{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":82,"total_is_capped":false,"direct_labels_cover":2,"predictions_cover":82,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"8abaf8d661e6","filters":{"venue":"Evaluation & the Health Professions"}},"results":[{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Bernard Charlin","is_ca":true},{"name":"Cees van der Vleuten","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1004936907630079,"gpt":0.5254536042443122,"spread":0.4249599134813042,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007475629,0.00007780342,0.0004125875,0.00007553025,0.0001077229,0.000001947237,0.00006725996,0.00006764568,0.000152867],"category_scores_gemma":[0.02828288,0.00004761865,0.00007337005,0.000303967,0.00009508114,0.00002336801,0.00003275862,0.0002966629,0.000002428985],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003519237,"about_ca_system_score_gemma":0.004706956,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005610876,"about_ca_topic_score_gemma":0.0001684031,"domain_scores_codex":[0.9974061,0.0007308756,0.0007802877,0.0001516157,0.0007406235,0.0001905256],"domain_scores_gemma":[0.9942755,0.004363343,0.0004640401,0.0003279378,0.00045543,0.0001137842],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003826296,0.00851727,0.4390731,0.001428976,0.0004121723,0.00001469236,0.04592658,0.06241105,0.002112856,0.1059688,0.007622115,0.322686],"study_design_scores_gemma":[0.01632683,0.001170417,0.9387864,0.02097092,0.0001376203,0.000004773437,0.003645034,0.00737332,0.000215076,0.01113577,0.0001309449,0.0001028629],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.962121,0.0006140629,0.00156059,0.03217416,0.0003010174,0.001975537,0.00002052725,0.0000173574,0.001215742],"genre_scores_gemma":[0.9961286,0.0001262263,0.002672793,0.0008253938,0.00003233995,0.0001056574,0.00003717862,0.00000805834,0.00006377699],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4997133,"threshold_uncertainty_score":0.9799023,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Thomas Rotter","is_ca":true},{"name":"Christopher Plishka","is_ca":true},{"name":"Adegboyega Lawal","is_ca":true},{"name":"Elizabeth Harrison","is_ca":true},{"name":"Nazmi Sari","is_ca":true},{"name":"Donna Goodridge","is_ca":true},{"name":"Rachel Flynn","is_ca":true},{"name":"James G. Chan","is_ca":true},{"name":"Michelle Fiander","is_ca":false},{"name":"Bozena Poksińska","is_ca":false},{"name":"Keith A. Willoughby","is_ca":true},{"name":"Leigh Kinsman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3840186299789496,"gpt":0.5973783700362918,"spread":0.2133597400573422,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.03361221,0.0002473028,0.0008900221,0.0002571214,0.003338675,0.00001708531,0.0003692278,0.0001334174,0.000433956],"category_scores_gemma":[0.0005238719,0.0001784303,0.00008121887,0.0005870603,0.00007996062,0.0003887379,0.0001505694,0.0004264822,0.0001596486],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00156449,"about_ca_system_score_gemma":0.005185088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002499195,"about_ca_topic_score_gemma":0.003707333,"domain_scores_codex":[0.9859534,0.007899615,0.003554093,0.0005093158,0.001293333,0.0007902559],"domain_scores_gemma":[0.9951494,0.0007703275,0.001596715,0.0008675164,0.001359128,0.0002569405],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001119237,0.0003127777,0.00002354868,0.7455627,0.00007915978,2.028577e-7,0.1856917,0.000005000334,0.000001735255,0.03554634,0.01284847,0.01981649],"study_design_scores_gemma":[0.004100814,0.0009808651,0.002219564,0.640892,0.0002245702,0.00000139166,0.3259367,0.002054166,0.00001396525,0.00798649,0.0151322,0.0004572231],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0170908,0.1062769,0.02507068,0.6595204,0.008013126,0.1810478,0.0002171297,0.0003155667,0.002447475],"genre_scores_gemma":[0.2727705,0.04615095,0.07265533,0.517468,0.0009391158,0.08260719,0.003599223,0.0002035715,0.003606119],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.2556797,"threshold_uncertainty_score":0.9979588,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Beverley Shea","is_ca":true},{"name":"David Moher","is_ca":true},{"name":"Ian D. Graham","is_ca":false},{"name":"Ba’ Pham","is_ca":true},{"name":"Peter Tugwell","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9446355596120252,"gpt":0.7095874658572304,"spread":0.2350480937547947,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6237694,0.000243478,0.007099913,0.0003012999,0.0004433496,0.0001408656,0.001525024,0.0000900759,0.01067722],"category_scores_gemma":[0.3007449,0.00007661202,0.00107818,0.002794458,0.0001218148,0.0002540565,0.0001242331,0.0004191804,0.0001697693],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000853097,"about_ca_system_score_gemma":0.0002776075,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003992135,"about_ca_topic_score_gemma":0.0003307006,"domain_scores_codex":[0.4831352,0.4463927,0.05537923,0.001017747,0.01355792,0.0005172031],"domain_scores_gemma":[0.9180993,0.02368963,0.0485283,0.006616124,0.002852791,0.0002138302],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00003676068,0.001610428,0.1668931,0.1189513,0.0001973977,1.408512e-7,0.05001942,0.0005622365,0.0006072525,0.003292631,0.5925928,0.06523652],"study_design_scores_gemma":[0.003816311,0.000367052,0.4361905,0.1621929,0.001764093,0.000009034839,0.02031208,0.3058308,0.00005323104,0.01791816,0.05079109,0.000754679],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.08651979,0.6729706,0.004825148,0.177311,0.001504403,0.05416443,0.00003125914,0.000007982043,0.002665347],"genre_scores_gemma":[0.9920588,0.002479959,0.0009606755,0.002124165,0.00003250911,0.0009977792,0.0000029766,0.000008938842,0.001334174],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.905539,"threshold_uncertainty_score":0.9902272,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"K. Ann McKibbon","is_ca":true},{"name":"Nancy L Wilczynski","is_ca":false},{"name":"R. Brian Haynes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8112156344634597,"gpt":0.7367197101535863,"spread":0.0744959243098734,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.04839377,0.0001857184,0.0003804736,0.000366749,0.005441389,0.00003100963,0.0003476802,0.0001535451,0.0001455565],"category_scores_gemma":[0.00421265,0.0001223346,0.00005144308,0.001461801,0.0002652905,0.0005129624,0.0001340567,0.001019062,0.0001070588],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002328689,"about_ca_system_score_gemma":0.0237516,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002288453,"about_ca_topic_score_gemma":0.004248257,"domain_scores_codex":[0.9856902,0.009710085,0.001484336,0.0004462409,0.001197493,0.001471623],"domain_scores_gemma":[0.9886871,0.00843921,0.000489988,0.0003663594,0.001826481,0.0001908708],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002555132,0.0001378994,0.004388391,0.002414042,0.00002424101,3.674173e-7,0.7210635,0.001616107,0.00005496251,0.1028506,0.1605084,0.006685943],"study_design_scores_gemma":[0.001248255,0.000200529,0.02803734,0.001495088,0.00000712422,4.933818e-7,0.9034892,0.00930554,0.000007583825,0.04917781,0.006860697,0.0001703254],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6560761,0.003951841,0.02861397,0.288047,0.004225225,0.01671515,0.00004290661,0.0001343628,0.002193421],"genre_scores_gemma":[0.9719552,0.0004020606,0.01301946,0.004002471,0.0009688166,0.0068309,0.0001236772,0.00003116547,0.002666212],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3158792,"threshold_uncertainty_score":0.9958534,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Stéphanie Baggio","is_ca":false},{"name":"Joseph Studer","is_ca":false},{"name":"Katia Iglesias","is_ca":false},{"name":"Jean‐Bernard Daeppen","is_ca":false},{"name":"Gerhard Gmel","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0648383612560048,"gpt":0.4448959682180539,"spread":0.3800576069620492,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005164455,0.0001436878,0.0002619839,0.0001884293,0.0003592133,0.000005058242,0.0002027866,0.000165054,0.006636363],"category_scores_gemma":[0.0003530668,0.00008024647,0.00006060529,0.0004621965,0.00008370245,0.00006480257,0.00003779081,0.0002937936,0.0008292928],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001499452,"about_ca_system_score_gemma":0.000147792,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001735263,"about_ca_topic_score_gemma":0.0001790381,"domain_scores_codex":[0.9950728,0.003086461,0.0005739111,0.0003505938,0.0004639952,0.0004522071],"domain_scores_gemma":[0.9982464,0.0006560241,0.0004577689,0.0004184617,0.000125398,0.00009592734],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0006180895,0.0006805135,0.03659626,0.00007336166,0.00008611631,0.000001752892,0.09684216,0.0000127386,0.003953916,0.01598871,0.07699449,0.7681519],"study_design_scores_gemma":[0.005834701,0.0007614599,0.9056399,0.00178418,0.00005548526,0.000008161741,0.01076696,0.0004244372,0.0001590163,0.05068279,0.0234693,0.0004136226],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6835883,0.0005098663,0.001092393,0.2553509,0.004407594,0.002586558,0.00001562124,0.0001541094,0.05229462],"genre_scores_gemma":[0.9934935,0.0001050287,0.0001051953,0.002456987,0.0003277379,0.0004152893,0.000009007877,0.00002017583,0.003067069],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8690436,"threshold_uncertainty_score":0.9999487,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Glynis M. Breakwell","is_ca":false},{"name":"Emanuele Fino","is_ca":false},{"name":"Rusi Jaspal","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4151495611428296,"gpt":0.5796821983329958,"spread":0.1645326371901663,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.007656274,0.0001464761,0.0001623544,0.00008109094,0.002621704,0.00004286091,0.0002247033,0.00007234051,0.0005143558],"category_scores_gemma":[0.0009075141,0.00007987682,0.00003173256,0.0006450366,0.0001169177,0.00006063873,0.0001655247,0.0004846366,0.00004306938],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001041443,"about_ca_system_score_gemma":0.006085379,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01848389,"about_ca_topic_score_gemma":0.03247897,"domain_scores_codex":[0.9912586,0.006804122,0.0006340846,0.0003677733,0.000538965,0.0003965091],"domain_scores_gemma":[0.9947872,0.004006493,0.0003592,0.0005085302,0.0001565372,0.0001820108],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004741926,0.0007806178,0.4872341,0.000105942,0.0001133644,0.000005960102,0.3204426,0.0002850535,0.00006388204,0.02339605,0.01668184,0.1504163],"study_design_scores_gemma":[0.002515705,0.00004937716,0.8054467,0.0001779716,0.00004771611,0.000004885982,0.04329635,0.0003496716,0.00004653095,0.007017854,0.1409045,0.0001427518],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.877888,0.001096482,0.001470939,0.1156256,0.00118319,0.002582235,0.00003916787,0.00003517995,0.00007929737],"genre_scores_gemma":[0.978641,0.0001131111,0.0001853351,0.01852114,0.00009797352,0.001221331,0.0004933658,0.00001717761,0.0007095304],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3182126,"threshold_uncertainty_score":0.9995492,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Janet Curran","is_ca":true},{"name":"Andrea Murphy","is_ca":true},{"name":"Syed Sibte Raza Abidi","is_ca":true},{"name":"Douglas