{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":119,"total_is_capped":false,"direct_labels_cover":7,"predictions_cover":119,"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"},"query_hash":"bd96c94426fe","filters":{"venue":"Research Synthesis Methods"}},"results":[{"id":"W2068264290","doi":"10.1002/jrsm.1123","title":"A scoping review of scoping reviews: advancing the approach and enhancing the consistency","year":2014,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3082,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Guelph; Public Health Agency of Canada","funders":"Ontario Ministry of Agriculture, Food and Rural Affairs","keywords":"Consistency (knowledge bases); Computer science; Management science; Data science; Artificial intelligence; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.923851977769396,"gpt":0.7277039479738097,"spread":0.1961480297955863,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8674546,0.001038748,0.03233648,0.001053986,0.001231888,0.001237589,0.007614791,0.0003332912,0.001921087],"category_scores_gemma":[0.7785432,0.0003363213,0.00714746,0.006299767,0.00103737,0.0002395186,0.00170997,0.001707473,0.0004517362],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001302364,"about_ca_system_score_gemma":0.001512423,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001762829,"about_ca_topic_score_gemma":0.00001686503,"domain_scores_codex":[0.3442413,0.6025283,0.03582132,0.003359887,0.01253693,0.001512236],"domain_scores_gemma":[0.1675581,0.7719051,0.0312148,0.02374607,0.00490973,0.0006662289],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[4.055199e-7,0.000009484009,5.320543e-7,0.3897209,0.0001654179,4.844143e-7,0.00005685872,1.866977e-7,0.00000125018,0.0003091281,0.001830547,0.6079048],"study_design_scores_gemma":[0.00001265201,0.00001099199,2.307412e-7,0.5386342,0.001300408,0.00003874799,0.0001469155,0.00007943699,0.000005345767,0.0002387714,0.4593396,0.0001926586],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[7.881358e-8,0.8912809,0.08701174,0.0003627573,0.0001505349,0.0130855,0.000007171405,0.000006399373,0.008094846],"genre_scores_gemma":[2.586249e-7,0.8728453,0.1229208,0.0001712135,0.0002188215,0.003064823,0.000002458247,0.00006128768,0.0007150216],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6077121,"threshold_uncertainty_score":0.9999089,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W1784175517","doi":"10.1002/jrsm.1164","title":"Methods to estimate the between‐study variance and its uncertainty in meta‐analysis","year":2015,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1615,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Medical Research Council; National Institute for Health and Care Research","keywords":"Meta-analysis; Variance (accounting); Econometrics; Computer science; Statistics; Statistical analysis; Mathematics; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.9521150158924461,"gpt":0.7527147048590068,"spread":0.1994003110334392,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8450038,0.0004814694,0.008835135,0.002719626,0.0004171867,0.001746516,0.004420181,0.0001418801,0.00245257],"category_scores_gemma":[0.5537968,0.0001948238,0.002405785,0.01772856,0.0001650094,0.0002987077,0.001209393,0.0006678993,0.0006768132],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001231278,"about_ca_system_score_gemma":0.0002223872,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004954842,"about_ca_topic_score_gemma":0.0004279155,"domain_scores_codex":[0.4987711,0.4780998,0.00820989,0.002654823,0.01113623,0.001128168],"domain_scores_gemma":[0.5094136,0.4671413,0.002897857,0.01255288,0.006174935,0.001819353],"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":[0.0001117012,0.0006145676,0.0395983,0.0001231983,0.1632858,0.00005949637,0.01284588,0.02119855,0.0006357852,0.00582408,0.007329979,0.7483727],"study_design_scores_gemma":[0.0006988874,0.0005737386,0.1174555,0.00005409095,0.2126344,0.00001613448,0.02394611,0.4008975,0.002602129,0.1030822,0.1363223,0.001716924],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04544069,0.01217517,0.9169139,0.01199687,0.000155994,0.005187104,0.00003937854,0.00002271925,0.008068215],"genre_scores_gemma":[0.3172601,0.00001993657,0.6779604,0.00008983394,0.00007303189,0.001092612,7.751697e-7,0.00002599786,0.003477296],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7466558,"threshold_uncertainty_score":0.9992898,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2885675882","doi":"10.1002/jrsm.1316","title":"A