{"meta":{"query_hash":"317a44b554f5","filters":{"venue":"Intelligent Medicine"},"cohort_total":8,"direct_labels_cover":0,"predictions_cover":8,"exported":8,"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/317a44b554f5","api":"https://metacan.xera.ac/api/v1/cohort?venue=Intelligent+Medicine"},"results":[{"id":"W4410351965","doi":"10.1016/j.imed.2025.04.002","title":"Advancement in blood pressure abnormality detection and interpretation using large language models","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Blood Pressure and Hypertension Studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Abnormality; Interpretation (philosophy); Natural language processing; Computer science; Linguistics; Psychology; Philosophy; Social psychology; Programming language","score_opus":0.02810964764701008,"score_gpt":0.3267108763144974,"score_spread":0.29860122866748734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410351965","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.72086567,0.17409973,0.09487202,0.002978486,0.00066153385,0.0011473674,0.000008822438,0.000112949085,0.005253445],"genre_scores_gemma":[0.99540305,0.001564881,0.00023858584,0.0021635545,0.000085153384,0.000018190618,0.0000063410175,0.000008320382,0.00051189837],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99896187,0.000049023085,0.00035811163,0.00025426364,0.000189113,0.00018764702],"domain_scores_gemma":[0.99950236,0.00008060131,0.0000599289,0.0001903267,0.00011142962,0.000055374512],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00035638246,0.00013612994,0.00041997136,0.00023591451,0.00005766212,0.000005592718,0.000039052975,0.00007335995,0.000046060115],"category_scores_gemma":[0.00010575314,0.00010617019,0.00003215819,0.00022041335,0.00006497368,0.00009085296,0.0000689626,0.00019012784,0.0000013114777],"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":[0.006707494,0.0041033323,0.08994627,0.007734116,0.009736826,0.0013548982,0.09573738,0.0037298456,0.28276834,0.0079005575,0.0024161995,0.48786476],"study_design_scores_gemma":[0.021808889,0.0024359112,0.039173394,0.013278042,0.02656569,0.0006520937,0.037246443,0.6044498,0.19280814,0.005632977,0.05477305,0.0011755738],"about_ca_topic_score_codex":0.00034283553,"about_ca_topic_score_gemma":0.00020711131,"teacher_disagreement_score":0.6007199,"about_ca_system_score_codex":0.000012374525,"about_ca_system_score_gemma":0.000021587502,"threshold_uncertainty_score":0.43294948},"labels":[],"label_agreement":null},{"id":"W4413015157","doi":"10.1016/j.imed.2025.07.001","title":"Artificial intelligence-based framework for Alzheimer’s disease diagnosis via video vision transformer","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":4,"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":"National Institute of General Medical Sciences; National Institutes of Health; Alzheimer's Disease Neuroimaging Initiative; National Heart, Lung, and Blood Institute; Foundation for the National Institutes of Health","keywords":"Magnetic resonance imaging; Artificial intelligence; Transformer; Computer science; Computer vision; Medicine; Radiology; Engineering; Electrical engineering","score_opus":0.06339457897162563,"score_gpt":0.41276219816813864,"score_spread":0.349367619196513,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413015157","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010925397,0.0039210073,0.9260588,0.05289047,0.0011261749,0.0031127124,0.000025298066,0.00013150166,0.0018086281],"genre_scores_gemma":[0.98659164,0.00075629324,0.0028477265,0.0070358263,0.00072228163,0.0013045497,0.00020034998,0.00004617338,0.00049514446],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99684554,0.00008893263,0.0008834361,0.00067092525,0.00082605885,0.00068508036],"domain_scores_gemma":[0.99699885,0.0013081046,0.00009334795,0.00049767544,0.0005158984,0.00058615475],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008379103,0.00037560143,0.0005896161,0.0006589408,0.000213992,0.000035816847,0.00025768223,0.00017455172,0.0033992496],"category_scores_gemma":[0.0012594647,0.0002846338,0.0003482444,0.00082999805,0.00045829083,0.00008892675,0.000031749507,0.00042024037,0.0001531327],"study_design_candidate":"design_other","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":[0.0049787117,0.0017663261,0.025564916,0.0005601952,0.0005089769,0.00008847338,0.00026913543,0.00002674488,0.0008507419,0.01736261,0.00821146,0.9398117],"study_design_scores_gemma":[0.0024024523,0.008939762,0.03101238,0.009668747,0.005882594,0.000010064161,0.0014030117,0.017614007,0.46381631,0.28954813,0.16866292,0.001039587],"about_ca_topic_score_codex":0.00004873681,"about_ca_topic_score_gemma":0.000013384552,"teacher_disagreement_score":0.9756663,"about_ca_system_score_codex":0.0001264211,"about_ca_system_score_gemma":0.00033818244,"threshold_uncertainty_score":0.9999606},"labels":[],"label_agreement":null},{"id":"W4413469574","doi":"10.1016/j.imed.2025.03.003","title":"A