326,0.0006118163,0.00005936289,0.0060763983,0.007118854,0.000036239802,0.0004353667],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.95990294,0.019951448,0.008425189,0.0021709565,0.008795086,0.00075441704],"domain_scores_gemma":[0.7358677,0.19507061,0.02992661,0.005371629,0.02930154,0.004461977],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.051165134,0.0003628034,0.0017187013,0.008450512,0.00081820955,0.0015726365,0.00176916,0.0009969301,0.0015810446],"category_scores_gemma":[0.18451017,0.00024919593,0.0028670821,0.0053113843,0.0014545345,0.0025098154,0.004557998,0.0013061041,0.00026064203],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0031790482,0.0010305295,0.69499254,0.023135653,0.0030373149,0.00034421933,0.013773737,0.0023732951,0.0004964032,0.0010023408,0.007173849,0.249461],"study_design_scores_gemma":[0.0006196507,0.00413082,0.9442818,0.010169892,0.0025108713,0.00076724874,0.00916476,0.005793914,0.0011982387,0.0016295362,0.019550288,0.0001828948],"about_ca_topic_score_codex":0.004407173,"about_ca_topic_score_gemma":0.009291796,"teacher_disagreement_score":0.051165134,"about_ca_system_score_codex":0.0034431608,"about_ca_system_score_gemma":0.0053186268,"threshold_uncertainty_score":0.2705903},"labels":[],"label_agreement":null},{"id":"W4389858138","doi":"10.2196/51535","title":"Health Care Professionals’ and Parents’ Perspectives on the Use of AI for Pain Monitoring in the Neonatal Intensive Care Unit: Multisite Qualitative Study","year":2023,"lang":"en","type":"article","venue":"JMIR AI","topic":"Pediatric Pain Management Techniques","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Mount Sinai Hospital; York University; Children's Hospital of Eastern Ontario","funders":"Medical Research Council; Canadian Institutes of Health Research; University of Ottawa","keywords":"Neonatal intensive care unit; Qualitative research; Health professionals; Medicine; Nursing; Intensive care unit; Health care; Unit (ring theory); Psychology; Intensive care medicine; Psychiatry; Sociology; Political science","score_opus":0.10829707079050663,"score_gpt":0.46681857333475785,"score_spread":0.3585215025442512,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389858138","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":[0.99305564,0.0007703967,0.0012427331,0.0024234627,0.000054309654,0.00019697257,0.00013221099,0.000008810809,0.00211539],"genre_scores_gemma":[0.9963541,0.0008839721,0.00074162206,0.0009822826,0.000018512466,0.00033473334,0.0000451013,0.000009471595,0.00063011545],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9829092,0.013136223,0.0005582064,0.0006575692,0.0008826749,0.0018561987],"domain_scores_gemma":[0.94395757,0.04628695,0.0027609218,0.0006775205,0.0035124873,0.0028046886],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021269249,0.0005446681,0.0008487388,0.0017475629,0.00802768,0.0036489763,0.0015807532,0.0022537229,0.0028007806],"category_scores_gemma":[0.040940795,0.0009819925,0.00050945074,0.0017257516,0.0072349724,0.0037050538,0.004946737,0.0031984064,0.00025232983],"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":[0.000020590633,0.000020843363,0.0026261497,0.0001089977,0.0000036853614,0.00042209387,0.9948131,0.0000142695035,0.00017161421,0.00023335728,0.00019013272,0.0013751106],"study_design_scores_gemma":[0.0000032241105,0.000031577532,0.0010143261,0.00013193165,0.0000028164695,0.00011830158,0.9973839,0.00003381284,0.00007555101,0.000072466326,0.0011262002,0.0000059665163],"about_ca_topic_score_codex":0.014897796,"about_ca_topic_score_gemma":0.015909676,"teacher_disagreement_score":0.021269249,"about_ca_system_score_codex":0.006427378,"about_ca_system_score_gemma":0.0067499205,"threshold_uncertainty_score":0.11248386},"labels":[],"label_agreement":null},{"id":"W4396670539","doi":"10.2196/55957","title":"Toward Clinical Generative AI: Conceptual Framework","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Generative grammar; Computer science; Artificial