comparison of heterogeneity variance estimators in simulated random‐effects meta‐analyses","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":1026,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Michael's Hospital","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Estimator; Meta-analysis; Random effects model; Variance (accounting); Econometrics; Statistics; Variance components; Computer science; Estimation; Mathematics; Economics; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.6474146937413818,"gpt":0.6854525192511277,"spread":0.03803782550974588,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02236141,0.0002797965,0.001963798,0.0005537727,0.0001634962,0.00005599243,0.0005677435,0.0002227799,0.0004566901],"category_scores_gemma":[0.1355134,0.000204033,0.0002899603,0.001517424,0.0007661665,0.00009889595,0.0002089316,0.0005259208,0.00002048288],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007538885,"about_ca_system_score_gemma":0.0001260695,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002296973,"about_ca_topic_score_gemma":0.00003856561,"domain_scores_codex":[0.9779407,0.01893757,0.0009715584,0.0005543361,0.0008652472,0.0007306358],"domain_scores_gemma":[0.8570858,0.1411158,0.0002399805,0.0007948888,0.0005513271,0.0002122692],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002154699,0.003336794,0.008820699,0.004135055,0.008067599,0.00009514541,0.002069505,0.0001745987,0.2083574,0.1830657,0.0007368717,0.5789859],"study_design_scores_gemma":[0.0004587591,0.0001965463,0.001295963,0.0002170085,0.0006916628,0.000001442721,0.00005030613,0.06695732,0.6171876,0.312678,0.00006592388,0.0001994331],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02582819,0.0005073066,0.971609,0.00006613418,0.00008925358,0.0005744242,0.00001385561,0.000047701,0.001264126],"genre_scores_gemma":[0.3884794,0.000009485066,0.611362,0.000005738482,0.00003519854,0.00007265902,2.876575e-7,0.00002331767,0.00001191104],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5787865,"threshold_uncertainty_score":0.8717685,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1911705068","doi":"10.1002/jrsm.1106","title":"Searching for grey literature for systematic reviews: challenges and benefits","year":2013,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":803,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Work & Health; University of Waterloo; University of Toronto","funders":"Workers Compensation Board of Manitoba","keywords":"Grey literature; Systematic review; Scope (computer science); Computer science; Process (computing); Data science; Management science; Information retrieval; MEDLINE; Engineering; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9427986778929087,"gpt":0.6848807056848404,"spread":0.2579179722080683,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6505072,0.0004181424,0.00634373,0.001137879,0.0005666972,0.003448298,0.002288557,0.0002083388,0.0006840758],"category_scores_gemma":[0.741722,0.0001847899,0.001972079,0.001228071,0.0001256443,0.0006703832,0.0003064251,0.0003449674,0.0007095443],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005148207,"about_ca_system_score_gemma":0.00006098198,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006923514,"about_ca_topic_score_gemma":0.000008196474,"domain_scores_codex":[0.85653,0.1276027,0.007962733,0.001864409,0.005043385,0.0009968068],"domain_scores_gemma":[0.4121601,0.5644368,0.003853676,0.007554207,0.01111043,0.0008848078],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000009501728,0.0000404325,0.00002044669,0.04460454,0.0002644362,5.061901e-7,0.001607521,0.00000365194,0.0004499888,0.0144771,0.01383623,0.9246857],"study_design_scores_gemma":[0.0006844105,0.000547046,0.001839974,0.05602794,0.0009688631,0.00007502478,0.01202042,0.0776349,0.002316679,0.3401588,0.5062315,0.00149437],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.004459862,0.6136814,0.311117,0.01190299,0.0003485329,0.0507247,0.0001022597,0.00003259572,0.007630589],"genre_scores_gemma":[0.009561552,0.01246366,0.9359971,0.0001123457,0.0003967708,0.02732175,0.000005876429,0.00007523696,0.01406571],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9231912,"threshold_uncertainty_score":0.9975862,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4210542220","doi":"10.1002/jrsm.1547","title":"Meta‐analysis