novel stacking-based classifier for identifying antifreeze protein using latent semantic analysis","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Machine Learning in Bioinformatics","field":"Biochemistry, Genetics and Molecular Biology","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":"University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Stacking; Latent semantic analysis; Computer science; Classifier (UML); Antifreeze; Probabilistic latent semantic analysis; Artificial intelligence; Identification (biology); Antifreeze protein; Pattern recognition (psychology); Computational biology; Machine learning; Chemistry; Biology; Biochemistry","score_opus":0.054640104164757244,"score_gpt":0.36262393535679227,"score_spread":0.30798383119203504,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413469574","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19122878,0.0003008422,0.80675864,0.0005693336,0.0002358281,0.0005425927,0.000011193902,0.000023030274,0.00032973412],"genre_scores_gemma":[0.9806766,0.00001511534,0.016727518,0.000769487,0.00016350279,0.00004053151,0.0002653205,0.000022078277,0.0013198587],"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99853975,0.000037264934,0.000555167,0.00033448962,0.00022828401,0.00030502933],"domain_scores_gemma":[0.99896467,0.00004687871,0.00021518675,0.0004760164,0.00022540746,0.00007185126],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005913666,0.00021801324,0.00034045576,0.00038440552,0.00012410227,0.000032542746,0.00026429477,0.00014963835,0.000060582563],"category_scores_gemma":[0.0005920931,0.00018095548,0.00020569548,0.00056936644,0.00012264418,0.00000425769,0.00009163072,0.00014580629,0.0000032587593],"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":[0.00017919831,0.00013728163,0.011998682,0.00064802566,0.0013307513,0.0000028405345,0.0001605024,0.024898984,0.95597774,0.0005010941,0.0009831828,0.0031817246],"study_design_scores_gemma":[0.0014786473,0.000366786,0.002127215,0.00064641895,0.0013603892,0.0000049230525,0.00020999859,0.6805404,0.2959221,0.00012238666,0.016827527,0.00039323763],"about_ca_topic_score_codex":0.00011002142,"about_ca_topic_score_gemma":0.000083093226,"teacher_disagreement_score":0.79003114,"about_ca_system_score_codex":0.00004799591,"about_ca_system_score_gemma":0.00010776944,"threshold_uncertainty_score":0.73791504},"labels":[],"label_agreement":null},{"id":"W4414420904","doi":"10.1016/j.imed.2025.05.010","title":"Enhancing echocardiographic artificial intelligence systems with large language models: QHAutoEF and the role of DeepSeek in future development","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Development (topology); Expert system; Field (mathematics); Key (lock); Applications of artificial intelligence","score_opus":0.015030931858124276,"score_gpt":0.2879344423799848,"score_spread":0.2729035105218605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414420904","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.565739,0.116192386,0.2817399,0.028598871,0.0012682752,0.004128036,0.000010193303,0.00023498537,0.0020883896],"genre_scores_gemma":[0.9957106,0.0013011573,0.0005285424,0.0020328413,0.00020618062,0.000116156836,0.000012216565,0.000023386896,0.000068900736],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9977042,0.00013757148,0.000864559,0.0004173105,0.00051609485,0.00036025222],"domain_scores_gemma":[0.99846435,0.00061628554,0.00016649222,0.00045980135,0.0001888253,0.00010424832],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016537583,0.00026817562,0.00078543025,0.0007000084,0.000084907515,0.000017385697,0.00019258836,0.00012495914,0.000021016072],"category_scores_gemma":[0.00026720017,0.00015839482,0.0000699363,0.0011583699,0.00036120787,0.00005629416,0.00009356215,0.00040118126,0.0000034110749],"study_design_candidate":"qualitative","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":[0.008655189,0.0022263655,0.04616064,0.006440003,0.00282851,0.0006976337,0.20364763,0.014123686,0.010105068,0.25012708,0.00085614994,0.45413205],"study_design_scores_gemma":[0.009549235,0.0016958638,0.008201187,0.042131037,0.002324842,0.00029030972,0.39425984,0.21502084,0.24240817,0.017081102,0.06548477,0.0015527926],"about_ca_topic_score_codex":0.0009191275,"about_ca_topic_score_gemma":0.0009744585,"teacher_disagreement_score":0.45257926,"about_ca_system_score_codex":0.00014316909,"about_ca_system_score_gemma":0.00027097447,"threshold_uncertainty_score":0.6459153},"labels":[],"label_agreement":null},{"id":"W7106482238","doi":"10.1016/j.imed.2025.07.005","title":"One-stop