intelligence","score_opus":0.35184315798997606,"score_gpt":0.5640545340335312,"score_spread":0.21221137604355517,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396670539","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0065503484,0.010593667,0.681112,0.077457786,0.0007615425,0.00083109934,0.0002839095,0.00036574315,0.22204393],"genre_scores_gemma":[0.4556413,0.009340191,0.50243974,0.010308153,0.0011254072,0.0027998616,0.00056675164,0.00021491146,0.017563725],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99158984,0.0060728765,0.00041362242,0.0006605566,0.0008444857,0.00041867295],"domain_scores_gemma":[0.9849273,0.010899993,0.0009493208,0.00095377007,0.0013544745,0.0009152434],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014317367,0.0012567329,0.00067044474,0.0046451516,0.0031868173,0.012994369,0.004849334,0.0068837386,0.007966433],"category_scores_gemma":[0.014807484,0.00075675186,0.0017212132,0.0026330047,0.027238106,0.010317343,0.005643415,0.0061158095,0.0015724059],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000031899747,0.000008224515,0.000074342024,0.000043672968,0.0000029093121,0.00004578625,0.0005977291,0.0008403438,0.000021338608,0.99550974,0.0005153963,0.002337378],"study_design_scores_gemma":[0.000011167397,0.000011709618,0.00006244513,0.0001716867,0.0000063704656,0.000101446545,0.00060167344,0.0053913994,0.0000625223,0.9724142,0.02115431,0.000011121866],"about_ca_topic_score_codex":0.006165125,"about_ca_topic_score_gemma":0.0034915295,"teacher_disagreement_score":0.014317367,"about_ca_system_score_codex":0.0084894085,"about_ca_system_score_gemma":0.011036401,"threshold_uncertainty_score":0.07571834},"labels":[],"label_agreement":null},{"id":"W4399207674","doi":"10.2196/56537","title":"Evaluating Literature Reviews Conducted by Humans Versus ChatGPT: Comparative Study","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Public Health; University of Ottawa; Ottawa Hospital; Queen's University; Canadian Medical Protective Association","funders":"","keywords":"Preprint; Computer science; World Wide Web","score_opus":0.5298481920861103,"score_gpt":0.6187488228927178,"score_spread":0.08890063080660748,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399207674","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.68095803,0.1992193,0.027793074,0.004303267,0.0016356672,0.059980907,0.008723745,0.0006248186,0.016761184],"genre_scores_gemma":[0.83063495,0.045884877,0.06306921,0.0023887504,0.0008388978,0.05243974,0.0032899284,0.00021603583,0.0012376464],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.57647187,0.26724324,0.09169594,0.011599758,0.051015045,0.0019741766],"domain_scores_gemma":[0.14501864,0.65842175,0.09263445,0.013682841,0.08644775,0.0037945327],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27173975,0.001368196,0.0051069097,0.030562488,0.001664261,0.0076553915,0.002672063,0.0021903599,0.004208188],"category_scores_gemma":[0.66000044,0.0011814327,0.0043252115,0.022528399,0.0028019487,0.008530358,0.006139343,0.0010178466,0.00096938567],"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":[0.014592042,0.0018492152,0.08645887,0.31041852,0.015836002,0.001670121,0.1063084,0.0013189971,0.003656352,0.0021854802,0.00772696,0.44797903],"study_design_scores_gemma":[0.010487005,0.036091886,0.36331782,0.28345135,0.046107862,0.006859279,0.13988212,0.007147506,0.010665665,0.006508641,0.08799723,0.001483629],"about_ca_topic_score_codex":0.0015299895,"about_ca_topic_score_gemma":0.0044088615,"teacher_disagreement_score":0.7282603,"about_ca_system_score_codex":0.008371261,"about_ca_system_score_gemma":0.010806108,"threshold_uncertainty_score":0.89807457},"labels":[],"label_agreement":null},{"id":"W4401910075","doi":"10.2196/57983","title":"Exploring Machine Learning Applications in Pediatric Asthma Management: Scoping Review","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Asthma and respiratory diseases","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"","keywords":"CINAHL; Logistic regression; Machine learning; Asthma; Receiver operating characteristic; Medicine; Artificial intelligence; Random forest; MEDLINE; Cochrane Library; Support vector machine; Predictive modelling; Sample size determination; Computer science; Meta-analysis; Statistics; Internal medicine; Psychological