of prevalence: <scp><i>I</i><sup>2</sup></scp> statistic and how to deal with heterogeneity","year":2022,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":524,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Meta-analysis; Statistics; Statistic; Point estimation; Econometrics; Study heterogeneity; Sample size determination; Confidence interval; Systematic review; Pooled variance; Publication bias; Summary statistics; Subgroup analysis; Demography; Medicine; Mathematics; MEDLINE; Internal medicine; Biology","retraction":null,"screen_n_in":null,"score":{"opus":0.8023315025989195,"gpt":0.6016847519460555,"spread":0.200646750652864,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4758545,0.000431367,0.007482494,0.003282752,0.0007229819,0.0009285659,0.003155712,0.00007685479,0.00657859],"category_scores_gemma":[0.1223074,0.0002214733,0.002759197,0.01182165,0.0003382347,0.0002281025,0.001261187,0.0005078195,0.00008736854],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009435392,"about_ca_system_score_gemma":0.0002236257,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006581377,"about_ca_topic_score_gemma":0.00003979616,"domain_scores_codex":[0.7508338,0.2230047,0.004736643,0.002552582,0.01783348,0.001038838],"domain_scores_gemma":[0.7217814,0.2646533,0.002838451,0.006538359,0.003221871,0.0009666217],"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":[0.0001555189,0.0008424343,0.06163869,0.001812426,0.6031743,0.000118156,0.00977348,0.05511849,0.003777042,0.004288578,0.04859291,0.2107079],"study_design_scores_gemma":[0.000465195,0.001352346,0.03228266,0.00005853674,0.3855777,0.00007317916,0.02209501,0.235455,0.009804004,0.005617616,0.3064538,0.0007648694],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.4130825,0.01181754,0.5665437,0.00182591,0.00003407725,0.00293589,0.001127553,0.00001965209,0.002613172],"genre_scores_gemma":[0.3339019,0.0001398012,0.6548268,0.0001121561,0.00002804386,0.001384778,0.00000575277,0.00003870302,0.009562159],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3535471,"threshold_uncertainty_score":0.9943295,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2604568423","doi":"10.1002/jrsm.1287","title":"Machine learning for identifying Randomized Controlled Trials: An evaluation and practitioner's guide","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":449,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"U.S. National Library of Medicine; Agency for Healthcare Research and Quality; National Institutes of Health; National Cancer Institute; Medical Research Council; McMaster University","keywords":"Computer science; Machine learning; Artificial intelligence; Randomized controlled trial; Workflow; Convolutional neural network; Support vector machine; Systematic review; Sensitivity (control systems); Data mining; MEDLINE; Medicine; Database","retraction":null,"screen_n_in":null,"score":{"opus":0.4261457634092655,"gpt":0.6226321823758852,"spread":0.1964864189666198,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5441753,0.0002309819,0.002271913,0.0008436915,0.001321518,0.0008525858,0.0007710583,0.0001641773,0.0002121673],"category_scores_gemma":[0.6465802,0.0001665332,0.0003798974,0.0005819079,0.000317977,0.0008118777,0.0003096261,0.000664958,0.00001420615],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001731778,"about_ca_system_score_gemma":0.0004580516,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000579313,"about_ca_topic_score_gemma":0.00003247378,"domain_scores_codex":[0.807717,0.1876723,0.00145528