{"meta":{"query_hash":"8abaf8d661e6","filters":{"venue":"Evaluation & the Health Professions"},"cohort_total":82,"direct_labels_cover":2,"predictions_cover":82,"exported":82,"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/8abaf8d661e6","api":"https://metacan.xera.ac/api/v1/cohort?venue=Evaluation+%26+the+Health+Professions"},"results":[{"id":"W1698167416","doi":"10.1177/0163278702238052","title":"The Biochemical Bases for Reward","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Pain Management and Placebo Effect","field":"Neuroscience","cited_by":55,"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 British Columbia","funders":"","keywords":"Nucleus accumbens; Placebo; Dopamine; Addiction; Medicine; Opioid; Neuroscience; Psychology; Pharmacology; Mechanism (biology); Psychiatry; Internal medicine","score_opus":0.3085821682452684,"score_gpt":0.4527774283355963,"score_spread":0.1441952600903279,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1698167416","genre_codex":"review","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":[0.07418536,0.30599043,0.09437354,0.22867265,0.009507664,0.00028400708,0.00049954536,0.00030490413,0.28618184],"genre_scores_gemma":[0.8268759,0.09911235,0.02215877,0.017717812,0.0035384875,0.0003211026,0.00015460335,0.000056522684,0.030064398],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9994747,0.0001606667,0.00003708945,0.00009893859,0.00015073117,0.00007794553],"domain_scores_gemma":[0.99926287,0.00022189452,0.00013125292,0.00009780753,0.00017286057,0.00011331814],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012164974,0.00051366707,0.00045175053,0.0006620529,0.0008930295,0.0021170196,0.000602239,0.0021558446,0.0042659263],"category_scores_gemma":[0.0022254009,0.0002442256,0.00050249166,0.00029864357,0.005509177,0.0022981076,0.0014904744,0.0030797643,0.0010253238],"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":[0.00034333204,0.00014434796,0.0010555591,0.00052716106,0.00007246333,0.00062120915,0.00046455237,0.0008322542,0.014370157,0.91838604,0.0054430338,0.057739954],"study_design_scores_gemma":[0.00013021455,0.00040597995,0.0025287676,0.00030004996,0.0000473175,0.0011453681,0.00022303154,0.0008323314,0.0091796415,0.84692335,0.1382287,0.000055261225],"about_ca_topic_score_codex":0.00067258964,"about_ca_topic_score_gemma":0.0005557103,"teacher_disagreement_score":0.0042659263,"about_ca_system_score_codex":0.0019409797,"about_ca_system_score_gemma":0.0014294451,"threshold_uncertainty_score":0.014270961},"labels":[],"label_agreement":null},{"id":"W1966473912","doi":"10.1177/0163278707311870","title":"When Is Knowledge Ripe for Primary Care?","year":2007,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hôpital Charles-Le Moyne; Université Laval; Université de Montréal","funders":"","keywords":"Primary care; Medicine; Psychology; Family medicine","score_opus":0.41085470618830544,"score_gpt":0.6224560121363097,"score_spread":0.2116013059480043,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966473912","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030734368,0.035427906,0.0022425607,0.88984215,0.0016027401,0.00007889663,0.00008824382,0.00004966722,0.039933525],"genre_scores_gemma":[0.86007524,0.020701393,0.0040128794,0.109296545,0.0024884853,0.00023010261,0.00010377602,0.00006652892,0.0030249692],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9270043,0.04912124,0.0026106364,0.004373503,0.010812102,0.006078349],"domain_scores_gemma":[0.7536125,0.17369643,0.014944811,0.0072944183,0.024476808,0.025975024],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06806767,0.00033347995,0.002010367,0.0038624902,0.010293906,0.028272195,0.0025845917,0.01218924,0.00879221],"category_scores_gemma":[0.19178596,0.0009869857,0.00089283846,0.0029167524,0.03533908,0.028286459,0.012280005,0.012275246,0.0013584872],"study_design_candidate":"theoretical_or_conceptual","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":[0.00037163464,0.00038654706,0.056049656,0.0064670253,0.00036696412,0.005095565,0.25950518,0.00015074936,0.0010049886,0.18465959,0.099171735,0.38677043],"study_design_scores_gemma":[0.00015245238,0.00046380828,0.029779663,0.01700121,0.00021525378,0.0030612499,0.40249854,0.00018173535,0.0006615444,0.25567967,0.29011536,0.00018955172],"about_ca_topic_score_codex":0.01037359,"about_ca_topic_score_gemma":0.012054489,"teacher_disagreement_score":0.93193233,"about_ca_system_score_codex":0.019675273,"about_ca_system_score_gemma":0.033765584,"threshold_uncertainty_score":0.35998052},"labels":[],"label_agreement":null},{"id":"W1972824546","doi":"10.1177/0163278703258104","title":"Assessment of Communication Barriers in Community Pharmacies","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Pharmaceutical Practices and Patient Outcomes","field":"Medicine","cited_by":26,"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 British Columbia; Michael Smith Health Research BC; Regina Qu'Appelle Health Region","funders":"","keywords":"Pharmacist; Pharmacy; Quality (philosophy); Variance (accounting); Variables; Multilevel model; Psychology; Graduation (instrument); Medical education; Medicine; Nursing; Family medicine; Applied psychology; Business; Statistics","score_opus":0.4565529397769157,"score_gpt":0.6137302823738618,"score_spread":0.15717734259694616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972824546","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":[0.9990232,0.00015218687,0.0001835712,0.00006897233,0.000002131208,0.00003710403,0.000023255143,0.0000044926783,0.00050510804],"genre_scores_gemma":[0.99910456,0.00011980303,0.00052612106,0.000026245674,0.0000051679517,0.000035040037,0.000034877507,0.0000017041007,0.00014651609],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9959149,0.0017145533,0.00044757238,0.00021431262,0.0013504964,0.000358174],"domain_scores_gemma":[0.97663957,0.010384684,0.008306159,0.00044142143,0.002776688,0.00145149],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034944671,0.00022288573,0.00033345897,0.0014913242,0.0006547871,0.0007710838,0.00036884253,0.00043537837,0.0017378712],"category_scores_gemma":[0.033334687,0.00021417672,0.00024074085,0.0006286501,0.00029461228,0.00079836167,0.00086247956,0.0004176427,0.00012629754],"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":[0.0006177335,0.0006122983,0.9266593,0.00042886546,0.00012425461,0.00037196605,0.00850897,0.00017387472,0.0017625958,0.00012215509,0.0002795241,0.060338497],"study_design_scores_gemma":[0.00003249275,0.0011097413,0.986331,0.00012233597,0.00006054956,0.0008537577,0.008624403,0.0007799729,0.00085096346,0.00012510184,0.0010799557,0.000029759807],"about_ca_topic_score_codex":0.003564297,"about_ca_topic_score_gemma":0.0046638562,"teacher_disagreement_score":0.003564297,"about_ca_system_score_codex":0.00075215526,"about_ca_system_score_gemma":0.0017799167,"threshold_uncertainty_score":0.018480778},"labels":[],"label_agreement":null},{"id":"W1973215149","doi":"10.1177/0163278714540681","title":"An 8-Item Short Form of the Inventory of Dimensions of Emerging Adulthood (IDEA) Among Young Swiss Men","year":2014,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Identity, Memory, and Therapy","field":"Psychology","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":"Centre for Addiction and Mental Health","funders":"Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Psychology; Young adult; Feeling; Scale (ratio); Psychopathology; Longitudinal study; Mental health; Clinical psychology; Developmental psychology; Psychiatry; Medicine; Social psychology","score_opus":0.07899338823122372,"score_gpt":0.4427404409925142,"score_spread":0.3637470527612905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1973215149","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":[0.9903122,0.00046534193,0.0006871713,0.00013108934,0.00004560114,0.0006356583,0.004273426,0.00003659053,0.0034128674],"genre_scores_gemma":[0.9782573,0.0006803719,0.0048196735,0.000069320275,0.000047224014,0.0017569159,0.008311964,0.000008231872,0.0060490887],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99973077,0.000084611376,0.000050401526,0.00003244325,0.000067111185,0.000034610046],"domain_scores_gemma":[0.9990368,0.00020622372,0.00035801486,0.000064574044,0.00021696316,0.00011748537],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007282068,0.00043871254,0.00024964302,0.001094605,0.0003789582,0.00037277237,0.00022661751,0.00026939617,0.0038212982],"category_scores_gemma":[0.0022407584,0.00012402037,0.00043705665,0.00059895415,0.00016056419,0.00036823036,0.0005555745,0.0004458381,0.00062105415],"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":[0.00061963254,0.00055200414,0.89204764,0.00032815197,0.00016440579,0.00039204952,0.002175398,0.0004382831,0.0044521266,0.0004794464,0.0039090896,0.09444181],"study_design_scores_gemma":[0.000047975704,0.0006930847,0.994262,0.000046056415,0.00003870528,0.00021035498,0.000602751,0.00027410695,0.0003563187,0.000121557816,0.0033318629,0.00001526787],"about_ca_topic_score_codex":0.0036195696,"about_ca_topic_score_gemma":0.0077751637,"teacher_disagreement_score":0.0038212982,"about_ca_system_score_codex":0.00025077473,"about_ca_system_score_gemma":0.00047392026,"threshold_uncertainty_score":0.012783468},"labels":[],"label_agreement":null},{"id":"W1974057428","doi":"10.1177/0163278707307924","title":"Assessment of Family Physicians' Performance Using Patient Charts","year":2007,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Quality and Management","field":"Health Professions","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Association des Médecins d'Urgence du Québec","funders":"","keywords":"Chart; Medicine; Inter-rater reliability; Recall; Concordance; Medical record; Reliability (semiconductor); Cohen's kappa; Family medicine; Psychology; Statistics; Rating scale","score_opus":0.4189002070409529,"score_gpt":0.6045903599396676,"score_spread":0.1856901528987147,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974057428","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":[0.98244053,0.0010175925,0.006874078,0.00030805307,0.000049054484,0.00054036157,0.0011840133,0.00025304954,0.0073332204],"genre_scores_gemma":[0.9896718,0.00048280583,0.008256093,0.00003564517,0.00004877719,0.00019361202,0.00055729214,0.000012732958,0.0007412917],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97792107,0.011803613,0.0023712101,0.00087295106,0.0066232104,0.00040795363],"domain_scores_gemma":[0.9328633,0.025145262,0.014540514,0.002664081,0.022986818,0.0018000606],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017802864,0.00046552622,0.00063000264,0.0046307687,0.0006062025,0.0013376488,0.0007109518,0.00040917817,0.0011961352],"category_scores_gemma":[0.07206656,0.00024515713,0.00046807327,0.001971614,0.0005599452,0.0011711116,0.0008284923,0.0004339313,0.0003976082],"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":[0.0007057644,0.00035479723,0.8581768,0.00035880986,0.0002677048,0.00033266458,0.0067428574,0.001536173,0.002186737,0.00025564776,0.003129346,0.12595268],"study_design_scores_gemma":[0.000064966,0.0015423069,0.9764382,0.00018163174,0.00009595118,0.00096223765,0.003075844,0.0074019963,0.0051245224,0.0002711875,0.0047474736,0.000093642884],"about_ca_topic_score_codex":0.00651631,"about_ca_topic_score_gemma":0.0064237,"teacher_disagreement_score":0.017802864,"about_ca_system_score_codex":0.00097586866,"about_ca_system_score_gemma":0.0013172306,"threshold_uncertainty_score":0.094151676},"labels":[],"label_agreement":null},{"id":"W1974235884","doi":"10.1177/0163278703255234","title":"Meta-Analyses of Cluster Randomization Trials","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Randomization; Randomized controlled trial; Cluster (spacecraft); Meta-analysis; Cluster randomised controlled trial; Statistical power; Sample size determination; Statistics; Restricted randomization; Computer science; Medicine; Econometrics; Mathematics; Surgery; Internal medicine","score_opus":0.9826344435159781,"score_gpt":0.7393840614099454,"score_spread":0.24325038210603267,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974235884","genre_codex":"review","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009516024,0.5148067,0.42186671,0.0035464468,0.010425147,0.026696853,0.0065599945,0.0021355515,0.0044465484],"genre_scores_gemma":[0.20313336,0.09116204,0.61716956,0.0037744374,0.0026660578,0.07572511,0.0036964081,0.0008765003,0.0017965434],"study_design_codex":"meta_analysis","study_design_gemma":"not_applicable","domain_scores_codex":[0.48797828,0.44476417,0.034341343,0.014800515,0.016850697,0.001265012],"domain_scores_gemma":[0.59892017,0.3438565,0.021869553,0.025688825,0.008861912,0.00080300006],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22655097,0.006945222,0.027013099,0.024508338,0.0015251102,0.0065689436,0.005907937,0.005179519,0.0069551454],"category_scores_gemma":[0.5416268,0.004013136,0.04805422,0.018678559,0.0021137665,0.0048885937,0.004891253,0.006838934,0.00094729144],"study_design_candidate":"not_applicable","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":[0.0063846395,0.00012067149,0.0042781252,0.110208556,0.7872416,0.00022223817,0.00033573544,0.009789738,0.00025561947,0.0117377415,0.008275717,0.061149642],"study_design_scores_gemma":[0.012308873,0.0018751292,0.005280366,0.029354887,0.8033152,0.0005406599,0.0002732975,0.023337195,0.0012223915,0.10129501,0.020764153,0.000432811],"about_ca_topic_score_codex":0.003321869,"about_ca_topic_score_gemma":0.004917764,"teacher_disagreement_score":0.77344906,"about_ca_system_score_codex":0.0051718513,"about_ca_system_score_gemma":0.0068927677,"threshold_uncertainty_score":0.9538004},"labels":[],"label_agreement":null},{"id":"W1974257054","doi":"10.1177/0163278703026002005","title":"Evaluability Assessment","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":40,"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 Calgary","funders":"","keywords":"Logic model; Process (computing); Process management; Outcome (game theory); Product (mathematics); Key (lock); Theory of change; Program evaluation; Reflection (computer programming); Psychology; Management science; Knowledge management; Medical education; Medicine; Nursing; Computer science; Political science; Sociology; Engineering","score_opus":0.5915797188746392,"score_gpt":0.6761750254876046,"score_spread":0.08459530661296533,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974257054","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":[0.11599539,0.0026894554,0.588427,0.0051095122,0.0010878941,0.023586791,0.0024678197,0.0017018716,0.25893426],"genre_scores_gemma":[0.61134464,0.0010438096,0.34556186,0.0009485179,0.00027251066,0.019335754,0.0022567713,0.00057453377,0.018661682],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7672253,0.14153636,0.024926983,0.007395183,0.055506364,0.0034097661],"domain_scores_gemma":[0.5687214,0.18593384,0.024782034,0.039725438,0.17714758,0.0036897769],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.21332447,0.0013086817,0.0011885896,0.011950475,0.0028216748,0.0063838335,0.0022281585,0.0011764617,0.0123301],"category_scores_gemma":[0.3802526,0.0003852777,0.001956029,0.0051832986,0.0021843049,0.0050754216,0.0063706296,0.001598445,0.0014008085],"study_design_candidate":"theoretical_or_conceptual","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":[0.00079412677,0.00070268154,0.031576466,0.0026104078,0.0004174062,0.00016111677,0.010902294,0.007214093,0.0025903974,0.06722936,0.016040642,0.8597609],"study_design_scores_gemma":[0.0006422613,0.0055916575,0.13069762,0.009171301,0.001392572,0.0009218263,0.030674336,0.06936972,0.02370923,0.20781037,0.5192372,0.0007818705],"about_ca_topic_score_codex":0.0023948413,"about_ca_topic_score_gemma":0.0027611784,"teacher_disagreement_score":0.21332447,"about_ca_system_score_codex":0.005779904,"about_ca_system_score_gemma":0.009333748,"threshold_uncertainty_score":0.97011095},"labels":[],"label_agreement":null},{"id":"W1985929520","doi":"10.1177/0163278710383561","title":"Implementation of a Panel of Service Users for the Evaluation of Mental Health Outpatient Services","year":2010,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; Douglas Mental Health University Institute","funders":"Canadian Mental Health Association; McGill University; Yale University","keywords":"Mental health; Service (business); Context (archaeology); Medicine; Nursing; Medical education; Psychiatry; Business; Marketing","score_opus":0.37791495132753533,"score_gpt":0.586986413016971,"score_spread":0.20907146168943563,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1985929520","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":[0.9728601,0.00026704822,0.00813072,0.0047923075,0.00041575124,0.0022020177,0.00011340265,0.00015038376,0.011068229],"genre_scores_gemma":[0.97021466,0.00022738795,0.014198161,0.0017780415,0.00044701426,0.0015201854,0.00049426913,0.00007200738,0.011048227],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9841575,0.010816346,0.00068727176,0.0007693461,0.0017163501,0.0018531535],"domain_scores_gemma":[0.97171074,0.0059932587,0.0037067274,0.0018747126,0.006896459,0.0098179635],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01841939,0.00060440396,0.00038561283,0.0008106561,0.008789148,0.0024674712,0.0017665195,0.0017319262,0.0072648134],"category_scores_gemma":[0.032126334,0.00063556,0.0006431172,0.00057952,0.0013612607,0.0010573621,0.005012336,0.0034285286,0.0020140046],"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":[0.0014010518,0.005457851,0.25395206,0.00056782545,0.000109121094,0.0076977685,0.20115931,0.001139248,0.04313719,0.0013033403,0.046142094,0.43793318],"study_design_scores_gemma":[0.00027038433,0.008235203,0.49377546,0.00047307168,0.00010061438,0.00608697,0.26761627,0.0058505647,0.009905723,0.0012255149,0.20591702,0.00054330286],"about_ca_topic_score_codex":0.0044865613,"about_ca_topic_score_gemma":0.010325867,"teacher_disagreement_score":0.01841939,"about_ca_system_score_codex":0.0022911625,"about_ca_system_score_gemma":0.0054557654,"threshold_uncertainty_score":0.09741217},"labels":[],"label_agreement":null},{"id":"W2019936193","doi":"10.1177/0163278713475868","title":"Evidence for the Validity of Grouped Self-Assessments in Measuring the Outcomes of Educational Programs","year":2013,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"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":"University of Saskatchewan","keywords":"Psychology; Correlation; Self-assessment; Applied psychology; Clinical psychology; Social psychology; Mathematics","score_opus":0.7640554176157064,"score_gpt":0.6509811233514293,"score_spread":0.11307429426427706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2019936193","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":[0.797658,0.031657062,0.095930785,0.0045580524,0.0012215169,0.0016307952,0.002027848,0.00024784383,0.06506808],"genre_scores_gemma":[0.97333956,0.0031072497,0.020965567,0.0005290544,0.00025406395,0.00043381713,0.0005498764,0.00008827569,0.0007324924],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.7354783,0.18943709,0.013489671,0.011103732,0.048931587,0.0015595782],"domain_scores_gemma":[0.18714167,0.6572934,0.048972808,0.05137894,0.052735727,0.002477536],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.28092933,0.0009126329,0.0011951409,0.0066418676,0.0013467708,0.004152265,0.003133726,0.0016680182,0.002065654],"category_scores_gemma":[0.5188504,0.00075482385,0.0026695062,0.00890469,0.0062307306,0.0047479435,0.0033802264,0.0018756408,0.00088096963],"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":[0.002398466,0.0011039858,0.70822674,0.0058530816,0.007585355,0.00010216832,0.0079572415,0.0037586847,0.000535763,0.009622407,0.002215206,0.2506409],"study_design_scores_gemma":[0.00028727716,0.004387881,0.94150877,0.0071068667,0.0027022357,0.00039250296,0.0041833357,0.00965141,0.0030435356,0.01453409,0.011979966,0.00022204117],"about_ca_topic_score_codex":0.0060603134,"about_ca_topic_score_gemma":0.008781861,"teacher_disagreement_score":0.28092933,"about_ca_system_score_codex":0.002209197,"about_ca_system_score_gemma":0.0029340212,"threshold_uncertainty_score":0.8867422},"labels":[],"label_agreement":null},{"id":"W2031744068","doi":"10.1177/0163278713511325","title":"Evidence-Based Health Care Management","year":2013,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":33,"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":"Agency for Healthcare Research and Quality","keywords":"Health care; Psychological intervention; Workforce; MEDLINE; Inclusion (mineral); Medicine; Relevance (law); Quality management; Evidence-based practice; Systematic review; Nursing; Business; Alternative medicine; Psychology; Political science; Marketing","score_opus":0.8318884867437546,"score_gpt":0.6073044842747833,"score_spread":0.22458400246897126,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2031744068","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":[0.00055083423,0.96753246,0.006069465,0.016975248,0.0021223077,0.0012053858,0.0003652383,0.00005370993,0.0051253405],"genre_scores_gemma":[0.03265433,0.897629,0.050598446,0.011244784,0.0029089514,0.0033241264,0.0010064694,0.000035644614,0.00059821125],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8638606,0.0747955,0.033237472,0.0050188215,0.021672532,0.0014151558],"domain_scores_gemma":[0.74185026,0.18638729,0.025365891,0.010258192,0.03339332,0.0027451573],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.110131145,0.0021208883,0.0064167106,0.025104873,0.0016477157,0.011705674,0.004916481,0.006052256,0.009235089],"category_scores_gemma":[0.28114316,0.0014608679,0.0057440284,0.015873978,0.0021526816,0.009339989,0.0066581685,0.00562483,0.0015088301],"study_design_candidate":"not_applicable","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":[0.00016461965,0.00019289664,0.002139715,0.42302057,0.0058419253,0.00025032242,0.0012161194,0.00072760985,0.00042555705,0.021568635,0.026444376,0.5180077],"study_design_scores_gemma":[0.0003430588,0.00025117517,0.0038312266,0.7856678,0.0070505957,0.0004011154,0.0008887825,0.0006422264,0.00043806608,0.02238098,0.17803147,0.00007353951],"about_ca_topic_score_codex":0.0041946024,"about_ca_topic_score_gemma":0.0074752257,"teacher_disagreement_score":0.110131145,"about_ca_system_score_codex":0.010642406,"about_ca_system_score_gemma":0.035702776,"threshold_uncertainty_score":0.5824361},"labels":[],"label_agreement":null},{"id":"W2031999169","doi":"10.1177/01632780022034507","title":"Quality Improvement in a School Health Program","year":2000,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"School Health and Nursing Education","field":"Health Professions","cited_by":7,"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 of Health Services and Policy Research","funders":"","keywords":"Referral; Nursing; Intervention (counseling); Program evaluation; Medical education; Plan (archaeology); Quality management; Process (computing); Service delivery framework; Medicine; Quality (philosophy); Psychology; Service (business); Business; Computer