automated diagnostic system for active sacroiliitis in three-dimensional magnetic resonance images using artificial intelligent models: a retrospective study","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Spondyloarthritis Studies and Treatments","field":"Medicine","cited_by":0,"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":"National Key Research and Development Program of China; Chinese People’s Liberation Army; Natural Science Foundation of Beijing Municipality","keywords":"Sacroiliitis; Magnetic resonance imaging; Intraclass correlation; Ankylosing spondylitis; Retrospective cohort study; Segmentation; Medical imaging; Quadrant (abdomen)","score_opus":0.04803214735726182,"score_gpt":0.32841779112571107,"score_spread":0.28038564376844927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7106482238","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9638068,0.02113039,0.006019569,0.00075847673,0.0012571346,0.005925078,0.00012747262,0.00035418523,0.00062085944],"genre_scores_gemma":[0.9975084,0.00031807827,0.0008505273,0.00017211089,0.00023108158,0.00065550563,0.00004624079,0.000052031563,0.00016600246],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9966312,0.00010301684,0.001075625,0.00090876553,0.00063630386,0.0006450786],"domain_scores_gemma":[0.99792415,0.00068917405,0.00017955618,0.0004956991,0.0005472288,0.00016418133],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0004386232,0.0004656295,0.001243259,0.0005673457,0.00024550533,0.000019885947,0.00016101486,0.00014693552,0.0000682765],"category_scores_gemma":[0.0010119089,0.00040970117,0.00015866841,0.0008924589,0.0002915189,0.00009399005,0.00014131416,0.00032994608,0.00001582031],"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":[0.02608295,0.027657153,0.261389,0.0027895577,0.0048297704,0.0076742456,0.016637651,0.005385269,0.006765204,0.032698113,0.0026638792,0.6054272],"study_design_scores_gemma":[0.033608086,0.026303347,0.5189238,0.03611842,0.0052367505,0.00037574593,0.037324555,0.29292294,0.025525866,0.020851482,0.0008129027,0.0019960923],"about_ca_topic_score_codex":0.0019248058,"about_ca_topic_score_gemma":0.0014847743,"teacher_disagreement_score":0.6034311,"about_ca_system_score_codex":0.0015777608,"about_ca_system_score_gemma":0.000196714,"threshold_uncertainty_score":0.9998355},"labels":[],"label_agreement":null},{"id":"W7117365168","doi":"10.1016/j.imed.2025.12.006","title":"Evaluation of artificial intelligence-based tool Covidence in literature screening for guideline updates: A prospective study","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Hamilton Health Sciences; Population Health Research Institute; McMaster University; University of British Columbia; Impact","funders":"Hamilton Health Sciences Foundation","keywords":"Guideline; Randomized controlled trial; Workload; Breast cancer; MEDLINE; Lung cancer screening; Stage (stratigraphy); Clinical Practice; Evidence-based medicine","score_opus":0.36637380526354874,"score_gpt":0.5644291767119313,"score_spread":0.19805537144838253,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117365168","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19407648,0.0011720848,0.7781123,0.016216658,0.0006610186,0.0094565,0.000025623101,0.00005627094,0.00022304586],"genre_scores_gemma":[0.984064,0.00003940028,0.012684625,0.0020978313,0.00032535833,0.0004758534,0.00023849429,0.0000210404,0.000053415508],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99420357,0.0003881651,0.0031704823,0.0005880389,0.0013711831,0.0002785765],"domain_scores_gemma":[0.9920776,0.0023187029,0.00053052406,0.00051120465,0.0044841967,0.000077800694],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.014628338,0.00025513093,0.00068148714,0.0007596163,0.000056913483,0.000019771285,0.00018838071,0.00012664331,0.00036617907],"category_scores_gemma":[0.0480476,0.00020657018,0.000121055986,0.0015208387,0.00014569784,0.00016315443,0.000052553034,0.0003703864,0.000005145141],"study_design_candidate":"design_other","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":[0.0066039786,0.002995394,0.0755818,0.00039025585,0.00044496925,0.000028779614,0.003604528,0.007335399,0.0023904417,0.002087172,0.0034565546,0.89508075],"study_design_scores_gemma":[0.0132567575,0.010210098,0.03145885,0.008452484,0.0055411975,0.000015116651,0.03617596,0.815287,0.03308838,0.043312386,0.0025984072,0.0006033724],"about_ca_topic_score_codex":0.000325865,"about_ca_topic_score_gemma":0.000525796,"teacher_disagreement_score":0.89447737,"about_ca_system_score_codex":0.00038658653,"about_ca_system_score_gemma":0.0008500788,"threshold_uncertainty_score":0.95997113},"labels":[],"label_agreement":null},{"id":"W7117367844","doi":"10.1016/j.imed.2025.12.008","title":"The