,0.0008930854,0.001576549,0.0006858312],"domain_scores_gemma":[0.6040453,0.3878423,0.001379948,0.001422274,0.00478087,0.0005292166],"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":[0.01669798,0.00006124123,0.00006934762,0.0001376212,0.000303536,0.000001764311,0.001508344,0.0002190794,0.004987935,0.01787479,0.0001562171,0.9579821],"study_design_scores_gemma":[0.04082707,0.0001922546,0.00006662342,0.0001059068,0.0001315213,0.000009932432,0.0001464597,0.9273446,0.00335088,0.02244986,0.00520885,0.0001660906],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0009053992,0.002201711,0.9851065,0.003921509,0.0003881909,0.00486666,0.000001899561,0.0001873618,0.002420809],"genre_scores_gemma":[0.09274568,0.0002945466,0.9020898,0.00007129309,0.0006605147,0.003561172,0.000006904919,0.00003676202,0.0005332786],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9578161,"threshold_uncertainty_score":0.9999786,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2106889389","doi":"10.1002/jrsm.46","title":"Pooling health‐related quality of life outcomes in meta‐analysis—a tutorial and review of methods for enhancing interpretability","year":2011,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":293,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hospital for Sick Children; McMaster University","funders":"","keywords":"Interpretability; Pooling; Strictly standardized mean difference; Meta-analysis; Odds; Odds ratio; Construct (python library); Psychological intervention; Statistics; Medicine; Computer science; Mathematics; Artificial intelligence; Logistic regression","retraction":null,"screen_n_in":null,"score":{"opus":0.9473026375794946,"gpt":0.738310881158621,"spread":0.2089917564208736,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8665285,0.0003648087,0.01955656,0.001862309,0.0001231049,0.00009477402,0.001906559,0.0001784567,0.005200711],"category_scores_gemma":[0.764193,0.0001850203,0.007081633,0.005350828,0.0003213858,0.0002382543,0.0004378083,0.0004148449,0.000009452427],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008380038,"about_ca_system_score_gemma":0.0004453088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001555747,"about_ca_topic_score_gemma":0.0002923259,"domain_scores_codex":[0.4431213,0.5137088,0.03438266,0.002251618,0.005604766,0.0009308822],"domain_scores_gemma":[0.3169543,0.6447713,0.0193309,0.01041427,0.007609172,0.0009200806],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004888079,0.001238409,0.05064848,0.05088737,0.2085486,0.000001076405,0.01115613,0.00003245477,0.005292914,0.0160296,0.0005398039,0.6551364],"study_design_scores_gemma":[0.00275166,0.001497268,0.2094546,0.008481305,0.282389,0.000008931129,0.01896462,0.04897263,0.07612929,0.3353286,0.01223252,0.003789627],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008800416,0.1570633,0.8288041,0.0009333528,0.0002045368,0.00309353,0.00005476479,0.00000704426,0.001038929],"genre_scores_gemma":[0.09814334,0.002349884,0.898811,0.00007397686,0.00001735764,0.000385935,0.000001462725,0.00001764636,0.0001993508],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6513467,"threshold_uncertainty_score":0.9957086,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4221118345","doi":"10.1002/jrsm.1557","title":"Graphical Representation of Overlap for <scp>OVErviews</scp>: <scp>GROOVE</scp> tool","year":2022,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":257,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact","funders":"Comisión Nacional de Investigación Científica y Tecnológica; Universitat Autònoma de Barcelona; Instituto de Salud Carlos III; Fondo Nacional de Desarrollo Científico y Tecnológico; Agencia Nacional de Investigación y Desarrollo; Ministerio de Economía y Competitividad","keywords":"Computer science; Representation (politics); Matrix (chemical analysis); Data mining; Groove (engineering); Information retrieval; Matrix