science","score_opus":0.26324780480908944,"score_gpt":0.6149232725493936,"score_spread":0.35167546774030417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2031999169","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":[0.9601101,0.0037211159,0.0071042893,0.0061294516,0.00026989935,0.0070514134,0.00043683083,0.0008533334,0.014323432],"genre_scores_gemma":[0.97267646,0.0015787584,0.020208394,0.0010982776,0.000096046446,0.0018278138,0.00048110003,0.000050264713,0.00198301],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.97784674,0.015093903,0.00088039687,0.0009144186,0.0033431284,0.0019213109],"domain_scores_gemma":[0.9820176,0.00671444,0.002573978,0.0013790492,0.003218092,0.004096937],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015978027,0.0003642661,0.00038488218,0.0014145148,0.002346798,0.0022581234,0.00089390925,0.0007637802,0.00397953],"category_scores_gemma":[0.027632637,0.00036564388,0.0007337102,0.0015973398,0.0007030331,0.0011200791,0.0021046174,0.0013039272,0.00026333294],"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":[0.0027779383,0.032035377,0.06464021,0.0020750056,0.0003127748,0.00022281577,0.0048823874,0.0030831664,0.0015899943,0.0019382365,0.01024376,0.87619835],"study_design_scores_gemma":[0.018969571,0.12738587,0.7291854,0.004001822,0.0017643794,0.000748317,0.012053228,0.012847541,0.010926988,0.0028899468,0.07901435,0.00021258915],"about_ca_topic_score_codex":0.018095441,"about_ca_topic_score_gemma":0.018675596,"teacher_disagreement_score":0.018095441,"about_ca_system_score_codex":0.0047296956,"about_ca_system_score_gemma":0.015324108,"threshold_uncertainty_score":0.08450085},"labels":[],"label_agreement":null},{"id":"W2042916727","doi":"10.1177/01632780022034660","title":"State Medicaid Quality Programs","year":2000,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","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":"Delmar (Canada)","funders":"","keywords":"Medicaid; Quality (philosophy); State (computer science); Medicine; Psychology; Computer science; Health care; Political science","score_opus":0.36991834278495056,"score_gpt":0.477743967270258,"score_spread":0.10782562448530741,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042916727","genre_codex":"other","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":[0.07327422,0.0014514874,0.00493155,0.06890023,0.0011883426,0.0019782933,0.02746819,0.0015094047,0.8192983],"genre_scores_gemma":[0.3107656,0.001112151,0.004687855,0.03876232,0.0014204435,0.0018023548,0.015205456,0.00032037444,0.6259234],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9922207,0.001020727,0.00023646899,0.0006182273,0.0040112413,0.0018925866],"domain_scores_gemma":[0.9861535,0.0033653108,0.0010052884,0.0012266272,0.0050398633,0.0032095101],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0063117594,0.00033039047,0.00057703647,0.002564264,0.0029857133,0.0049251644,0.0016832054,0.0023975507,0.07842839],"category_scores_gemma":[0.017220005,0.00063815573,0.0009155715,0.0021144887,0.00072815,0.0018103471,0.0023866473,0.002615359,0.0072359387],"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":[0.00025531626,0.0008953815,0.03000804,0.00014258994,0.0000877545,0.000051769777,0.00034859343,0.0009759097,0.00070575194,0.10500176,0.7406534,0.12087374],"study_design_scores_gemma":[0.00040036,0.00026456438,0.16634311,0.00024287985,0.00017329588,0.00012043254,0.0006077815,0.0044956002,0.0017917231,0.012412527,0.8130772,0.00007048517],"about_ca_topic_score_codex":0.09873855,"about_ca_topic_score_gemma":0.2453152,"teacher_disagreement_score":0.09873855,"about_ca_system_score_codex":0.009957184,"about_ca_system_score_gemma":0.03455418,"threshold_uncertainty_score":0.2623691},"labels":[],"label_agreement":null},{"id":"W2044889022","doi":"10.1177/016327870002300205","title":"Contextual Influences on Learning Application in Practice","year":2000,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"U.S. Public Health Service","keywords":"Logistic regression; Context (archaeology); Process (computing); Psychology; Sample (material); Affect (linguistics); Medical education; Rehabilitation; Applied psychology; Professional development; Nursing; Medicine; Computer science; Pedagogy","score_opus":0.12526926570819355,"score_gpt":0.5024420030738176,"score_spread":0.37717273736562407,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2044889022","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":[0.9915554,0.0004513904,0.0017668589,0.00043957026,0.000014254604,0.00008396942,0.000037785518,0.000026700072,0.005624085],"genre_scores_gemma":[0.9990728,0.000104201456,0.0006551368,0.000024447161,0.0000052568926,0.00002075813,0.000011643856,0.0000040416935,0.00010172814],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9835535,0.011363361,0.00074054365,0.0009490799,0.0021952994,0.0011982458],"domain_scores_gemma":[0.95949996,0.026920535,0.006741891,0.0014150018,0.0025480867,0.0028744983],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053238915,0.00020884027,0.00026025713,0.0010151815,0.0016320694,0.0026375041,0.0003082702,0.00041257998,0.0018301961],"category_scores_gemma":[0.045066368,0.0002439568,0.00027315516,0.0008154053,0.0017368358,0.00097271346,0.002408599,0.0006099853,0.00013270238],"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":[0.000710992,0.0014725734,0.811598,0.00042438434,0.00023068226,0.000777999,0.040683325,0.0018763308,0.0034250831,0.0036481174,0.00051757204,0.134635],"study_design_scores_gemma":[0.000026204636,0.0006241587,0.9707742,0.00022411632,0.00010677434,0.00022680288,0.02049706,0.00085513823,0.001197428,0.0016241707,0.0037874565,0.00005661708],"about_ca_topic_score_codex":0.006359914,"about_ca_topic_score_gemma":0.012477631,"teacher_disagreement_score":0.006359914,"about_ca_system_score_codex":0.002239247,"about_ca_system_score_gemma":0.0030854312,"threshold_uncertainty_score":0.028155744},"labels":[],"label_agreement":null},{"id":"W2051569330","doi":"10.1177/0163278702250097","title":"Development Of The Approaches To Work And Workplace Climate Questionnaires For Physicians","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare professionals’ stress and burnout","field":"Health Professions","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Queen's University","funders":"","keywords":"Work environment; Work (physics); Organisation climate; Medical education; Psychology; Medicine; Job satisfaction; Engineering; Social psychology","score_opus":0.34624733533811514,"score_gpt":0.48290605602911757,"score_spread":0.13665872069100243,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2051569330","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":[0.85861063,0.00089498697,0.02585166,0.0031677634,0.00044387518,0.07969631,0.011978347,0.00052921753,0.018827142],"genre_scores_gemma":[0.6493015,0.0021087935,0.1469364,0.0019510109,0.00042742447,0.17414452,0.010542751,0.00016222654,0.014425377],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99444515,0.0018582107,0.0013466148,0.00023194683,0.0015765061,0.0005416057],"domain_scores_gemma":[0.96975774,0.012326245,0.005108524,0.0014717347,0.009075023,0.0022606892],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015714878,0.00058351783,0.0009962546,0.0026151126,0.00087340595,0.00090559584,0.0010593942,0.0008309219,0.0037640152],"category_scores_gemma":[0.035856124,0.00085595006,0.0011929031,0.0015783987,0.00062365574,0.001101439,0.0016955659,0.0022157833,0.0012445742],"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":[0.0017224137,0.007920977,0.448583,0.0019959467,0.0002611656,0.000641466,0.029872699,0.0041018627,0.008582279,0.0046279137,0.045721795,0.44596863],"study_design_scores_gemma":[0.0006011168,0.0028212413,0.9181138,0.00028575913,0.000052481235,0.00020299022,0.0053511565,0.003265184,0.0017524788,0.0016006789,0.0657927,0.00016029619],"about_ca_topic_score_codex":0.006790527,"about_ca_topic_score_gemma":0.012805212,"teacher_disagreement_score":0.015714878,"about_ca_system_score_codex":0.0023695335,"about_ca_system_score_gemma":0.0037404792,"threshold_uncertainty_score":0.08310926},"labels":[],"label_agreement":null},{"id":"W2052657185","doi":"10.1177/0163278703261202","title":"Recruitment Strategies","year":2004,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Reproductive Health and Contraception","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; University of Toronto","funders":"","keywords":"Pharmacy; Warrant; Pharmacist; Nursing; Community pharmacy; Pharmacy practice; Medicine; Medical education; Business; Family medicine","score_opus":0.32978561976890386,"score_gpt":0.5261916604227772,"score_spread":0.19640604065387335,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052657185","genre_codex":"protocol","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":[0.0324861,0.0043734238,0.16248856,0.012372443,0.004975652,0.62329555,0.004594915,0.0023092614,0.15310414],"genre_scores_gemma":[0.040096484,0.0035946418,0.25491047,0.012531838,0.0010020274,0.58830285,0.0028940835,0.0005627853,0.09610481],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.95986843,0.024571672,0.004682604,0.002117411,0.005526772,0.0032330896],"domain_scores_gemma":[0.97882736,0.007265242,0.0012198157,0.0030200318,0.0072067445,0.002460803],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.047100257,0.0021156983,0.0014856014,0.0056542754,0.0048900256,0.0036604533,0.00538129,0.0043914462,0.101021014],"category_scores_gemma":[0.072788715,0.0014838623,0.0013274698,0.0028430775,0.0013081303,0.0029041767,0.0073791207,0.0032162105,0.07371052],"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":[0.0014578911,0.0035808887,0.0059749507,0.004587375,0.000059803035,0.0015315133,0.018605072,0.00079176115,0.0034333777,0.024607735,0.13065705,0.80471265],"study_design_scores_gemma":[0.0014206661,0.0033513065,0.0057605724,0.0036428722,0.000109168715,0.0013023419,0.01388523,0.0020233933,0.0034581402,0.01814698,0.9467465,0.00015283185],"about_ca_topic_score_codex":0.0016311,"about_ca_topic_score_gemma":0.0031401988,"teacher_disagreement_score":0.101021014,"about_ca_system_score_codex":0.0030716804,"about_ca_system_score_gemma":0.012995702,"threshold_uncertainty_score":0.33794898},"labels":[],"label_agreement":null},{"id":"W2052904783","doi":"10.1177/0163278709338570","title":"Bridging the Gap: Knowledge Seeking and Sharing in a Virtual Community of Emergency Practice","year":2009,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Systems and Technology","field":"Business, Management and Accounting","cited_by":71,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University; Izaak Walton Killam Health Centre","funders":"Associated Medical Services","keywords":"Bridging (networking); Medical education; Medicine; Resource (disambiguation); Quality (philosophy); Knowledge management; Computer science","score_opus":0.2013394292551039,"score_gpt":0.45463960418971,"score_spread":0.2533001749346061,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052904783","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":[0.99006456,0.00009108235,0.0019512312,0.0013217833,0.000016488959,0.00013234068,0.0000076803435,0.000021747002,0.006393132],"genre_scores_gemma":[0.996718,0.000054142718,0.002491368,0.00014597735,0.000012570809,0.00008829781,0.0000068735753,0.000002893098,0.00047973625],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9901412,0.008158999,0.0001732906,0.0003217095,0.0006091928,0.0005956856],"domain_scores_gemma":[0.98309475,0.011135355,0.0013899513,0.00085566624,0.0005690885,0.0029552188],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009274451,0.00018862075,0.0002821742,0.001209755,0.0037208318,0.0037043674,0.0009829239,0.0009702603,0.0047156066],"category_scores_gemma":[0.022035716,0.0002454308,0.00022660347,0.0006308684,0.002525468,0.0044373027,0.0067026853,0.00052637374,0.00029143787],"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":[0.0013456519,0.007665155,0.089525335,0.0010583751,0.00009110578,0.0021012162,0.38550884,0.0009087531,0.0057560587,0.013578542,0.005606081,0.48685494],"study_design_scores_gemma":[0.0012558296,0.00996554,0.11739887,0.0011851481,0.00026914114,0.0032540981,0.74919575,0.011280218,0.0042046793,0.03152467,0.070257865,0.00020813549],"about_ca_topic_score_codex":0.0007680825,"about_ca_topic_score_gemma":0.0012911347,"teacher_disagreement_score":0.009274451,"about_ca_system_score_codex":0.00085582724,"about_ca_system_score_gemma":0.002943914,"threshold_uncertainty_score":0.049048603},"labels":[],"label_agreement":null},{"id":"W2053560656","doi":"10.1177/0163278712444287","title":"Prevention Education Effects on Fundamental Memory Processes","year":2012,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":6,"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 British Columbia","funders":"National Institute on Drug Abuse; National Institute on Alcohol Abuse and Alcoholism","keywords":"Session (web analytics); Conceptualization; Denial; Task (project management); Psychology; Random assignment; Cognitive psychology; Applied psychology; Computer science; Medicine; Artificial intelligence; Psychotherapist","score_opus":0.132832182757019,"score_gpt":0.4853538985500621,"score_spread":0.35252171579304314,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2053560656","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":[0.99943906,0.00008137261,0.000071882736,0.00003178742,0.0000024982814,0.000021084668,0.0000056062236,0.0000044545004,0.00034233916],"genre_scores_gemma":[0.99899906,0.00013679499,0.00045105384,0.00002561661,0.0000043363834,0.000020932406,0.0000228011,0.0000012649188,0.00033821538],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9994242,0.00025562182,0.00003055819,0.00006854929,0.00012061381,0.00010061345],"domain_scores_gemma":[0.99667656,0.001894655,0.00058250525,0.00021474138,0.00021981998,0.00041166515],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010012935,0.00031007171,0.00024857587,0.00031342905,0.00045704335,0.000413443,0.00029445504,0.0002856625,0.0020694141],"category_scores_gemma":[0.0065715765,0.00018128184,0.00022406936,0.00012197577,0.0005006598,0.00030533728,0.0006787676,0.0005634446,0.000095591684],"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":[0.00934287,0.04265127,0.4396903,0.0008339525,0.00030064437,0.00028886186,0.011667466,0.0011728083,0.030731378,0.000894152,0.0005458224,0.46188053],"study_design_scores_gemma":[0.00033557857,0.021993758,0.9621337,0.00011611224,0.00021560625,0.00013364245,0.0019017252,0.000735956,0.010324489,0.0005804275,0.0015038335,0.000025115802],"about_ca_topic_score_codex":0.0038628203,"about_ca_topic_score_gemma":0.007477689,"teacher_disagreement_score":0.0038628203,"about_ca_system_score_codex":0.00042190898,"about_ca_system_score_gemma":0.0009157324,"threshold_uncertainty_score":0.007680714},"labels":[],"label_agreement":null},{"id":"W2058893968","doi":"10.1177/016327870202500206","title":"Individualizing Treatment Decisions","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":69,"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 Toronto","funders":"","keywords":"Process (computing); Intervention (counseling); Computer science; Quality (philosophy); Management science; Process management; Risk analysis (engineering); Psychology; Medicine; Nursing; Engineering","score_opus":0.7496509901992127,"score_gpt":0.5668166863411787,"score_spread":0.182834303858034,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2058893968","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":[0.07912305,0.015947955,0.3521984,0.33918464,0.0035455872,0.0041447706,0.00060468836,0.0014754485,0.2037755],"genre_scores_gemma":[0.5591344,0.0108125685,0.39909565,0.020172706,0.0023244834,0.0019253596,0.0004490014,0.00017008674,0.0059157424],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9612141,0.027981207,0.002440058,0.0016314796,0.0052944394,0.0014386543],"domain_scores_gemma":[0.9494358,0.038764335,0.003652666,0.0026514195,0.003082125,0.0024135443],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03925864,0.0011283015,0.0011861377,0.0037009863,0.002527632,0.0067883474,0.0024148917,0.004188313,0.010194013],"category_scores_gemma":[0.1054075,0.000487138,0.0015163111,0.0011257216,0.0029397826,0.006218846,0.005491147,0.0068529067,0.0021399485],"study_design_candidate":"theoretical_or_conceptual","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":[0.00038900628,0.0009236855,0.0125252595,0.0007264612,0.00041852557,0.00036511972,0.0062945895,0.008076996,0.0009580982,0.07893682,0.056410603,0.8339749],"study_design_scores_gemma":[0.00058789115,0.0010411248,0.014392748,0.0055509945,0.000534153,0.0013146248,0.009430106,0.015587229,0.0030612615,0.6914713,0.25667474,0.0003539476],"about_ca_topic_score_codex":0.0018434024,"about_ca_topic_score_gemma":0.0037356508,"teacher_disagreement_score":0.03925864,"about_ca_system_score_codex":0.0038423124,"about_ca_system_score_gemma":0.010997601,"threshold_uncertainty_score":0.20762205},"labels":[],"label_agreement":null},{"id":"W2060526915","doi":"10.1177/0163278702250082","title":"A Validity Study Of Expert Judgment Procedures For Setting Cutoff Scores On High-Stakes Credentialing Examinations Using Cluster Analysis","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":21,"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 Calgary; Royal College of Physicians and Surgeons of Canada; Scanimetrics (Canada)","funders":"","keywords":"Cutoff; Licensure; Credentialing; Cluster (spacecraft); Categorization; Clinical judgment; Test (biology); Psychology; Medical physics; Statistics; Medicine; Computer science; Artificial intelligence; Medical education; Mathematics","score_opus":0.2742225804547531,"score_gpt":0.5179446169931036,"score_spread":0.24372203653835045,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2060526915","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":[0.83738506,0.000391745,0.15298733,0.00045204748,0.00016907671,0.0012268499,0.0001749376,0.00017398092,0.0070388904],"genre_scores_gemma":[0.9595755,0.00005437357,0.03929709,0.00005950716,0.000032218682,0.00055730675,0.00017456534,0.000039208357,0.00021029246],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.77511215,0.16766907,0.010388376,0.008682983,0.036360312,0.0017870336],"domain_scores_gemma":[0.35635564,0.5288065,0.025630645,0.032993097,0.054308135,0.0019060229],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17572579,0.0006286231,0.0006838363,0.0048751007,0.0017617606,0.002454167,0.0019991514,0.0012602456,0.00095480535],"category_scores_gemma":[0.53401935,0.0005000977,0.0012107025,0.00225809,0.004028382,0.002579726,0.002456891,0.0011072403,0.0003692811],"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":[0.0026918277,0.0006596837,0.80367434,0.00051520934,0.0015199851,0.00012428913,0.01908545,0.0072996765,0.0021152731,0.015467138,0.002162136,0.1446851],"study_design_scores_gemma":[0.0010121345,0.0036126713,0.70712215,0.0010201909,0.0006800433,0.0010643598,0.016162986,0.22141576,0.013792354,0.026621358,0.006979767,0.00051626243],"about_ca_topic_score_codex":0.0046265726,"about_ca_topic_score_gemma":0.0050236443,"teacher_disagreement_score":0.17572579,"about_ca_system_score_codex":0.0025423937,"about_ca_system_score_gemma":0.0037447375,"threshold_uncertainty_score":0.9293379},"labels":[],"label_agreement":null},{"id":"W2060962391","doi":"10.1177/01632780122034975","title":"Location of Death in Canada","year":2001,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Global Health Care Issues","field":"Health Professions","cited_by":55,"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 Lethbridge; University of Alberta","funders":"","keywords":"Medicine; Cause of death; Health care; Demography; Emergency medicine; Population; Place of death; Hospital care; Medical emergency; Environmental health; Palliative care; Disease; Nursing","score_opus":0.22030024043533963,"score_gpt":0.5384405973095842,"score_spread":0.3181403568742445,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2060962391","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":[0.87609535,0.013648618,0.00057985075,0.004631082,0.000374568,0.000109414716,0.042808518,0.00015217204,0.06160042],"genre_scores_gemma":[0.9770622,0.0052391314,0.00033831713,0.0003481731,0.000047033907,0.000021335938,0.008170073,0.0000276802,0.008745983],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99902904,0.00005003161,0.000049320806,0.00009769106,0.00038047868,0.0003934375],"domain_scores_gemma":[0.9970017,0.00008660282,0.00043658717,0.000045008586,0.0018133369,0.0006168556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000363123,0.0002379332,0.0002995995,0.0041315425,0.0029410154,0.0018070731,0.0011206357,0.00039192764,0.006910701],"category_scores_gemma":[0.002903739,0.00018748452,0.0007185538,0.006774607,0.00049667066,0.00052481215,0.0013283563,0.00069536694,0.00052369334],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019922978,0.00003573918,0.93462574,0.00022510828,0.00009780137,0.0004586906,0.0017771234,0.0005083254,0.00015516357,0.0021262416,0.017424148,0.042366736],"study_design_scores_gemma":[0.000014293163,0.000030657084,0.9754834,0.00014717647,0.000041782438,0.00038999648,0.0040000323,0.00034546468,0.00011576761,0.00018637453,0.019203406,0.00004169892],"about_ca_topic_score_codex":0.9948167,"about_ca_topic_score_gemma":0.9971976,"teacher_disagreement_score":0.035240598,"about_ca_system_score_codex":0.035240598,"about_ca_system_score_gemma":0.039854772,"threshold_uncertainty_score":0.25568962},"labels":[],"label_agreement":null},{"id":"W2096652611","doi":"10.1177/0163278702025003005","title":"A Qualitative Evaluation of an HIV/AIDS Respite Care Service in Victoria, Canada","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":3,"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 Victoria","funders":"","keywords":"Respite care; Participatory action research; Nursing; Qualitative research; Participatory evaluation; Action research; Medicine; Human immunodeficiency virus (HIV); Action (physics); Service (business); Psychology; Sociology; Family medicine; Pedagogy; Business; Social science","score_opus":0.533127960404624,"score_gpt":0.5721523280408563,"score_spread":0.03902436763623229,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096652611","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":[0.9627014,0.000695346,0.0017142572,0.003119387,0.000129947,0.0044084922,0.00048272536,0.000044376608,0.026704036],"genre_scores_gemma":[0.9802918,0.0009719781,0.0036781258,0.0012515488,0.00002633389,0.001740681,0.00017623839,0.00004678277,0.011816475],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.97440207,0.01667841,0.0006029716,0.000825053,0.0031058867,0.00438556],"domain_scores_gemma":[0.97319645,0.010553866,0.0009525962,0.00050092005,0.009103625,0.005692488],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017310183,0.00064203196,0.0007548195,0.0016558085,0.021644961,0.004577446,0.0030975745,0.0015495871,0.0048982534],"category_scores_gemma":[0.027449757,0.0007639985,0.000420117,0.0026780039,0.006923249,0.0013705916,0.005352293,0.0019111462,0.00037676032],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041386523,0.00077608833,0.0045791725,0.0016583257,0.000017174747,0.0022664613,0.9418623,0.0004497707,0.0021065427,0.0015460212,0.0034602957,0.040863924],"study_design_scores_gemma":[0.00005929074,0.0005389761,0.0069260946,0.00053966587,0.00001802398,0.000140773,0.9718338,0.0002382961,0.00076909026,0.00017570618,0.018723972,0.000036377714],"about_ca_topic_score_codex":0.62900716,"about_ca_topic_score_gemma":0.8561637,"teacher_disagreement_score":0.37099284,"about_ca_system_score_codex":0.068852775,"about_ca_system_score_gemma":0.10288271,"threshold_uncertainty_score":0.74635535},"labels":[],"label_agreement":null},{"id":"W2097743932","doi":"10.1177/01632780122034777","title":"Cost-Effectiveness