Sinclair","is_ca":true},{"name":"Patrick J. McGrath","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2013394292551039,"gpt":0.45463960418971,"spread":0.2533001749346061,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.01452764,0.0001013081,0.0001876819,0.0001791943,0.001400121,0.00002609498,0.0002884004,0.00007450375,0.00003554368],"category_scores_gemma":[0.001896572,0.0000630375,0.00002627366,0.0007183667,0.00003732273,0.0004280784,0.00022756,0.0007321754,0.00001489173],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007880875,"about_ca_system_score_gemma":0.0001935085,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005470217,"about_ca_topic_score_gemma":0.003767921,"domain_scores_codex":[0.9977943,0.0008915435,0.000616132,0.0001510237,0.0002683783,0.0002786394],"domain_scores_gemma":[0.9983398,0.0002969081,0.0005889239,0.0004254395,0.0003352448,0.00001367337],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00006855054,0.0005199466,0.02524086,0.0007388212,0.00002615043,8.39983e-7,0.03040092,0.0001412971,0.0002121815,0.1774295,0.006466807,0.7587541],"study_design_scores_gemma":[0.002078446,0.0002305341,0.5351833,0.002534142,0.0001091236,0.00001254204,0.09591894,0.244232,0.000007048726,0.06203302,0.05723321,0.0004276232],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7874646,0.001772068,0.00124769,0.1994195,0.001000859,0.002239702,0.000001216818,0.00009699557,0.006757348],"genre_scores_gemma":[0.9967101,0.00007036193,0.00003107878,0.002757041,0.000263676,0.0000764666,0.000005933515,0.000009170137,0.00007622564],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7583265,"threshold_uncertainty_score":0.9998999,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Sharon E. Straus","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7496509901992127,"gpt":0.5668166863411787,"spread":0.182834303858034,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003901696,0.0003576562,0.0005016015,0.000299861,0.01170693,0.00002345945,0.0008388149,0.0003062239,0.006523229],"category_scores_gemma":[0.003336311,0.000231277,0.0001525694,0.00100904,0.0001474733,0.0002883844,0.0004102529,0.001386699,0.004068967],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002278263,"about_ca_system_score_gemma":0.002115117,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007957347,"about_ca_topic_score_gemma":0.0009702352,"domain_scores_codex":[0.9807968,0.01405255,0.001911129,0.0005738621,0.001435819,0.001229853],"domain_scores_gemma":[0.9886672,0.006137116,0.001277984,0.002427274,0.0009588287,0.0005316081],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007491931,0.0008488523,0.03020268,0.0001875082,0.0001236998,0.00000117497,0.3748831,0.0001631566,0.00002795089,0.01157322,0.3138927,0.268021],"study_design_scores_gemma":[0.006795194,0.0009464609,0.05995185,0.002741501,0.0001972861,0.000008537279,0.1471287,0.03777258,0.00001079371,0.01340464,0.7302851,0.0007573513],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.08134463,0.04545986,0.001222935,0.7738498,0.01129529,0.0356143,0.0003345532,0.001435276,0.04944338],"genre_scores_gemma":[0.9600905,0.002388041,0.002471728,0.01753759,0.0005255638,0.009379888,0.0002264668,0.00008686122,0.007293382],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8787459,"threshold_uncertainty_score":0.9967065,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Placebo; Detection theory; Placebo response; Psychology; Medicine; Computer science; Alternative medicine","authors":[{"name":"Lorraine G. Allan","is_ca":true},{"name":"Shepard Siegel","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1063051789484242,"gpt":0.3885802900338601,"spread":0.2822751110854359,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02526054,0.0001240883,0.0002147768,0.0002201573,0.001242024,0.00001805409,0.0003517331,0.00005421731,0.001404394],"category_scores_gemma":[0.004104517,0.00006217928,0.0001889962,0.002053984,0.0001085076,0.00008830559,0.00009262085,0.0003077866,0.00008548943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001070674,"about_ca_system_score_gemma":0.00006412953,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001863152,"about_ca_topic_score_gemma":0.00005794569,"domain_scores_codex":[0.9658037,0.03251348,0.000335971,0.0002604538,0.0008233586,0.0002630162],"domain_scores_gemma":[0.9904423,0.008534139,0.0003992452,0.0005190855,0.00005294888,0.00005225239],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001218246,0.0006075523,0.01105995,0.0004403206,0.0007538564,0.000001188513,0.02415222,0.02872009,0.09553571,0.008364186,0.01868363,0.8104631],"study_design_scores_gemma":[0.0009664292,0.0002047334,0.02155037,0.0001004878,0.001004728,8.267576e-7,0.0002818481,0.9605556,0.01223264,0.001855268,0.00111664,0.0001304837],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8949475,0.0007286171,0.04329368,0.02837561,0.002605011,0.01516476,0.00003786741,0.0002653381,0.01458165],"genre_scores_gemma":[0.9959263,0.00002490021,0.000007631084,0.001821576,0.00004484367,0.0003659267,0.000002430553,0.000009267717,0.001797075],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9318355,"threshold_uncertainty_score":0.9995084,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Leonard A. Jason","is_ca":false},{"name":"Abigail Brown","is_ca":false},{"name":"Erin Clyne","is_ca":false},{"name":"Lindsey Bartgis","is_ca":false},{"name":"Meredyth Evans","is_ca":false},{"name":"Molly Brown","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2653346418392165,"gpt":0.4143584613140521,"spread":0.1490238194748356,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004928148,0.0006004106,0.0009873863,0.000449536,0.003160773,0.0000731512,0.0003776012,0.0003875886,0.003004464],"category_scores_gemma":[0.001217906,0.0004624397,0.0002788249,0.0009209267,0.0007160748,0.0004173263,0.0002176131,0.001245448,0.0001260364],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001880657,"about_ca_system_score_gemma":0.007969979,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008424159,"about_ca_topic_score_gemma":0.0001656314,"domain_scores_codex":[0.9926659,0.001582987,0.001534434,0.001074035,0.00134392,0.001798754],"domain_scores_gemma":[0.9950549,0.001401083,0.000642573,0.001129958,0.0008802876,0.0008912031],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002520083,0.007117623,0.0162091,0.1475606,0.006120484,0.01436925,0.1494073,0.0006758734,0.007824398,0.1465081,0.04182949,0.4598577],"study_design_scores_gemma":[0.04555411,0.03464517,0.4800388,0.05151174,0.003475232,0.1751988,0.03775107,0.1201707,0.000384214,0.04468715,0.002461737,0.004121354],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9021216,0.03390783,0.01487999,0.02054366,0.001808185,0.0233647,0.0006225996,0.0005800735,0.002171359],"genre_scores_gemma":[0.9892802,0.00123394,0.003372238,0.0001687803,0.000145797,0.004474018,0.0004652196,0.0001176261,0.000742241],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4638297,"threshold_uncertainty_score":0.9997827,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Ba’ Pham","is_ca":true},{"name":"Robert W. Platt","is_ca":true},{"name":"Laura McAuley","is_ca":false},{"name":"Terry P. Klassen","is_ca":true},{"name":"David Moher","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9114822169183275,"gpt":0.6483470983849812,"spread":0.2631351185333464,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.2204158,0.0002075552,0.00116319,0.0003433804,0.001107024,0.0005161504,0.0009738376,0.00008540813,0.02082449],"category_scores_gemma":[0.05491303,0.00008300105,0.0003512658,0.002269625,0.0000417886,0.0002505558,0.0001651731,0.0002127523,0.006650096],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009129872,"about_ca_system_score_gemma":0.0004085095,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005223769,"about_ca_topic_score_gemma":0.0001386631,"domain_scores_codex":[0.9528763,0.03322266,0.006244765,0.0009574579,0.006289085,0.0004097848],"domain_scores_gemma":[0.9848874,0.00501352,0.003659312,0.003345341,0.00272419,0.0003701866],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001360915,0.0000360488,0.007595188,0.00002579749,0.00004399107,1.49588e-7,0.007789609,0.0000214092,0.00002078635,0.0006274172,0.4477071,0.5361189],"study_design_scores_gemma":[0.0005894657,0.0001672815,0.2228165,0.0002354556,0.0002147177,0.00001480324,0.00960767,0.03320975,0.00001273924,0.03270817,0.7001718,0.0002516263],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2835796,0.005004563,0.04457092,0.6451732,0.0008562177,0.008773009,0.00002582924,0.00002508756,0.01199158],"genre_scores_gemma":[0.916829,0.0002808602,0.008921178,0.0217975,0.0001855126,0.001284937,0.00001551791,0.00002120156,0.05066423],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6332495,"threshold_uncertainty_score":0.9941233,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Nigel Mantou Lou","is_ca":true},{"name":"Tina Montreuil","is_ca":true},{"name":"Liane S. Feldman","is_ca":true},{"name":"Gerald M. Fried","is_ca":true},{"name":"Mélanie Lavoie‐Tremblay","is_ca":true},{"name":"Farhan Bhanji","is_ca":true},{"name":"Heather Kennedy","is_ca":true},{"name":"Pepa Kaneva","is_ca":true},{"name":"Susan Drouin","is_ca":true},{"name":"Jason M. Harley","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09507221901161102,"gpt":0.4609919768550074,"spread":0.3659197578433964,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.007418953,0.0004234056,0.0009662348,0.0005135641,0.004340397,0.00003641504,0.0003053455,0.0004515615,0.0001556359],"category_scores_gemma":[0.001312692,0.0003486345,0.0001130517,0.0009062432,0.0001823434,0.0002957105,0.0004160605,0.001212851,0.00001120145],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002464509,"about_ca_system_score_gemma":0.007547505,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.09252362,"about_ca_topic_score_gemma":0.2707559,"domain_scores_codex":[0.98616,0.007992139,0.002541241,0.001176214,0.000908641,0.001221724],"domain_scores_gemma":[0.9929615,0.003013267,0.001232824,0.0008678321,0.001115876,0.0008087636],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006081061,0.000679837,0.5649629,0.02898267,0.0001529323,0.00001541434,0.3505602,0.000087383,0.0001427508,0.01422441,0.005326502,0.03425688],"study_design_scores_gemma":[0.004213327,0.0001385358,0.7884754,0.007529504,0.00007550404,0.000004195335,0.1898807,0.002506177,0.00000230248,0.005783799,0.00111464,0.0002759839],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5444776,0.007451443,0.000181406,0.4347554,0.000529301,0.01196868,0.0004091641,0.00008061529,0.0001463961],"genre_scores_gemma":[0.9748498,0.004954078,0.000979295,0.007313793,0.0001901442,0.01012458,0.0006827146,0.00006902411,0.0008365373],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4303722,"threshold_uncertainty_score":0.9998966,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Raúl de la Fuente‐Fernández","is_ca":true},{"name":"A. Jon