representation","retraction":null,"screen_n_in":null,"score":{"opus":0.8564511360551862,"gpt":0.6584873612399715,"spread":0.1979637748152147,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.520696,0.0004466122,0.005032916,0.002175115,0.0009567327,0.0008084142,0.004674645,0.0001968038,0.003465037],"category_scores_gemma":[0.732603,0.0002763394,0.004343123,0.008245128,0.0003301699,0.0004242652,0.001306896,0.0007974177,0.0003135567],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001588951,"about_ca_system_score_gemma":0.0003114165,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006074711,"about_ca_topic_score_gemma":0.000008550353,"domain_scores_codex":[0.7001514,0.2503756,0.01549563,0.0037634,0.02829946,0.001914502],"domain_scores_gemma":[0.3005836,0.6754956,0.006862586,0.01092818,0.005449584,0.0006805441],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003030766,0.0006905023,0.01293915,0.0006329712,0.001300294,0.00001715587,0.002295292,0.0005167964,0.02492379,0.04398733,0.5621125,0.350554],"study_design_scores_gemma":[0.0003134067,0.0002461937,0.008521835,0.00005467782,0.0003319354,0.00001908771,0.01112796,0.01822026,0.01877631,0.09997199,0.8423187,0.00009765067],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2634029,0.007606428,0.6655095,0.00145955,0.001166988,0.008203508,0.000542524,0.00005147653,0.05205715],"genre_scores_gemma":[0.2227294,0.0007397193,0.700323,0.0003989688,0.0006419328,0.007561858,0.00005115587,0.0001852878,0.06736869],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4251199,"threshold_uncertainty_score":0.9999689,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3210017165","doi":"10.1002/jrsm.1535","title":"Introduction to PRISMA 2020 and implications for research synthesis methodologists","year":2021,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":256,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Systematic review; Statement (logic); Computer science; Key (lock); MEDLINE; Research methodology; Data science; Management science; Epistemology; Medicine; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9629191545952714,"gpt":0.7585722339411163,"spread":0.2043469206541552,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7612048,0.0003351567,0.003493674,0.001728825,0.001287423,0.002201197,0.002620728,0.0002609942,0.006379274],"category_scores_gemma":[0.9152082,0.0002001577,0.001124074,0.008558288,0.0004335091,0.0002843113,0.001116491,0.0006844754,0.00116316],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002081325,"about_ca_system_score_gemma":0.0004687501,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002948513,"about_ca_topic_score_gemma":0.00003390164,"domain_scores_codex":[0.6141402,0.3629912,0.007743736,0.004493065,0.008902233,0.001729568],"domain_scores_gemma":[0.189859,0.7723045,0.001594782,0.01311707,0.02173536,0.001389284],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003207567,0.00007822228,0.0005420893,0.0001111789,0.0002068213,0.000002659835,0.0002404726,0.00001649638,0.03969796,0.01204875,0.2723987,0.6746245],"study_design_scores_gemma":[0.00006113097,0.00006939285,0.0112085,0.00005297543,0.0001415618,0.00004301073,0.002137848,0.001163695,0.06777566,0.08954083,0.8275459,0.0002594857],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.006650538,0.002633395,0.8049579,0.1776075,0.0002826613,0.002567519,0.00007828387,0.00001987025,0.005202318],"genre_scores_gemma":[0.009192144,0.0003989231,0.9700305,0.0001092582,0.001103885,0.003515395,0.000004358129,0.00004377317,0.01560171],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.674365,"threshold_uncertainty_score":0.9996145,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3088093190","doi":"10.1002/jrsm.1458","title":"The effect direction plot revisited: Application of the 2019 Cochrane Handbook guidance on alternative synthesis methods","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":249,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Plot (graphics); Metric (unit); Computer science; Forest plot; Visualization; Meta-analysis; Systematic review; Data mining; Sign (mathematics); Data science; Artificial intelligence; Statistics; MEDLINE; Medicine; Mathematics; Operations