{"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":"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":[0.08185802,0.013721573,0.88202155,0.003747406,0.0006671362,0.00021283339,0.004578678,0.008712685,0.0044801887],"genre_scores_gemma":[0.5026759,0.009203463,0.46826637,0.0013575741,0.0010513209,0.00025216234,0.011323365,0.0006985828,0.0051712706],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9973435,0.0009458668,0.00024972312,0.00069409254,0.0006630715,0.000103868035],"domain_scores_gemma":[0.9904873,0.0062636747,0.00044481875,0.0009990856,0.0016441204,0.00016104692],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033182907,0.0017451016,0.0013715397,0.0019838032,0.00039838642,0.002560916,0.0016317026,0.0009866073,0.0021768876],"category_scores_gemma":[0.010634019,0.00064372824,0.0024008993,0.0013227421,0.0004539166,0.0028013345,0.0011258416,0.0026687637,0.002460507],"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":[0.0004717264,0.00070731575,0.027503625,0.0007516509,0.00094151724,0.00042443376,0.00025359183,0.03524577,0.028599406,0.004700886,0.012083021,0.88831705],"study_design_scores_gemma":[0.000061888684,0.00033074745,0.01434537,0.00015046132,0.00063927943,0.00058838225,0.00016871675,0.9410908,0.015185869,0.0098485835,0.017461818,0.00012813682],"about_ca_topic_score_codex":0.009151259,"about_ca_topic_score_gemma":0.009440244,"teacher_disagreement_score":0.009151259,"about_ca_system_score_codex":0.00061843486,"about_ca_system_score_gemma":0.0019403361,"threshold_uncertainty_score":0.018195987},"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":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04166543,0.0012539914,0.9499325,0.00089291815,0.000102532584,0.00013281121,0.00040544264,0.0022697628,0.003344693],"genre_scores_gemma":[0.7773074,0.00067068264,0.21538992,0.00048593915,0.000094137096,0.00017898221,0.000828938,0.00007872683,0.0049653198],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9997756,0.000046593035,0.000012246157,0.000074928146,0.00005109081,0.000039465383],"domain_scores_gemma":[0.9997459,0.00009356661,0.00002776895,0.000018261044,0.00009164546,0.0000227877],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00071809994,0.0008518476,0.0005500134,0.0007698965,0.00029178426,0.00072115654,0.0012384746,0.001011275,0.001762242],"category_scores_gemma":[0.0013715416,0.00026026013,0.00091597025,0.00037973173,0.00046381695,0.00066309737,0.0006967524,0.0011516913,0.0004907418],"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":[0.00021961705,0.00018555511,0.0024368535,0.00010130453,0.00010593929,0.00032337202,0.000077694,0.70947474,0.008338499,0.010437357,0.0055101314,0.262789],"study_design_scores_gemma":[0.000004138509,0.000018268947,0.00010808124,0.0000041162502,0.000007071582,0.000027151067,0.0000029431264,0.99712425,0.0007617309,0.0016275462,0.00031214723,0.0000025939016],"about_ca_topic_score_codex":0.015331264,"about_ca_topic_score_gemma":0.011349177,"teacher_disagreement_score":0.015331264,"about_ca_system_score_codex":0.0013028664,"about_ca_system_score_gemma":0.0012350014,"threshold_uncertainty_score":0.03048408},"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":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14954488,0.0023880412,0.8348994,0.00066354213,0.00029936695,0.00026223148,0.0020719555,0.0061222445,0.0037483214],"genre_scores_gemma":[0.74659276,0.0010810344,0.24036382,0.0004754508,0.00024092721,0.00035195728,0.006378648,0.00013051227,0.0043848916],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99935466,0.00006912087,0.000052004238,0.00015362304,0.0002450487,0.00012551196],"domain_scores_gemma":[0.9995003,0.00014613042,0.00006172264,0.000039395865,0.00021364292,0.000038869843],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009052598,0.0011304003,0.001203795,0.0027035563,0.00054749515,0.0007996335,0.0010667539,0.001145305,0.0015102265],"category_scores_gemma":[0.0012835108,0.00021312578,0.0015783611,0.0014834682,0.00035160567,0.0015010035,0.0006858178,0.0010849057,0.001089797],"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":[0.00066839904,0.0006214498,0.02