intervention","score_opus":0.044860687219786914,"score_gpt":0.34476741181105014,"score_spread":0.2999067245912632,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401910075","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00022683835,0.9983132,0.00023298342,0.00044044587,0.00011352962,0.00013441946,0.00010210521,0.000008186848,0.00042829697],"genre_scores_gemma":[0.0021722543,0.99638855,0.000612539,0.00028071756,0.000108506334,0.0002530419,0.0000965962,0.000004336786,0.00008345215],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.99243915,0.002604747,0.0029224716,0.0004394555,0.0013840357,0.00021011556],"domain_scores_gemma":[0.91836184,0.07098677,0.0060950825,0.0006211312,0.0036502325,0.00028509556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013175174,0.0015706656,0.004293295,0.017674325,0.0007788846,0.004159782,0.0020569183,0.0031917032,0.005711391],"category_scores_gemma":[0.069930814,0.0010445436,0.0054918625,0.018325992,0.0010567356,0.0032137746,0.002043281,0.0018444037,0.0006780767],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009477016,0.000029100664,0.00062647555,0.7596857,0.0019291221,0.0001241964,0.00032043905,0.0004870861,0.00013794634,0.0011379432,0.0051456825,0.23028153],"study_design_scores_gemma":[0.000032181608,0.000047359063,0.00082915777,0.96083707,0.004580578,0.00017094858,0.00015098567,0.000173029,0.000098864344,0.0007250015,0.03233979,0.000015011236],"about_ca_topic_score_codex":0.006496639,"about_ca_topic_score_gemma":0.010937973,"teacher_disagreement_score":0.017674325,"about_ca_system_score_codex":0.0037240074,"about_ca_system_score_gemma":0.01465968,"threshold_uncertainty_score":0.06967783},"labels":[],"label_agreement":null},{"id":"W4402340921","doi":"10.2196/57673","title":"The Utility and Implications of Ambient Scribes in Primary Care","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vector Institute; Dalhousie University; University of Alberta; McMaster University","funders":"","keywords":"Preprint; Primary care; Environmental science; Computer science; World Wide Web; Medicine","score_opus":0.06706918653510571,"score_gpt":0.4581921410834211,"score_spread":0.39112295454831536,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402340921","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":[0.61232054,0.008824556,0.13992587,0.09192103,0.0007467412,0.00065954373,0.00085148466,0.0011970237,0.14355314],"genre_scores_gemma":[0.97855616,0.0012836119,0.01573597,0.0020616706,0.00015977923,0.00015732473,0.00006761228,0.00005773012,0.0019200874],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9763016,0.020202499,0.000562767,0.00070277747,0.0016262864,0.00060418306],"domain_scores_gemma":[0.92674804,0.06287652,0.0024657259,0.002634213,0.0033875613,0.0018880096],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014523583,0.0002775992,0.0003385076,0.0009110751,0.0026878128,0.005254934,0.0009320932,0.00129093,0.00803459],"category_scores_gemma":[0.05652786,0.00027529462,0.0007456533,0.0010497441,0.0034146185,0.003210851,0.0038399398,0.001678984,0.00091157743],"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":[0.0012186489,0.0008670737,0.11169467,0.0021281077,0.00016439792,0.002233705,0.18179339,0.005388612,0.0039281077,0.087982684,0.02399175,0.5786088],"study_design_scores_gemma":[0.00041742538,0.0040216153,0.16861418,0.0049906448,0.0005285296,0.0049065803,0.28811574,0.044077545,0.0063066874,0.23309547,0.24435934,0.0005662921],"about_ca_topic_score_codex":0.005398442,"about_ca_topic_score_gemma":0.006529681,"teacher_disagreement_score":0.014523583,"about_ca_system_score_codex":0.0030020871,"about_ca_system_score_gemma":0.002625872,"threshold_uncertainty_score":0.07680893},"labels":[],"label_agreement":null},{"id":"W4405934475","doi":"10.2196/55277","title":"Creation of Scientific Response Documents for Addressing Product Medical Information Inquiries: Mixed Method Approach Using Artificial Intelligence","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Pfizer (Canada)","funders":"","keywords":"Preprint; Product (mathematics); Data science; Pharmaceutical industry; Computer science; Business; World Wide