of a Lottery for Increasing Physicians’ Responses to a Mail Survey","year":2001,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":32,"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 de Sherbrooke","funders":"","keywords":"Lottery; Respondent; Randomized controlled trial; Family medicine; Medicine; Significant difference; Psychology; Statistics; Mathematics; Surgery","score_opus":0.7326192029808507,"score_gpt":0.6326923150652278,"score_spread":0.09992688791562288,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097743932","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":[0.90361834,0.0030997957,0.023613544,0.0062083807,0.0028755376,0.04902284,0.00076915,0.0009303356,0.009862058],"genre_scores_gemma":[0.95554876,0.000495222,0.01305163,0.0013429255,0.00055953977,0.027208667,0.00017610041,0.000030345516,0.0015867912],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.87992126,0.11035994,0.0028022674,0.0019366685,0.0033187643,0.0016610653],"domain_scores_gemma":[0.79589564,0.17608288,0.013114066,0.006993772,0.0025360568,0.005377523],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05887738,0.0018823283,0.0033670892,0.0031215765,0.0009946224,0.001519928,0.002399043,0.006662036,0.014708214],"category_scores_gemma":[0.15157066,0.0014245436,0.0024490664,0.0015725693,0.0032395378,0.0032178022,0.002173678,0.0035011503,0.0012128644],"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":[0.806098,0.045073405,0.005263474,0.0031739478,0.0018567536,0.00012859472,0.0006450859,0.009237759,0.00085285836,0.0048632747,0.003735349,0.11907148],"study_design_scores_gemma":[0.67916936,0.2599884,0.0100770015,0.0010645369,0.0020176321,0.00015583982,0.00040637417,0.03389214,0.0019144776,0.0070506614,0.003994405,0.00026930997],"about_ca_topic_score_codex":0.0012600397,"about_ca_topic_score_gemma":0.00081999885,"teacher_disagreement_score":0.94112265,"about_ca_system_score_codex":0.002616745,"about_ca_system_score_gemma":0.0029394317,"threshold_uncertainty_score":0.31137705},"labels":[],"label_agreement":null},{"id":"W2100835342","doi":"10.1177/0163278703261213","title":"Multivariate Tests of Means in Independent Groups Designs","year":2004,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":13,"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 Manitoba","funders":"","keywords":"Multivariate statistics; Multivariate analysis; Statistics; Mathematics; Medicine","score_opus":0.49141777104692247,"score_gpt":0.5819482732124225,"score_spread":0.09053050216550002,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2100835342","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":[0.020534134,0.0010260942,0.95885575,0.00097152486,0.00077289547,0.0017503623,0.0010746531,0.0012128534,0.0138017405],"genre_scores_gemma":[0.20358472,0.0010337171,0.7792195,0.000723862,0.00065591355,0.011583526,0.00086822105,0.00031015402,0.0020203982],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8899647,0.08350471,0.0041825464,0.005986104,0.015226747,0.0011351625],"domain_scores_gemma":[0.68175286,0.27516353,0.016005805,0.016267607,0.009458062,0.0013521542],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06949889,0.0012270188,0.0018983695,0.0038088884,0.0012633115,0.0029140383,0.002390473,0.002104965,0.015607462],"category_scores_gemma":[0.33931473,0.00066236686,0.0016289575,0.0047796774,0.0047958586,0.004510424,0.0037436206,0.0043455623,0.0020701105],"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":[0.0015462895,0.0006821024,0.012836477,0.0022653139,0.0012061883,0.00032283983,0.0030135359,0.012442,0.0022757181,0.49374977,0.018646544,0.45101315],"study_design_scores_gemma":[0.0009289065,0.005812666,0.021601703,0.001361119,0.00051335455,0.0006402667,0.0020433634,0.0862652,0.0077593476,0.80576265,0.067035995,0.00027543033],"about_ca_topic_score_codex":0.0008184856,"about_ca_topic_score_gemma":0.0005332375,"teacher_disagreement_score":0.06949889,"about_ca_system_score_codex":0.0013462642,"about_ca_system_score_gemma":0.004286077,"threshold_uncertainty_score":0.36754966},"labels":[],"label_agreement":null},{"id":"W2100881081","doi":"10.1177/0163278713518942","title":"Notes From the Field","year":2014,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Central Venous Catheters and Hemodialysis","field":"Health Professions","cited_by":8,"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 British Columbia; University of Calgary","funders":"","keywords":"Modalities; Checklist; Modality (human–computer interaction); Video recording; Reliability (semiconductor); Video quality; Medicine; Medical physics; Computer science; Psychology; Multimedia; Artificial intelligence; Cognitive psychology; Operations management; Engineering","score_opus":0.1514922505540342,"score_gpt":0.487047537602202,"score_spread":0.3355552870481678,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2100881081","genre_codex":"other","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":[0.032950517,0.0077165714,0.007756724,0.04524455,0.056091905,0.0039943387,0.025046604,0.0013289801,0.81986976],"genre_scores_gemma":[0.054433048,0.0039503872,0.0071927146,0.011996561,0.0038045228,0.0016106644,0.0053696395,0.0004422376,0.91120017],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99793124,0.0006125469,0.000293503,0.00026089067,0.00070107204,0.00020076953],"domain_scores_gemma":[0.99164456,0.0028806035,0.00034265575,0.0010575553,0.0032377767,0.000836848],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002373806,0.0007191189,0.00041223873,0.0020700267,0.003049903,0.0015781102,0.0013303725,0.001454551,0.21268335],"category_scores_gemma":[0.014518452,0.00023775404,0.0003258503,0.0011876746,0.00059394614,0.0011337136,0.0025393772,0.0024816366,0.059901062],"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":[0.00021577996,0.00010285636,0.00073140796,0.0004244037,0.000005626249,0.0013547711,0.003325468,0.00005022901,0.0017802196,0.0017630429,0.8867002,0.103546016],"study_design_scores_gemma":[0.000016948075,0.000032259388,0.0012756098,0.000267067,0.0000027457372,0.0002381835,0.0028019212,0.00002696234,0.0005020371,0.00045838775,0.99436176,0.00001626387],"about_ca_topic_score_codex":0.011294558,"about_ca_topic_score_gemma":0.017132377,"teacher_disagreement_score":0.21268335,"about_ca_system_score_codex":0.0017927625,"about_ca_system_score_gemma":0.0045205927,"threshold_uncertainty_score":0.7114967},"labels":[],"label_agreement":null},{"id":"W2101902841","doi":"10.1177/0163278710393955","title":"Assessing Costs and Potential Returns of Evidence-Based Programs for Seniors","year":2010,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":10,"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":"National Center for Chronic Disease Prevention and Health Promotion; Centers for Disease Control and Prevention; AGE-WELL","keywords":"Public health; Psychological intervention; Cost–benefit analysis; Economic evaluation; Return on investment; Program evaluation; Cost effectiveness; Medicine; Actuarial science; Environmental health; Public economics; Business; Nursing; Economics; Risk analysis (engineering); Political science","score_opus":0.682390433676244,"score_gpt":0.5636710224035907,"score_spread":0.11871941127265329,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2101902841","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":[0.8708357,0.011913854,0.0594377,0.01033093,0.00032977184,0.004848421,0.003209518,0.0001563376,0.03893769],"genre_scores_gemma":[0.98379713,0.001823135,0.012549262,0.00018560115,0.000044312084,0.0006512704,0.00028150328,0.000007351975,0.00066033413],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9661894,0.026997682,0.001866072,0.00043787833,0.003605716,0.00090322405],"domain_scores_gemma":[0.9043171,0.08320734,0.005672834,0.0020571176,0.0036196813,0.0011259082],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03896425,0.0011127117,0.001287303,0.0042186757,0.00044911695,0.0029624824,0.0008660437,0.0015380372,0.0036446014],"category_scores_gemma":[0.15311207,0.00045269934,0.0027469688,0.0032971874,0.00097542425,0.0028954067,0.0024031755,0.0014353914,0.00016033927],"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":[0.007837874,0.002139264,0.16155744,0.0037493836,0.006055889,0.0003208766,0.00045936764,0.29777044,0.00057113863,0.07794436,0.003679014,0.43791497],"study_design_scores_gemma":[0.003170775,0.02572187,0.22744654,0.00689361,0.013435051,0.0009853583,0.0036044093,0.51727855,0.0054856553,0.17736726,0.018132577,0.00047837722],"about_ca_topic_score_codex":0.0036547938,"about_ca_topic_score_gemma":0.003592375,"teacher_disagreement_score":0.03896425,"about_ca_system_score_codex":0.005281201,"about_ca_system_score_gemma":0.004257883,"threshold_uncertainty_score":0.20606512},"labels":[],"label_agreement":null},{"id":"W2104638892","doi":"10.1177/0163278709356192","title":"Identifying the Unauthorized Use of Examination Material","year":2009,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Sherbrooke; Medical Council of Canada","funders":"","keywords":"Outlier; Computer science; Psychology; Artificial intelligence","score_opus":0.8765739870349761,"score_gpt":0.6346960718444731,"score_spread":0.24187791519050306,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2104638892","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":[0.90095925,0.00067770766,0.07934961,0.0012290475,0.00008838254,0.0018600314,0.0007311356,0.0002448937,0.014859929],"genre_scores_gemma":[0.9436345,0.00053787715,0.051829413,0.00019780885,0.000034365865,0.00061408523,0.0002901956,0.00003998748,0.0028215502],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9581514,0.02090407,0.006048871,0.002067918,0.011642921,0.001184687],"domain_scores_gemma":[0.7716456,0.14537348,0.0403411,0.029353436,0.012133299,0.0011530084],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.028332112,0.00038909685,0.0006921263,0.003911686,0.0012515438,0.002611327,0.0012820764,0.0010517667,0.0038577465],"category_scores_gemma":[0.18894723,0.00041758438,0.0009110569,0.0019743275,0.0016939114,0.0032258308,0.0046048136,0.0012356903,0.00095664704],"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":[0.0006644851,0.00056133827,0.55416745,0.0008371225,0.00022105937,0.0018008612,0.015829038,0.0018308918,0.0063628927,0.01681808,0.0029100773,0.39799666],"study_design_scores_gemma":[0.00014166259,0.0027153874,0.71751076,0.003434724,0.0005786498,0.015290515,0.019435182,0.042349253,0.082332104,0.055547483,0.060235187,0.00042896744],"about_ca_topic_score_codex":0.0012107625,"about_ca_topic_score_gemma":0.001600581,"teacher_disagreement_score":0.9989482,"about_ca_system_score_codex":0.00079269323,"about_ca_system_score_gemma":0.0025653117,"threshold_uncertainty_score":0.14983636},"labels":[],"label_agreement":null},{"id":"W2105174743","doi":"10.1177/0163278706297344","title":"Sampling Bias in an International Internet Survey of Diversion Programs in the Criminal Justice System","year":2007,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","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":"Lawson Health Research Institute","funders":"","keywords":"The Internet; Sampling frame; Terminology; Criminal justice; Sampling (signal processing); Sample (material); Sampling bias; Internet privacy; Computer science; World Wide Web; Psychology; Medical education; Public relations; Political science; Medicine; Sample size determination; Statistics; Environmental health; Telecommunications; Criminology","score_opus":0.8713175713779645,"score_gpt":0.6396185879627105,"score_spread":0.23169898341525397,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105174743","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":[0.90460944,0.0012644217,0.04737484,0.004251293,0.00032271713,0.013926557,0.0005944752,0.00012826668,0.027528025],"genre_scores_gemma":[0.96292734,0.0005869414,0.02062946,0.004351279,0.00019609217,0.009250446,0.000435989,0.000050368653,0.0015721498],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.6469906,0.30776536,0.019722342,0.0039594267,0.018068446,0.0034937253],"domain_scores_gemma":[0.8029818,0.12963243,0.03058032,0.019129092,0.016076071,0.0016002438],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.18003,0.0005391609,0.0006475408,0.0032013631,0.0020037815,0.0028215996,0.0012563247,0.0013095202,0.001660257],"category_scores_gemma":[0.286569,0.00064555975,0.00074902695,0.0043272492,0.003047141,0.0021520949,0.0030242486,0.0014657073,0.0005442773],"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":[0.0005828744,0.00092339766,0.8343915,0.00062981574,0.00025231062,0.00035614346,0.041199185,0.0007303639,0.00074484537,0.008666204,0.0077010076,0.10382235],"study_design_scores_gemma":[0.00044451683,0.0033944927,0.86234283,0.0033133875,0.00052783143,0.0019290339,0.06292177,0.012443625,0.0058234367,0.010924262,0.035740647,0.00019423723],"about_ca_topic_score_codex":0.006251474,"about_ca_topic_score_gemma":0.005675048,"teacher_disagreement_score":0.81997,"about_ca_system_score_codex":0.0022782094,"about_ca_system_score_gemma":0.0027023905,"threshold_uncertainty_score":0.95210105},"labels":[],"label_agreement":null},{"id":"W2113224507","doi":"10.1177/0163278702025001005","title":"The Contribution of Handsearching European General Health Care Journals to the Cochrane Controlled Trials Register","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cochrane","funders":"","keywords":"MEDLINE; Cochrane Library; Medicine; Systematic review; Randomized controlled trial; Meta-analysis; Family medicine; Cochrane collaboration; Internal medicine","score_opus":0.9027067621472988,"score_gpt":0.6861497762688541,"score_spread":0.21655698587844474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113224507","genre_codex":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018619217,0.70262605,0.08980293,0.106271304,0.018585771,0.024621353,0.0050201723,0.0015384135,0.032914806],"genre_scores_gemma":[0.13047352,0.42682657,0.31565064,0.045911666,0.013979476,0.051533148,0.008365663,0.0016473245,0.0056119515],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.13701256,0.5243719,0.2525538,0.01193615,0.07188444,0.0022411486],"domain_scores_gemma":[0.026622197,0.7321842,0.075579286,0.06778408,0.09423509,0.0035951226],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6795925,0.004615126,0.017513078,0.10202609,0.004732605,0.021103242,0.010082907,0.010192872,0.013047953],"category_scores_gemma":[0.9238434,0.005077818,0.008396727,0.08163504,0.008622702,0.030249543,0.019686367,0.008774115,0.004665864],"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":[0.0019999861,0.00030099822,0.014553687,0.1728041,0.005057372,0.0006117711,0.004840927,0.0005883536,0.0013212397,0.0076020355,0.04003371,0.7502858],"study_design_scores_gemma":[0.0050079986,0.0021707104,0.042574536,0.52226174,0.033161744,0.0037647455,0.0047628554,0.0037566223,0.0036598498,0.033294044,0.34451288,0.0010722661],"about_ca_topic_score_codex":0.014275028,"about_ca_topic_score_gemma":0.026372116,"teacher_disagreement_score":0.3204075,"about_ca_system_score_codex":0.012928332,"about_ca_system_score_gemma":0.08121572,"threshold_uncertainty_score":0.39511955},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2116016506","doi":"10.1177/0163278710393954","title":"“It’s a Feel. That’s What a Lot of Our Evidence Would Consist of ”: Public Health Practitioners’ Perspectives on Evidence","year":2011,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Public Health Policies and Education","field":"Health Professions","cited_by":37,"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 British Columbia; Simon Fraser University; Island Health; Vancouver Coastal Health; University of Victoria","funders":"Canadian Institutes of Health Research","keywords":"Grassroots; Evidence-based practice; Tacit knowledge; Public health; Public relations; Discipline; Work (physics); Medical education; Medicine; Psychology; Nursing; Sociology; Alternative medicine; Political science; Knowledge management; Politics; Social science","score_opus":0.7601847510600535,"score_gpt":0.6064561193109729,"score_spread":0.15372863174908058,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2116016506","genre_codex":"commentary","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":[0.020032119,0.039131165,0.007123019,0.9152282,0.0018118,0.000085894695,0.000032893924,0.000027357439,0.01652754],"genre_scores_gemma":[0.701961,0.039445095,0.020841714,0.23084706,0.0031196773,0.00037872788,0.00006201232,0.00014410623,0.0032005261],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.5998191,0.33518016,0.017501608,0.0060642,0.033758216,0.0076767937],"domain_scores_gemma":[0.33693218,0.6046368,0.012042699,0.006513566,0.032526407,0.007348372],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.29452863,0.0007988715,0.002212422,0.00905102,0.017477578,0.03865499,0.0050221314,0.022324858,0.0014768604],"category_scores_gemma":[0.33223665,0.0022402264,0.0014551677,0.0070969765,0.088645145,0.032657176,0.022546329,0.04500762,0.0003549901],"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":[0.00009344652,0.00007687946,0.0041465214,0.0031928427,0.00021462866,0.0035475418,0.7490045,0.0003840304,0.00083661254,0.15661195,0.032012425,0.04987865],"study_design_scores_gemma":[0.00007103544,0.00014098955,0.001703656,0.011991163,0.000115074625,0.0020037775,0.635605,0.00035324093,0.0005413988,0.0962496,0.25106376,0.00016126192],"about_ca_topic_score_codex":0.018147243,"about_ca_topic_score_gemma":0.016241133,"teacher_disagreement_score":0.7054714,"about_ca_system_score_codex":0.029881772,"about_ca_system_score_gemma":0.04991816,"threshold_uncertainty_score":0.8699718},"labels":[],"label_agreement":null},{"id":"W2119103520","doi":"10.1177/0163278702238054","title":"A Signal Detection Theory Analysis of the Placebo Effect","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Pain Management and Placebo Effect","field":"Neuroscience","cited_by":69,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Placebo; Detection theory; Placebo response; Psychology; Medicine; Computer science; Alternative medicine","score_opus":0.10630517894842421,"score_gpt":0.38858029003386013,"score_spread":0.2822751110854359,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2119103520","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":[0.0102313375,0.0033457528,0.94833463,0.007370354,0.0005883656,0.0001708321,0.00013118416,0.00016077587,0.029666757],"genre_scores_gemma":[0.7980998,0.005076476,0.17004752,0.0043722163,0.0022504844,0.000618973,0.00013434535,0.0001110507,0.019289063],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9950854,0.002691309,0.00017873805,0.00067731837,0.0011351969,0.00023189529],"domain_scores_gemma":[0.9720681,0.024140732,0.0010734296,0.00085182313,0.0016417075,0.00022431252],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01122827,0.0011374581,0.0012879297,0.0017399564,0.00069189613,0.002468664,0.0014080254,0.0025598346,0.011730485],"category_scores_gemma":[0.03842112,0.00046615992,0.0015129636,0.0009539238,0.004259355,0.0033342869,0.0011801972,0.0035035901,0.0010432096],"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":[0.00016988502,0.000054436274,0.00051528495,0.00026111602,0.00010680032,0.00019336292,0.00015102727,0.017234208,0.001381695,0.9460642,0.002118804,0.031749126],"study_design_scores_gemma":[0.000090136185,0.00022092309,0.0005865531,0.00009650173,0.00007097401,0.00023684507,0.000036924357,0.08775645,0.0007934398,0.9060444,0.004036341,0.00003039662],"about_ca_topic_score_codex":0.00094669807,"about_ca_topic_score_gemma":0.0003250098,"teacher_disagreement_score":0.011730485,"about_ca_system_score_codex":0.0019706094,"about_ca_system_score_gemma":0.0016044718,"threshold_uncertainty_score":0.059381425},"labels":[],"label_agreement":null},{"id":"W2133192384","doi":"10.1177/0163278702025001008","title":"A Comparison of the Quality of Cochrane Reviews and Systematic Reviews Published in Paper-Based Journals","year":2002,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":110,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Children's Hospital of Eastern Ontario; Institute of Population and Public Health; University of Ottawa","funders":"","keywords":"Systematic review; Quality (philosophy); Cochrane collaboration; MEDLINE; Medicine; Medical education; Psychology; Political science","score_opus":0.9446355596120252,"score_gpt":0.7095874658572304,"score_spread":0.23504809375479474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133192384","genre_codex":"review","genre_gemma":"empirical","domain_codex":"reporting","domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06309073,0.8820808,0.02216774,0.0055889776,0.003404017,0.00837094,0.0060226894,0.00025458308,0.009019559],"genre_scores_gemma":[0.55875576,0.33757326,0.064743504,0.0037124355,0.0030771098,0.024469953,0.006292014,0.0003420832,0.0010338317],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.28192326,0.29635265,0.30343884,0.010214938,0.10644132,0.0016289694],"domain_scores_gemma":[0.085716754,0.6784736,0.14812095,0.015296899,0.06976106,0.002630746],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.42259964,0.0020478228,0.01274055,0.08956971,0.0017359358,0.014573694,0.0030107198,0.003444249,0.0032406852],"category_scores_gemma":[0.854129,0.002155375,0.01477384,0.091852814,0.0049339053,0.009875611,0.0071950993,0.0023389715,0.00046494152],"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":[0.0072631785,0.00038873623,0.11322865,0.39710718,0.16500212,0.00056780665,0.009574705,0.002062085,0.00078647496,0.008588951,0.0062116734,0.28921852],"study_design_scores_gemma":[0.010964102,0.006909023,0.3333362,0.33236644,0.17599186,0.0054296562,0.012817503,0.0034246158,0.002728647,0.036808662,0.07800059,0.0012228277],"about_ca_topic_score_codex":0.002871848,"about_ca_topic_score_gemma":0.006059153,"teacher_disagreement_score":0.5774003,"about_ca_system_score_codex":0.010288583,"about_ca_system_score_gemma":0.013940272,"threshold_uncertainty_score":0.71203744},"labels":[],"label_agreement":null},{"id":"W2134265312","doi":"10.1177/0163278709338561","title":"Medical Record Review Conduction Model for Improving Interrater Reliability of Abstracting Medical-Related Information","year":2009,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Electronic Health Records Systems","field":"Health Professions","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Inter-rater reliability; Reliability (semiconductor); Medical record; Medical information; Psychology; Medicine; Family medicine; Internal medicine; Developmental