management","retraction":null,"screen_n_in":null,"score":{"opus":0.4509388357373815,"gpt":0.5858464710871348,"spread":0.1349076353497533,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.160489,0.0002965036,0.001442231,0.0003057177,0.0009259196,0.0001459416,0.001344829,0.0002086908,0.0001366169],"category_scores_gemma":[0.1955704,0.0002247999,0.000375604,0.001002821,0.0004352261,0.0002298522,0.0002457504,0.0006544964,0.0006425647],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006701124,"about_ca_system_score_gemma":0.0001356068,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008064705,"about_ca_topic_score_gemma":0.00001588186,"domain_scores_codex":[0.9641169,0.03052108,0.003122295,0.001016515,0.000521369,0.0007018409],"domain_scores_gemma":[0.8775634,0.1179167,0.002264894,0.001658358,0.0003157175,0.00028102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001471723,0.000423002,0.01464776,0.004702932,0.001899874,0.00000208106,0.005237706,0.002723954,0.01857985,0.2252155,0.08500356,0.6400921],"study_design_scores_gemma":[0.0009605295,0.0008222762,0.02232224,0.002212933,0.0001158351,0.000006294071,0.0007589794,0.1451671,0.4254179,0.0241795,0.3769547,0.001081698],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008642462,0.03240076,0.8238878,0.1106467,0.001025052,0.006396451,0.0005250332,0.000188042,0.01628773],"genre_scores_gemma":[0.3073798,0.0123259,0.6504108,0.00878112,0.004684348,0.01166122,0.0000259969,0.0006250059,0.004105767],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6390104,"threshold_uncertainty_score":0.9167075,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2126629865","doi":"10.1002/jrsm.1078","title":"Non‐randomized studies as a source of complementary, sequential or replacement evidence for randomized controlled trials in systematic reviews on the effects of interventions","year":2013,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":245,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; McMaster University","funders":"","keywords":"Generalizability theory; Randomized controlled trial; Systematic review; Credibility; Context (archaeology); Evidence-based medicine; Evidence-based practice; Psychology; External validity; Health care; Psychological intervention; MEDLINE; Population; Medicine; Applied psychology; Social psychology; Alternative medicine; Nursing; Epistemology; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9453333605578149,"gpt":0.7340988391264817,"spread":0.2112345214313333,"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":[0.9554551,0.0004841941,0.06025628,0.001614239,0.0002302151,0.0004102075,0.00254633,0.00008920331,0.006438646],"category_scores_gemma":[0.9905246,0.0001405495,0.01997388,0.001754282,0.0008574391,0.0001935918,0.0003894188,0.0002636039,0.0002007133],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001310156,"about_ca_system_score_gemma":0.0001736546,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002827828,"about_ca_topic_score_gemma":0.00003637737,"domain_scores_codex":[0.09125347,0.8506657,0.04901547,0.001120321,0.007360904,0.0005840711],"domain_scores_gemma":[0.002716952,0.9670022,0.02233332,0.003914325,0.003897296,0.0001358988],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.6919625,0.0008961497,0.00003056571,0.1972146,0.0318323,0.000002700152,0.006988965,0.00005848141,0.006351719,0.008902861,0.03730305,0.01845605],"study_design_scores_gemma":[0.6969717,0.0009954284,0.000008255412,0.1843392,0.01625329,0.00000421969,0.01372368,0.03644885,0.0150033,0.03516182,0.0006500593,0.0004402213],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"methods","genre_gemma":"protocol","genre_scores_codex":[0.02683709,0.09317076,0.5409442,0.005885655,0.0007354317,0.3316314,0.00001123465,0.000008965911,0.0007753532],"genre_scores_gemma":[0.3345582,0.01813303,0.2269926,0.000284599,0.0002046543,0.3952168,0.000002962571,0.00008167576,0.02452541],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.3139515,"threshold_uncertainty_score":0.9944696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2884972765","doi":"10.1002/jrsm.1313","title":"Using systematic reviews in guideline development: The GRADE approach","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":243,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Grading (engineering); Guideline; Systematic