Web; Medicine","score_opus":0.08223819874697301,"score_gpt":0.4333640955664769,"score_spread":0.3511258968195039,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405934475","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":[0.08052166,0.0020545744,0.8090539,0.0020474293,0.00025750103,0.09260872,0.002617089,0.0018184684,0.009020653],"genre_scores_gemma":[0.083686665,0.00040998362,0.84953916,0.00072143425,0.0000734612,0.06303244,0.00079674105,0.00017731631,0.0015628745],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7187058,0.24257779,0.015394613,0.010063341,0.012079695,0.0011788092],"domain_scores_gemma":[0.34273317,0.57273257,0.027236553,0.02234714,0.033132076,0.0018185723],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2037963,0.002600614,0.0024800661,0.013011793,0.0040618656,0.009418691,0.0065068863,0.003295638,0.011553617],"category_scores_gemma":[0.36346903,0.0025207873,0.0035614937,0.009136298,0.002697691,0.006212263,0.008782143,0.0034140314,0.0030232512],"study_design_candidate":"simulation_or_modeling","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":[0.0042920727,0.0046049673,0.020907091,0.018744105,0.0017144034,0.0006420361,0.07038901,0.00584955,0.006216813,0.020977357,0.007059176,0.83860344],"study_design_scores_gemma":[0.010665579,0.0158632,0.06884232,0.017277695,0.0063753286,0.002056511,0.12332664,0.35651654,0.062992774,0.16171768,0.17201495,0.0023508254],"about_ca_topic_score_codex":0.0046310984,"about_ca_topic_score_gemma":0.007775874,"teacher_disagreement_score":0.2037963,"about_ca_system_score_codex":0.0069043287,"about_ca_system_score_gemma":0.012208075,"threshold_uncertainty_score":0.98186094},"labels":[],"label_agreement":null},{"id":"W4407274801","doi":"10.2196/64279","title":"Deep Learning Models to Predict Diagnostic and Billing Codes Following Visits to a Family Medicine Practice: Development and Validation Study","year":2025,"lang":"en","type":"article","venue":"JMIR AI","topic":"Electronic Health Records Systems","field":"Health Professions","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":"Queen's University","funders":"","keywords":"Preprint; Medicine; Artificial intelligence; Computer science; World Wide Web","score_opus":0.0838726322423534,"score_gpt":0.46960321165286545,"score_spread":0.38573057941051203,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407274801","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":[0.9539824,0.0009250207,0.040698566,0.00050122617,0.00006733549,0.00038630038,0.0011059861,0.0006118219,0.0017212941],"genre_scores_gemma":[0.95607394,0.00034631006,0.039435487,0.00015560756,0.000020919533,0.0002733674,0.0026620403,0.000033817665,0.0009985126],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9982956,0.0007335492,0.00016465042,0.00032259192,0.00032298872,0.00016062616],"domain_scores_gemma":[0.986822,0.008398366,0.00067168387,0.000737966,0.0030733047,0.0002966529],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0073415264,0.0012698881,0.0006891415,0.0012621924,0.00036078537,0.0008629872,0.0013349658,0.0010986091,0.0011241843],"category_scores_gemma":[0.015188421,0.00053228106,0.0009801708,0.0008611541,0.0003922311,0.0009404814,0.0011611965,0.0018054078,0.00037220333],"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":[0.00078552234,0.0030210465,0.17125253,0.00023893613,0.0007299521,0.00024563007,0.00022612927,0.60106885,0.0017788209,0.00077245134,0.00499838,0.21488176],"study_design_scores_gemma":[0.000055255245,0.00030424516,0.0072639827,0.000048174996,0.00006203409,0.00004042298,0.00003856746,0.9902457,0.0012716228,0.0002987395,0.0003594653,0.00001184497],"about_ca_topic_score_codex":0.023169292,"about_ca_topic_score_gemma":0.013695222,"teacher_disagreement_score":0.023169292,"about_ca_system_score_codex":0.001890494,"about_ca_system_score_gemma":0.002882798,"threshold_uncertainty_score":0.046068847},"labels":[],"label_agreement":null},{"id":"W4409430890","doi":"10.2196/65481","title":"AI-Powered Drug Classification and Indication Mapping for Pharmacoepidemiologic Studies: Prompt Development and Validation","year":2025,"lang":"en","type":"article","venue":"JMIR AI","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Preprint; Computer science; World Wide