psychology","score_opus":0.17414370332611645,"score_gpt":0.5366436959399579,"score_spread":0.3624999926138414,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134265312","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":[0.02757074,0.0009190796,0.9456509,0.002383454,0.00033822964,0.013649479,0.00043652405,0.0027075806,0.006344069],"genre_scores_gemma":[0.13221894,0.00048056885,0.8508326,0.0006609738,0.00017327507,0.013437223,0.0005142203,0.0002420151,0.0014401901],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.4661978,0.4419437,0.031570427,0.015577798,0.0427511,0.001959275],"domain_scores_gemma":[0.3149949,0.44178274,0.058473863,0.048810367,0.13284661,0.003091486],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.38292578,0.0021711562,0.0025677434,0.007971213,0.0029423789,0.0057509504,0.0053616758,0.0024851228,0.0025488897],"category_scores_gemma":[0.53883225,0.0018605759,0.0035326688,0.0061298707,0.0037383083,0.009083599,0.006219733,0.003221905,0.003060449],"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":[0.002400611,0.0014874657,0.13399315,0.005749167,0.0019077974,0.00039436924,0.019798962,0.028278533,0.005643853,0.040847193,0.022973891,0.73652494],"study_design_scores_gemma":[0.0024784713,0.010853656,0.10255241,0.007117652,0.0035063003,0.0028967508,0.007848274,0.69037664,0.021279965,0.07653934,0.07310417,0.0014463525],"about_ca_topic_score_codex":0.0057944725,"about_ca_topic_score_gemma":0.008154671,"teacher_disagreement_score":0.61707425,"about_ca_system_score_codex":0.006480835,"about_ca_system_score_gemma":0.019422213,"threshold_uncertainty_score":0.76096237},"labels":[],"label_agreement":null},{"id":"W2135260555","doi":"10.1177/016327870102400202","title":"Is There a “Best” Way to Detect and Minimize Publication Bias?","year":2001,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":58,"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; University of Alberta; McGill University; Children's Hospital of Eastern Ontario","funders":"","keywords":"Funnel plot; Publication bias; Statistics; Meta-analysis; Robustness (evolution); Standard error; Reliability (semiconductor); Forest plot; Mathematics; Econometrics; Medicine; Computer science; Confidence interval; Internal medicine","score_opus":0.9114822169183275,"score_gpt":0.6483470983849812,"score_spread":0.26313511853334637,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2135260555","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":[0.06566873,0.22509636,0.6721869,0.018851914,0.0024873202,0.010182352,0.0017977941,0.0017929297,0.0019356192],"genre_scores_gemma":[0.256259,0.02095489,0.70732075,0.0032502867,0.000731418,0.010559223,0.0004855364,0.0002599603,0.0001789575],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.54684895,0.35506207,0.072920986,0.008941449,0.015020078,0.0012064622],"domain_scores_gemma":[0.3684985,0.5362231,0.042037003,0.03165667,0.020153133,0.0014315702],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.48051253,0.005309495,0.016953278,0.019256039,0.0020284131,0.008509585,0.004248428,0.006989868,0.002306955],"category_scores_gemma":[0.7024023,0.0030181159,0.02351847,0.012003341,0.004162392,0.0115055125,0.0048236935,0.003521599,0.00067222305],"study_design_candidate":"theoretical_or_conceptual","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":[0.012479113,0.00050527876,0.057213724,0.115650356,0.26014772,0.00070713955,0.0035611086,0.008389769,0.0032691786,0.008879373,0.0054690377,0.52372825],"study_design_scores_gemma":[0.028289082,0.015116335,0.05887859,0.05657689,0.6331253,0.0029346903,0.002350297,0.045779966,0.017282698,0.11260337,0.024762549,0.0023002585],"about_ca_topic_score_codex":0.00080473133,"about_ca_topic_score_gemma":0.0019714998,"teacher_disagreement_score":0.5194875,"about_ca_system_score_codex":0.003183314,"about_ca_system_score_gemma":0.0055009695,"threshold_uncertainty_score":0.6406205},"labels":[],"label_agreement":null},{"id":"W2135855338","doi":"10.1177/0163278711424281","title":"Contrasting Case Definitions for Chronic Fatigue Syndrome, Myalgic Encephalomyelitis/Chronic Fatigue Syndrome and Myalgic Encephalomyelitis","year":2011,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Fibromyalgia and Chronic Fatigue Syndrome Research","field":"Medicine","cited_by":61,"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 Institute of Allergy and Infectious Diseases; U.S. Public Health Service","keywords":"Chronic fatigue syndrome; Encephalomyelitis; Medicine; Cognition; Population; Physical therapy; Psychology; Psychiatry; Multiple sclerosis","score_opus":0.26533464183921646,"score_gpt":0.4143584613140521,"score_spread":0.14902381947483562,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2135855338","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":[0.96690714,0.012560506,0.0049379524,0.002563498,0.0005157789,0.0009371079,0.000525552,0.000020479483,0.011031936],"genre_scores_gemma":[0.98562163,0.002598787,0.006289443,0.0015353241,0.00031385597,0.0015184677,0.0018113533,0.000013470312,0.000297593],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9640866,0.025943,0.003714822,0.0011523016,0.0037743028,0.0013289638],"domain_scores_gemma":[0.9249743,0.057681434,0.010751655,0.0017617585,0.0037196402,0.0011112252],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.020846905,0.00046584278,0.0007738243,0.004295087,0.001074875,0.0012854554,0.0012030293,0.00095795497,0.0008124409],"category_scores_gemma":[0.080406,0.0002235617,0.0006663685,0.0026521601,0.0019723685,0.0014492301,0.0028642155,0.001089319,0.00015115172],"study_design_candidate":"theoretical_or_conceptual","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":[0.0017843215,0.00041758842,0.8864582,0.001193843,0.00030495765,0.003317836,0.014780827,0.00095679867,0.00097488775,0.025385983,0.007858854,0.056565873],"study_design_scores_gemma":[0.000510343,0.0013297048,0.90548426,0.0020849912,0.00026124527,0.027332122,0.02288558,0.0025234567,0.00079516653,0.013128172,0.023529135,0.00013584968],"about_ca_topic_score_codex":0.0070687667,"about_ca_topic_score_gemma":0.012820495,"teacher_disagreement_score":0.9791531,"about_ca_system_score_codex":0.001954273,"about_ca_system_score_gemma":0.0015974408,"threshold_uncertainty_score":0.110250294},"labels":[],"label_agreement":null},{"id":"W2138928670","doi":"10.1177/0163278711425041","title":"A Comparison of Empirical Ranking Methods of Frequency and Severity Ratings of Clinical Presentations","year":2011,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","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":"Health Sciences Centre; University of Calgary","funders":"","keywords":"Ranking (information retrieval); Statistics; Psychology; Medicine; MEDLINE; Econometrics; Clinical psychology; Computer science; Mathematics; Artificial intelligence; Biology","score_opus":0.893414992052289,"score_gpt":0.7377671514789872,"score_spread":0.15564784057330183,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138928670","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":[0.6435356,0.004562436,0.33068556,0.0010849145,0.0005143899,0.0017884658,0.0010912062,0.00059336517,0.016144069],"genre_scores_gemma":[0.9085012,0.0006964726,0.08849127,0.000097625525,0.00012313173,0.00080712646,0.0004901586,0.00007305875,0.0007199694],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7082744,0.24139306,0.010089016,0.0051638274,0.034149844,0.0009298501],"domain_scores_gemma":[0.26345524,0.6761554,0.017676327,0.014593705,0.027519075,0.00060034206],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1339853,0.0009208838,0.0010043886,0.008798724,0.0007608257,0.0025371416,0.0014870132,0.00092150347,0.0019250914],"category_scores_gemma":[0.4594454,0.0005082301,0.0014420885,0.005370594,0.0019616634,0.003054311,0.0018910113,0.0010475348,0.0005414248],"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":[0.0025153693,0.0009047591,0.47032908,0.0022239122,0.0031968553,0.00012943747,0.011964267,0.012817279,0.0012897016,0.015299257,0.004441934,0.47488824],"study_design_scores_gemma":[0.0012201304,0.006131861,0.6722809,0.0020821919,0.0009824446,0.0011375778,0.019977037,0.24547864,0.004559177,0.032644086,0.012627786,0.0008782769],"about_ca_topic_score_codex":0.0025381178,"about_ca_topic_score_gemma":0.0032409765,"teacher_disagreement_score":0.8660147,"about_ca_system_score_codex":0.0017981157,"about_ca_system_score_gemma":0.0015313625,"threshold_uncertainty_score":0.70859045},"labels":[],"label_agreement":null},{"id":"W2145344077","doi":"10.1177/01632780022034633","title":"An Eclectic Model for Evaluating Web Based Continuing Medical Education Courseware Systems","year":2000,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Medical education; Continuing medical education; Computer science; Continuing education; Medicine; Multimedia; World Wide Web","score_opus":0.1326738158357816,"score_gpt":0.5477048575227006,"score_spread":0.41503104168691896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2145344077","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":[0.10591976,0.0009441265,0.8486038,0.0022756606,0.000065520246,0.017557295,0.000551553,0.0005050632,0.023577202],"genre_scores_gemma":[0.21327381,0.00026182694,0.77233917,0.00033387868,0.000009553249,0.012953026,0.00020804786,0.000021019021,0.00059962965],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8613574,0.112106554,0.007236283,0.0032368382,0.015235871,0.00082710997],"domain_scores_gemma":[0.8155071,0.15604247,0.007892456,0.0030059335,0.016554948,0.0009970278],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.072029546,0.002149621,0.0013363824,0.00759126,0.0011614304,0.005449115,0.0025465386,0.0019495228,0.0028914309],"category_scores_gemma":[0.15931861,0.0009938523,0.0020296723,0.003983382,0.002983724,0.0072833314,0.0037085828,0.002089635,0.0006922127],"study_design_candidate":"theoretical_or_conceptual","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":[0.0041373856,0.005107104,0.0504876,0.007697228,0.0017900799,0.00040623287,0.014687595,0.113652036,0.010925256,0.2683053,0.0046233716,0.5181809],"study_design_scores_gemma":[0.0038439685,0.013639208,0.021976165,0.0037416755,0.0015328357,0.00052977627,0.0058548604,0.7001426,0.016535833,0.21396914,0.017759215,0.0004747145],"about_ca_topic_score_codex":0.002276438,"about_ca_topic_score_gemma":0.0024687578,"teacher_disagreement_score":0.072029546,"about_ca_system_score_codex":0.01026552,"about_ca_system_score_gemma":0.0076980144,"threshold_uncertainty_score":0.38093323},"labels":[],"label_agreement":null},{"id":"W2147972149","doi":"10.1177/0163278704267036","title":"Evaluating Institutionalization by Comparing the Use of Health Services before and after Admission to a Long-Term-Care Facility","year":2004,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Alberta Health; University of Alberta","funders":"","keywords":"Long-term care; Medicine; Disadvantaged; Institutionalisation; Ambulatory care; Hospital admission; Health care; Population; Type of service; Gerontology; Family medicine; Nursing; Environmental health; Business; Service (business); Psychiatry","score_opus":0.3600581615283008,"score_gpt":0.5633035789381972,"score_spread":0.20324541740989638,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2147972149","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":[0.0017731896,0.9977048,0.000036711655,0.000046215187,0.000025038255,0.000028820616,0.00008158146,0.0000013578748,0.00030229322],"genre_scores_gemma":[0.011558596,0.98785675,0.00025024646,0.00006035502,0.00001989669,0.00004312873,0.00013100797,6.7143714e-7,0.00007935833],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9986191,0.0005164919,0.00032751885,0.00014333449,0.0003398487,0.000053690856],"domain_scores_gemma":[0.9944634,0.0031930269,0.0011764584,0.000082635175,0.0009946667,0.000089810565],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0038753452,0.0005631775,0.0024064516,0.0051125893,0.00021692697,0.001059419,0.0007887524,0.00053336687,0.0012648604],"category_scores_gemma":[0.008401362,0.00025746826,0.001521615,0.008066801,0.00039081764,0.0006357409,0.00039341804,0.00034958287,0.0002294649],"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":[0.00043382443,0.00012226723,0.02559291,0.16277842,0.0036525377,0.00014181949,0.00024227946,0.00020638708,0.0003146526,0.00033908364,0.0030325828,0.8031432],"study_design_scores_gemma":[0.00061715976,0.0043056994,0.39272454,0.3026702,0.049832392,0.0036262516,0.0030992716,0.0005211417,0.0024320127,0.0009450842,0.2390778,0.00014844848],"about_ca_topic_score_codex":0.01524092,"about_ca_topic_score_gemma":0.03464457,"teacher_disagreement_score":0.01524092,"about_ca_system_score_codex":0.0011102146,"about_ca_system_score_gemma":0.0033048182,"threshold_uncertainty_score":0.030304372},"labels":[],"label_agreement":null},{"id":"W2152382065","doi":"10.1177/0163278713488117","title":"Generalizing the Findings From Group Dynamics–Based Physical Activity Research to Practice Settings","year":2013,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Physical Activity and Health","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Group dynamic; Group (periodic table); Dynamics (music); Psychology; Medicine; Social psychology; Pedagogy; Chemistry","score_opus":0.5393916580775326,"score_gpt":0.6279639342730506,"score_spread":0.088572276195518,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2152382065","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":[0.015472814,0.88097745,0.02883696,0.007864584,0.0028763283,0.044522516,0.00552383,0.00020340492,0.01372215],"genre_scores_gemma":[0.18482208,0.6330137,0.072660364,0.020714076,0.001542798,0.07758681,0.0073569375,0.00025411064,0.002049176],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.77531147,0.10171829,0.07970438,0.0101416875,0.029996619,0.003127677],"domain_scores_gemma":[0.5390009,0.34877092,0.03307857,0.024236107,0.053809654,0.001103852],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19015189,0.002515989,0.009646486,0.04933267,0.0016553088,0.009532754,0.0033738248,0.0031685429,0.009942214],"category_scores_gemma":[0.5257026,0.002454739,0.012757319,0.034261093,0.0031202429,0.009873411,0.007468815,0.0027204135,0.0018660227],"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":[0.00031631012,0.00008823069,0.0043556173,0.87485504,0.014694392,0.00026272706,0.0034389372,0.00028972913,0.0007676585,0.0025977185,0.0034032674,0.09493038],"study_design_scores_gemma":[0.0007348816,0.0011045038,0.02035072,0.8530848,0.057350762,0.0006033205,0.0050774817,0.00035437188,0.0010561545,0.0072312104,0.05294931,0.000102445454],"about_ca_topic_score_codex":0.008209965,"about_ca_topic_score_gemma":0.02332727,"teacher_disagreement_score":0.19015189,"about_ca_system_score_codex":0.007464896,"about_ca_system_score_gemma":0.029965553,"threshold_uncertainty_score":0.9986869},"labels":[],"label_agreement":null},{"id":"W2156926979","doi":"10.1177/0163278706293400","title":"Developing Optimal Search Strategies for Retrieving Qualitative Studies in PsycINFO","year":2006,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"PsycINFO; Qualitative research; Search engine indexing; Computer science; MEDLINE; Information retrieval; Qualitative property; Psychology; Machine learning","score_opus":0.8112156344634597,"score_gpt":0.7367197101535863,"score_spread":0.0744959243098734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2156926979","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":[0.030783102,0.050352823,0.48017657,0.015176481,0.0015035941,0.35833886,0.031232648,0.006442388,0.025993502],"genre_scores_gemma":[0.025928786,0.016173316,0.74321985,0.00079269096,0.00012736344,0.20881078,0.0032689117,0.00034691303,0.001331403],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.787705,0.13552624,0.06538658,0.0026537988,0.0075249323,0.0012033836],"domain_scores_gemma":[0.45334768,0.4882955,0.019959193,0.012901694,0.024001794,0.0014940392],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.18193303,0.0030430327,0.0066285324,0.057260565,0.0023322008,0.007711102,0.0048962617,0.003759201,0.041318543],"category_scores_gemma":[0.5119656,0.0037038235,0.0047778725,0.05075593,0.0021482327,0.0127018215,0.0068928525,0.0026990024,0.009696496],"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":[0.0022327108,0.00033173707,0.0024897037,0.22736558,0.001373743,0.0010089355,0.014057999,0.0022202395,0.0028345776,0.0138229355,0.032259423,0.7000023],"study_design_scores_gemma":[0.01420475,0.0034370606,0.015612455,0.39160487,0.015443579,0.0025870318,0.049281742,0.021470968,0.015934853,0.11888734,0.34972274,0.0018126161],"about_ca_topic_score_codex":0.0049668476,"about_ca_topic_score_gemma":0.0105070565,"teacher_disagreement_score":0.81806695,"about_ca_system_score_codex":0.0064690374,"about_ca_system_score_gemma":0.020676395,"threshold_uncertainty_score":0.96216536},"labels":[],"label_agreement":null},{"id":"W2164610426","doi":"10.1177/0163278704267043","title":"Standardized Assessment of Reasoning in Contexts of Uncertainty","year":2004,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":153,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Psychology; Standardized test; Medical physics; Computer science; Natural language processing; Medical education; Mathematics education; Medicine","score_opus":0.10049369076300795,"score_gpt":0.5254536042443122,"score_spread":0.42495991348130424,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164610426","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":[0.8314853,0.0019099315,0.13251308,0.0006171973,0.00028744238,0.005166412,0.0009864346,0.0005617639,0.026472505],"genre_scores_gemma":[0.8751742,0.001336359,0.11744935,0.00012443382,0.000063231084,0.0025556902,0.0010317771,0.000039049904,0.0022258116],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98858345,0.006035696,0.0021575857,0.0005238253,0.0023512281,0.00034824124],"domain_scores_gemma":[0.9782613,0.009486073,0.0041559525,0.0018858679,0.0053582205,0.0008526546],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008924841,0.0004959799,0.00079223217,0.0033456488,0.0004517167,0.0014668907,0.0006789284,0.00061500655,0.0018601444],"category_scores_gemma":[0.03761058,0.00021557235,0.00077283056,0.0015769642,0.001376806,0.0017941335,0.002114507,0.0009496404,0.00041407428],"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":[0.0019401367,0.0016290679,0.24937943,0.0020690635,0.00030022155,0.0010420264,0.011222789,0.008847052,0.022519927,0.022441573,0.007561297,0.6710474],"study_design_scores_gemma":[0.0007091098,0.011845827,0.720677,0.0019132548,0.00043109307,0.006771906,0.018753594,0.046841003,0.059624948,0.08833673,0.04345666,0.00063886656],"about_ca_topic_score_codex":0.00069429097,"about_ca_topic_score_gemma":0.0014681023,"teacher_disagreement_score":0.008924841,"about_ca_system_score_codex":0.00084355415,"about_ca_system_score_gemma":0.002585997,"threshold_uncertainty_score":0.047199667},"labels":[],"label_agreement":null},{"id":"W2169790552","doi":"10.1177/0163278713491267","title":"The Effect of an Organizational Network for Patient Safety on Safety Event Reporting","year":2013,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Patient Safety and Medication Errors","field":"Health Professions","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":"Mount Sinai Hospital; Institute for Clinical Evaluative Sciences; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Canadian Patient Safety Institute","keywords":"Patient safety; Event (particle physics); Psychology; Medicine; Medical emergency; Business; Health care; Political science","score_opus":0.09708794570485801,"score_gpt":0.48589048263401063,"score_spread":0.3888025369291526,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169790552","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":[0.9894962,0.00041262197,0.0011822241,0.002192275,0.000078244826,0.0002800436,0.00007624824,0.00003647923,0.0062456923],"genre_scores_gemma":[0.9972518,0.00016511431,0.0018019978,0.0002288785,0.00006102894,0.0001303096,0.000045215183,0.0000036779934,0.00031207787],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97190213,0.019693322,0.0012263557,0.00094067416,0.004916921,0.0013205046],"domain_scores_gemma":[0.81397974,0.12189312,0.034667518,0.0071246577,0.009524676,0.012810278],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023821626,0.0003152866,0.00022889329,0.0012656697,0.0012753117,0.0014576154,0.0011438727,0.0007532042,0.0028127946],"category_scores_gemma":[0.09566262,0.00019110978,0.00063245115,0.0007931057,0.0009441256,0.002471553,0.0034970385,0.0011369647,0.00020632251],"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":[0.0016250487,0.005783353,0.73633724,0.00037479683,0.00030572322,0.00021052246,0.002286668,0.0032047788,0.001027126,0.001004415,0.0013017777,0.24653853],"study_design_scores_gemma":[0.0001193795,0.0062970524,0.9828718,0.00028801415,0.00017180813,0.00019960947,0.0031941717,0.0037420208,0.0007355596,0.00042707555,0.001913653,0.000039803963],"about_ca_topic_score_codex":0.0055435756,"about_ca_topic_score_gemma":0.010271002,"teacher_disagreement_score":0.023821626,"about_ca_system_score_codex":0.0031635934,"about_ca_system_score_gemma":0.0050234324,"threshold_uncertainty_score":0.12598228},"labels":[],"label_agreement":null},{"id":"W2337224824","doi":"10.1177/0163278715616440","title":"How Should Alcohol Problems Be Conceptualized? Causal Indicators Within the Rutgers Alcohol Problem Index","year":2015,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Substance Abuse Treatment and Outcomes","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":"Université Laval","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Institute on Alcohol Abuse and Alcoholism","keywords":"Causal model; Structural equation modeling; Latent variable; Alcohol; Psychology; Construct (python library); Index (typography); Econometrics; Social psychology; Statistics; Mathematics; Computer science","score_opus":0.31414295753857585,"score_gpt":0.45322424653099425,"score_spread":0.1390812889924184,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2337224824","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":[0.7767884,0.006466554,0.12236485,0.0666586,0.0007195705,0.0010396057,0.0008155692,0.00019853034,0.024948178],"genre_scores_gemma":[0.98726094,0.00093138777,0.010313943,0.0005852794,0.00008539899,0.00034020498,0.00018495628,0.000015141722,0.00028274613],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99182713,0.0058195256,0.00045381935,0.000488073,0.0010176527,0.00039385405],"domain_scores_gemma":[0.98259133,0.0098954765,0.0038677794,0.0011989868,0.0016699613,0.00077649753],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013049513,0.0007642658,0.0008135895,0.0053884024,0.00084441784,0.003555231,0.0022628768,0.0010713383,0.0021992412],"category_scores_gemma":[0.047510818,0.0005986408,0.0012204865,0.003754224,0.006335147,0.0064785755,0.0026153927,0.003091352,0.00015514871],"study_design_candidate":"theoretical_or_conceptual","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":[0.00011675007,0.00041679307,0.6116331,0.00037781993,0.0003838018,0.0001891971,0.01515398,0.0047191996,0.00019279917,0.2800502,0.0033059334,0.08346053],"study_design_scores_gemma":[0.0000989952,0.00037245383,0.46892968,0.0014831737,0.00031452478,0.00043060526,0.03444036,0.03646934,0.00033413494,0.44525644,0.011679427,0.00019081478],"about_ca_topic_score_codex":0.01693601,"about_ca_topic_score_gemma":0.01665091,"teacher_disagreement_score":0.01693601,"about_ca_system_score_codex":0.0039918423,"about_ca_system_score_gemma":0.0038513842,"threshold_uncertainty_score":0.06901324},"labels":[],"label_agreement":null},{"id":"W2341010248","doi":"10.1177/0163278715605358","title":"Using