Stoessl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3085821682452684,"gpt":0.4527774283355963,"spread":0.1441952600903279,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.01264534,0.00008785175,0.00008469184,0.00003182487,0.002278792,0.00003891438,0.0003028478,0.00003586837,0.0002503207],"category_scores_gemma":[0.009917819,0.00004385122,0.00005676776,0.0002483948,0.00008742815,0.00006329167,0.00006054293,0.0001740549,0.0002098766],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006769402,"about_ca_system_score_gemma":0.00008679089,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004304238,"about_ca_topic_score_gemma":0.0000060406,"domain_scores_codex":[0.9939538,0.00475481,0.0002441554,0.0002273727,0.0004763613,0.0003434712],"domain_scores_gemma":[0.9898585,0.009490938,0.0001531111,0.0003741855,0.00005685897,0.00006637524],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005291218,0.0000691524,0.00007393723,0.000047338,0.000004558874,1.247022e-7,0.001525961,0.00002917685,0.003230738,0.004443391,0.9269995,0.06352322],"study_design_scores_gemma":[0.001197276,0.0001980157,0.0007354274,0.0001529528,0.00004610611,0.000001761744,0.0007800647,0.4368752,0.005211961,0.01190271,0.5426976,0.0002009048],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.02418722,0.001923233,0.01512479,0.9339123,0.003235337,0.01341258,0.00003711193,0.0002814613,0.00788599],"genre_scores_gemma":[0.9730264,0.0003492669,0.0001692513,0.01679533,0.0002555472,0.001748223,0.000009570664,0.00001628379,0.007630152],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9488392,"threshold_uncertainty_score":0.9990201,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Donna M. Wilson","is_ca":false},{"name":"Herbert C. Northcott","is_ca":false},{"name":"Corrine D. Truman","is_ca":false},{"name":"Susan L. Smith","is_ca":false},{"name":"Marjorie Anderson","is_ca":true},{"name":"Robin L. Fainsinger","is_ca":false},{"name":"Michael Stingl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2203002404353396,"gpt":0.5384405973095842,"spread":0.3181403568742445,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007771941,0.0001661749,0.0003831574,0.0001430582,0.001171741,0.000002061055,0.000320347,0.0001315668,0.0007907471],"category_scores_gemma":[0.001619187,0.000118184,0.00002629361,0.001388884,0.00003727758,0.0001159417,0.0001076126,0.0008443394,0.0001868162],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.004737309,"about_ca_system_score_gemma":0.0414936,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9196478,"about_ca_topic_score_gemma":0.9800003,"domain_scores_codex":[0.990787,0.00542238,0.001522609,0.0002926268,0.001146998,0.0008284065],"domain_scores_gemma":[0.995487,0.001353203,0.0008653234,0.0006479801,0.001418016,0.000228489],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002333719,0.000172609,0.8498515,0.0014055,0.00001812544,0.000002919572,0.020817,0.001398547,0.00001498229,0.008246666,0.1069375,0.01090121],"study_design_scores_gemma":[0.001168677,0.00007252469,0.9502289,0.002084529,0.00001496826,0.000001720652,0.02280225,0.005251708,0.000003551683,0.002435337,0.01581263,0.0001232185],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8708813,0.004553019,0.0003608862,0.1059668,0.003497469,0.007653249,0.00002116122,0.00006446205,0.007001608],"genre_scores_gemma":[0.99031,0.0002963551,0.000193191,0.00735429,0.0001579725,0.0008024289,0.00004336531,0.00002231077,0.0008201068],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1194287,"threshold_uncertainty_score":0.9990833,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"François Goulet","is_ca":true},{"name":"André Jacques","is_ca":false},{"name":"Robert Gagnon","is_ca":false},{"name":"Pierre Racette","is_ca":false},{"name":"William J. Sieber","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4189002070409529,"gpt":0.6045903599396676,"spread":0.1856901528987147,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.02303459,0.0002031432,0.0004203795,0.0001940928,0.00337293,0.00000478316,0.0002573306,0.0001639235,0.0002899207],"category_scores_gemma":[0.0002421613,0.0001456294,0.00008464073,0.0006371916,0.00009992249,0.0001876356,0.0002118462,0.0009940572,0.00006424569],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00105465,"about_ca_system_score_gemma":0.00330965,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007384409,"about_ca_topic_score_gemma":0.0002033704,"domain_scores_codex":[0.9903169,0.004971887,0.00186187,0.0003552841,0.001495302,0.0009987337],"domain_scores_gemma":[0.9958805,0.0008397636,0.00147118,0.000750449,0.0008525203,0.0002056026],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0002456249,0.001291298,0.05378916,0.005457427,0.0001634718,0.000001825365,0.1316383,0.001501641,0.002221373,0.03665697,0.02824578,0.7387872],"study_design_scores_gemma":[0.001474689,0.0005566836,0.8419961,0.003080551,0.00007822247,6.246112e-7,0.09015826,0.03039747,0.00006287033,0.000935908,0.03096384,0.0002947558],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9495943,0.0003094324,0.01172987,0.01932334,0.003381027,0.0071634,0.00002575495,0.00009415956,0.008378726],"genre_scores_gemma":[0.980508,0.0001493809,0.002415508,0.01573557,0.0003019872,0.0003948865,0.00004358169,0.00003014289,0.0004209135],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7882069,"threshold_uncertainty_score":0.9979246,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Stéphanie Baggio","is_ca":false},{"name":"Katia Iglesias","is_ca":false},{"name":"Joseph Studer","is_ca":false},{"name":"Gerhard Gmel","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07899338823122372,"gpt":0.4427404409925142,"spread":0.3637470527612905,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007570974,0.0001389822,0.0003319339,0.0001427328,0.0006189767,0.000004155813,0.0004555591,0.000119273,0.0003033942],"category_scores_gemma":[0.0003539517,0.00007918062,0.0001348067,0.0003745935,0.0001915282,0.0001220895,0.00007873288,0.0003069289,0.000006847898],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008585282,"about_ca_system_score_gemma":0.0004243451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009987766,"about_ca_topic_score_gemma":0.00110107,"domain_scores_codex":[0.9942869,0.003510012,0.0008985428,0.0002367192,0.0007569721,0.0003108682],"domain_scores_gemma":[0.9972396,0.0002807617,0.0008294223,0.001059182,0.0004724232,0.0001185383],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000253374,0.001572681,0.6252281,0.0004677991,0.0002947375,2.106884e-7,0.2927277,0.0002026331,0.00497734,0.02694219,0.01936066,0.02797258],"study_design_scores_gemma":[0.001209114,0.0003444139,0.9456481,0.0004196024,0.000129191,0.000001916837,0.02855213,0.006576344,0.000661201,0.01605224,0.0002727186,0.0001330367],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9887299,0.0004219942,0.0003875969,0.003948001,0.002103543,0.001865859,0.00001531428,0.00002616099,0.002501626],"genre_scores_gemma":[0.998633,0.00004986728,0.00006938561,0.0003646149,0.000119315,0.0001748371,0.00002130178,0.00002531581,0.0005424042],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.32042,"threshold_uncertainty_score":0.4760731,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Wilfreda E. Thurston","is_ca":true},{"name":"J. R. Graham","is_ca":true},{"name":"Jennifer Hatfield","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5915797188746392,"gpt":0.6761750254876046,"spread":0.08459530661296533,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.1274638,0.0001989081,0.0003349382,0.000201099,0.002327385,0.0001674545,0.0006986897,0.00009120511,0.02334952],"category_scores_gemma":[0.00942204,0.0001067358,0.0001420122,0.001363204,0.0001351858,0.0004355494,0.0001071699,0.0005319179,0.001362804],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006456125,"about_ca_system_score_gemma":0.007972847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002632121,"about_ca_topic_score_gemma":0.0001463199,"domain_scores_codex":[0.9713912,0.01789584,0.001804691,0.0007213061,0.00757612,0.0006108524],"domain_scores_gemma":[0.9922036,0.002994475,0.000854035,0.001690903,0.001984538,0.0002724582],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00005772055,0.001034169,0.09281694,0.00004270346,0.0000637919,4.277761e-7,0.01462052,0.009110641,0.0001442622,0.2203509,0.2620541,0.3997038],"study_design_scores_gemma":[0.001634029,0.0002721365,0.4281308,0.00006088473,0.00004263349,0.000005947124,0.01268862,0.07883973,0.00004351903,0.2390808,0.2389294,0.0002714373],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2822975,0.001327295,0.04665357,0.4299949,0.01064037,0.01255815,0.00003988227,0.0002657029,0.2162227],"genre_scores_gemma":[0.9841757,0.0000360718,0.003065691,0.006749018,0.00009651526,0.0007625415,0.00001876804,0.00001418563,0.005081502],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7018782,"threshold_uncertainty_score":0.9994147,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Lisa Engel","is_ca":true},{"name":"Courtney Henderson","is_ca":true},{"name":"Jennifer Fergenbaum","is_ca":true},{"name":"Angela Colantonio","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1741437033261165,"gpt":0.5366436959399579,"spread":0.3624999926138414,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06332566,0.0002633521,0.0007739954,0.0001708868,0.002037395,0.000007147449,0.0004883563,0.0006874586,0.001729531],"category_scores_gemma":[0.02607086,0.0001739646,0.0001589916,0.0005236985,0.00008942809,0.0007076383,0.00009686926,0.002460734,0.0001180887],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00132506,"about_ca_system_score_gemma":0.0177492,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000643905,"about_ca_topic_score_gemma":0.0005698749,"domain_scores_codex":[0.9854307,0.006129148,0.004775287,0.000396634,0.002240728,0.001027531],"domain_scores_gemma":[0.9907531,0.002434823,0.003403849,0.0008145968,0.002061871,0.0005318302],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003094159,0.0003207148,0.0009334308,0.02199121,0.00004177471,2.053731e-7,0.02391578,0.0002429859,0.00006331685,0.002916973,0.1185128,0.8307514],"study_design_scores_gemma":[0.003163513,0.0006725871,0.002825503,0.02293143,0.0001206593,0.000009888012,0.004455037,0.9432403,0.000008026074,0.009305133,0.01294967,0.0003182454],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.07644817,0.005895386,0.08620338,0.7769985,0.009546249,0.04047022,0.00007053117,0.0005289253,0.003838702],"genre_scores_gemma":[0.9494341,0.002567216,0.00156362,0.04112134,0.0004519281,0.003497976,0.0003451199,0.00004180989,0.0009768674],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9429973,"threshold_uncertainty_score":0.9998406,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Joan Higgins","is_ca":true},{"name":"Karen Strange","is_ca":true},{"name":"Jennifer Scarr","is_ca":true},{"name":"Michael J. Pennock","is_ca":true},{"name":"Victoria Barr","is_ca":true},{"name":"Ann Yew","is_ca":true},{"name":"Janine Drummond","is_ca":true},{"name":"Jennifer Terpstra","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7601847510600535,"gpt":0.6064561193109729,"spread":0.1537286317490806,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04051687,0.0003609624,0.0008475863,0.0005620185,0.003347806,0.00003246187,0.0007194716,0.0004444827,0.00139476],"category_scores_gemma":[0.01443523,0.000257062,0.000174223,0.00132217,0.0003068133,0.001972008,0.0002390138,0.00226045,0.0002099801],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002105245,"about_ca_system_score_gemma":0.03074572,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008151313,"about_ca_topic_score_gemma":0.001406817,"domain_scores_codex":[0.9749851,0.01833375,0.002475451,0.0007072416,0.00198668,0.001511799],"domain_scores_gemma":[0.9854616,0.00460073,0.004633739,0.001552979,0.002929786,0.0008211841],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002612518,0.00110183,0.006389258,0.001998227,0.00007617001,2.142247e-7,0.7578809,0.00001068766,0.00002636895,0.01550693,0.2022658,0.01448238],"study_design_scores_gemma":[0.0008579326,0.001170432,0.08810456,0.01376437,0.00004755575,0.000003420625,0.8777211,0.0004099557,0.00001488579,0.001270726,0.01639428,0.0002407402],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.03796115,0.01015347,0.00170922,0.9333067,0.005486871,0.009030391,0.00005369213,0.0001258858,0.002172643],"genre_scores_gemma":[0.9394299,0.01360174,0.0007696982,0.04223453,0.0004692128,0.002086638,0.00003417753,0.00005131711,0.001322741],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9014688,"threshold_uncertainty_score":0.9999881,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Shayna