Web","score_opus":0.09121621506466121,"score_gpt":0.4358894887411843,"score_spread":0.3446732736765231,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409430890","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09485007,0.0012411048,0.86387277,0.005563004,0.0003519966,0.006888592,0.007176157,0.01578036,0.004275937],"genre_scores_gemma":[0.24866062,0.00057846354,0.7385209,0.0014108641,0.00015404576,0.0041651526,0.0056460914,0.00020657881,0.0006572948],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9719478,0.02064604,0.0021471842,0.0017793484,0.003087459,0.00039221073],"domain_scores_gemma":[0.7334935,0.20566376,0.010340305,0.019522203,0.028468536,0.0025116603],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0960719,0.0014093602,0.0012100005,0.003613746,0.0008371374,0.0031776556,0.0017691388,0.0015051307,0.0047613983],"category_scores_gemma":[0.25102592,0.00061051233,0.0012539314,0.0021198227,0.0014165309,0.004098518,0.0035331522,0.0026338915,0.0025482895],"study_design_candidate":"simulation_or_modeling","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":[0.0033914978,0.0017124448,0.11388418,0.0040972116,0.00041266432,0.00039553223,0.002224835,0.044098604,0.011763178,0.020069242,0.029250149,0.76870036],"study_design_scores_gemma":[0.0018047934,0.004066752,0.043964725,0.002047427,0.00055436615,0.0008000725,0.0026672203,0.7382117,0.043959983,0.08868788,0.072996035,0.00023904622],"about_ca_topic_score_codex":0.0026785973,"about_ca_topic_score_gemma":0.002358551,"teacher_disagreement_score":0.0960719,"about_ca_system_score_codex":0.0022266028,"about_ca_system_score_gemma":0.007919908,"threshold_uncertainty_score":0.50808287},"labels":[],"label_agreement":null},{"id":"W4410629063","doi":"10.2196/60885","title":"Predicting Episodes of Hypovigilance in Intensive Care Units Using Routine Physiological Parameters and Artificial Intelligence: Derivation Study","year":2025,"lang":"en","type":"article","venue":"JMIR AI","topic":"Intensive Care Unit Cognitive Disorders","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":"Université Laval; Thales (Canada); Centre intégré de santé et de services sociaux de Chaudière-Appalaches","funders":"","keywords":"Preprint; Artificial intelligence; Computer science; Psychology","score_opus":0.06642024970766383,"score_gpt":0.3681563449388794,"score_spread":0.3017360952312156,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410629063","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":[0.99802256,0.000027548698,0.0011471934,0.00003123249,0.0000026399237,0.00006652994,0.0005384143,0.0000126664145,0.00015115777],"genre_scores_gemma":[0.99451584,0.00004456351,0.0021194168,0.000044172204,0.0000043356254,0.000071210925,0.0030207206,0.000007343858,0.00017233186],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99812084,0.0010522262,0.00013531838,0.0002895408,0.00026713894,0.00013491095],"domain_scores_gemma":[0.9840715,0.008534453,0.0015855712,0.0020171334,0.0030209615,0.0007703999],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005150118,0.000609677,0.00045551732,0.0007984412,0.0005398495,0.0008127704,0.0010138568,0.0004789588,0.0010747336],"category_scores_gemma":[0.02140229,0.0003223243,0.0009124259,0.0010359362,0.00046400973,0.00039292852,0.0009950077,0.00078181666,0.00037643398],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00052811136,0.00043397758,0.9878011,0.000022545992,0.00022542842,0.0001267851,0.00016478372,0.0031224373,0.00017110043,0.00008827457,0.0004557789,0.006859781],"study_design_scores_gemma":[0.00013371793,0.0007603227,0.917669,0.000022370281,0.000116043266,0.00030789353,0.0003178406,0.079585165,0.00031598928,0.00019804791,0.0005394967,0.000034025936],"about_ca_topic_score_codex":0.11747619,"about_ca_topic_score_gemma":0.08333305,"teacher_disagreement_score":0.11747619,"about_ca_system_score_codex":0.0022504164,"about_ca_system_score_gemma":0.002954295,"threshold_uncertainty_score":0.23358482},"labels":[],"label_agreement":null},{"id":"W4415110146","doi":"10.2196/63767","title":"Deep Learning Models to Screen Electronic Health Records for Breast and Colorectal Cancer Progression: Performance Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR AI","topic":"Machine Learning in Healthcare","field":"Computer Science","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":"University of Manitoba; Research Institute in Oncology and Hematology; University of Waterloo; CancerCare Manitoba","funders":"","keywords":"Deep learning; Health records; Chart; Electronic health record; Colorectal cancer; Breast cancer","score_opus":0.029887741233579457,"score_gpt":0.39742023292361667,"score_spread":0.3675324916900372,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415110146","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":[0.9689915,0.0042949147,0.01743037,0.0015730682,0.00022539457,0.00024895713,0.0032732666,0.001250314,0.0027122076],"genre_scores_gemma":[0.97617733,0.0007761124,0.01394452,0.00027773925,0.00006600157,0.00012855197,0.0071726693,0.000033353368,0.0014238161],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9976743,0.0010599503,0.00024114203,0.00040015232,0.00035805884,0.0002662901],"domain_scores_gemma":[0.9922462,0.004411339,0.0005953178,0.00058687106,0.001745552,0.00041475485],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0062376577,0.0017788649,0.0010080443,0.0013563514,0.0003954811,0.0009725347,0.0012374028,0.0014241678,0.0013448291],"category_scores_gemma":[0.0122296335,0.00037815617,0.0010343223,0.0009903267,0.00041422847,0.0011470453,0.0012098184,0.0016688132,0.0006255285],"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":[0.004996336,0.004619701,0.21046802,0.0006847588,0.0017860531,0.00030926036,0.00018577595,0.43086007,0.002446053,0.0008343996,0.016987741,0.32582188],"study_design_scores_gemma":[0.00013883667,0.0009960731,0.010457184,0.000059827307,0.00017948885,0.00009219719,0.000067498855,0.9854295,0.0014497056,0.00045014624,0.00065550127,0.000023941184],"about_ca_topic_score_codex":0.03605272,"about_ca_topic_score_gemma":0.025316441,"teacher_disagreement_score":0.03605272,"about_ca_system_score_codex":0.0020603403,"about_ca_system_score_gemma":0.00249195,"threshold_uncertainty_score":0.07168573},"labels":[],"label_agreement":null},{"id":"W4415749046","doi":"10.2196/75932","title":"A Multiagent Summarization and Auto-Evaluation Framework for Medical Text: Development and Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR AI","topic":"Topic Modeling","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Cleveland Clinic","keywords":"Automatic summarization; Adaptability; Key (lock); Dependency (UML); Scalability; Salient","score_opus":0.04394454882875393,"score_gpt":0.3929848069328207,"score_spread":0.34904025810406675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415749046","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.22077188,0.003995159,0.7247828,0.0010925147,0.0003285797,0.005189935,0.0016043342,0.03416481,0.008069908],"genre_scores_gemma":[0.36929476,0.0006552628,0.6213665,0.0002773967,0.00009121255,0.0015300399,0.0030557322,0.0004395978,0.0032894928],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9953988,0.0025636435,0.000334377,0.0004788035,0.0010855275,0.00013884684],"domain_scores_gemma":[0.9894757,0.005917603,0.0005641454,0.00063704257,0.0029468855,0.00045860917],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0098526515,0.0013177802,0.00083049043,0.0014999991,0.0005052766,0.0014523413,0.0020116013,0.0012378196,0.0038287952],"category_scores_gemma":[0.016052479,0.00037240563,0.00081901456,0.0007320606,0.00044781997,0.0019537618,0.001183569,0.0012961321,0.0011765368],"study_design_candidate":"simulation_or_modeling","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":[0.0029879722,0.003011505,0.008228617,0.0024809223,0.00085282477,0.0005432708,0.0015118858,0.121773235,0.028529124,0.00462729,0.02035703,0.8050963],"study_design_scores_gemma":[0.00038120395,0.0020016057,0.0041467985,0.00013213982,0.00026133435,0.00020121627,0.0002729616,0.9674982,0.0137625765,0.0014505046,0.009821441,0.00007002616],"about_ca_topic_score_codex":0.009059133,"about_ca_topic_score_gemma":0.008049098,"teacher_disagreement_score":0.99014735,"about_ca_system_score_codex":0.0019240975,"about_ca_system_score_gemma":0.0018427969,"threshold_uncertainty_score":0.05210638},"labels":[],"label_agreement":null}]}