Automated Scoring to Evaluate Written Responses in English and French on a High-Stakes Clinical Competency Examination","year":2015,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Topic Modeling","field":"Computer Science","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada; University of Alberta","funders":"","keywords":"Reliability (semiconductor); Computer science; Stage (stratigraphy); Artificial intelligence; Natural language processing; Machine learning; Medical education; Medicine","score_opus":0.4259430117257569,"score_gpt":0.5111785923759283,"score_spread":0.08523558065017145,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2341010248","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":[0.6752609,0.00026206634,0.31115165,0.0003782534,0.000084145344,0.0013281782,0.0011179007,0.0029546642,0.0074623763],"genre_scores_gemma":[0.8023966,0.00011047092,0.19365028,0.00009104976,0.00005532152,0.0006816608,0.0012379544,0.000081060156,0.0016955002],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9780075,0.015243071,0.0011838032,0.001559166,0.0034731831,0.0005332525],"domain_scores_gemma":[0.94899255,0.029779624,0.00484361,0.0025517584,0.013023042,0.0008094408],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01581016,0.0009760304,0.000615637,0.0041752513,0.0006104189,0.0017592685,0.00085164106,0.0007321174,0.0014886765],"category_scores_gemma":[0.04321626,0.00021185991,0.00052771426,0.0017047095,0.0004591133,0.00084376644,0.0014613988,0.00059934845,0.00095676736],"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":[0.00060364464,0.0006139239,0.20006397,0.0003416812,0.0002860421,0.00023050136,0.0035822382,0.018117178,0.02887038,0.0015056113,0.004552653,0.74123216],"study_design_scores_gemma":[0.00023356895,0.0016788689,0.38499275,0.00018933832,0.00023782569,0.0009300447,0.0038607027,0.5331036,0.05838478,0.004374868,0.011615702,0.00039784028],"about_ca_topic_score_codex":0.010518738,"about_ca_topic_score_gemma":0.021808386,"teacher_disagreement_score":0.01581016,"about_ca_system_score_codex":0.0011125676,"about_ca_system_score_gemma":0.0019218543,"threshold_uncertainty_score":0.0836131},"labels":[],"label_agreement":null},{"id":"W2511580120","doi":"10.1177/0163278716663602","title":"Emerging Adulthood: A Time of Changes in Psychosocial Well-Being","year":2016,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Psychological Well-being and Life Satisfaction","field":"Psychology","cited_by":95,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Addiction and Mental Health","funders":"Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Psychosocial; Psychology; Young adult; Mental health; Cohort; Clinical psychology; Well-being; Developmental psychology; Medicine; Psychiatry","score_opus":0.0648383612560048,"score_gpt":0.44489596821805394,"score_spread":0.38005760696204915,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2511580120","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":[0.9977952,0.00082649104,0.00018451465,0.00012564854,0.00001170717,0.000013619972,0.00017525362,0.0000037628206,0.0008638462],"genre_scores_gemma":[0.9988444,0.00052907836,0.00022940038,0.00004206019,0.0000082511515,0.000013189213,0.00016399963,0.0000012158813,0.00016833351],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.999819,0.00005754281,0.000017318875,0.000036457684,0.000026240758,0.00004356752],"domain_scores_gemma":[0.99936944,0.000100901576,0.00021965218,0.000026588245,0.0000775497,0.00020585668],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000555152,0.00015348612,0.00027807843,0.000430585,0.00056130806,0.0009864795,0.0001701183,0.00034024878,0.00095636223],"category_scores_gemma":[0.0014323551,0.00010513302,0.00021731223,0.00043535055,0.00034919995,0.0009219939,0.0010433662,0.0006710415,0.00009184234],"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":[0.00030029262,0.0001279971,0.96877116,0.0000952916,0.000045023,0.0007439841,0.0061361985,0.00006308291,0.001896237,0.0008036477,0.00024264136,0.020774415],"study_design_scores_gemma":[0.0000021026392,0.00018748065,0.99257094,0.000037433994,0.00001423065,0.00032600807,0.0058810734,0.00004952393,0.00012465783,0.0002216858,0.00057922205,0.0000056862236],"about_ca_topic_score_codex":0.004041059,"about_ca_topic_score_gemma":0.009211772,"teacher_disagreement_score":0.004041059,"about_ca_system_score_codex":0.00035091568,"about_ca_system_score_gemma":0.00045896802,"threshold_uncertainty_score":0.008035064},"labels":[],"label_agreement":null},{"id":"W2522164456","doi":"10.1177/0163278716665882","title":"Organizational Capabilities for Integrating Care","year":2016,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Toronto Rehabilitation Institute; Sinai Health System; Lunenfeld-Tanenbaum Research Institute; Cancer Care Ontario; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Psychological intervention; Context (archaeology); Knowledge management; MEDLINE; Psychology; Integrated care; Survey instrument; Process management; Medicine; Applied psychology; Computer science; Nursing; Health care; Business","score_opus":0.10478814752706225,"score_gpt":0.5213973918583964,"score_spread":0.41660924433133417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2522164456","genre_codex":"other","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":[0.2999664,0.07646947,0.08973648,0.059162326,0.0010116901,0.0026001884,0.0011014958,0.00028893573,0.4696629],"genre_scores_gemma":[0.95243645,0.014931496,0.027966896,0.0018672291,0.00011593851,0.00067156553,0.0002634603,0.000012183256,0.0017347304],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9775194,0.013166435,0.0026545043,0.00079455646,0.00423036,0.0016347589],"domain_scores_gemma":[0.9521856,0.023834791,0.011311758,0.0032075664,0.005782012,0.00367827],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017268216,0.0003513118,0.0003670095,0.004079426,0.0023635726,0.0058249137,0.00097189087,0.0012619176,0.0031728926],"category_scores_gemma":[0.04751604,0.00034517373,0.0010484993,0.0026482553,0.0060460772,0.005207233,0.008312704,0.0019357657,0.0002598815],"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":[0.00006780539,0.00019272543,0.06714394,0.006647775,0.0003685863,0.00046507505,0.035544086,0.0018569466,0.000802535,0.45293748,0.009919679,0.42405334],"study_design_scores_gemma":[0.000080453894,0.00045956962,0.19089022,0.021587728,0.00049995637,0.0022920538,0.06393157,0.0015364842,0.0012488972,0.3241453,0.39309323,0.00023467641],"about_ca_topic_score_codex":0.0038949598,"about_ca_topic_score_gemma":0.0062530087,"teacher_disagreement_score":0.017268216,"about_ca_system_score_codex":0.0041742427,"about_ca_system_score_gemma":0.015263931,"threshold_uncertainty_score":0.09132409},"labels":[],"label_agreement":null},{"id":"W2533150155","doi":"10.1177/0163278716674247","title":"A Factor Analytic Investigation of the Person-in-Recovery and Provider Versions of the Revised Recovery Self-Assessment (RSA-R)","year":2016,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Waypoint Centre for Mental Health Care","funders":"","keywords":"Confirmatory factor analysis; Mental health; Autonomy; Psychology; Scale (ratio); Mental illness; Component (thermodynamics); Exploratory factor analysis; Psychometrics; Mental health care; Clinical psychology; Psychiatry; Structural equation modeling; Computer science","score_opus":0.27597465581174196,"score_gpt":0.45750430557905436,"score_spread":0.1815296497673124,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2533150155","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":[0.9916044,0.00009963172,0.00631043,0.00014983189,0.000024136045,0.0004899571,0.00023505217,0.00001894964,0.0010676285],"genre_scores_gemma":[0.9867014,0.000076665914,0.011816714,0.000034999564,0.000012627438,0.0005127438,0.00053412543,0.000010923031,0.00029980243],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99258316,0.0043961736,0.000584486,0.00048314163,0.0016424521,0.00031057248],"domain_scores_gemma":[0.9718606,0.018103808,0.0023575819,0.0018242215,0.0052478462,0.0006059828],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016338266,0.0004964205,0.00047724298,0.0017806685,0.00064091745,0.0008773404,0.00062741677,0.00023610947,0.001462816],"category_scores_gemma":[0.056020204,0.0003068972,0.0011314772,0.0020862592,0.0008850873,0.0010062016,0.0010810703,0.0009933837,0.00016015532],"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":[0.00044360696,0.00034141674,0.8906339,0.00013691233,0.0002944704,0.00009524408,0.010619564,0.00069010677,0.001433274,0.002141513,0.0007383299,0.092431806],"study_design_scores_gemma":[0.00010660839,0.0011582143,0.972879,0.00014124179,0.000190023,0.0002751818,0.01089477,0.009033541,0.0010043655,0.0012797675,0.0029831408,0.00005410818],"about_ca_topic_score_codex":0.016414823,"about_ca_topic_score_gemma":0.025195226,"teacher_disagreement_score":0.016414823,"about_ca_system_score_codex":0.0016482922,"about_ca_system_score_gemma":0.004674247,"threshold_uncertainty_score":0.08640611},"labels":[],"label_agreement":null},{"id":"W2614651567","doi":"10.1177/0163278717709562","title":"Evaluating the Effectiveness of Complex, Multi-component, Dynamic, Community-Based Injury Prevention Interventions: A Statistical Framework","year":2017,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Health Policy Implementation Science","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":"McMaster University; Population Health Research Institute","funders":"Medical Research Council; University of South Africa; South African Medical Research Council","keywords":"Psychological intervention; Pooling; Component (thermodynamics); Computer science; Management science; Conceptual framework; Intervention (counseling); Promotion (chess); Risk analysis (engineering); Data mining; Process management; Artificial intelligence; Medicine; Engineering; Sociology; Political science; Nursing","score_opus":0.8577686624068719,"score_gpt":0.7766059520045184,"score_spread":0.08116271040235346,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2614651567","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03301355,0.0029297085,0.95120484,0.003098,0.00024996293,0.0054911063,0.00092030526,0.0005038028,0.0025886982],"genre_scores_gemma":[0.35603413,0.0010056711,0.62437147,0.0006605379,0.00021834693,0.016635729,0.0006421757,0.00009725074,0.00033469376],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.64306617,0.3144427,0.012873868,0.009057579,0.019334577,0.0012250528],"domain_scores_gemma":[0.5118638,0.4408412,0.019950375,0.016662996,0.009486845,0.0011947483],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25731513,0.002557301,0.0049367365,0.011617608,0.0018869998,0.00523628,0.004660988,0.0021436634,0.0026643455],"category_scores_gemma":[0.43389902,0.001167188,0.0065099574,0.0078790095,0.008101271,0.006825991,0.0055859005,0.003459489,0.00025441532],"study_design_candidate":"theoretical_or_conceptual","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":[0.0019225166,0.0017522301,0.07959885,0.0072224946,0.025981406,0.0001900086,0.0023950695,0.17520185,0.0016135682,0.34352568,0.004484995,0.35611132],"study_design_scores_gemma":[0.0012563127,0.009305217,0.04145889,0.0024245447,0.0083783185,0.00014693808,0.0019188544,0.50787574,0.0033460755,0.4102561,0.013303532,0.00032949203],"about_ca_topic_score_codex":0.0070038815,"about_ca_topic_score_gemma":0.0040733824,"teacher_disagreement_score":0.74268484,"about_ca_system_score_codex":0.0061889295,"about_ca_system_score_gemma":0.012657867,"threshold_uncertainty_score":0.9158627},"labels":[],"label_agreement":null},{"id":"W2762779240","doi":"10.1177/0163278717734282","title":"Variables Associated With Perceived Work Role Performance Among Professionals in Multidisciplinary Mental Health Teams Overall and in Primary Care and Specialized Service Teams, Respectively","year":2017,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; McGill University; Douglas Mental Health University Institute","funders":"Fonds de Recherche du Québec - Santé","keywords":"Context (archaeology); Mental health; Multidisciplinary approach; Psychology; Team effectiveness; Service (business); Applied psychology; Multidisciplinary team; Sample (material); Nursing; Knowledge management; Medicine; Business; Marketing; Psychiatry","score_opus":0.04318118310479186,"score_gpt":0.4348069271015254,"score_spread":0.39162574399673356,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2762779240","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":[0.99944335,0.00006741688,0.0000363206,0.00007546984,0.0000034654304,0.0000085076945,0.000077175835,0.0000010193119,0.0002872825],"genre_scores_gemma":[0.99972874,0.000026335383,0.000039367795,0.0000133163085,0.0000021141093,0.0000057708826,0.000088855304,4.891156e-7,0.00009497294],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9988727,0.0002983425,0.00008754932,0.000100810656,0.00029060352,0.00035000197],"domain_scores_gemma":[0.99548066,0.0008346192,0.0012454945,0.000115357114,0.0008549202,0.0014690118],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0015435485,0.00024253741,0.0002291337,0.00088435906,0.0010080592,0.00096111395,0.00043774818,0.00039383286,0.0013818641],"category_scores_gemma":[0.0052767266,0.00014322298,0.00037144404,0.00090721773,0.0006089122,0.00037412255,0.0008598345,0.0006209923,0.00013256035],"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":[0.000018235674,0.000042518914,0.997413,0.0000067646674,0.000012999817,0.000019358351,0.00071078603,0.000029238321,0.0000439724,0.000007653123,0.00007045491,0.0016249926],"study_design_scores_gemma":[0.0000019475842,0.000030353085,0.99785405,0.0000057664697,0.0000028735215,0.000011998679,0.0019067127,0.00010032345,0.000008758537,0.000005939426,0.00006919239,0.0000020603536],"about_ca_topic_score_codex":0.28572747,"about_ca_topic_score_gemma":0.38261265,"teacher_disagreement_score":0.28572747,"about_ca_system_score_codex":0.0029806304,"about_ca_system_score_gemma":0.002287763,"threshold_uncertainty_score":0.5681287},"labels":[],"label_agreement":null},{"id":"W2796820415","doi":"10.1177/0163278718756992","title":"What Is Lean Management in Health Care? Development of an Operational Definition for a Cochrane Systematic Review","year":2018,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Quality and Management","field":"Health Professions","cited_by":119,"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 Northern British Columbia; University of Alberta; University of Saskatchewan; Queen's University","funders":"University of Saskatchewan","keywords":"Operationalization; Lean manufacturing; Process management; Health care; Lean project management; Computer science; Operations management; Knowledge management; Business; Engineering; Political science","score_opus":0.38401862997894964,"score_gpt":0.5973783700362918,"score_spread":0.21335974005734215,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2796820415","genre_codex":"review","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0026337055,0.6914331,0.16276726,0.05206465,0.0049279584,0.06844211,0.0036072854,0.00045984905,0.01366409],"genre_scores_gemma":[0.029950695,0.20414156,0.5998705,0.011619327,0.001320345,0.15067425,0.001626622,0.0001475496,0.00064907334],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.5108236,0.2860997,0.14767727,0.011537007,0.040702775,0.0031597354],"domain_scores_gemma":[0.52205664,0.37372047,0.044309665,0.012150085,0.043410555,0.0043526003],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2609779,0.005157702,0.017566718,0.092316926,0.005064725,0.024567138,0.009101472,0.008161593,0.004116424],"category_scores_gemma":[0.5216776,0.004618003,0.017923092,0.053842455,0.011831292,0.022688657,0.014578913,0.009210997,0.0008800249],"study_design_candidate":"not_applicable","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":[0.0003552742,0.00011960178,0.0023390758,0.49034607,0.006142351,0.00037326704,0.005000976,0.0044928244,0.0003919578,0.1819279,0.022422533,0.28608823],"study_design_scores_gemma":[0.00048664922,0.00025502132,0.0017224812,0.7819607,0.0075627402,0.0004366424,0.0025453148,0.0034899309,0.00052240863,0.09098206,0.10975331,0.00028268615],"about_ca_topic_score_codex":0.008124486,"about_ca_topic_score_gemma":0.010763827,"teacher_disagreement_score":0.73902214,"about_ca_system_score_codex":0.034589313,"about_ca_system_score_gemma":0.07229333,"threshold_uncertainty_score":0.91134584},"labels":[],"label_agreement":null},{"id":"W2801003465","doi":"10.1177/0163278718772887","title":"A Network Analysis Perspective to Implementation: The Example of Health Links to Promote Coordinated Care","year":2018,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Systems and Technology","field":"Business, Management and Accounting","cited_by":24,"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 Toronto","funders":"","keywords":"Social network analysis; Formative assessment; Psychological intervention; Influencer marketing; Health care; Process (computing); Knowledge management; Social network (sociolinguistics); Bridging (networking); Network analysis; Centrality; Computer science; Process management; Intervention (counseling); Workgroup; Psychology; Business; Nursing; Medicine; Social media; Computer security; Political science; World Wide Web; Engineering","score_opus":0.12441310473074768,"score_gpt":0.45341539886124727,"score_spread":0.3290022941304996,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2801003465","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":[0.028940607,0.009027739,0.40330797,0.1639339,0.0009960404,0.000635112,0.0002914929,0.00029244518,0.39257464],"genre_scores_gemma":[0.80089617,0.005790293,0.17189533,0.0052123102,0.00045495515,0.0011490542,0.00011011324,0.00024746032,0.014244263],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.97271186,0.023414457,0.00033279476,0.0010327396,0.0015366328,0.000971487],"domain_scores_gemma":[0.97614026,0.018362071,0.0011454152,0.0019084967,0.0015857703,0.0008579525],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028843338,0.0011809962,0.00067691837,0.0053956723,0.008141349,0.011616706,0.0026067283,0.0048822286,0.0077914773],"category_scores_gemma":[0.030747699,0.0006345463,0.0011498011,0.0043874183,0.038252514,0.016250163,0.0085157985,0.0059843906,0.00056303316],"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":[0.000012627756,0.000019429048,0.00050228025,0.00013411515,0.000015492757,0.000081554754,0.0055917166,0.0015741078,0.00006385198,0.98002464,0.0017231226,0.010257051],"study_design_scores_gemma":[0.000039867573,0.00005755852,0.0008839253,0.00057153404,0.0000405789,0.00010534625,0.0074991873,0.0075878887,0.00033648615,0.88479793,0.098048165,0.000031507996],"about_ca_topic_score_codex":0.029007545,"about_ca_topic_score_gemma":0.024227796,"teacher_disagreement_score":0.029007545,"about_ca_system_score_codex":0.017230904,"about_ca_system_score_gemma":0.013735925,"threshold_uncertainty_score":0.15254003},"labels":[],"label_agreement":null},{"id":"W2883166157","doi":"10.1177/0163278718784998","title":"Interactive Video Gaming Improves Functional Balance in Poststroke Individuals: Meta-Analysis of Randomized Controlled Trials","year":2018,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Stroke Rehabilitation and Recovery","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Berg Balance Scale; Funnel plot; Balance (ability); Timed Up and Go test; Meta-analysis; Portuguese; Physical medicine and rehabilitation; Physical therapy; Randomized controlled trial; Video game; Confidence interval; Statistic; Scale (ratio); MEDLINE; Publication bias; Medicine; Psychology; Multimedia; Computer science; Statistics","score_opus":0.3527344878480664,"score_gpt":0.5404014187464845,"score_spread":0.18766693089841813,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2883166157","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":[0.004225165,0.994611,0.0002759256,0.00010212607,0.00011091369,0.000329529,0.0001385903,0.000012398896,0.00019431979],"genre_scores_gemma":[0.10595658,0.8890632,0.002215757,0.0006172845,0.0002790645,0.0011983936,0.00037706504,0.00001725764,0.00027548047],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9947813,0.0025292865,0.0016440436,0.00036861756,0.000501536,0.00017511912],"domain_scores_gemma":[0.9869221,0.010316929,0.0017136872,0.00024243628,0.000654854,0.00014996684],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009147575,0.002358701,0.014568852,0.004811225,0.00042813987,0.00246216,0.0015984061,0.0018081416,0.0034279095],"category_scores_gemma":[0.024969475,0.00096476864,0.020591902,0.00456448,0.00046326924,0.0012924568,0.0009247062,0.0016375382,0.00017784057],"study_design_candidate":"meta_analysis","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":[0.0054991576,0.0001255769,0.0020555102,0.55931216,0.3937024,0.0001463386,0.00009952024,0.00040396804,0.00035090675,0.00016239242,0.0005882509,0.037553784],"study_design_scores_gemma":[0.0021975304,0.0007890129,0.0033271075,0.04107147,0.94970495,0.00011941345,0.00005802152,0.00016567616,0.0002092103,0.00020892957,0.002126978,0.000021655804],"about_ca_topic_score_codex":0.0031591272,"about_ca_topic_score_gemma":0.00798385,"teacher_disagreement_score":0.014568852,"about_ca_system_score_codex":0.0017488311,"about_ca_system_score_gemma":0.0026913441,"threshold_uncertainty_score":0.048377573},"labels":[],"label_agreement":null},{"id":"W2888782080","doi":"10.1177/0163278718796153","title":"A Qualitative Systematic Review of Public–Private Partnership in Promoting Physical Activity","year":2018,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Urban Transport and Accessibility","field":"Social Sciences","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":"University of Alberta","funders":"National Research Foundation of Korea","keywords":"General partnership; Public relations; Pledge; Systematic review; Business; Grey literature; Public–private partnership; Promotion (chess); Qualitative research; Private sector; Inclusion (mineral); Qualitative property; Political