A. Rusticus","is_ca":true},{"name":"Derek Wilson","is_ca":true},{"name":"Oscar Casiro","is_ca":true},{"name":"Chris Y. Lovato","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2534548523223802,"gpt":0.5345699707893247,"spread":0.2811151184669445,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.05803698,0.0002208128,0.0005323096,0.0001508743,0.002314823,0.00001260064,0.0002563045,0.0001194864,0.0003137229],"category_scores_gemma":[0.005360039,0.0001246042,0.00006694784,0.0006818206,0.0002290438,0.0001199241,0.0002132122,0.001205007,0.00002889308],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001228038,"about_ca_system_score_gemma":0.008991117,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005145238,"about_ca_topic_score_gemma":0.00003041715,"domain_scores_codex":[0.978686,0.01570125,0.002459538,0.0004509536,0.002246679,0.0004555362],"domain_scores_gemma":[0.9940372,0.001012472,0.003677332,0.0008049724,0.0003549362,0.0001131348],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001812697,0.00130422,0.2892075,0.001522385,0.0001110289,8.134847e-9,0.0820551,0.00333711,0.001337889,0.001278465,0.001985071,0.61768],"study_design_scores_gemma":[0.001110514,0.0006121446,0.952419,0.001617226,0.00003543806,0.000002347301,0.03415883,0.005049761,0.0002689915,0.0002480414,0.004346016,0.0001316503],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9147292,0.001186847,0.001804232,0.07494533,0.000883603,0.006334526,0.000004687353,0.00002081675,0.00009078905],"genre_scores_gemma":[0.9909497,0.000236717,0.002830271,0.002578224,0.00006953618,0.0005861915,0.0004379718,0.00003293619,0.002278407],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6632116,"threshold_uncertainty_score":0.998984,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Mirou Jaana","is_ca":true},{"name":"Smruti Vartak","is_ca":false},{"name":"Marcia M. Ward","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8318884867437546,"gpt":0.6073044842747833,"spread":0.2245840024689713,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.08321084,0.0007575064,0.004373332,0.0009449323,0.002399324,0.0001802597,0.001253119,0.0004611374,0.002531515],"category_scores_gemma":[0.003034604,0.0006325037,0.0006953708,0.0009487448,0.0001291904,0.0005602142,0.0002580803,0.001146608,0.01454875],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.008271531,"about_ca_system_score_gemma":0.009373018,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009582398,"about_ca_topic_score_gemma":0.0001699973,"domain_scores_codex":[0.9714228,0.01337594,0.01166281,0.001494482,0.0007368459,0.001307138],"domain_scores_gemma":[0.9790491,0.003919754,0.01356257,0.002469889,0.0003658333,0.0006327997],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003935187,0.00009620631,0.00002079892,0.07176144,0.00016089,2.384846e-7,0.003690479,0.0000490563,5.030108e-10,0.01925011,0.2976002,0.6073667],"study_design_scores_gemma":[0.0004702562,0.0001375818,0.00009408399,0.04942964,0.0001076367,0.000002185241,0.002939385,0.0008340721,4.405768e-9,0.001806702,0.9436823,0.0004961622],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000001449598,0.8232974,0.001637718,0.1512547,0.00275378,0.01931532,0.0003867878,0.0001347043,0.001218206],"genre_scores_gemma":[0.00002219752,0.9351998,0.00332351,0.04517805,0.0007344743,0.01229675,0.0009811237,0.0001466217,0.002117472],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6460822,"threshold_uncertainty_score":0.9996126,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Université de Sherbrooke","funders":"","keywords":"Lottery; Respondent; Randomized controlled trial; Family medicine; Medicine; Significant difference; Psychology; Statistics; Mathematics; Surgery","authors":[{"name":"Geneviève Baron","is_ca":true},{"name":"Philippe De Wals","is_ca":true},{"name":"François Milord","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7326192029808507,"gpt":0.6326923150652278,"spread":0.09992688791562288,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3506264,0.00009728451,0.0002744358,0.0001368091,0.001812334,0.00001264161,0.0002417019,0.0001053489,0.00008962939],"category_scores_gemma":[0.1012302,0.0000746693,0.00006536295,0.0008221738,0.0001925161,0.0001064409,0.00004683504,0.0001630919,0.0000185255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002403379,"about_ca_system_score_gemma":0.002473291,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004768381,"about_ca_topic_score_gemma":0.006777842,"domain_scores_codex":[0.6830932,0.3150364,0.0004523502,0.0002951855,0.0006128314,0.0005100303],"domain_scores_gemma":[0.8921484,0.1065672,0.0002462217,0.0002655263,0.0006670531,0.0001055774],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.1924492,0.0009400184,0.4241445,0.0002515914,0.0001523539,0.000001449521,0.1802794,0.0002998811,0.003668158,0.003736384,0.01013442,0.1839426],"study_design_scores_gemma":[0.0006153149,0.0001656921,0.9870083,0.0002730595,0.00001857333,7.045286e-7,0.005903748,0.0001073137,0.00009951749,0.001978015,0.003745243,0.00008451325],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9573432,0.0001711253,0.01914072,0.01655921,0.00093654,0.005567711,0.00006743372,0.0000367043,0.0001773586],"genre_scores_gemma":[0.994332,0.00004786113,0.001832555,0.001705185,0.0001246985,0.001344742,0.00003398868,0.00001400312,0.0005650218],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5628638,"threshold_uncertainty_score":0.9994872,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"John R. Kirby","is_ca":true},{"name":"M. Dianne Delva","is_ca":false},{"name":"Christopher K. Knapper","is_ca":false},{"name":"Richard Birtwhistle","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3462473353381151,"gpt":0.4829060560291176,"spread":0.1366587206910024,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.01228125,0.0002426949,0.0004159186,0.00009606588,0.006791714,0.00001030902,0.0003128526,0.0002087181,0.00008922775],"category_scores_gemma":[0.001810329,0.0001332387,0.00008252767,0.0006719488,0.0001060991,0.00008741248,0.0002362641,0.0007401459,0.00002697108],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003486211,"about_ca_system_score_gemma":0.004129717,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003182445,"about_ca_topic_score_gemma":0.0006085825,"domain_scores_codex":[0.991286,0.005330988,0.001285403,0.000426497,0.0007679765,0.0009031559],"domain_scores_gemma":[0.99595,0.001848441,0.0008764433,0.000622876,0.0004668567,0.0002353192],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0006278097,0.0006737032,0.1034797,0.0043379,0.0001269754,1.253416e-7,0.2369156,0.0002899174,0.0001355515,0.1337333,0.04292605,0.4767534],"study_design_scores_gemma":[0.004366877,0.0002732259,0.6206792,0.04052262,0.0001754832,0.000001511545,0.1088092,0.001505831,0.0003700313,0.01659179,0.2058062,0.0008980566],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5306227,0.005514775,0.01187227,0.410761,0.007409145,0.03164591,0.0001220224,0.0001819319,0.001870225],"genre_scores_gemma":[0.9654987,0.0001056198,0.01756059,0.007787348,0.0002235234,0.007078129,0.0000302744,0.0000519802,0.001663779],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5171995,"threshold_uncertainty_score":0.9945013,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Syed Latifi","is_ca":true},{"name":"Mark J. Gierl","is_ca":true},{"name":"André-Philippe Boulais","is_ca":true},{"name":"André F. De Champlain","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4259430117257569,"gpt":0.5111785923759283,"spread":0.08523558065017145,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02472413,0.0001435911,0.0002513095,0.0002915477,0.0004350209,0.00008071755,0.0004135164,0.00009024476,0.000006940654],"category_scores_gemma":[0.004626111,0.0001070976,0.00002367801,0.0006442445,0.00003660342,0.0003505099,0.0002619512,0.0003733555,0.00001690986],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003996163,"about_ca_system_score_gemma":0.00190256,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003620968,"about_ca_topic_score_gemma":0.0002046623,"domain_scores_codex":[0.990899,0.006439669,0.0008124844,0.00051456,0.0009741726,0.0003600595],"domain_scores_gemma":[0.997155,0.001016417,0.0002799221,0.0006547849,0.0006488694,0.0002450464],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002498534,0.0009157947,0.04606875,0.0001723596,0.0000486635,0.00001074794,0.2239785,0.09749638,0.0003017388,0.03811464,0.004803735,0.5878389],"study_design_scores_gemma":[0.0009312623,0.0002233674,0.1956016,0.0003692461,0.0000072313,0.000001543951,0.001018688,0.7986625,0.000007824549,0.003001827,0.00007117724,0.0001037157],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9359041,0.0001630628,0.04517967,0.01475417,0.001849411,0.001670988,0.000002431377,0.0003077133,0.0001684982],"genre_scores_gemma":[0.9526242,0.00001025455,0.04553491,0.001446029,0.0001621856,0.0001312918,0.000004415741,0.00001095353,0.00007576171],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7011662,"threshold_uncertainty_score":0.856894,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Saskatchewan","funders":"University of Saskatchewan","keywords":"Psychology; Correlation; Self-assessment; Applied psychology; Clinical psychology; Social psychology; Mathematics","authors":[{"name":"Marcel D’Eon","is_ca":true},{"name":"Krista Trinder","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7640554176157064,"gpt":0.6509811233514293,"spread":0.1130742942642771,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0486081,0.000130406,0.0002894218,0.0001299077,0.0008927021,0.00007169443,0.00103765,0.00005248429,0.001077815],"category_scores_gemma":[0.006584603,0.0000517177,0.0001282632,0.0008862115,0.0001229121,0.0005083616,0.0001430466,0.0002435325,0.0000496569],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002124421,"about_ca_system_score_gemma":0.003252528,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003419926,"about_ca_topic_score_gemma":0.0005935377,"domain_scores_codex":[0.9907905,0.003745806,0.001513669,0.0002857424,0.003364509,0.0002997776],"domain_scores_gemma":[0.9774145,0.0182735,0.001305989,0.0009160389,0.002025337,0.00006461882],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003754543,0.0005722886,0.8401332,0.00007166668,0.00005809484,4.774261e-9,0.0128044,0.001010135,0.00006463919,0.005576507,0.005803326,0.1338681],"study_design_scores_gemma":[0.000512785,0.0001220281,0.8980064,0.0001629812,0.00003427896,2.685553e-7,0.006454256,0.06970942,0.00002539808,0.02457193,0.0003447131,0.00005551933],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7293256,0.0005438737,0.002422104,0.2557138,0.001376498,0.01029663,0.000007575048,0.0000142099,0.0002997844],"genre_scores_gemma":[0.993075,0.00009887759,0.00147115,0.0009809423,0.00006361611,0.003812935,0.000007729868,0.000008016662,0.0004817729],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2637494,"threshold_uncertainty_score":0.9998353,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Elan