review; Certainty; Psychological intervention; Computer science; Confounding; MEDLINE; Psychology; Medicine; Engineering; Nursing; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9791960199110384,"gpt":0.7442177456352804,"spread":0.234978274275758,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8632087,0.0004009545,0.006124179,0.001454276,0.0007396342,0.001252597,0.005380542,0.0001476986,0.00419356],"category_scores_gemma":[0.5803725,0.0001457631,0.001240051,0.007095107,0.0004089928,0.0002504916,0.0006408031,0.0005147315,0.003194722],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002146268,"about_ca_system_score_gemma":0.0002974696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004163822,"about_ca_topic_score_gemma":0.00003321728,"domain_scores_codex":[0.4680173,0.4888417,0.02563643,0.002165818,0.01402295,0.001315849],"domain_scores_gemma":[0.7202591,0.2385099,0.01099318,0.02147537,0.007983257,0.0007791892],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004523493,0.0008680836,0.003813802,0.02255713,0.001450422,0.00002468022,0.01653108,0.0002495658,0.008574547,0.01387889,0.2171574,0.7148492],"study_design_scores_gemma":[0.0001504673,0.00003134293,0.0006250985,0.003678597,0.0002763234,0.00006173483,0.005792775,0.2457056,0.00381623,0.009697546,0.7295892,0.0005751073],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.003095205,0.0106696,0.9546632,0.0008827338,0.000231737,0.006155991,0.00000215092,0.000009875254,0.02428955],"genre_scores_gemma":[0.0168255,0.00008884347,0.9780542,0.0001413535,0.000239678,0.0008224319,0.000001083725,0.00002944359,0.003797495],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7142741,"threshold_uncertainty_score":0.9998475,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2888732220","doi":"10.1002/jrsm.1319","title":"Methods to calculate uncertainty in the estimated overall effect size from a random‐effects meta‐analysis","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":238,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Michael's Hospital","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Meta-analysis; Random effects model; Confidence interval; Statistics; Point estimation; Econometrics; Outcome (game theory); Computer science; Prediction interval; Interval (graph theory); Mathematics; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.8407238715168954,"gpt":0.6957608599089178,"spread":0.1449630116079776,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7887173,0.0007643484,0.01282462,0.002584667,0.0006359837,0.002338636,0.006534556,0.0002948964,0.02045905],"category_scores_gemma":[0.7272811,0.0002963113,0.008801485,0.01770295,0.0004370976,0.0002709666,0.0007380999,0.0008292895,0.00259054],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000162511,"about_ca_system_score_gemma":0.000151438,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002330832,"about_ca_topic_score_gemma":0.0004556978,"domain_scores_codex":[0.4336022,0.5458371,0.006649616,0.002548591,0.01011578,0.001246757],"domain_scores_gemma":[0.1137426,0.8687429,0.002095988,0.01182265,0.002846172,0.0007497158],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001272629,0.0005026762,0.003355661,0.0002806561,0.2524453,0.0001576395,0.007280054,0.006063807,0.03924012,0.001410022,0.02301968,0.6649718],"study_design_scores_gemma":[0.001962008,0.0007775958,0.03683319,0.0001826001,0.2340284,0.00001426482,0.001001357,0.5341312,0.04805265,0.06889893,0.07243551,0.001682243],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.08176427,0.005100101,0.8960487,0.004057253,0.0002752377,0.005534825,0.00005530518,0.00003330248,0.007131015],"genre_scores_gemma":[0.244452,0.00002890831,0.7513862,0.0005753125,0.0001698537,0.001842739,0.000003744257,0.00004009022,0.001501239],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6632895,"threshold_uncertainty_score":0.9999489,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3006701470","doi":"10.1002/jrsm.1399","title":"Innovations in framework synthesis as a systematic review method","year":2020,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":223,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Ontario