335673,0.00026211952,0.00027206203,0.0003046712,0.000103354534,0.037251547,0.043669105,0.003366091,0.01567265,0.87445176],"study_design_scores_gemma":[0.000032402786,0.00017075613,0.0049718944,0.000024463594,0.000095517964,0.00020940464,0.000056642573,0.97490656,0.013468343,0.0030808107,0.0029447654,0.000038405058],"about_ca_topic_score_codex":0.004653409,"about_ca_topic_score_gemma":0.00393213,"teacher_disagreement_score":0.004653409,"about_ca_system_score_codex":0.0005761566,"about_ca_system_score_gemma":0.0012757921,"threshold_uncertainty_score":0.009252608},"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":"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":[0.102173574,0.0017194931,0.8744817,0.0065587275,0.00023492186,0.00021153645,0.0011786591,0.0065042246,0.006937224],"genre_scores_gemma":[0.6167194,0.0010484318,0.37564862,0.0010202115,0.000099528756,0.00009951689,0.0014297457,0.0003439325,0.0035905659],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9983658,0.0007668007,0.00012710085,0.00020850671,0.00041993678,0.000111908295],"domain_scores_gemma":[0.99051183,0.006090508,0.00036446724,0.0010264581,0.0017881282,0.0002184897],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004643289,0.0006349461,0.0007587181,0.00078702,0.00033115735,0.0023174204,0.0013235688,0.00085356034,0.0030290845],"category_scores_gemma":[0.01734503,0.00032682318,0.00069508835,0.000554444,0.0006549194,0.00570621,0.0017472958,0.0019058164,0.0008849263],"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":[0.0004942567,0.0007383151,0.0139262155,0.00046312556,0.00024618357,0.0002746169,0.000716763,0.17741346,0.0103733875,0.02242035,0.010689071,0.7622442],"study_design_scores_gemma":[0.000032306634,0.00013003219,0.0007325218,0.00005987185,0.000049693856,0.00006989468,0.00013772276,0.96313775,0.0056855506,0.02402387,0.0059133335,0.000027490914],"about_ca_topic_score_codex":0.00982189,"about_ca_topic_score_gemma":0.011020279,"teacher_disagreement_score":0.00982189,"about_ca_system_score_codex":0.0009885348,"about_ca_system_score_gemma":0.0018994416,"threshold_uncertainty_score":0.024556339},"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":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9975231,0.00008815207,0.001746181,0.000014903748,0.0000039550177,0.00006107443,0.00031009322,0.0000352713,0.00021723336],"genre_scores_gemma":[0.9972792,0.00007892919,0.0015571263,0.00001819873,0.00000706016,0.000037237674,0.00083182426,0.000008612259,0.00018176019],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99898046,0.00030612116,0.00013245511,0.00027855538,0.00022174846,0.00008071593],"domain_scores_gemma":[0.9972572,0.0008032309,0.00039919087,0.000401518,0.0009300932,0.00020880142],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001868079,0.0004066076,0.00038716246,0.0016530376,0.00026564026,0.00056704617,0.00048346893,0.00039687622,0.00072393526],"category_scores_gemma":[0.004111532,0.0002944755,0.0004892367,0.00064506277,0.00036192112,0.00041867615,0.0004524864,0.00022763922,0.0004861817],"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":[0.00044404092,0.00034642842,0.9816082,0.000037595644,0.00007970734,0.0007223033,0.00027730296,0.0006533945,0.0013017489,0.00003707082,0.00024773972,0.014244444],"study_design_scores_gemma":[0.00008198794,0.0026435459,0.9410029,0.00003197987,0.00029730744,0.003202529,0.0009960182,0.04595459,0.0043740873,0.00009956235,0.0012520371,0.00006346712],"about_ca_topic_score_codex":0.0058455034,"about_ca_topic_score_gemma":0.004960543,"teacher_disagreement_score":0.0058455034,"about_ca_system_score_codex":0.0005670473,"about_ca_system_score_gemma":0.00056884636,"threshold_uncertainty_score":0.011622965},"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":"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":[0.9399798,0.0054279957,0.027217466,0.0010466533,0.00025378334,0.02015818,0.002331426,0.00055697974,0.0030277802],"genre_scores_gemma":[0.91873294,0.001269464,0.05771295,0.0007834434,0.0001690714,0.018838117,0.001862772,0.00013773686,0.0004933624],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.81071144,0