Paluck","is_ca":true},{"name":"Lawrence W. Green","is_ca":false},{"name":"C. James Frankish","is_ca":true},{"name":"David Fielding","is_ca":true},{"name":"Beth E. Haverkamp","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4565529397769157,"gpt":0.6137302823738618,"spread":0.1571773425969462,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01160953,0.00007751572,0.000206854,0.00007903788,0.0006104611,0.000004752751,0.0001254503,0.0000408546,0.0007541631],"category_scores_gemma":[0.001643312,0.00005037359,0.00004250888,0.0003208435,0.0000961437,0.0001156627,0.00004720657,0.000912003,0.000006364364],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000214299,"about_ca_system_score_gemma":0.001291458,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001405578,"about_ca_topic_score_gemma":0.00006064861,"domain_scores_codex":[0.9903376,0.008423516,0.0004943684,0.00007749292,0.0004893132,0.000177744],"domain_scores_gemma":[0.9975661,0.001292869,0.0003149176,0.0004973598,0.0001914347,0.0001373188],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002202727,0.001615414,0.9295918,0.0006698496,0.0001037574,4.505743e-7,0.02707923,0.0005222762,0.001025676,0.02350762,0.004204401,0.0114593],"study_design_scores_gemma":[0.006791061,0.0004711139,0.8930646,0.0008206668,0.0002140418,0.000008269106,0.02940995,0.04079385,0.0004390539,0.005263167,0.02256125,0.0001630194],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9287705,0.001207077,0.0002689153,0.04690792,0.0003526606,0.003200135,0.000008098107,0.00002645866,0.01925817],"genre_scores_gemma":[0.9940799,0.0003851003,0.0008586218,0.004321653,0.000006837743,0.0001553689,0.00003342768,0.000007508396,0.0001515922],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06530932,"threshold_uncertainty_score":0.8257558,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Oswin Chang","is_ca":true},{"name":"Anne Holbrook","is_ca":true},{"name":"Simran Lohit","is_ca":true},{"name":"Jiawen Deng","is_ca":true},{"name":"Janice Hua Xu","is_ca":true},{"name":"Munil Lee","is_ca":true},{"name":"Alan Cheng","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5094499951362133,"gpt":0.6813970224575658,"spread":0.1719470273213526,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04067347,0.0003765496,0.00280261,0.0003505718,0.0009563105,0.00003411699,0.0004983243,0.0005154255,0.0003546014],"category_scores_gemma":[0.096321,0.0002385513,0.0003274413,0.001723745,0.0005524022,0.0001320893,0.0002673382,0.001824003,0.0000197574],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007589548,"about_ca_system_score_gemma":0.0271391,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002480724,"about_ca_topic_score_gemma":0.00006416397,"domain_scores_codex":[0.9861183,0.005757631,0.004230952,0.0007445618,0.002702968,0.0004455663],"domain_scores_gemma":[0.9829693,0.009019163,0.003371202,0.001241865,0.002885288,0.0005132084],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00000850021,0.0003537631,0.0007483434,0.2270221,0.0001928565,3.144776e-7,0.0005658865,1.84293e-7,5.110616e-8,0.0001451282,0.008015409,0.7629474],"study_design_scores_gemma":[0.001530933,0.0002708233,0.02105534,0.9515122,0.002724603,0.00003137513,0.002139133,0.000890596,4.52673e-8,0.0005275166,0.01905285,0.0002646025],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0003210772,0.9578369,0.0008608055,0.0114025,0.001679999,0.02771154,0.00004939705,0.00009133305,0.0000464848],"genre_scores_gemma":[0.0001779428,0.9851156,0.004164389,0.003541547,0.0005367947,0.005395295,0.0009145808,0.00006362426,0.00009016453],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7626828,"threshold_uncertainty_score":0.9878286,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Emma Somer","is_ca":true},{"name":"Christian Gische","is_ca":false},{"name":"Milica Miočević","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8558120885164818,"gpt":0.6339693109664204,"spread":0.2218427775500614,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01124955,0.000101982,0.0002293079,0.0002309088,0.001236366,0.00001632005,0.0001390814,0.00006186387,0.0004191046],"category_scores_gemma":[0.0002580773,0.00007647407,0.00003313069,0.0007731707,0.00002631902,0.0001091515,0.0001713259,0.0003163252,0.000001448736],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002709111,"about_ca_system_score_gemma":0.00008337891,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006622564,"about_ca_topic_score_gemma":0.0005080984,"domain_scores_codex":[0.9942652,0.004301952,0.0005555343,0.0004101584,0.0002109115,0.0002562238],"domain_scores_gemma":[0.9985221,0.0008291741,0.0002268807,0.0003250527,0.00004553897,0.00005122761],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006755972,0.002082057,0.03253248,0.00003860905,0.0001648884,0.00001419122,0.1512327,0.05893359,0.001434363,0.0007866995,0.0007806342,0.7513242],"study_design_scores_gemma":[0.001271318,0.0001967232,0.01438363,0.00001972322,0.0001094629,0.00001292011,0.0300737,0.9518106,0.00000403083,0.001835688,0.0001684613,0.0001137209],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6938836,0.005496883,0.285517,0.01054688,0.0009003761,0.003482119,0.00004186871,0.00003734034,0.0000938622],"genre_scores_gemma":[0.9840828,0.00000771276,0.01325439,0.0004565378,0.00003956606,0.0018719,0.0002260019,0.000009596097,0.00005153208],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.892877,"threshold_uncertainty_score":0.9509249,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Reza Yousefi‐Nooraie","is_ca":true},{"name":"Sobia Khan","is_ca":true},{"name":"Jennifer Gutberg","is_ca":true},{"name":"G. Ross Baker","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1244131047307477,"gpt":0.4534153988612473,"spread":0.3290022941304996,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.008077638,0.0001629801,0.0004136827,0.0003856675,0.002324105,0.00004907215,0.0003864041,0.0001087336,0.0003704809],"category_scores_gemma":[0.0003105705,0.00009822914,0.00008315076,0.004518888,0.00006489259,0.000135753,0.0002603605,0.0003369238,0.0001350314],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004472995,"about_ca_system_score_gemma":0.0009643426,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.09456066,"about_ca_topic_score_gemma":0.06041166,"domain_scores_codex":[0.9968522,0.000722888,0.0008708727,0.0004020737,0.0005705707,0.0005813775],"domain_scores_gemma":[0.9960176,0.0000992915,0.0008088807,0.0007067877,0.002315209,0.00005220762],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001549365,0.0001747866,0.03560171,0.0006573678,0.0005749886,4.574644e-7,0.1266671,0.001121116,0.00002935676,0.1824918,0.1776675,0.4748589],"study_design_scores_gemma":[0.001345858,0.0008575168,0.2940094,0.0005992355,0.0004475555,0.000001285837,0.2066697,0.005741929,0.0000131993,0.01501432,0.4748664,0.0004336245],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2137204,0.0009344517,0.0125365,0.7575582,0.001503214,0.01232302,0.00002694012,0.0002161682,0.001181139],"genre_scores_gemma":[0.972708,0.000003847884,0.000512558,0.02484807,0.000976152,0.0007701585,0.00007061581,0.00002099563,0.0000895372],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7589877,"threshold_uncertainty_score":0.9989747,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hôpital Charles-Le Moyne; Université Laval; Université de Montréal","funders":"","keywords":"Primary care; Medicine; Psychology; Family medicine","authors":[{"name":"Marie‐Dominique Beaulieu","is_ca":true},{"name":"Michelle Proulx","is_ca":true},{"name":"Guy Jobin","is_ca":true},{"name":"Marianne Kugler","is_ca":true},{"name":"Françis Gossard","is_ca":true},{"name":"Jean‐Louis Denis","is_ca":true},{"name":"Danielle Larouche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4108547061883054,"gpt":0.6224560121363097,"spread":0.2116013059480043,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.03463121,0.0002054989,0.0003257577,0.000288838,0.008558424,0.00001873458,0.0005513927,0.0002590017,0.00145082],"category_scores_gemma":[0.00190333,0.0001342695,0.0001011182,0.0007529593,0.0001541923,0.0002585271,0.0001689572,0.001097568,0.001102385],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00231462,"about_ca_system_score_gemma":0.01994672,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000466924,"about_ca_topic_score_gemma":0.001607712,"domain_scores_codex":[0.991906,0.00323502,0.001318846,0.0005193673,0.001184389,0.001836366],"domain_scores_gemma":[0.9914725,0.004227714,0.0005951556,0.0007644711,0.00215906,0.0007810383],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000141421,0.0001234984,0.004647434,0.001387571,0.000009724081,6.758248e-8,0.2191284,0.000001451611,0.00004250101,0.001238301,0.635853,0.1374266],"study_design_scores_gemma":[0.001870641,0.0003590748,0.1284276,0.0008089244,0.00002780946,7.697135e-7,0.0905575,0.001203677,0.00002412604,0.01382607,0.7626631,0.0002308146],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2085918,0.01507055,0.04208319,0.5326749,0.03058045,0.06377327,0.0002967007,0.0005927325,0.1063364],"genre_scores_gemma":[0.8295516,0.0005482888,0.01247385,0.0946051,0.006774689,0.01381976,0.001003229,0.0001349746,0.04108849],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6209598,"threshold_uncertainty_score":0.9996754,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Group dynamic; Group (periodic table); Dynamics (music); Psychology; Medicine; Social psychology; Pedagogy; Chemistry","authors":[{"name":"Samantha M. Harden","is_ca":false},{"name":"Shauna M. Burke","is_ca":true},{"name":"Amanda M. Haile","is_ca":false},{"name":"Paul A. Estabrooks","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5393916580775326,"gpt":0.6279639342730506,"spread":0.088572276195518,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01520105,0.0004466074,0.001717071,0.000244371,0.003274278,0.000100057,0.0005947629,0.000336794,0.0001350906],"category_scores_gemma":[0.006458374,0.0002337626,0.0003973455,0.001536977,0.0001992071,0.0003025381,0.000400652,0.005016021,0.0009398329],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002814051,"about_ca_system_score_gemma":0.007198912,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002519583,"about_ca_topic_score_gemma":0.000196279,"domain_scores_codex":[0.979364,0.0153648,0.0006948462,0.0008659774,0.002670252,0.00104011],"domain_scores_gemma":[0.9805742,0.01571881,0.0007914028,0.001459754,0.000863101,0.0005927171],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00008196206,0.001621601,0.000001030467,0.004264103,0.0001008828,9.934043e-7,0.002519207,0.000003474071,0.00001140404,0.00107046,0.03291081,0.9574141],"study_design_scores_gemma":[0.0005647482,0.000757186,0.0001902332,0.01837322,0.0009206885,0.000005966468,0.001068401,0.01727636,0.000002407412,0.001220105,0.9593397,0.0002809933],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.005376168,0.4409452,0.001290488,0.5213374,0.001270162,0.02879411,0.0002217199,0.0001740155,0.0005907034],"genre_scores_gemma":[0.008801686,0.9454523,0.00254346,0.02099024,0.004526498,0.014261,0.001448785,0.0002825102,0.001693486],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9571331,"threshold_uncertainty_score":0.9998381,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Steve