science; MEDLINE; Psychology; Sociology; Finance","score_opus":0.6015343105248565,"score_gpt":0.638706201338082,"score_spread":0.03717189081322547,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2888782080","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":[0.014759139,0.9460261,0.008319893,0.0065636626,0.00090095366,0.01281641,0.0014549885,0.000042356452,0.009116437],"genre_scores_gemma":[0.1653182,0.7691904,0.02716125,0.0047609042,0.00024072052,0.030474048,0.00085060037,0.000031390082,0.0019724583],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9296064,0.052048206,0.010088735,0.0015708446,0.0053919544,0.001293903],"domain_scores_gemma":[0.90642655,0.07441256,0.0057563297,0.0016010184,0.010854977,0.0009486087],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.064857915,0.0011779638,0.0038015728,0.015934376,0.0028879123,0.0041949153,0.0020423282,0.002374433,0.004871685],"category_scores_gemma":[0.115206204,0.00111654,0.0028547675,0.01796934,0.002436933,0.0063595404,0.0048781205,0.0014641156,0.0004503304],"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":[0.00023986514,0.0000876317,0.001928117,0.75690585,0.0011807877,0.0011005548,0.035558887,0.00034487536,0.00050826545,0.010110066,0.007603387,0.18443172],"study_design_scores_gemma":[0.00016301843,0.0002866954,0.0027923551,0.84961057,0.003701143,0.00090851437,0.044068363,0.00016979386,0.00045567815,0.003737359,0.09402765,0.00007877735],"about_ca_topic_score_codex":0.008269913,"about_ca_topic_score_gemma":0.026266424,"teacher_disagreement_score":0.064857915,"about_ca_system_score_codex":0.0127524985,"about_ca_system_score_gemma":0.044748228,"threshold_uncertainty_score":0.34300554},"labels":[],"label_agreement":null},{"id":"W2896039089","doi":"10.1177/0163278718801439","title":"Psychometric Testing of the Evidence-Based Practice Nursing Leadership Scale and the Work Environment Scale After Cross-Cultural Adaptation in Mainland China","year":2018,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Mainland China; Scale (ratio); Nursing; Context (archaeology); Health care; Psychology; Medicine; China; Political science; Geography","score_opus":0.516755752520595,"score_gpt":0.5683442077082915,"score_spread":0.0515884551876965,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2896039089","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":[0.99913675,0.000042323012,0.00019113107,0.000042874082,0.000010992289,0.000088880304,0.000050854058,0.0000035501569,0.00043267524],"genre_scores_gemma":[0.9980172,0.00007762682,0.00090493646,0.00005350198,0.0000068844324,0.0002885501,0.00029485478,0.0000039037886,0.00035250426],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9983242,0.00046812263,0.00028415286,0.00021502377,0.00054643344,0.00016198466],"domain_scores_gemma":[0.9958449,0.0011174752,0.0005571156,0.00044312706,0.0015359045,0.0005014535],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053602397,0.0004979624,0.00048552314,0.0011617811,0.00076720095,0.00050036807,0.00045550888,0.00036915758,0.00090828113],"category_scores_gemma":[0.009998125,0.00032033125,0.00093323155,0.0012213093,0.00073957717,0.0005747429,0.0009863798,0.0005574806,0.0001548407],"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":[0.00017495212,0.0006263451,0.94031835,0.000108050095,0.00023990714,0.00022912194,0.0057869065,0.0005742984,0.0020883826,0.00018496328,0.0006252968,0.049043603],"study_design_scores_gemma":[0.000039010756,0.0002806148,0.99704915,0.000021739275,0.000040332085,0.00004839744,0.0010308872,0.0007658895,0.00029007797,0.00004571706,0.0003785066,0.000009676193],"about_ca_topic_score_codex":0.016652651,"about_ca_topic_score_gemma":0.022030417,"teacher_disagreement_score":0.016652651,"about_ca_system_score_codex":0.0014044625,"about_ca_system_score_gemma":0.0027552356,"threshold_uncertainty_score":0.033111453},"labels":[],"label_agreement":null},{"id":"W2907960444","doi":"10.1177/0163278718820410","title":"Construct Validity and Internal Consistency of the Catalan Version of the Pain Self-Efficacy Questionnaire in Young People With Chronic Pain","year":2018,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Fundación Grünenthal España","keywords":"Construct validity; Discriminant validity; Chronic pain; Coping (psychology); Convergent validity; Psychology; Catalan; Clinical psychology; Concurrent validity; Physical therapy; Internal consistency; Psychometrics; Medicine; Psychiatry","score_opus":0.027107725078756514,"score_gpt":0.3574962764806622,"score_spread":0.33038855140190565,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2907960444","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":[0.9985524,0.00019133397,0.00030471073,0.00003559422,0.000022155229,0.00008508111,0.00017236182,0.000004565511,0.0006318978],"genre_scores_gemma":[0.99803776,0.00014492444,0.00096895476,0.000038819595,0.000011536916,0.00016507637,0.00042017063,0.0000033626893,0.0002093612],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9977836,0.0007134595,0.00041541652,0.0001989637,0.00072043145,0.0001680725],"domain_scores_gemma":[0.9900108,0.004405625,0.0019626042,0.00046813337,0.0027571737,0.00039561072],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0067070713,0.00027119726,0.0005160287,0.001313149,0.00038919036,0.0007282801,0.00032045375,0.00039633556,0.000882265],"category_scores_gemma":[0.013116068,0.00028700999,0.00097340555,0.0006360472,0.0004002799,0.00038104446,0.00073482946,0.00050522486,0.00019038391],"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":[0.00013207793,0.0001276638,0.98817724,0.000073801144,0.0001609816,0.00004896419,0.0008263528,0.00012883956,0.0004998397,0.00004949391,0.00016149641,0.009613201],"study_design_scores_gemma":[0.00001914011,0.00014937279,0.99855953,0.000026002488,0.000034817513,0.0000728357,0.00045996698,0.0003188878,0.00009016982,0.00004484672,0.00021820611,0.0000062365866],"about_ca_topic_score_codex":0.0021312106,"about_ca_topic_score_gemma":0.004078638,"teacher_disagreement_score":0.0067070713,"about_ca_system_score_codex":0.00038010225,"about_ca_system_score_gemma":0.00051584264,"threshold_uncertainty_score":0.035470784},"labels":[],"label_agreement":null},{"id":"W2920775009","doi":"10.1177/0163278719834339","title":"Evaluating the Quality of Health Professions Learning Environments: Development and Validation of the Health Education Learning Environment Survey (HELES)","year":2019,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kwantlen Polytechnic University; University of British Columbia","funders":"University of British Columbia","keywords":"Accreditation; Learning environment; Medical education; Psychology; Socialization; Context (archaeology); Quality (philosophy); Social environment; Applied psychology; Medicine; Mathematics education; Social psychology","score_opus":0.2534548523223802,"score_gpt":0.5345699707893247,"score_spread":0.2811151184669445,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2920775009","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":[0.9745603,0.00017081534,0.013141208,0.0003679297,0.000050167582,0.0060044173,0.0014002776,0.00008570414,0.004219091],"genre_scores_gemma":[0.94453317,0.00030423963,0.039432194,0.0003208655,0.000043022563,0.01080088,0.0031700912,0.000036999205,0.0013586385],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.988374,0.00559519,0.0020919833,0.00047317866,0.0030426222,0.00042301966],"domain_scores_gemma":[0.9728679,0.0120241735,0.00427674,0.0015487899,0.00808956,0.0011928866],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02319739,0.000365776,0.0004542198,0.0025542802,0.0004975769,0.0007484458,0.0005990147,0.0003830173,0.00073648297],"category_scores_gemma":[0.032626763,0.00027629352,0.0006143364,0.0014499928,0.00066490466,0.00089485466,0.0021651285,0.000758511,0.00024059079],"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":[0.00025029,0.0015226913,0.8642355,0.00019443422,0.0001547295,0.00004957503,0.0025922796,0.00095556595,0.0017255316,0.0007014718,0.0012151444,0.12640268],"study_design_scores_gemma":[0.000095378535,0.0016210973,0.9878165,0.000077644545,0.000036119294,0.00005128987,0.0019780695,0.0019009515,0.001601807,0.00033312995,0.004458839,0.000029200379],"about_ca_topic_score_codex":0.0024746174,"about_ca_topic_score_gemma":0.004496702,"teacher_disagreement_score":0.02319739,"about_ca_system_score_codex":0.00077147805,"about_ca_system_score_gemma":0.0028161698,"threshold_uncertainty_score":0.12268102},"labels":[],"label_agreement":null},{"id":"W3108466076","doi":"10.1177/0163278720975833","title":"Use of Critical Items in Determining Point-of-Care Ultrasound Competence","year":2020,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":8,"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; University of Calgary","funders":"","keywords":"Intraclass correlation; Inter-rater reliability; Medicine; Competence (human resources); Confidence interval; Physical therapy; Medical physics; Psychology; Rating scale; Psychometrics; Internal medicine; Clinical psychology; Social psychology","score_opus":0.348185947162498,"score_gpt":0.5173826674688528,"score_spread":0.16919672030635474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3108466076","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":[0.8312319,0.004326311,0.14485615,0.0005959614,0.00047049168,0.0052191694,0.0007601595,0.00028983373,0.0122499345],"genre_scores_gemma":[0.860582,0.0010534314,0.1337527,0.00017120165,0.00009403451,0.0033211359,0.00043451006,0.000049848564,0.00054112356],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9107982,0.050804973,0.015469936,0.0036046559,0.017549137,0.0017729839],"domain_scores_gemma":[0.77105695,0.15439187,0.023441734,0.0071708383,0.040459424,0.0034791662],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07112083,0.0014547728,0.0012879113,0.01105484,0.0006769639,0.001621069,0.0018579589,0.001367021,0.001823999],"category_scores_gemma":[0.17868954,0.00057593174,0.0016281456,0.0030565728,0.0015593197,0.0029225063,0.003536525,0.0012853275,0.0005736399],"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":[0.002105229,0.0005701325,0.6501827,0.0045458353,0.0011137287,0.0011175595,0.012224233,0.0043918267,0.014276847,0.002913401,0.0033197252,0.30323872],"study_design_scores_gemma":[0.00038641208,0.009338813,0.8812558,0.004958732,0.00079987344,0.007796287,0.010673211,0.030720713,0.02503021,0.00991433,0.018456846,0.0006688188],"about_ca_topic_score_codex":0.0007991734,"about_ca_topic_score_gemma":0.001360172,"teacher_disagreement_score":0.07112083,"about_ca_system_score_codex":0.0010312039,"about_ca_system_score_gemma":0.0018081159,"threshold_uncertainty_score":0.37612736},"labels":[],"label_agreement":null},{"id":"W3111715355","doi":"10.1177/0163278720978814","title":"A Systematic Review of the Outcome Measures Used to Evaluate Interprofessional Learning by Health Care Professional Students During Clinical Experiences","year":2020,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Interprofessional education; Medical education; Inclusion (mineral); Health care; Intervention (counseling); MEDLINE; Medicine; Scale (ratio); Psychology; Perception; Nursing","score_opus":0.3172032332576435,"score_gpt":0.658786603852949,"score_spread":0.34158337059530547,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3111715355","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":[0.0011463419,0.9958448,0.00043467345,0.00025267596,0.00014350929,0.0010062899,0.0007270211,0.00001033056,0.00043440948],"genre_scores_gemma":[0.016643535,0.9777474,0.0017351673,0.0005294692,0.00006512299,0.0025535438,0.0005480027,0.000010422303,0.00016737024],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9713733,0.009653597,0.012184301,0.0015060994,0.0048486907,0.00043412115],"domain_scores_gemma":[0.9183974,0.061538596,0.011351047,0.0013377386,0.00693959,0.00043557663],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02272372,0.002201462,0.009248156,0.012540845,0.0012406021,0.0037682,0.0020940297,0.0022506712,0.00428439],"category_scores_gemma":[0.11376198,0.0011993096,0.009875551,0.015102913,0.0015214063,0.0030240072,0.0021060822,0.0017314085,0.0004745633],"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":[0.00015775123,0.000015154005,0.0007049667,0.9617591,0.006942627,0.000052786956,0.00018826767,0.000057990837,0.000093565824,0.00020943495,0.00085074845,0.028967664],"study_design_scores_gemma":[0.00017615818,0.00016445419,0.0033814905,0.91263396,0.069680445,0.00023703455,0.00026156535,0.00004854946,0.00025551495,0.00028413266,0.012844217,0.000032514934],"about_ca_topic_score_codex":0.009621858,"about_ca_topic_score_gemma":0.030160848,"teacher_disagreement_score":0.02272372,"about_ca_system_score_codex":0.006276965,"about_ca_system_score_gemma":0.019940551,"threshold_uncertainty_score":0.12017596},"labels":[],"label_agreement":null},{"id":"W3119424768","doi":"10.1177/0163278720983416","title":"The COVID-19 Preventive Behaviors Index: Development and Validation in Two Samples From the United Kingdom","year":2021,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"COVID-19 and Mental Health","field":"Psychology","cited_by":80,"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":"Trent University; Nottingham Trent University","keywords":"Coronavirus disease 2019 (COVID-19); Scale (ratio); Psychology; Confirmatory factor analysis; Index (typography); Exploratory factor analysis; Applied psychology; Clinical psychology; Structural equation modeling; Psychometrics; Medicine; Statistics; Computer science","score_opus":0.4151495611428296,"score_gpt":0.5796821983329958,"score_spread":0.16453263719016625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3119424768","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":[0.9978842,0.00006700185,0.00064798753,0.00005002386,0.000007208229,0.00033810496,0.00033931373,0.000004609027,0.0006613448],"genre_scores_gemma":[0.9927246,0.00019628838,0.00350612,0.0001383962,0.000009855143,0.001025553,0.0014997191,0.000020184429,0.0008792334],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99666494,0.0014322964,0.00055795495,0.00028529024,0.00074033247,0.00031923273],"domain_scores_gemma":[0.99092054,0.0023388364,0.001420096,0.0010486662,0.0036869405,0.0005848636],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004800579,0.0004620985,0.0005587382,0.0013268067,0.0008718034,0.0008788024,0.0006168178,0.0006203419,0.0016273755],"category_scores_gemma":[0.018029375,0.0005987418,0.0006092852,0.0013077733,0.00095389335,0.00069715886,0.002193954,0.0009838697,0.0006756055],"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":[0.0007203193,0.0007184162,0.94835526,0.00019611114,0.00005970805,0.0003886578,0.019914636,0.00025098288,0.0022573676,0.00036010315,0.0013911795,0.025387336],"study_design_scores_gemma":[0.00008272636,0.0005169915,0.9912956,0.00006367815,0.000025390387,0.00027935117,0.005132068,0.00032430128,0.0005786089,0.000049626375,0.001633107,0.00001850355],"about_ca_topic_score_codex":0.06276742,"about_ca_topic_score_gemma":0.06856572,"teacher_disagreement_score":0.06276742,"about_ca_system_score_codex":0.0021953639,"about_ca_system_score_gemma":0.0017752665,"threshold_uncertainty_score":0.12480414},"labels":[],"label_agreement":null},{"id":"W3125690464","doi":"10.1177/0163278721989547","title":"Multilevel Latent Class Profile Analysis: An Application to Stage-Sequential Patterns of Alcohol Use in a Sample of Canadian Youth","year":2021,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Alcohol Consumption and Health Effects","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; National Research Foundation of Korea","keywords":"Latent class model; Multilevel model; Longitudinal study; Binge drinking; Psychology; Sample (material); Class (philosophy); Longitudinal data; Environmental health; Poison control; Demography; Human factors and ergonomics; Statistics; Medicine; Computer science; Mathematics; Sociology","score_opus":0.41929450095601434,"score_gpt":0.5001983318422347,"score_spread":0.08090383088622038,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3125690464","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":[0.77187806,0.0007217836,0.20716467,0.0016725641,0.00012281777,0.0039899717,0.009859395,0.00090421335,0.0036865277],"genre_scores_gemma":[0.8557931,0.000344091,0.13653938,0.00008293507,0.000025585137,0.0019186777,0.004019192,0.0000649027,0.0012120998],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9958269,0.0019357073,0.0002211813,0.0006174849,0.0008922767,0.0005064711],"domain_scores_gemma":[0.99393255,0.001959496,0.00072856096,0.0008710283,0.0020086595,0.0004996164],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008934116,0.0007143428,0.0010409763,0.0044987877,0.0046557887,0.0015334944,0.0022660152,0.0006559704,0.002821601],"category_scores_gemma":[0.026876707,0.00048502008,0.0025274388,0.006849776,0.00089733384,0.0007628,0.0023759222,0.001612784,0.0002239786],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046597546,0.00029923365,0.7721233,0.00024881904,0.0008807682,0.00023314211,0.009487048,0.007511228,0.00080060743,0.01162108,0.0076471074,0.18868169],"study_design_scores_gemma":[0.0001285387,0.00018136148,0.78838587,0.00016481936,0.0003271073,0.00016002286,0.0077399383,0.18099162,0.00051534525,0.012947571,0.008316915,0.00014083221],"about_ca_topic_score_codex":0.8767906,"about_ca_topic_score_gemma":0.8890676,"teacher_disagreement_score":0.12320942,"about_ca_system_score_codex":0.011332437,"about_ca_system_score_gemma":0.020506782,"threshold_uncertainty_score":0.24787003},"labels":[],"label_agreement":null},{"id":"W3157516646","doi":"10.1177/01632787211012742","title":"Evaluations of Healthcare Providers’ Perceived Support From Personal, Hospital, and System Resources: Implications for Well-Being and Management in Healthcare in Montreal, Quebec, During COVID-19","year":2021,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare professionals’ stress and burnout","field":"Health Professions","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Helpfulness; Burnout; Health care; Personal protective equipment; Nursing; Medicine; Distress; Preparedness; Pandemic; Stress management; Psychology; Coronavirus disease 2019 (COVID-19); Clinical psychology; Disease","score_opus":0.09507221901161102,"score_gpt":0.46099197685500737,"score_spread":0.36591975784339636,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3157516646","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":[0.9972504,0.00019465305,0.000051745294,0.0004897364,0.0000069925995,0.000067985675,0.00021227915,0.0000028043276,0.0017234135],"genre_scores_gemma":[0.999052,0.00017150679,0.000117917065,0.00011059054,0.0000046358755,0.00003297805,0.00011818952,0.0000013260958,0.00039092157],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99795204,0.00083937333,0.00012433008,0.000088572626,0.0004459834,0.00054971874],"domain_scores_gemma":[0.9915802,0.0015432866,0.0018031788,0.00011857895,0.0021225433,0.002832138],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0038697238,0.00023550243,0.00022648447,0.0006802392,0.0026594363,0.0014335943,0.00070051383,0.0003213624,0.0024328285],"category_scores_gemma":[0.009659905,0.00023715952,0.00027096077,0.00084628223,0.0010485427,0.0004758482,0.0011109298,0.0006403118,0.00009802646],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011773283,0.00021835127,0.9664724,0.00010634368,0.00006610925,0.00011125194,0.017042395,0.00013643772,0.00048817662,0.00006167027,0.0013719739,0.013807236],"study_design_scores_gemma":[0.00000828695,0.00010616231,0.9774203,0.000057940768,0.000011650795,0.000011290283,0.021352075,0.00015760737,0.00006561524,0.000007351051,0.0007909994,0.00001073121],"about_ca_topic_score_codex":0.86136377,"about_ca_topic_score_gemma":0.9437826,"teacher_disagreement_score":0.13863623,"about_ca_system_score_codex":0.017244855,"about_ca_system_score_gemma":0.014888847,"threshold_uncertainty_score":0.27890533},"labels":[],"label_agreement":null},{"id":"W3161491150","doi":"10.1177/01632787211018244","title":"What Does the Public Prefer Versus What Is the Public Ready to Forgo?","year":2021,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":3,"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":"Israel National Institute for Health Policy Research","keywords":"Rationing; Quarter (Canadian coin); Population; Medicine; Health care; Public health; Sample (material); Family medicine; Psychology; Business; Actuarial science; Environmental health; Nursing; Economics","score_opus":0.8326587515855285,"score_gpt":0.6383953077165719,"score_spread":0.19426344386895655,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3161491150","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":[0.95782715,0.0011890183,0.00074212084,0.018487569,0.000077296914,0.000059172475,0.00011518788,0.000011367139,0.02149125],"genre_scores_gemma":[0.9981791,0.000327886,0.000107710875,0.0010893457,0.000047972866,0.000012578939,0.000017854032,0.0000018387371,0.00021589716],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928803,0.0041298815,0.00027864394,0.00030348104,0.0013017049,0.0011060719],"domain_scores_gemma":[0.9749299,0.013167271,0.0073353886,0.0007608867,0.0018953809,0.0019112561],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01596557,0.00024251455,0.00040603275,0.0016957751,0.0014318274,0.005638504,0.0004938379,0.002157232,0.0037383758],"category_scores_gemma":[0.035041217,0.00024103989,0.0005447909,0.0011436564,0.004282391,0.005035153,0.0016654363,0.0021393867,0.0004534694],"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":[0.0005256707,0.0008739737,0.8345018,0.00053305493,0.00023632319,0.00037307182,0.034224898,0.00028023784,0.0007390374,0.009857427,0.0023395452,0.1155149],"study_design_scores_gemma":[0.000097802775,0.0011611249,0.71494025,0.0010255836,0.0002561647,0.0010395449,0.24990432,0.0013307438,0.0008636297,0.012146808,0.017106356,0.00012766523],"about_ca_topic_score_codex":0.0041421466,"about_ca_topic_score_gemma":0.0052872617,"teacher_disagreement_score":0.01596557,"about_ca_system_score_codex":0.003027881,"about_ca_system_score_gemma":0.0024321931,"threshold_uncertainty_score":0.084434986},"labels":[],"label_agreement":null},{"id":"W3195518440","doi":"10.1177/01632787211040859","title":"Interdisciplinary Health Care Evaluation Instruments: A Review of Psychometric Evidence","year":2021,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"PsycINFO; CINAHL; Teamwork; MEDLINE; Burnout; Health care; Inclusion (mineral); Medical education; Applied psychology; Medicine; Evidence-based medicine; Psychology; Psychometrics; Nursing; Clinical psychology; Psychological intervention; Social psychology","score_opus":0.4038604056452354,"score_gpt":0.6740222997443843,"score_spread":0.2701618940991489,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3195518440","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":[0.00021107688,0.9991166,0.00012345797,0.00019060007,0.000046396144,0.000071391434,0.000040875828,0.0000036800313,0.00019597524],"genre_scores_gemma":[0.005815994,0.9921748,0.0011728413,0.0003306621,0.000054267013,0.0002951326,0.00010204608,0.000004228559,0.00005006254],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9749835,0.010125864,0.0074770893,0.0013192197,0.005693165,0.00040113702],"domain_scores_gemma":[0.8389195,0.13444129,0.011343145,0.0016713826,0.012950547,0.0006741357],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.041887656,0.0015777901,0.005775798,0.0140708955,0.0009285384,0.0034044548,0.0023115603,0.0019947526,0.0030671791],"category_scores_gemma":[0.12838055,0.0010227584,0.0041697016,0.016957697,0.0016073548,0.0031413229,0.0022849273,0.0016791199,0.00045337324],"study_design_candidate":"systematic_review","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":[0.00019567258,0.00009656459,0.0024509474,0.44268337,0.0020556042,0.000047841655,0.00031250963,0.00016845774,0.000079895675,0.0008605001,0.0045039235,0.5465446],"study_design_scores_gemma":[0.00016156348,0.00025733016,0.012502865,0.92260987,0.009491761,0.00039772844,0.00067279,0.00023862328,0.00020877364,0.0010754436,0.05232832,0.000054918328],"about_ca_topic_score_codex":0.007149867,"about_ca_topic_score_gemma":0.012319478,"teacher_disagreement_score":0.041887656,"about_ca_system_score_codex":0.006176852,"about_ca_system_score_gemma":0.012756298,"threshold_uncertainty_score":0.22152573},"labels":[],"label_agreement":null},{"id":"W4210342638","doi":"10.1177/01632787211067532","title":"Network