.12919237,0.031313904,0.008344717,0.018827075,0.0016104908],"domain_scores_gemma":[0.24394864,0.604658,0.05845639,0.037915114,0.051227316,0.0037945653],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19568731,0.0013144892,0.0027522407,0.004678082,0.0011866208,0.0043439902,0.0024843283,0.0026785415,0.002972109],"category_scores_gemma":[0.49121165,0.0018725548,0.0053676497,0.0045239823,0.0018805474,0.006287077,0.0033003353,0.0025007923,0.0011594064],"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":[0.0781291,0.023636932,0.56890196,0.016122587,0.011047169,0.00035018357,0.0099917585,0.0071685384,0.0012848681,0.0019496738,0.0047109853,0.27670625],"study_design_scores_gemma":[0.04508097,0.1660003,0.5874168,0.012068772,0.03150672,0.0017251286,0.006007577,0.10629141,0.0114325145,0.0039909035,0.027230093,0.0012487958],"about_ca_topic_score_codex":0.0033620943,"about_ca_topic_score_gemma":0.0031497844,"teacher_disagreement_score":0.8043127,"about_ca_system_score_codex":0.0025782946,"about_ca_system_score_gemma":0.0068752943,"threshold_uncertainty_score":0.99186075},"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":"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":[0.40972653,0.24842066,0.15571156,0.019763248,0.0038782894,0.0717438,0.041156873,0.0054253777,0.04417367],"genre_scores_gemma":[0.59599847,0.021582693,0.3300968,0.002823829,0.001056306,0.040553123,0.006306697,0.00042631896,0.0011558397],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.54762614,0.29288775,0.10541129,0.011126515,0.041664336,0.0012840115],"domain_scores_gemma":[0.17270695,0.70490146,0.07180505,0.015165924,0.03359388,0.0018266462],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2576095,0.0023975677,0.008291963,0.03296618,0.0011873365,0.008884719,0.0032125714,0.0026839874,0.0059431363],"category_scores_gemma":[0.69671756,0.0019492117,0.012450668,0.019444672,0.0017951825,0.009149944,0.0060157785,0.002124287,0.0012395905],"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":[0.021402733,0.0007878057,0.09233755,0.1636501,0.05595041,0.00027160783,0.004025036,0.004347207,0.0026004445,0.0026793913,0.020429008,0.63151866],"study_design_scores_gemma":[0.04501571,0.018079648,0.3113245,0.16896227,0.21819867,0.0030548612,0.004756244,0.06404837,0.015144419,0.02587681,0.12233655,0.0032019245],"about_ca_topic_score_codex":0.0028146827,"about_ca_topic_score_gemma":0.0060289945,"teacher_disagreement_score":0.7423905,"about_ca_system_score_codex":0.0030755887,"about_ca_system_score_gemma":0.010384231,"threshold_uncertainty_score":0.9154997},"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":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92832977,0.008907656,0.034142066,0.0014109235,0.00034650095,0.00055085146,0.017303001,0.0035461516,0.005463005],"genre_scores_gemma":[0.9531087,0.0014238943,0.02315383,0.00034536724,0.00018664321,0.00018328281,0.02045921,0.0003258192,0.0008131273],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9891544,0.0054734834,0.0010310482,0.0016573465,0.002451223,0.00023241412],"domain_scores_gemma":[0.9452004,0.031772308,0.0061715436,0.008881486,0.006860558,0.0011137382],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014798188,0.00087959116,0.00055054104,0.00209451,0.00030247724,0.0023499832,0.0012861119,0.0006706559,0.0011420851],"category_scores_gemma":[0.084071636,0.0004039992,0.0012928151,0.0013174778,0.0007951773,0.0017430816,0.0015968257,0.0011537299,0.0012132977],"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":[0.0033072722,0.000568698,0.45311385,0.0023974772,0.0012871766,0.0008183372,0.0038111403,0.020034667,0.006215787,0.0010711409,0.028859986,0.4785145],"study_design_scores_gemma":[0.00058546546,0.0060847146,0.62609744,0.0016344965,0.0025589631,0.0067218053,0.0059633083,0.21749435,0.040349096,0.0059463703,0.08581115,0.0007528334],"about_ca_topic_score_codex":0.0025651788,"about_ca_topic_score_gemma":0.0027964185,"teacher_disagreement_score":0.014798188,"about_ca_system_score_codex":0.0009624206,"about_ca_system_score_gemma":0.00073411694,"threshold_uncertainty_score":0.078261256},"labels":[],"label_agreement":null}]}