McDonald","is_ca":false},{"name":"Carol Lefebvre","is_ca":false},{"name":"Gerd Antes","is_ca":false},{"name":"Daniel Galandi","is_ca":false},{"name":"Peter C Gøtzsche","is_ca":false},{"name":"C. Hammarquist","is_ca":false},{"name":"Margaret Haugh","is_ca":true},{"name":"Jos Kleijnen","is_ca":false},{"name":"Marjan Loep","is_ca":false},{"name":"Vanna Pistotti","is_ca":false},{"name":"Alric Rüther","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9027067621472988,"gpt":0.6861497762688541,"spread":0.2165569858784447,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7434521,0.0002488366,0.004066582,0.000240458,0.004614689,0.0007058276,0.001915001,0.00004783098,0.003353871],"category_scores_gemma":[0.2524427,0.00006736177,0.001589697,0.001283191,0.00009753383,0.000126456,0.0001839184,0.0004932449,0.0009124283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002030584,"about_ca_system_score_gemma":0.0006996856,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004571725,"about_ca_topic_score_gemma":0.0003725207,"domain_scores_codex":[0.265754,0.681035,0.03161985,0.001123185,0.01932349,0.001144463],"domain_scores_gemma":[0.9216914,0.04418385,0.02195436,0.004814373,0.006909066,0.0004469781],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001467135,0.00006531895,0.00037122,0.00004580048,0.0003035509,2.822181e-7,0.0341827,0.001176009,0.00004862703,0.003399244,0.7366878,0.2235727],"study_design_scores_gemma":[0.0121903,0.0009732521,0.03439388,0.002494371,0.001242684,0.0000253819,0.0532103,0.1494053,0.00002183108,0.008003298,0.7375559,0.0004835388],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01504758,0.08917017,0.03196894,0.8392494,0.002203037,0.01866514,0.00005548677,0.00001174633,0.00362849],"genre_scores_gemma":[0.9767211,0.001274299,0.0003975178,0.01059444,0.0006435441,0.0005065692,0.0000168743,0.00001792267,0.009827708],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9616736,"threshold_uncertainty_score":0.9998655,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Claudio Violato","is_ca":true},{"name":"Anthony Marini","is_ca":true},{"name":"Curtis Lee","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2742225804547531,"gpt":0.5179446169931036,"spread":0.2437220365383505,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007039423,0.0001514601,0.0004476477,0.0002477428,0.0009197303,0.00001972801,0.0000879876,0.00007744652,0.0001237825],"category_scores_gemma":[0.05547954,0.00009825001,0.0001505311,0.0006421781,0.00004608755,0.0000467998,0.00002920849,0.0002054401,0.000002339639],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002169346,"about_ca_system_score_gemma":0.001238441,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000330589,"about_ca_topic_score_gemma":0.0002751138,"domain_scores_codex":[0.9957862,0.001798196,0.0008665644,0.0003519753,0.000903458,0.0002935344],"domain_scores_gemma":[0.9874585,0.01058614,0.0006238053,0.0004856473,0.0006910925,0.0001547855],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003157023,0.04915012,0.4374299,0.001952075,0.005890951,0.000006628532,0.2515596,0.1555144,0.002042566,0.007061516,0.04126181,0.04497346],"study_design_scores_gemma":[0.02145904,0.00746136,0.6430835,0.01159138,0.01069771,0.000008987932,0.09268834,0.2071869,0.002023793,0.002765286,0.0003371076,0.0006966353],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9783287,0.0001981181,0.008767989,0.006943514,0.0004839996,0.005147293,0.00001751359,0.00003547583,0.00007737245],"genre_scores_gemma":[0.9937349,0.00002517626,0.003478941,0.001614707,0.0001319688,0.0007561183,0.00007217855,0.00001966764,0.0001663652],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2056535,"threshold_uncertainty_score":0.9524766,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Barna Konkolÿ Thege","is_ca":true},{"name":"Elke Ham","is_ca":true},{"name":"Laura C. Ball","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.275974655811742,"gpt":0.4575043055790544,"spread":0.1815296497673124,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.006528256,0.000204466,0.0003860672,0.0001366964,0.00180702,0.000004250809,0.0003585506,0.0001648083,0.0003019047],"category_scores_gemma":[0.0004352488,0.00008551837,0.0001305307,0.0006558484,0.0001637759,0.0002400972,0.0002495444,0.0005951573,0.00001615106],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001358192,"about_ca_system_score_gemma":0.005545238,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008634587,"about_ca_topic_score_gemma":0.0007379639,"domain_scores_codex":[0.9885322,0.008182285,0.001343391,0.0003403194,0.001070109,0.0005316596],"domain_scores_gemma":[0.9946797,0.002182167,0.001725021,0.0008304492,0.0003904472,0.0001922023],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001587263,0.000192959,0.9513249,0.002562963,0.00004725162,5.752985e-8,0.02829535,0.00001222485,0.0005847838,0.0008276612,0.004686605,0.0113065],"study_design_scores_gemma":[0.00216934,0.0002979369,0.9773446,0.008432593,0.00007724841,3.082313e-7,0.00782052,0.0009278454,0.00008805659,0.002152105,0.0005724414,0.000117043],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9256708,0.0004232095,0.00005175591,0.06093865,0.002017061,0.01046535,0.0001360034,0.00002280585,0.0002743741],"genre_scores_gemma":[0.991701,0.0003812487,0.0002295706,0.005778124,0.00009834788,0.001124287,0.00002096951,0.00002086267,0.0006455681],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06603023,"threshold_uncertainty_score":0.9994925,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Jenna M. Evans","is_ca":true},{"name":"Agnes Grudniewicz","is_ca":true},{"name":"G. Ross Baker","is_ca":true},{"name":"Walter P. Wodchis","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1047881475270623,"gpt":0.5213973918583964,"spread":0.4166092443313342,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004449488,0.0002018974,0.0002476207,0.0001345523,0.005471401,0.00001444117,0.0002891707,0.0001922357,0.004206997],"category_scores_gemma":[0.00482965,0.0001008813,0.00007497919,0.0004809631,0.0001076004,0.0002306778,0.00009393699,0.0003982629,0.0004526979],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001414181,"about_ca_system_score_gemma":0.01275126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009310928,"about_ca_topic_score_gemma":0.0008945094,"domain_scores_codex":[0.9940827,0.003254005,0.001004824,0.0003732443,0.0007148625,0.0005704003],"domain_scores_gemma":[0.9914122,0.003256954,0.0006608493,0.0004825035,0.004022563,0.0001649163],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.0004283647,0.0001678866,0.01854112,0.0006537794,0.00005503167,6.438859e-8,0.1975939,0.00001114007,0.001713737,0.4414339,0.2912978,0.04810336],"study_design_scores_gemma":[0.005627776,0.0004913331,0.03491079,0.004312313,0.0001024097,0.000001889116,0.5980074,0.001682442,0.0002462188,0.1325164,0.2214563,0.0006447539],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1221898,0.001317949,0.04800009,0.7689605,0.01935351,0.02240754,0.0005868939,0.0005369804,0.01664663],"genre_scores_gemma":[0.9583078,0.00002978921,0.002739592,0.01147851,0.00117189,0.007368845,0.0003473953,0.00005452407,0.01850166],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8361179,"threshold_uncertainty_score":0.9967033,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Kathleen Hartford","is_ca":true},{"name":"Robert Carey","is_ca":true},{"name":"James D. Mendonça","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8713175713779645,"gpt":0.6396185879627105,"spread":0.231698983415254,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.333678,0.00007140493,0.0001422671,0.0001796157,0.0005476238,0.00001839676,0.0005177846,0.0001100886,0.00005674861],"category_scores_gemma":[0.01784343,0.00004464085,0.0000253179,0.0006633981,0.0001676796,0.0001737555,0.00006033479,0.0003655287,0.000007913681],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000323738,"about_ca_system_score_gemma":0.0009033782,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01979237,"about_ca_topic_score_gemma":0.106899,"domain_scores_codex":[0.8539656,0.1427273,0.0009884841,0.000336171,0.001430097,0.0005523303],"domain_scores_gemma":[0.9812663,0.01774039,0.0003295177,0.0002071341,0.0003961388,0.00006048127],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002715181,0.0003436526,0.5538439,0.00008130745,0.000007860111,0.000001813904,0.4013624,0.00007897879,0.00001249859,0.005643553,0.0001174567,0.03579151],"study_design_scores_gemma":[0.0002870229,0.00007259916,0.8193777,0.0001751923,0.00001722192,7.084049e-7,0.1788548,0.0009549428,0.000002705507,0.0001032966,0.0001156388,0.00003816873],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9910954,0.00008256231,0.002718996,0.002810286,0.001586801,0.001200334,0.000006038445,0.00001719194,0.000482355],"genre_scores_gemma":[0.998597,0.00002122498,0.0007669135,0.0003589887,0.0000871848,0.00005291367,0.00005515396,0.000004417777,0.00005625384],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3158346,"threshold_uncertainty_score":0.9904297,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Milica Miočević","is_ca":true},{"name":"Mariola Moeyaert","is_ca":false},{"name":"Axel Mayer","is_ca":false},{"name":"Amanda Kay Montoya","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5170953304107206,"gpt":0.5221333027951004,"spread":0.005037972384379752,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00482609,0.0001145933,0.0001893243,0.0001674947,0.001991554,0.00001482848,0.0001773625,0.00004566887,0.02980458],"category_scores_gemma":[0.0001149286,0.00006680726,0.00006446663,0.00161111,0.00004692833,0.00004504462,0.0001318468,0.0004218437,0.0002856503],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000500398,"about_ca_system_score_gemma":0.00007075758,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000716897,"about_ca_topic_score_gemma":0.00169524,"domain_scores_codex":[0.993635,0.004718751,0.000452786,0.0003594543,0.0005291627,0.0003048319],"domain_scores_gemma":[0.9990798,0.0002056279,0.0002027592,0.0003664567,0.00007307941,0.00007225114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004132277,0.001747289,0.004750806,0.000003138807,0.0001235184,0.00002332367,0.1355281,0.002599261,0.0002245445,0.0007848216,0.7743402,0.07946184],"study_design_scores_gemma":[0.00222864,0.002482381,0.2956623,0.00001012637,0.0003919789,0.0001402541,0.2861677,0.0009214816,0.00004711789,0.001076218,0.4104234,0.0004483518],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6862055,0.0008772864,0.0003351496,0.3001269,0.008113033,0.003120485,0.00003781785,0.00004723828,0.001136593],"genre_scores_gemma":[0.9870103,0.000004143832,0.00006494056,0.003809897,0.004382894,0.002953737,0.00007447718,0.000009156267,0.00169046],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3639168,"threshold_uncertainty_score":0.9993077,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Vilma Ferreira","is_ca":false},{"name":"Nelson Carvas","is_ca":false},{"name":"Mariana Cunha Artilheiro","is_ca":false},{"name":"José Eduardo Pompeu","is_ca":false},{"name":"Syed Ahmed Hassan","is_ca":true},{"name":"Karina Tamy