Meta-Analysis for Single-Case Design Studies: An Illustration","year":2022,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Institute for Public Safety Research and Treatment; University of Regina","funders":"KU Leuven","keywords":"Meta-analysis; Psychological intervention; Relevance (law); Computer science; Network analysis; Single-subject design; Focus (optics); Data science; Psychology; Data mining; Medicine; Psychotherapist; Engineering","score_opus":0.9621373876294022,"score_gpt":0.6045769076283983,"score_spread":0.3575604800010038,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4210342638","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018955029,0.017306672,0.96392596,0.00573606,0.0012173752,0.004218349,0.0009786224,0.0008784813,0.003842937],"genre_scores_gemma":[0.058479823,0.0073593967,0.91351867,0.002100753,0.000414615,0.016609253,0.00034368696,0.00022245853,0.0009513709],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.7070489,0.27386034,0.006925986,0.0053021302,0.006327303,0.00053527247],"domain_scores_gemma":[0.66695267,0.30723003,0.006266276,0.013790792,0.0053089224,0.00045125274],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19088998,0.003068887,0.0057147057,0.0076796124,0.0013813088,0.00413657,0.0051352484,0.0045924443,0.01298567],"category_scores_gemma":[0.3488391,0.0015306125,0.017405806,0.007947219,0.002082828,0.0039424975,0.0048498726,0.005743895,0.0009490467],"study_design_candidate":"not_applicable","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":[0.0032062014,0.00027297478,0.007470551,0.05764444,0.04349621,0.0020530266,0.002116088,0.10385306,0.0013508998,0.41920972,0.045660168,0.31366664],"study_design_scores_gemma":[0.0027574638,0.00065625366,0.0026582736,0.014839114,0.015206944,0.0015435535,0.0003121801,0.29630798,0.0014063199,0.5670476,0.09691088,0.00035337565],"about_ca_topic_score_codex":0.006759563,"about_ca_topic_score_gemma":0.0069169085,"teacher_disagreement_score":0.80911005,"about_ca_system_score_codex":0.0050958036,"about_ca_system_score_gemma":0.0073369476,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4210538766","doi":"10.1177/01632787211073194","title":"Causal Mediation Analysis in Single Case Experimental Designs: Introduction to the Special Issue","year":2022,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":18,"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","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Mediation; Causal inference; Causal model; Computer science; Causal analysis; Outcome (game theory); Variable (mathematics); Psychology; Econometrics; Management science; Data science; Cognitive psychology; Risk analysis (engineering); Medicine; Statistics; Engineering; Mathematics; Social science; Sociology","score_opus":0.5170953304107206,"score_gpt":0.5221333027951004,"score_spread":0.005037972384379752,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4210538766","genre_codex":"editorial","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012389558,0.25467634,0.2837864,0.069344886,0.34676468,0.0027144665,0.0022174784,0.0018058263,0.037450932],"genre_scores_gemma":[0.010007095,0.2540771,0.16679108,0.046469882,0.4711093,0.009421301,0.0025721076,0.003034313,0.03651776],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9826664,0.009948386,0.0015585835,0.0015352615,0.0040043937,0.00028693993],"domain_scores_gemma":[0.86191815,0.120241605,0.0035404263,0.004288181,0.008865482,0.0011461987],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.021848459,0.0019292571,0.002990239,0.004918621,0.0011455696,0.0041641816,0.003016188,0.0057959505,0.029540274],"category_scores_gemma":[0.083340995,0.0012851764,0.0034009959,0.0050895824,0.0028508732,0.005125392,0.0030481943,0.010036798,0.012223701],"study_design_candidate":"theoretical_or_conceptual","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":[0.00007814716,0.00020117172,0.00054286595,0.00489645,0.00022628854,0.00017806161,0.00027128248,0.0007932759,0.0005313591,0.03323901,0.65406966,0.3049725],"study_design_scores_gemma":[0.000042173164,0.00018651021,0.0010313524,0.0034592694,0.0000923628,0.00040782394,0.00009097336,0.0014609572,0.00027712793,0.042586338,0.95029646,0.000068681424],"about_ca_topic_score_codex":0.0008415563,"about_ca_topic_score_gemma":0.0010436286,"teacher_disagreement_score":0.97815156,"about_ca_system_score_codex":0.0022936403,"about_ca_system_score_gemma":0.0025556535,"threshold_uncertainty_score":0.11554706},"labels":[],"label_agreement":null},{"id":"W4214485800","doi":"10.1177/01632787211070811","title":"A Randomization Permutation Test for Single Subject Mediation","year":2022,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institute on Drug Abuse","keywords":"Mediation; Resampling; Permutation (music); Type I and type II errors; Confidence interval; Test (biology); Randomization; Statistical hypothesis testing; Statistics; Outcome (game theory); Computer science; Mathematics; Clinical trial; Medicine; Internal medicine","score_opus":0.7565974041509078,"score_gpt":0.6584428920063305,"score_spread":0.09815451214457727,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4214485800","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":[0.016748719,0.00042856683,0.96913177,0.0020231414,0.0007411499,0.0017267241,0.00066712266,0.0007332931,0.0077994876],"genre_scores_gemma":[0.40934816,0.00041846544,0.56801695,0.0026689274,0.0008176665,0.013645307,0.001109435,0.000418228,0.0035569188],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8656085,0.106711656,0.0040533314,0.008720597,0.013035056,0.0018707814],"domain_scores_gemma":[0.68348,0.27284,0.009725401,0.024330646,0.00819797,0.0014258977],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.079790786,0.0010618733,0.0027494822,0.0023514715,0.001855392,0.00257408,0.0034039533,0.0032659664,0.018933961],"category_scores_gemma":[0.34749854,0.0006909177,0.003424521,0.002771113,0.0054111355,0.005530209,0.0033756064,0.006179487,0.0019693598],"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":[0.0024527544,0.0007015163,0.019275678,0.0011953547,0.0020820773,0.0011501218,0.0016402105,0.01291887,0.002385816,0.5497217,0.018734118,0.3877418],"study_design_scores_gemma":[0.0019544451,0.0040540206,0.012485999,0.0004484418,0.0008054164,0.0018896605,0.0005832619,0.14523236,0.0045745294,0.79631096,0.031352244,0.00030873643],"about_ca_topic_score_codex":0.0010854749,"about_ca_topic_score_gemma":0.0005444382,"teacher_disagreement_score":0.079790786,"about_ca_system_score_codex":0.0016180474,"about_ca_system_score_gemma":0.006402113,"threshold_uncertainty_score":0.42197907},"labels":[],"label_agreement":null},{"id":"W4214524885","doi":"10.1177/01632787211071136","title":"Methods for Modeling Autocorrelation and Handling Missing Data in Mediation Analysis in Single Case Experimental Designs (SCEDs)","year":2022,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Missing data; Autocorrelation; Autoregressive model; Econometrics; Statistics; Type I and type II errors; Ordinary least squares; Computer science; Generalized least squares; Mediation; Imputation (statistics); Data mining; Mathematics","score_opus":0.8558120885164818,"score_gpt":0.6339693109664204,"score_spread":0.22184277755006143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4214524885","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011822933,0.00045267638,0.9945386,0.00033651842,0.00013204261,0.0025399749,0.00013366701,0.00024017719,0.00044415522],"genre_scores_gemma":[0.02305304,0.00046177203,0.9529706,0.0002745954,0.00009320834,0.022623608,0.000117952426,0.00007810162,0.00032708194],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.42375985,0.53808606,0.013216603,0.009362355,0.014612339,0.00096280855],"domain_scores_gemma":[0.29359215,0.6153424,0.029799161,0.048395716,0.012107324,0.00076323247],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.43034732,0.003390539,0.0041050958,0.0057073683,0.0021151104,0.0034913816,0.008722352,0.004842865,0.012026791],"category_scores_gemma":[0.5927532,0.002992553,0.0071758456,0.006849335,0.0052517396,0.0058617974,0.0054925787,0.0067158765,0.001395101],"study_design_candidate":"theoretical_or_conceptual","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":[0.001573161,0.000980997,0.013200638,0.008849276,0.0071958588,0.00039099538,0.0057389126,0.02463168,0.0012729445,0.33841118,0.009238814,0.5885155],"study_design_scores_gemma":[0.003527006,0.0046370286,0.013073984,0.007408552,0.0044714855,0.0006687437,0.0011891294,0.2547024,0.0050899778,0.6442693,0.060339663,0.0006227474],"about_ca_topic_score_codex":0.002141176,"about_ca_topic_score_gemma":0.0030551981,"teacher_disagreement_score":0.5696527,"about_ca_system_score_codex":0.0029809775,"about_ca_system_score_gemma":0.0077476297,"threshold_uncertainty_score":0.7024832},"labels":[],"label_agreement":null},{"id":"W4311582197","doi":"10.1177/01632787221146564","title":"Development and Assessment of the Personal Emotional Capital Questionnaire for Adolescents","year":2022,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","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 British Columbia, Okanagan Campus; University of British Columbia","funders":"","keywords":"Cronbach's alpha; Psychology; Clinical psychology; Confirmatory factor analysis; Convergent validity; Anxiety; Mental health; Discriminant validity; Scale (ratio); DASS; Social psychology; Developmental psychology; Psychometrics; Structural equation modeling; Psychiatry; Statistics; Internal consistency; Mathematics","score_opus":0.09733802043755943,"score_gpt":0.4316695662325183,"score_spread":0.33433154579495883,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311582197","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":[0.9871848,0.0006000242,0.0032537538,0.00030551542,0.00008556857,0.002682085,0.0022418948,0.00003404387,0.003612325],"genre_scores_gemma":[0.9453394,0.0010984961,0.03853007,0.00027740875,0.000055310735,0.00660804,0.0048688264,0.00001981255,0.0032026744],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99870014,0.00036545246,0.0002797213,0.00009872775,0.00044768656,0.00010823957],"domain_scores_gemma":[0.9980058,0.00040029924,0.000654952,0.00009443144,0.0006317264,0.00021273513],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028154547,0.00038682047,0.00056605716,0.0010573413,0.00038723703,0.0005263476,0.0005674936,0.00039021755,0.0016618605],"category_scores_gemma":[0.0049436092,0.00030977593,0.00085774006,0.0006097834,0.00016666549,0.00044925744,0.0007720256,0.0011603086,0.00030964013],"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":[0.00037491933,0.0020108602,0.8045954,0.00054385053,0.00026553302,0.0003578249,0.002614137,0.0010541801,0.0046881884,0.0011153375,0.006503591,0.17587608],"study_design_scores_gemma":[0.00009359554,0.00081549765,0.98675483,0.00011875219,0.000082041246,0.00045485352,0.0007827688,0.0015988871,0.0008910892,0.00039470522,0.007978042,0.000035065572],"about_ca_topic_score_codex":0.0012273331,"about_ca_topic_score_gemma":0.0032533386,"teacher_disagreement_score":0.0028154547,"about_ca_system_score_codex":0.00041693763,"about_ca_system_score_gemma":0.0009112348,"threshold_uncertainty_score":0.014889717},"labels":[],"label_agreement":null},{"id":"W4360804759","doi":"10.1177/01632787231165797","title":"Comparability of Objective Structured Clinical Examinations (OSCEs) and Written Tests for Assessing Medical School Students’ Competencies: A Scoping Review","year":2023,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Comparability; Medical education; MEDLINE; Medicine; Educational measurement; Psychology; Inclusion (mineral); Family medicine; Curriculum; Social psychology; Pedagogy","score_opus":0.5094499951362133,"score_gpt":0.6813970224575658,"score_spread":0.17194702732135259,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4360804759","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":[0.003173995,0.9937716,0.000542113,0.00043121586,0.00025544773,0.00079263694,0.00039722642,0.000008794437,0.0006269509],"genre_scores_gemma":[0.03858388,0.9538448,0.0027168347,0.0008534929,0.0002825973,0.0026581571,0.00087889674,0.000024341682,0.00015709539],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.91772044,0.02261743,0.041915283,0.005457126,0.011470782,0.00081895344],"domain_scores_gemma":[0.6566264,0.27927002,0.034465052,0.005125289,0.023441616,0.0010715962],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.077649355,0.0019744816,0.009624435,0.042698007,0.0013143928,0.007140236,0.0035511944,0.004414778,0.003538078],"category_scores_gemma":[0.30122083,0.002086011,0.009724085,0.027375385,0.002599507,0.0065365206,0.0034625593,0.0018045041,0.000549465],"study_design_candidate":"not_applicable","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":[0.00048549494,0.00008054265,0.0045241797,0.86714864,0.013446922,0.00015876473,0.0007351134,0.00019470988,0.00045216174,0.0005541727,0.0010326626,0.11118656],"study_design_scores_gemma":[0.00015064368,0.0003709099,0.008302468,0.9449887,0.033459667,0.00031226448,0.0008635065,0.00014680332,0.0003735307,0.0003414767,0.010645016,0.000045093868],"about_ca_topic_score_codex":0.0057995706,"about_ca_topic_score_gemma":0.011940305,"teacher_disagreement_score":0.077649355,"about_ca_system_score_codex":0.0059262975,"about_ca_system_score_gemma":0.017711153,"threshold_uncertainty_score":0.41065395},"labels":[],"label_agreement":null},{"id":"W4378782990","doi":"10.1177/01632787231180275","title":"Real Patient Participation in Workplace-Based Assessment of Health Professional Trainees: A Scoping Review","year":2023,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","field":"Medicine","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":"University of British Columbia","funders":"","keywords":"CINAHL; PsycINFO; MEDLINE; Context (archaeology); Health care; Medicine; Intervention (counseling); Medical education; Nursing; Psychology; Family medicine; Psychological intervention","score_opus":0.41688649064158395,"score_gpt":0.6574099080820286,"score_spread":0.24052341744044464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4378782990","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00051877816,0.9983114,0.00013009597,0.00024435396,0.00011089475,0.0001717442,0.000048767193,0.00000319444,0.00046078488],"genre_scores_gemma":[0.004066075,0.99466646,0.00042909713,0.00028824309,0.00005199131,0.00033620658,0.000065901106,0.000002714254,0.000093370945],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9863012,0.0058783395,0.0044028405,0.0006042379,0.0025005622,0.0003127507],"domain_scores_gemma":[0.93393815,0.055051487,0.0058293245,0.0008348561,0.0039985557,0.00034759424],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.017559271,0.0011691599,0.0041671325,0.011203976,0.0009414738,0.004492377,0.0018911905,0.0030310408,0.0033255478],"category_scores_gemma":[0.06250061,0.0009539429,0.0035763208,0.014181224,0.0011962166,0.0037678594,0.0020311698,0.0015513203,0.00054005586],"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":[0.00009703891,0.000060372244,0.00065624353,0.7684337,0.001433114,0.00012774796,0.0011309803,0.00010717245,0.00016458132,0.0006306694,0.0027306245,0.2244277],"study_design_scores_gemma":[0.000029746943,0.00009647519,0.0015849832,0.96682537,0.004151033,0.00027166118,0.0008839504,0.000042127318,0.000119365795,0.00022196435,0.025754059,0.000019236591],"about_ca_topic_score_codex":0.0062578605,"about_ca_topic_score_gemma":0.016482802,"teacher_disagreement_score":0.9824407,"about_ca_system_score_codex":0.0036103947,"about_ca_system_score_gemma":0.013630481,"threshold_uncertainty_score":0.09286344},"labels":[],"label_agreement":null},{"id":"W4399736709","doi":"10.1177/01632787241263370","title":"Consistency Between Administrative Health Records and Self-Reported Health Status and Health Care Use Among Indigenous Wayuu Health Insurance Enrollees: La Guajira, Colombia","year":2024,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Public Health and Social Inequalities","field":"Social Sciences","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":"University of Manitoba; University of Winnipeg","funders":"Inter-American Development Bank","keywords":"Health care; Indigenous; Medicine; Medical record; Population; Family medicine; Demography; Environmental health; Gerontology; Political science","score_opus":0.2684487602612849,"score_gpt":0.5248483726769652,"score_spread":0.25639961241568027,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399736709","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":[0.9974245,0.0001948256,0.00018204896,0.00007768652,0.00000609442,0.00004880963,0.0010997638,0.0000072710072,0.00095912116],"genre_scores_gemma":[0.99875784,0.0000883657,0.00020371468,0.000023924415,0.000004940765,0.000049989078,0.00068446813,0.000002904061,0.00018400684],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9978015,0.0008668986,0.00028269863,0.00034402107,0.00043472744,0.00027014036],"domain_scores_gemma":[0.99499667,0.0014710895,0.0019332349,0.0004924103,0.0008765072,0.00023005456],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002558694,0.0002845714,0.000333925,0.0018754926,0.0006653005,0.0013714051,0.0007418943,0.00022730784,0.0010184171],"category_scores_gemma":[0.009938933,0.00021067464,0.00028701822,0.0018503562,0.00046467944,0.00041375821,0.00092819246,0.0002743817,0.00013497024],"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":[0.00003815189,0.000025485364,0.99579155,0.00003910923,0.00005310619,0.000030533476,0.0014019231,0.000028159946,0.00012353317,0.000023203318,0.0001471026,0.0022981376],"study_design_scores_gemma":[0.0000021254195,0.000011793542,0.997696,0.000018191497,0.000015661155,0.000026889902,0.0018268271,0.00011820674,0.000047377285,0.000008106024,0.00022423036,0.000004588245],"about_ca_topic_score_codex":0.31435052,"about_ca_topic_score_gemma":0.39782938,"teacher_disagreement_score":0.31435052,"about_ca_system_score_codex":0.0018257448,"about_ca_system_score_gemma":0.0009216715,"threshold_uncertainty_score":0.6250416},"labels":[],"label_agreement":null},{"id":"W4401579654","doi":"10.1177/01632787241273911","title":"Physician-Led Synchronous Telemedicine Compared to Face-To-Face Care in Primary Care: A Systematic Review","year":2024,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Telemedicine and Telehealth Implementation","field":"Medicine","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":"Impact; McMaster University","funders":"McMaster University","keywords":"Telemedicine; PsycINFO; MEDLINE; Medicine; Health care; Cochrane Library; Systematic review; Family medicine; Face-to-face; Checklist; Meta-analysis; Grey literature; Psychology","score_opus":0.1268864395004697,"score_gpt":0.5116125494483785,"score_spread":0.3847261099479088,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401579654","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":[0.0019086619,0.99665797,0.00013605597,0.00018257054,0.000101662525,0.0005540231,0.00020448257,0.000008858951,0.0002457444],"genre_scores_gemma":[0.031115608,0.96595466,0.0008761892,0.00046950715,0.000079636884,0.0012148034,0.0001456867,0.000006587592,0.00013723849],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.99141073,0.003766503,0.0025885843,0.00058963994,0.0013891889,0.00025528547],"domain_scores_gemma":[0.9697953,0.023537576,0.004017493,0.00033166815,0.0019533045,0.00036462385],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009358569,0.0011398607,0.007897437,0.0055467337,0.00071353896,0.0028078344,0.0016508271,0.0017675633,0.005011958],"category_scores_gemma":[0.041812386,0.0010007332,0.0070604715,0.0077275774,0.00078864384,0.0020978833,0.0013783028,0.0013946659,0.00027758916],"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":[0.00033854705,0.000027308353,0.00046284022,0.96339357,0.0067527588,0.00004883553,0.0001897301,0.000073005576,0.00008426396,0.000120811594,0.0006613606,0.027846979],"study_design_scores_gemma":[0.0006855805,0.0006115452,0.004631651,0.9078159,0.074854836,0.0002457382,0.0004605002,0.00012222148,0.00023657059,0.00020566864,0.010095449,0.000034463304],"about_ca_topic_score_codex":0.010830391,"about_ca_topic_score_gemma":0.030674748,"teacher_disagreement_score":0.010830391,"about_ca_system_score_codex":0.00541248,"about_ca_system_score_gemma":0.014135605,"threshold_uncertainty_score":0.049493432},"labels":[],"label_agreement":null},{"id":"W4402384985","doi":"10.1177/01632787241281745","title":"The Use of Contribution Analysis in Evaluating Health Interventions: A Scoping Review","year":2024,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":3,"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 Ottawa; Bruyère; NOSM University; Université de Montréal","funders":"Canadian Institutes of Health Research","keywords":"Psychological intervention; Stakeholder; Health promotion; Stakeholder engagement; Psychology; Management science; Medicine; Medical education; Applied psychology; Public health; Public relations; Nursing; Political science; Engineering","score_opus":0.9519790173690221,"score_gpt":0.8310903154610878,"score_spread":0.12088870190793433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402384985","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021411735,0.8981906,0.05039139,0.0048650806,0.0016841562,0.034228962,0.0014265815,0.00021997222,0.0068520554],"genre_scores_gemma":[0.031372894,0.7710162,0.123174235,0.0015901851,0.00047818877,0.07071878,0.0009562545,0.00011912281,0.0005742201],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.65624034,0.2236424,0.080942154,0.0067364313,0.030991064,0.0014476619],"domain_scores_gemma":[0.38841385,0.53201264,0.027100256,0.0153228305,0.036070064,0.0010802901],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.32900405,0.004418265,0.013637393,0.058952715,0.0036590954,0.0133519005,0.0059181787,0.0052620377,0.006252232],"category_scores_gemma":[0.5421055,0.0036742894,0.014645945,0.044587888,0.0054101655,0.01408361,0.008860867,0.0037889045,0.0010176242],"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":[0.00014974241,0.000048235368,0.00073471636,0.73425424,0.006504545,0.00012496242,0.0021239752,0.000960847,0.00024328085,0.010136288,0.003165482,0.24155374],"study_design_scores_gemma":[0.00017102303,0.00020016845,0.0007485069,0.93770754,0.01352171,0.0001337138,0.0010766387,0.0009198805,0.00036494812,0.008860835,0.036205623,0.000089469235],"about_ca_topic_score_codex":0.009927313,"about_ca_topic_score_gemma":0.018875899,"teacher_disagreement_score":0.67099595,"about_ca_system_score_codex":0.01617443,"about_ca_system_score_gemma":0.060484935,"threshold_uncertainty_score":0.8274575},"labels":[],"label_agreement":null},{"id":"W4402937957","doi":"10.1177/01632787241286912","title":"Professionalism