performance of Covidence: An artificial intelligence-based tool for title and abstract screening in a breast cancer evidence-based clinical practice guideline","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","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":"Hamilton Health Sciences; Population Health Research Institute; McMaster University; Impact","funders":"Hamilton Health Sciences Foundation","keywords":"Workload; Clinical Practice; Guideline; Clinical trial; Breast cancer; Stage (stratigraphy); Identification (biology)","score_opus":0.361299736082622,"score_gpt":0.5694698143195533,"score_spread":0.20817007823693134,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117367844","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3039172,0.006091665,0.32468235,0.35795844,0.0025293413,0.0038429494,0.000058006957,0.00009807799,0.0008219838],"genre_scores_gemma":[0.9699946,0.0028305936,0.013964089,0.012301708,0.0006435313,0.000084851476,0.000036828475,0.000019694198,0.00012411708],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99695015,0.00012008222,0.002041653,0.00030460823,0.00038012792,0.00020336988],"domain_scores_gemma":[0.98938286,0.008951554,0.00043809775,0.00033424006,0.00080590014,0.00008737372],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0060548726,0.00014081296,0.0003486572,0.00015093881,0.00007523613,0.000017484264,0.0001512207,0.00010113918,0.0003911078],"category_scores_gemma":[0.016096542,0.000095463896,0.0000703209,0.00033210355,0.00030983955,0.00019827446,0.00002805274,0.00032210044,0.000008270874],"study_design_candidate":"design_other","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":[0.0047288723,0.0002651539,0.030174853,0.00022351916,0.00006214444,0.000006935985,0.000053118372,0.0006797067,0.00045108498,0.0005774733,0.009380101,0.95339704],"study_design_scores_gemma":[0.0037522868,0.0041823382,0.08483238,0.011733177,0.0015824108,0.000044028286,0.003927205,0.6526166,0.010698559,0.0012956908,0.2247519,0.00058338384],"about_ca_topic_score_codex":0.0004666362,"about_ca_topic_score_gemma":0.000107714106,"teacher_disagreement_score":0.9528136,"about_ca_system_score_codex":0.00006395066,"about_ca_system_score_gemma":0.000683297,"threshold_uncertainty_score":0.9921913},"labels":[],"label_agreement":null},{"id":"W7117408127","doi":"10.1016/j.imed.2025.11.003","title":"From radiology findings to artificial intelligence-powered impressions: A retrospective study on the comparative performance of recent large language models","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","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":false,"ca_institutions":"Hamilton Health Sciences; Juravinski Hospital; Population Health Research Institute; McMaster University","funders":"","keywords":"Ranking (information retrieval); Metric (unit); Test (biology); Retrospective cohort study","score_opus":0.22780496597945366,"score_gpt":0.4691366938755107,"score_spread":0.24133172789605706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117408127","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97094214,0.0005140388,0.007157937,0.014599387,0.001419455,0.0023518666,0.000024799634,0.000051834522,0.0029385232],"genre_scores_gemma":[0.99623495,0.00040030133,0.00010213893,0.0022521152,0.00047679033,0.0001946701,0.000041644176,0.00001656964,0.00028082714],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9970758,0.00024173851,0.0011744279,0.00056666567,0.00050975627,0.00043158853],"domain_scores_gemma":[0.9974094,0.0008787311,0.0001905825,0.000756762,0.00057950756,0.00018499183],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0011318139,0.000292076,0.0007896841,0.0004323026,0.00022129944,0.000011460242,0.0003448054,0.00014636174,0.0013226423],"category_scores_gemma":[0.001274916,0.00018205939,0.00008740757,0.0010751113,0.00023810264,0.00006217109,0.000097156844,0.0005885749,0.00012157558],"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":[0.010607959,0.006805842,0.053950414,0.00018007938,0.0012547802,0.00005629655,0.69596326,0.0009671041,0.026626369,0.03704454,0.022466179,0.1440772],"study_design_scores_gemma":[0.00025916187,0.010750454,0.017704772,0.0024422915,0.0004171225,0.000009700557,0.5204818,0.013800851,0.4082715,0.023568293,0.0018574533,0.0004365816],"about_ca_topic_score_codex":0.0011961643,"about_ca_topic_score_gemma":0.00018954986,"teacher_disagreement_score":0.3816451,"about_ca_system_score_codex":0.00033643728,"about_ca_system_score_gemma":0.00022260286,"threshold_uncertainty_score":0.9995903},"labels":[],"label_agreement":null}]}