Kasawara","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3527344878480664,"gpt":0.5404014187464845,"spread":0.1876669308984181,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.1031497,0.0005298425,0.02738995,0.002323218,0.0002594551,0.0000253383,0.0002084071,0.0003853765,0.002907624],"category_scores_gemma":[0.06623359,0.000228072,0.01324445,0.001753104,0.0002593644,0.0001239481,0.00008416449,0.0009353874,0.00003640287],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005907778,"about_ca_system_score_gemma":0.005103511,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006827036,"about_ca_topic_score_gemma":0.00003663477,"domain_scores_codex":[0.9512223,0.03906344,0.007056168,0.0006201059,0.001664314,0.0003736689],"domain_scores_gemma":[0.920006,0.06832861,0.009278367,0.0008021281,0.001415005,0.0001698545],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0416774,0.001016952,0.00007888349,0.01068269,0.7880272,7.041146e-7,0.005846219,0.00005188787,0.000004968008,0.0004515611,0.006054261,0.1461072],"study_design_scores_gemma":[0.1089402,0.0002585904,0.0006013262,0.006582193,0.8526877,0.000004133214,0.001339707,0.0022576,5.25773e-7,0.0002520102,0.02684206,0.0002340335],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001034518,0.9652926,0.0005351793,0.005027512,0.001416465,0.02707894,0.0002221922,0.00002692763,0.0002967291],"genre_scores_gemma":[0.001345911,0.978178,0.0007647928,0.0009603032,0.000373251,0.01529932,0.0009776783,0.00004715753,0.002053622],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1458732,"threshold_uncertainty_score":0.9980038,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Elena Castarlenas","is_ca":false},{"name":"Ester Solé","is_ca":false},{"name":"Santiago Galán","is_ca":false},{"name":"Mélanie Racine","is_ca":true},{"name":"Mark P. Jensen","is_ca":false},{"name":"Jordi Miró","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02710772507875651,"gpt":0.3574962764806622,"spread":0.3303885514019057,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01054278,0.00009605902,0.0002022688,0.00005438123,0.0004245303,0.000003432035,0.0001155682,0.00006527326,0.00004353662],"category_scores_gemma":[0.003256429,0.00004132512,0.00005941421,0.0003705773,0.0004777708,0.00004303344,0.00006706836,0.0002744008,0.000001478617],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002530233,"about_ca_system_score_gemma":0.001620446,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008078495,"about_ca_topic_score_gemma":0.005077169,"domain_scores_codex":[0.9930228,0.005716456,0.0004331097,0.0001629515,0.0005120803,0.0001526018],"domain_scores_gemma":[0.9980467,0.000554914,0.0004880994,0.000422283,0.0004326269,0.00005535231],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000134671,0.000334636,0.9608741,0.001547191,0.00002923197,4.146797e-8,0.02648087,0.00001460552,0.0008433774,0.0004013887,0.0009154064,0.008424412],"study_design_scores_gemma":[0.001427647,0.0006939606,0.9830239,0.003803347,0.00004802212,0.00000440122,0.002888894,0.00765594,0.00007017051,0.0003027738,0.00003638852,0.00004459262],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9724236,0.0002091869,0.0002630623,0.02405537,0.0003774164,0.002513619,0.000004665085,0.00001394141,0.0001391678],"genre_scores_gemma":[0.9992752,0.00002311779,0.0001286406,0.0003749689,0.00006830411,0.00005685287,0.000007701017,0.000008045984,0.00005716437],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02685164,"threshold_uncertainty_score":0.3898485,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Ana Barbosa Mendes","is_ca":false},{"name":"Laleh Jamshidi","is_ca":true},{"name":"Wim Van Den Noortgate","is_ca":false},{"name":"Belén Fernández‐Castilla","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9621373876294022,"gpt":0.6045769076283983,"spread":0.3575604800010038,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01373995,0.0001839261,0.0006286413,0.00008191556,0.005129575,0.00001725732,0.0002166149,0.00006523974,0.006343467],"category_scores_gemma":[0.0001483876,0.000105381,0.0003877248,0.001074785,0.00008547561,0.00006981878,0.00009664596,0.0003526006,0.00002655238],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001873123,"about_ca_system_score_gemma":0.0001000589,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009043199,"about_ca_topic_score_gemma":0.0001997066,"domain_scores_codex":[0.9882233,0.009666027,0.0006765084,0.000483348,0.0004901814,0.0004607061],"domain_scores_gemma":[0.9971353,0.001366193,0.0005187066,0.000547604,0.0003281521,0.000104022],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.001666576,0.005148502,0.0009592158,0.00005538576,0.1067731,0.00004593227,0.1531072,0.110662,0.00004101082,0.02197559,0.5224603,0.07710524],"study_design_scores_gemma":[0.005460039,0.0152114,0.01450889,0.00001599031,0.3913358,0.0002923162,0.3588255,0.01184735,0.000003992975,0.131752,0.0691999,0.001546808],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.09069163,0.1084809,0.2647069,0.4700805,0.01647307,0.04512149,0.0007762178,0.0008781813,0.002791191],"genre_scores_gemma":[0.9675796,0.00001537437,0.002336967,0.009670774,0.0002758494,0.01737251,0.0001287961,0.00001907784,0.00260104],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.876888,"threshold_uncertainty_score":0.9961656,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Donna M. Wilson","is_ca":true},{"name":"Corrine D. Truman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3600581615283008,"gpt":0.5633035789381972,"spread":0.2032454174098964,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.008873312,0.0005625881,0.001736053,0.0002860594,0.005445765,0.00005064161,0.0004608134,0.0004581959,0.0002273133],"category_scores_gemma":[0.0006026546,0.0003134937,0.0002466887,0.001043471,0.0001476509,0.0002668083,0.0004930027,0.001361007,0.00006663294],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002133253,"about_ca_system_score_gemma":0.01195243,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001444704,"about_ca_topic_score_gemma":0.002310835,"domain_scores_codex":[0.981622,0.01246659,0.00278938,0.0007714047,0.001517661,0.0008329464],"domain_scores_gemma":[0.9935117,0.001136349,0.002782075,0.001085524,0.001096449,0.0003879198],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007090767,0.00008845997,0.00295205,0.07603737,0.00006191234,1.73526e-7,0.05818616,0.00009491698,1.500158e-7,0.00007032397,0.002359815,0.8600777],"study_design_scores_gemma":[0.002338807,0.0008652522,0.07176334,0.3371301,0.001447943,0.0000106698,0.0159541,0.0005478844,1.897763e-7,0.0003282039,0.5687938,0.0008196479],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.004657417,0.9659845,0.001118016,0.007662283,0.001787899,0.0180887,0.0005955342,0.00009372806,0.00001190713],"genre_scores_gemma":[0.02004615,0.9634762,0.0007630354,0.002652591,0.0003596766,0.006172348,0.006009537,0.00009878715,0.0004217269],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8592581,"threshold_uncertainty_score":0.9999317,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Nicole A. Guitar","is_ca":true},{"name":"Denise M. Connelly","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3172032332576435,"gpt":0.658786603852949,"spread":0.3415833705953055,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03211322,0.001133653,0.005876993,0.0003548357,0.00776325,0.00004922673,0.003144566,0.0007939216,0.001281141],"category_scores_gemma":[0.01571473,0.0005704559,0.001274531,0.002715978,0.0002527686,0.0002432849,0.001924998,0.006324335,0.0006519324],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003420131,"about_ca_system_score_gemma":0.05241195,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001098799,"about_ca_topic_score_gemma":0.0002194871,"domain_scores_codex":[0.8936896,0.08670352,0.010749,0.001392585,0.006170989,0.00129436],"domain_scores_gemma":[0.9781016,0.004342331,0.01186297,0.001909807,0.002863558,0.00091979],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001372782,0.0003450961,0.0007645487,0.8210474,0.0003330495,4.395718e-7,0.09219693,0.00000153794,6.91203e-7,0.00008569743,0.06921546,0.01587193],"study_design_scores_gemma":[0.0007896528,0.0002379329,0.0003191492,0.7721159,0.0009018894,0.000004250574,0.09145311,0.0000190575,3.095833e-7,0.00001554368,0.1336869,0.0004563029],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0007025297,0.8955278,0.00005867546,0.04395845,0.01401798,0.04530136,0.000209049,0.0001475644,0.00007659083],"genre_scores_gemma":[0.002882259,0.9089698,0.0001198591,0.02928618,0.0008871263,0.04789533,0.0008325708,0.0002385057,0.008888419],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.08441199,"threshold_uncertainty_score":0.9996747,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Multivariate statistics; Multivariate analysis; Statistics; Mathematics; Medicine","authors":[{"name":"Lisa M. Lix","is_ca":true},{"name":"H. J. Keselman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4914177710469225,"gpt":0.5819482732124225,"spread":0.09053050216550002,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006694595,0.0001173295,0.000273202,0.00008254808,0.000252271,0.000004297207,0.0001541766,0.00007618697,0.00007637215],"category_scores_gemma":[0.004024371,0.00007520948,0.00003885465,0.0002638164,0.00005567106,0.00008258323,0.00006226161,0.0003420231,0.000008548511],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002799412,"about_ca_system_score_gemma":0.0007226314,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001339546,"about_ca_topic_score_gemma":0.000507754,"domain_scores_codex":[0.9964553,0.001753668,0.0006907729,0.0002088444,0.0005909691,0.0003004135],"domain_scores_gemma":[0.9970437,0.001940502,0.0003641452,0.0003483467,0.0002059216,0.00009738276],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00007349725,0.0006976477,0.0001882582,0.000212942,0.00001762743,0.000001243162,0.02161974,0.005667036,0.001001642,0.9388376,0.0001654251,0.03151728],"study_design_scores_gemma":[0.001369265,0.0001103233,0.009895837,0.0003845,0.00002075462,0.00000126209,0.001294556,0.007358546,0.0001139003,0.9793531,0.00001828897,0.00007969184],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02756122,0.00009061475,0.9626775,0.006869418,0.0002474732,0.002137446,0.00001957982,0.00003097416,0.000365773],"genre_scores_gemma":[0.6794113,0.00001425699,0.3199545,0.0002797276,0.0000239788,0.0002343191,0.000004703502,0.00001499034,0.00006217144],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6518501,"threshold_uncertainty_score":0.4817839,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Allan Donner","is_ca":true},{"name":"Gilda Piaggio","is_ca":false},{"name":"José Villar","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9826344435159781,"gpt":0.7393840614099454,"spread":0.2432503821060327,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7658501,0.0002534486,0.006590441,0.0004108519,0.0006848677,0.0001674277,0.0009140946,0.00009053113,0.058003],"category_scores_gemma":[0.3518569,0.00007940295,0.003615049,0.002082428,0.00006413905,0.0002141016,0.00005968118,0.000194171,0.001006228],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005931285,"about_ca_system_score_gemma":0.0009591514,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001995569,"about_ca_topic_score_gemma":0.00003448983,"domain_scores_codex":[0.2910846,0.6532568,0.03672332,0.001234753,0.01719006,0.0005103946],"domain_scores_gemma":[0.8580644,0.08284257,0.04227762,0.007983935,0.008488576,0.0003429164],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002279728,0.0004387474,0.001337763,0.0002007878,0.01415791,1.745734e-7,0.0116642,0.01861874,0.0003783023,0.03238976,0.8876939,0.03289177],"study_design_scores_gemma":[0.01597034,0.0003377426,0.005690604,0.0002555487,0.1032898,0.0000136653,0.01817508,0.1806452,0.001147557,0.3489054,0.3247136,0.0008553893],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01751489,0.02948214,0.7714603,0.1245071,0.003552488,0.02856991,0.00008236613,0.00002442437,0.02480639],"genre_scores_gemma":[0.9658439,0.0001005689,0.007672969,0.00374261,0.00009268682,0.001123773,0.00003040004,0.00001788215,0.02137519],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.948329,"threshold_uncertainty_score":0.9997716,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Judith