in Residency Training: The Learning Environment for Professionalism Survey","year":2024,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","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":true,"ca_institutions":"Children's Hospital of Eastern Ontario; Ottawa Hospital; University of Ottawa","funders":"University of Ottawa","keywords":"Bonferroni correction; Univariate; Medical education; Metric (unit); Psychology; Residency training; Medicine; Computer science; Multivariate statistics; Operations management; Statistics","score_opus":0.4312295384084632,"score_gpt":0.5618547651496061,"score_spread":0.1306252267411429,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402937957","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":[0.9786426,0.00022142581,0.0010401323,0.0019269565,0.00003771103,0.0011165442,0.009741696,0.00007204752,0.0072010155],"genre_scores_gemma":[0.97171766,0.00056661374,0.00638238,0.0011364489,0.000055200504,0.0025204998,0.011353185,0.000038255384,0.0062297652],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964606,0.0009581107,0.0005489153,0.00014904446,0.0012763137,0.0006069741],"domain_scores_gemma":[0.9904028,0.0010197702,0.003035419,0.0003227171,0.0025316312,0.0026875157],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004763342,0.00017037544,0.00036603902,0.0011493164,0.00091961416,0.00072965,0.0006769639,0.0005121007,0.0018992237],"category_scores_gemma":[0.008000965,0.00032305557,0.00049428036,0.0014205951,0.00042960615,0.0008129017,0.0019983964,0.0014429973,0.00088147505],"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":[0.0001481306,0.000710593,0.954527,0.00017277621,0.00004839447,0.00007667454,0.00200926,0.0003422043,0.00047332761,0.00016425029,0.012915314,0.028411917],"study_design_scores_gemma":[0.000015938906,0.00024164002,0.9912417,0.000035102013,0.000004968308,0.00005482467,0.0010947346,0.0002592998,0.00012179962,0.000056103934,0.0068583223,0.000015587719],"about_ca_topic_score_codex":0.034642726,"about_ca_topic_score_gemma":0.108659215,"teacher_disagreement_score":0.034642726,"about_ca_system_score_codex":0.0026358024,"about_ca_system_score_gemma":0.004391048,"threshold_uncertainty_score":0.06888217},"labels":[],"label_agreement":null},{"id":"W4403129632","doi":"10.1177/01632787241288225","title":"Health Care Utilization and Perceived Quality of Care in a Colombian Indigenous Health Organization","year":2024,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Indigenous Health and Education","field":"Health Professions","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":"University of Manitoba; University of Winnipeg","funders":"Inter-American Development Bank","keywords":"Indigenous; Health care; Service (business); Medicine; Business; Environmental health; Political science; Marketing","score_opus":0.1924636436143436,"score_gpt":0.5547069293277638,"score_spread":0.3622432857134202,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403129632","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":[0.99923444,0.00008969002,0.000015117866,0.0000685823,0.0000019660622,0.000006986255,0.000049659426,5.605757e-7,0.00053289696],"genre_scores_gemma":[0.99965394,0.000095434385,0.000031319185,0.000026607635,0.000002458096,0.000005890596,0.000046423705,3.969448e-7,0.00013753894],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9995741,0.00015223233,0.000026087906,0.000037748698,0.000058308964,0.0001514706],"domain_scores_gemma":[0.9989214,0.00019705083,0.0003827796,0.000035748715,0.00016519826,0.00029791947],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007016212,0.00017638605,0.00016022283,0.00058272114,0.00089621363,0.0009498322,0.00028802865,0.00021360976,0.0013546016],"category_scores_gemma":[0.0020416947,0.00011186452,0.00021536587,0.00045122308,0.00043304905,0.00031578503,0.00075665023,0.0004879915,0.00008308311],"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":[0.00004570059,0.0002122754,0.9898253,0.000035991252,0.00003215852,0.00008937217,0.0055157123,0.00003357596,0.00020284709,0.000055557844,0.00021507464,0.0037363945],"study_design_scores_gemma":[0.0000041310855,0.000057958758,0.98691756,0.000027243697,0.000011882724,0.000039168975,0.01246008,0.00010801996,0.000031941974,0.000011131456,0.0003258687,0.000004964508],"about_ca_topic_score_codex":0.22151242,"about_ca_topic_score_gemma":0.35466358,"teacher_disagreement_score":0.22151242,"about_ca_system_score_codex":0.0020188564,"about_ca_system_score_gemma":0.0010503953,"threshold_uncertainty_score":0.4404462},"labels":[],"label_agreement":null},{"id":"W4403899198","doi":"10.1177/01632787241286911","title":"Who’s at the Table? A Scoping Review of Stakeholder Engagement in Medical Education Program Evaluation","year":2024,"lang":"en","type":"review","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","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":"Université de Montréal; University of Ottawa","funders":"","keywords":"Stakeholder; Viewpoints; Stakeholder engagement; Medical education; Accountability; Inclusion (mineral); Public relations; Stakeholder analysis; Psychology; Program evaluation; Best practice; Medicine; Political science","score_opus":0.4900200572623418,"score_gpt":0.6423792961658585,"score_spread":0.1523592389035167,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403899198","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00045006882,0.99338585,0.00087581866,0.0032263831,0.0005186058,0.00064869237,0.0000971674,0.000011301695,0.00078605197],"genre_scores_gemma":[0.006289864,0.9877908,0.0022767312,0.0014956115,0.00017115673,0.0016522853,0.00012559361,0.000012851159,0.00018513181],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9045786,0.05677706,0.02400518,0.0022405107,0.011215178,0.0011834049],"domain_scores_gemma":[0.7443438,0.2043609,0.020411557,0.0039341394,0.025664426,0.0012852112],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0897622,0.0016889909,0.00686567,0.01982121,0.0022420038,0.0073371422,0.0026351807,0.0047280374,0.0033498218],"category_scores_gemma":[0.25886342,0.00168137,0.0067193536,0.019762123,0.0028161283,0.009713851,0.004155306,0.0031681226,0.00070591364],"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":[0.00016690171,0.00004929833,0.0006727431,0.67252773,0.0023656215,0.00011873314,0.0019643784,0.00021815926,0.00027854886,0.0027181075,0.008628824,0.31029096],"study_design_scores_gemma":[0.000032660868,0.00004849725,0.00048213048,0.9580816,0.0028058502,0.000077625686,0.0007131212,0.000061631676,0.0001157361,0.0009311331,0.03662793,0.000022004622],"about_ca_topic_score_codex":0.00962323,"about_ca_topic_score_gemma":0.02475935,"teacher_disagreement_score":0.9102378,"about_ca_system_score_codex":0.007885991,"about_ca_system_score_gemma":0.036514018,"threshold_uncertainty_score":0.47471356},"labels":[],"label_agreement":null},{"id":"W4404471000","doi":"10.1177/01632787231217000","title":"Estimating Mediation Effects in ABAB Reversal Designs","year":2024,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Behavioral and Psychological Studies","field":"Psychology","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":"McGill University","funders":"National Institute on Drug Abuse","keywords":"Mediation; Intervention (counseling); Outcome (game theory); Psychology; Psychological intervention; Causal inference; Quasi-experiment; Inference; Variable (mathematics); Computer science; Econometrics; Medicine; Mathematics; Psychiatry; Artificial intelligence","score_opus":0.5129719809319946,"score_gpt":0.5496520292627413,"score_spread":0.03668004833074667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404471000","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":[0.03711773,0.0012908018,0.9369856,0.00037791533,0.0005849885,0.017604837,0.0012151324,0.0010037634,0.003819209],"genre_scores_gemma":[0.097392365,0.00034259754,0.8504048,0.00034880888,0.00009675797,0.049845442,0.0005994518,0.00010814585,0.0008616085],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6450679,0.3242086,0.008547202,0.010442004,0.010341671,0.0013926742],"domain_scores_gemma":[0.50160223,0.42686713,0.02433856,0.035665985,0.01084116,0.00068494125],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.25727004,0.0032220385,0.0036257333,0.0036819566,0.0015767772,0.0026519482,0.0047321115,0.0037963542,0.01217283],"category_scores_gemma":[0.4087277,0.0024748836,0.006909578,0.0030344995,0.003677602,0.0034076092,0.003771205,0.0049658604,0.0017617968],"study_design_candidate":"theoretical_or_conceptual","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":[0.020442273,0.0045088823,0.047199752,0.009871484,0.008303499,0.00063570787,0.0046780757,0.035987604,0.0061109583,0.23546374,0.009319614,0.61747843],"study_design_scores_gemma":[0.013770399,0.037127126,0.03349995,0.0033302733,0.004863556,0.0008328303,0.0017186368,0.41803908,0.0148514835,0.41793862,0.053324394,0.0007035905],"about_ca_topic_score_codex":0.0016098713,"about_ca_topic_score_gemma":0.0012276792,"teacher_disagreement_score":0.25727004,"about_ca_system_score_codex":0.0023854498,"about_ca_system_score_gemma":0.0032510292,"threshold_uncertainty_score":0.9159183},"labels":[],"label_agreement":null},{"id":"W4405221032","doi":"10.1177/01632787241307031","title":"Validity and Reliability of the Turkish Version of the Low Back Activity Confidence Scale (LoBACS)","year":2024,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Musculoskeletal pain and rehabilitation","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":"","keywords":"Cronbach's alpha; Construct validity; Exploratory factor analysis; Confirmatory factor analysis; Psychology; Concurrent validity; Physical therapy; Reliability (semiconductor); Low back pain; Clinical psychology; Turkish; Face validity; Medicine; Psychometrics; Internal consistency; Structural equation modeling; Statistics","score_opus":0.05356067794905782,"score_gpt":0.40889571549370596,"score_spread":0.3553350375446481,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405221032","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":[0.99524975,0.00038867985,0.00081707607,0.00012219755,0.00006731851,0.00011152316,0.000313562,0.000017560027,0.0029123593],"genre_scores_gemma":[0.9963576,0.000237874,0.0019984536,0.000051636092,0.00002535148,0.00024629576,0.0005342145,0.000013470427,0.00053512],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9957349,0.0014832373,0.00067148753,0.0004218265,0.0014413649,0.00024718296],"domain_scores_gemma":[0.99392974,0.0015868127,0.0009903038,0.00034043696,0.0029092357,0.00024353889],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0050861835,0.00038557313,0.00054149213,0.0011652336,0.00041769192,0.0006922698,0.00040117747,0.00030656205,0.0012830232],"category_scores_gemma":[0.0131911775,0.00024079849,0.00081896887,0.00078624336,0.00059202174,0.00043097325,0.0009815033,0.00041273359,0.0004866331],"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":[0.0008891231,0.00041983364,0.856464,0.00048017592,0.00039924323,0.0003398631,0.0074158106,0.0009775775,0.004364851,0.00054252386,0.003057784,0.1246492],"study_design_scores_gemma":[0.000090115434,0.0005015354,0.99075586,0.00017773564,0.000105467,0.0004819786,0.0016105049,0.0017544432,0.0008279234,0.00026975616,0.003382434,0.00004221761],"about_ca_topic_score_codex":0.0032627597,"about_ca_topic_score_gemma":0.004197227,"teacher_disagreement_score":0.0050861835,"about_ca_system_score_codex":0.00062376424,"about_ca_system_score_gemma":0.00075800484,"threshold_uncertainty_score":0.026898623},"labels":[],"label_agreement":null},{"id":"W4406036650","doi":"10.1177/01632787241311360","title":"Commentary: Review of Mandatory Maintenance of Certification in the USA","year":2025,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Innovations in Medical Education","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":"McGill University; Jewish General Hospital","funders":"","keywords":"Certification; Maintenance of Certification; Medical education; Medicine; Quality (philosophy); Quality assurance; Lifelong learning; Nursing; Psychology; Political science; Pedagogy","score_opus":0.14278096329458848,"score_gpt":0.5272376957501023,"score_spread":0.3844567324555138,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406036650","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00016858341,0.3959781,0.00019345278,0.5079366,0.09202599,0.00006195636,0.00070936,0.0000333178,0.002892707],"genre_scores_gemma":[0.004601545,0.24987546,0.00047591396,0.66885537,0.07340059,0.00020047344,0.000370469,0.00005416441,0.0021660167],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97982556,0.0075709764,0.004355249,0.0014637442,0.0060550775,0.0007292483],"domain_scores_gemma":[0.81388575,0.118105285,0.015391714,0.0014750127,0.047816377,0.0033259108],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.016816555,0.0010971699,0.0032908674,0.0051631588,0.001852599,0.003177147,0.0043990538,0.015478318,0.008039353],"category_scores_gemma":[0.13900992,0.00092669873,0.002841534,0.0060966373,0.0034572575,0.0042962967,0.002035675,0.00975827,0.00292892],"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":[0.00006133628,0.000006448079,0.00008472199,0.019131294,0.00015512154,0.00017012283,0.00019618476,0.000035741294,0.000044256372,0.0011573208,0.95803785,0.020919576],"study_design_scores_gemma":[0.000103053644,0.00004544614,0.0010054345,0.126333,0.0005828356,0.00067609147,0.00033172965,0.00005102951,0.0001199479,0.0019498116,0.8687528,0.000048896374],"about_ca_topic_score_codex":0.027682146,"about_ca_topic_score_gemma":0.044911176,"teacher_disagreement_score":0.98318344,"about_ca_system_score_codex":0.010476152,"about_ca_system_score_gemma":0.028007664,"threshold_uncertainty_score":0.088935494},"labels":[],"label_agreement":null},{"id":"W4409084607","doi":"10.1177/01632787251331713","title":"Factors Associated With the Workforce Participation Intentions of Australian Primary Health Care Nurses and Midwives","year":2025,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Global Health Workforce Issues","field":"Health Professions","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":"Workforce; Rurality; Nursing; Quarter (Canadian coin); Job satisfaction; Scope of practice; Medicine; Logistic regression; Scale (ratio); Explanatory model; Health care; Workforce planning; Work (physics); Psychology; Medical education; Rural area; Social psychology","score_opus":0.21612117327343333,"score_gpt":0.5374015665716465,"score_spread":0.3212803932982132,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409084607","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":[0.99838305,0.00014069979,0.000044401906,0.00033003473,0.000009430533,0.000015819234,0.00006439371,0.0000010233081,0.0010111799],"genre_scores_gemma":[0.9994717,0.00009078976,0.00005227393,0.000051236468,0.0000049047344,0.0000186236,0.000055500463,6.698478e-7,0.00025428427],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9987382,0.00042573418,0.000118365104,0.00012372421,0.00032026434,0.00027369874],"domain_scores_gemma":[0.99334407,0.0024937144,0.0021137323,0.00019939433,0.00063222816,0.001216847],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028643776,0.00016993219,0.00022208282,0.00043466242,0.00060655334,0.00089746085,0.0004320714,0.0005668849,0.0039257477],"category_scores_gemma":[0.013441894,0.00023386876,0.00037766673,0.00047628404,0.00051489426,0.0003884242,0.001041563,0.000804302,0.00035436844],"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":[0.000025555562,0.00010921309,0.9931798,0.000040670973,0.000013285313,0.000051632218,0.0027006008,0.000034425648,0.00006532298,0.000026026939,0.00013673645,0.0036167116],"study_design_scores_gemma":[0.0000013033225,0.00005885178,0.99747086,0.00003454192,0.0000042653883,0.0000337434,0.0021004502,0.00007790626,0.000011827213,0.000022803528,0.00018138449,0.00000207688],"about_ca_topic_score_codex":0.029891789,"about_ca_topic_score_gemma":0.02543483,"teacher_disagreement_score":0.029891789,"about_ca_system_score_codex":0.000458541,"about_ca_system_score_gemma":0.0016681199,"threshold_uncertainty_score":0.059435606},"labels":[],"label_agreement":null},{"id":"W4410594100","doi":"10.1177/01632787251341460","title":"Evaluation of the Effects of the Strengthening Families Program in Quebec Adolescents and Parents Living in Challenging Family Conditions","year":2025,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","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 du Québec à Trois-Rivières","funders":"","keywords":"Intervention (counseling); Multivariate analysis; Psychology; Medicine; Clinical psychology; Longitudinal study; Developmental psychology; Psychiatry","score_opus":0.05327230922564062,"score_gpt":0.4147753659340637,"score_spread":0.3615030567084231,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410594100","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":[0.99855393,0.00010127862,0.000042467196,0.00015060307,0.0000071991094,0.00042496587,0.000110195964,0.000007944936,0.00060147606],"genre_scores_gemma":[0.99727434,0.00020603689,0.00074995053,0.00008607992,0.000011260357,0.00080246985,0.00018102494,0.0000022035595,0.0006865587],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99836665,0.0008598339,0.000051405907,0.00010097785,0.00023106192,0.00039010725],"domain_scores_gemma":[0.9976835,0.0004372557,0.00033132586,0.00009781046,0.00041867341,0.0010314339],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022834484,0.00043802432,0.00039409162,0.0004096227,0.0019492556,0.0004756754,0.00091685867,0.000412696,0.001605167],"category_scores_gemma":[0.003464673,0.00022520195,0.0004674138,0.00042694228,0.000587166,0.00032134648,0.0006846328,0.00060441525,0.00009250322],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.022600288,0.0746808,0.44800514,0.0008957333,0.0006687716,0.0009976854,0.008726999,0.003525216,0.0053305626,0.00047268163,0.0047117206,0.4293844],"study_design_scores_gemma":[0.0026305232,0.032608658,0.956643,0.00012788446,0.00019504924,0.0000637833,0.0036724664,0.00094292406,0.001071434,0.00004684482,0.0019713049,0.000026179561],"about_ca_topic_score_codex":0.6470448,"about_ca_topic_score_gemma":0.83142537,"teacher_disagreement_score":0.35295522,"about_ca_system_score_codex":0.008933259,"about_ca_system_score_gemma":0.016546782,"threshold_uncertainty_score":0.7100677},"labels":[],"label_agreement":null},{"id":"W4411348215","doi":"10.1177/01632787251348594","title":"Adaptation of the Sinclair Compassion Questionnaire Into Turkish: A Validity and Reliability Study","year":2025,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Mindfulness and Compassion Interventions","field":"Psychology","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":"University of Calgary","funders":"","keywords":"Cronbach's alpha; Confirmatory factor analysis; Psychology; Convergent validity; Exploratory factor analysis; Validity; Content validity; Reliability (semiconductor); Item analysis; Scale (ratio); Clinical psychology; Concurrent validity; Compassion; Turkish; Psychometrics; Structural equation modeling; Statistics; Internal consistency; Mathematics","score_opus":0.13492284466682755,"score_gpt":0.48164690358716256,"score_spread":0.346724058920335,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411348215","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":[0.9950134,0.00020249565,0.0018689755,0.00014867324,0.000077860466,0.0006627708,0.00024588607,0.000020638054,0.0017592832],"genre_scores_gemma":[0.9863107,0.00036457577,0.009852736,0.0001798856,0.000042095544,0.0019161996,0.00062071485,0.00002647113,0.00068664516],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9968971,0.0012072839,0.00050694676,0.00024830268,0.00091020437,0.0002300993],"domain_scores_gemma":[0.9938408,0.0017212745,0.0008058939,0.00037513732,0.002961374,0.00029547594],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0065463223,0.00040044976,0.0005583756,0.0013728543,0.0007361829,0.0006180751,0.00041336988,0.0003056023,0.0014491852],"category_scores_gemma":[0.011871414,0.00035164488,0.000831504,0.000990885,0.00057684234,0.00060381263,0.0011068485,0.0006266662,0.00037803108],"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":[0.0006630281,0.0018078776,0.7651763,0.0010015328,0.00024566028,0.000877393,0.02347773,0.0009839848,0.006412824,0.00090945937,0.0037960608,0.19464813],"study_design_scores_gemma":[0.0001930746,0.0021500273,0.9642024,0.00039828476,0.00012632446,0.0016281856,0.0144598065,0.0029020521,0.0020797008,0.0004933813,0.011264321,0.00010255963],"about_ca_topic_score_codex":0.0016996929,"about_ca_topic_score_gemma":0.002857879,"teacher_disagreement_score":0.0065463223,"about_ca_system_score_codex":0.0007554751,"about_ca_system_score_gemma":0.0019383369,"threshold_uncertainty_score":0.034620643},"labels":[],"label_agreement":null},{"id":"W4413298878","doi":"10.1177/01632787251368457","title":"Factors Influencing the Reporting of Notifiable Conduct in Health Professionals in Australia","year":2025,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Medical Malpractice and Liability Issues","field":"Health Professions","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":"Child, Adolescent and Family Mental Health","funders":"","keywords":"Misconduct; Sexual misconduct; Health care; Theory of planned behavior; Medicine; Psychology; Health professionals; Professional conduct; Agency (philosophy); Family medicine; Control (management); Criminology","score_opus":0.558608968909628,"score_gpt":0.6383305645727432,"score_spread":0.07972159566311521,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413298878","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":[0.99859136,0.00017389168,0.00013437937,0.0005312071,0.000005542871,0.000029005532,0.000013453919,0.0000019260117,0.0005190862],"genre_scores_gemma":[0.9993486,0.00017786078,0.00021242755,0.000096316595,0.000004544819,0.000017679899,0.000016065065,8.7407363e-7,0.00012570767],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98998815,0.0050703483,0.0013100252,0.00038778526,0.002600617,0.0006431916],"domain_scores_gemma":[0.93444365,0.026996963,0.024715269,0.0012521151,0.0081018945,0.0044900174],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.007823891,0.00013708719,0.00024294037,0.00095929933,0.0009564884,0.00082670775,0.00040890818,0.0005584166,0.0011355955],"category_scores_gemma":[0.07347739,0.00031571128,0.00018940201,0.0006650292,0.0010923222,0.00057801214,0.0012272408,0.0007629225,0.00012833023],"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":[0.000033150718,0.00010519405,0.9780286,0.00008935471,0.000015550402,0.00011068445,0.014092652,0.000033546992,0.00027337848,0.000046952282,0.00020665131,0.006964416],"study_design_scores_gemma":[0.0000042358156,0.00016217849,0.98378444,0.00010713345,0.000009094305,0.0002703809,0.014588683,0.0003086684,0.00010556901,0.00007434411,0.00057504943,0.000010211895],"about_ca_topic_score_codex":0.027772348,"about_ca_topic_score_gemma":0.03373496,"teacher_disagreement_score":0.99944156,"about_ca_system_score_codex":0.0014706091,"about_ca_system_score_gemma":0.0046083294,"threshold_uncertainty_score":0.05522138},"labels":[],"label_agreement":null}]}