M. Ottoson","is_ca":false},{"name":"Ilse Magdalene Patterson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1252692657081936,"gpt":0.5024420030738176,"spread":0.3771727373656241,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008775681,0.0001191528,0.0001343089,0.0001347697,0.0008523372,0.00002154421,0.0001718974,0.00009513408,0.006394889],"category_scores_gemma":[0.0004011181,0.00008490238,0.00002452133,0.0004122399,0.00005033839,0.0001752231,0.00001403597,0.0006087571,0.003943699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000202208,"about_ca_system_score_gemma":0.000376482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001969889,"about_ca_topic_score_gemma":0.000147118,"domain_scores_codex":[0.9949103,0.003304678,0.0005149881,0.0002781297,0.0006734174,0.0003184836],"domain_scores_gemma":[0.9984655,0.0007466216,0.000288863,0.0002830541,0.0001536443,0.00006234235],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003958991,0.0002431255,0.01260931,0.00001076608,0.00002136454,3.899022e-7,0.09296952,0.004854877,0.00001054696,0.004385212,0.006140038,0.878359],"study_design_scores_gemma":[0.00204606,0.000394485,0.7046607,0.0001891687,0.0000293085,0.000004624698,0.02480381,0.01853568,0.000003520111,0.0009424075,0.2481776,0.0002127314],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9048194,0.0003414712,0.0001090398,0.03094785,0.0003651453,0.002455303,0.000001463239,0.0000802005,0.06088014],"genre_scores_gemma":[0.9860048,0.00006570706,0.00007867957,0.00567758,0.0001502377,0.001351534,0.00007805405,0.00001368071,0.006579705],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8781462,"threshold_uncertainty_score":0.9968318,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Michel Perreault","is_ca":true},{"name":"Jeff Renaud","is_ca":true},{"name":"Francine Bourassa","is_ca":true},{"name":"Louise Beauchesne","is_ca":false},{"name":"André Mpiana","is_ca":true},{"name":"Sylvain Bernier","is_ca":true},{"name":"Diana Milton","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3779149513275353,"gpt":0.586986413016971,"spread":0.2090714616894356,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0260702,0.0002472153,0.0005673588,0.0002347349,0.002699183,0.000004081779,0.000482169,0.0002195634,0.001016827],"category_scores_gemma":[0.0006986762,0.0001585167,0.0001373288,0.0007755657,0.0001260802,0.0002374564,0.0001612348,0.0008325411,0.00001457833],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006321804,"about_ca_system_score_gemma":0.009659035,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01574241,"about_ca_topic_score_gemma":0.05758538,"domain_scores_codex":[0.9863855,0.007035334,0.002995358,0.0003904915,0.002530289,0.0006629745],"domain_scores_gemma":[0.9853486,0.003083678,0.005381514,0.001012695,0.004987189,0.0001863612],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006791018,0.0004012381,0.4379303,0.01071726,0.0002610931,9.742071e-9,0.3926761,0.001039078,0.002289022,0.005774389,0.01063591,0.1375965],"study_design_scores_gemma":[0.004846773,0.0006724868,0.7614496,0.001370773,0.0002140929,5.653596e-7,0.1981436,0.02947874,0.000344792,0.002030947,0.001284844,0.0001628119],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8803167,0.0006499858,0.0004292827,0.08844677,0.0046644,0.02447791,0.0009073028,0.00004011464,0.00006752048],"genre_scores_gemma":[0.9894498,0.0001431138,0.0008899714,0.00408817,0.0001530357,0.004643509,0.0005555929,0.00004154904,0.00003523715],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3235193,"threshold_uncertainty_score":0.9998964,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Younghan Lee","is_ca":false},{"name":"Lira Yun","is_ca":true},{"name":"Kim Mi-Lyang","is_ca":false},{"name":"Marvin Washington","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6015343105248565,"gpt":0.638706201338082,"spread":0.03717189081322547,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07291903,0.0002826562,0.002444208,0.0001371548,0.0009295091,0.00003062512,0.0007274984,0.0001936862,0.0001503738],"category_scores_gemma":[0.01010431,0.0001606861,0.0004168655,0.001654602,0.0004217125,0.0003105794,0.00007698166,0.0007428336,0.00004261771],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008007906,"about_ca_system_score_gemma":0.007875999,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003257753,"about_ca_topic_score_gemma":0.0009977859,"domain_scores_codex":[0.9426337,0.05309204,0.001620006,0.0004735103,0.001616228,0.0005645457],"domain_scores_gemma":[0.9922498,0.003048539,0.003229251,0.0007078896,0.0005855772,0.0001789566],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000001378274,0.0002342865,0.00001144227,0.8558236,0.0000335622,8.602343e-8,0.1064713,4.962696e-9,9.185852e-9,0.001014294,0.0001701902,0.03623992],"study_design_scores_gemma":[0.0001179159,0.00004322765,0.00006465796,0.979265,0.0006777136,9.844535e-8,0.005614163,0.00003974101,6.935833e-8,0.002024272,0.01196499,0.0001881996],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0003429009,0.9757816,0.00002309733,0.006792314,0.0002233657,0.01609758,0.00002269751,0.00004075847,0.000675713],"genre_scores_gemma":[0.009705725,0.9862268,0.00002569233,0.0001821003,0.0002183503,0.003288803,0.00006129541,0.00002376244,0.0002674818],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1234414,"threshold_uncertainty_score":0.998234,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Thomas R. Miller","is_ca":false},{"name":"Justin B. Dickerson","is_ca":false},{"name":"Matthew Lee Smith","is_ca":false},{"name":"Marcia G. Ory","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.682390433676244,"gpt":0.5636710224035907,"spread":0.1187194112726533,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06658603,0.000167914,0.0006422583,0.0002192587,0.0009550924,0.0001034752,0.0002474591,0.0001836872,0.0001912887],"category_scores_gemma":[0.01052482,0.000136245,0.0001110362,0.0002490944,0.0001689701,0.0006853758,0.00005073035,0.0003980294,0.00004663618],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003311144,"about_ca_system_score_gemma":0.001685824,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003916592,"about_ca_topic_score_gemma":0.0003249225,"domain_scores_codex":[0.9940875,0.001558638,0.003163369,0.0004776659,0.0002697991,0.0004430474],"domain_scores_gemma":[0.9926735,0.002724712,0.003363274,0.0006017089,0.0004202253,0.0002165439],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000675203,0.002542995,0.2674546,0.01223302,0.0004957075,8.588447e-7,0.05326312,0.001696693,0.001815413,0.2629029,0.1413279,0.2555915],"study_design_scores_gemma":[0.006990351,0.0009919355,0.3112635,0.004304702,0.0001434964,0.00001382249,0.02179368,0.5474011,0.0001265754,0.07755981,0.02829166,0.001119374],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5759618,0.001978264,0.01912287,0.3920179,0.002787755,0.007769632,0.0001141943,0.000067332,0.0001801898],"genre_scores_gemma":[0.9789342,0.00005770652,0.01244116,0.006868721,0.0003594073,0.001066418,0.00007546019,0.00003110139,0.0001658104],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5457044,"threshold_uncertainty_score":0.9978099,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Vernon Curran","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1326738158357816,"gpt":0.5477048575227006,"spread":0.415031041686919,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0160975,0.0001908994,0.0003260271,0.0002080819,0.001185961,0.00004188995,0.0002413586,0.0002362097,0.001830981],"category_scores_gemma":[0.006345584,0.0001382003,0.000070975,0.0006730071,0.0001049892,0.0001860416,0.0000121789,0.0006149998,0.00005064607],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007666315,"about_ca_system_score_gemma":0.02028211,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004979192,"about_ca_topic_score_gemma":0.00002811022,"domain_scores_codex":[0.994611,0.001437121,0.001005477,0.0004189311,0.002056577,0.000470899],"domain_scores_gemma":[0.9961731,0.0003781266,0.0004086999,0.0007460993,0.00198503,0.0003089326],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002610207,0.001300317,0.0006719213,0.000522197,0.0000288194,1.41927e-7,0.007557155,0.01208447,0.0001267033,0.001559937,0.06853143,0.9073559],"study_design_scores_gemma":[0.002274245,0.0003441704,0.001783582,0.001911766,0.0001502293,0.00001227444,0.003810543,0.987298,0.000004231777,0.0006085338,0.001673667,0.0001287661],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6914392,0.001238048,0.06345192,0.2234458,0.003736375,0.01514749,0.00002382949,0.0003600731,0.001157283],"genre_scores_gemma":[0.9647161,0.0000204711,0.009414728,0.01758569,0.0007590237,0.004657528,0.0007248139,0.00005161066,0.002070087],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9752135,"threshold_uncertainty_score":0.9990815,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Hosung Kang","is_ca":true},{"name":"Cecilia Flores‐Sandoval","is_ca":true},{"name":"Benson Law","is_ca":true},{"name":"Shannon L. Sibbald","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4038604056452354,"gpt":0.6740222997443843,"spread":0.2701618940991489,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.03427641,0.000802891,0.003432412,0.001061133,0.00442041,0.00002613301,0.001104092,0.0006662707,0.008483192],"category_scores_gemma":[0.008142996,0.0005229973,0.0007471762,0.006138555,0.0001434263,0.0004409624,0.000982355,0.003128056,0.0005648377],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.007586078,"about_ca_system_score_gemma":0.1251594,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001605332,"about_ca_topic_score_gemma":0.0002880146,"domain_scores_codex":[0.9354462,0.05225867,0.006120687,0.001183581,0.003936095,0.00105478],"domain_scores_gemma":[0.9775575,0.003148263,0.008866466,0.002276314,0.007701692,0.0004497322],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00001887036,0.0001380962,0.00001026869,0.1551868,0.00009064038,7.889019e-8,0.009677656,7.692489e-7,4.032202e-8,0.0003131802,0.1503164,0.6842472],"study_design_scores_gemma":[0.0003942919,0.0001876269,0.0001099629,0.5196106,0.000673588,0.000003667897,0.01582336,0.00005896937,4.295763e-8,0.0001313125,0.4627479,0.0002586864],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000006579286,0.9099842,0.0001420994,0.04997557,0.01031079,0.02756201,0.0002148682,0.00008170566,0.001722146],"genre_scores_gemma":[0.0001653638,0.96118,0.0004801232,0.01310075,0.0007423024,0.01859497,0.004144706,0.0001193135,0.001472491],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6839885,"threshold_uncertainty_score":0.9997222,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}