Tech University; McMaster University Medical Centre","funders":"","keywords":"Management science; Computer science; Systematic review; Stakeholder; Data science; MEDLINE; Political science; Public relations; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.9235373563940749,"gpt":0.8552043114246342,"spread":0.0683330449694407,"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","research_integrity","insufficient_payload"],"category_scores_codex":[0.2142847,0.0009443407,0.01054147,0.003461773,0.001588513,0.0001005539,0.003303678,0.00139119,0.004642937],"category_scores_gemma":[0.7660407,0.0007300405,0.0007598571,0.01851394,0.0002987423,0.0003132119,0.001296466,0.008244838,0.007903744],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003183292,"about_ca_system_score_gemma":0.01442674,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005297597,"about_ca_topic_score_gemma":0.00003723281,"domain_scores_codex":[0.7251812,0.2592979,0.007791292,0.001797982,0.002941224,0.002990387],"domain_scores_gemma":[0.2288699,0.759663,0.003803784,0.00413263,0.0019422,0.00158856],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000002373106,0.00002210651,7.750122e-7,0.6454921,0.00009829953,0.00002077223,0.0006475742,2.668613e-8,3.973871e-7,0.004894453,0.003398665,0.3454224],"study_design_scores_gemma":[0.00002722028,0.00001643385,5.934724e-7,0.4863464,0.0004824621,0.00001086397,0.0005003662,0.0000110823,0.000002059328,0.001161819,0.5111428,0.0002979434],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[2.837548e-8,0.9234039,0.03801204,0.0131006,0.0003249968,0.01893254,0.0002104708,0.0001820044,0.005833401],"genre_scores_gemma":[6.644196e-8,0.6681702,0.280933,0.004375245,0.0002755316,0.04561097,0.000006594608,0.0001613152,0.0004670511],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5517559,"threshold_uncertainty_score":0.9999052,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2062040186","doi":"10.1002/jrsm.1056","title":"Issues relating to study design and risk of bias when including non‐randomized studies in systematic reviews on the effects of interventions","year":2012,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":187,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute of Population and Public Health; University of Ottawa","funders":"National Institutes of Health; National Institute for Health and Care Research; Scottish Government","keywords":"Randomized controlled trial; Psychological intervention; Systematic review; Randomized experiment; Research design; Clinical study design; Psychology; Computer science; MEDLINE; Management science; Medicine; Applied psychology; Clinical trial; Social science; Psychiatry; Sociology; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.8578401848948041,"gpt":0.6814907782416452,"spread":0.1763494066531589,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1995873,0.0002061033,0.002215457,0.0005687288,0.0001573099,0.00001619782,0.0003410796,0.00006435932,0.000009840491],"category_scores_gemma":[0.7106343,0.0001116578,0.0002161324,0.0005723989,0.0003179969,0.0001402721,0.0004589206,0.0004154517,0.000002575943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001095085,"about_ca_system_score_gemma":0.00001303186,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008435213,"about_ca_topic_score_gemma":0.000008057466,"domain_scores_codex":[0.908967,0.08822193,0.001587461,0.0002467271,0.0005817629,0.0003950866],"domain_scores_gemma":[0.4823799,0.515647,0.0008099584,0.0007706991,0.0003119553,0.00008048552],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.004785346,0.004890371,0.00652895,0.4168761,0.005664874,0.00001318371,0.3759509,0.00002849405,0.03606245,0.06121167,0.000935411,0.0870522],"study_design_scores_gemma":[0.003631867,0.001076778,0.0002279887,0.1733731,0.001142244,0.000002403213,0.02792336,0.0006734966,0.4339803,0.3576271,0.000005203264,0.000336062],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2950661,0.01765103,0.6652485,0.000103642,0.00006339275,0.02162465,0.000001536903,0.00005669953,0.0001844473],"genre_scores_gemma":[0.3559909,0.002715484,0.6370519,0.00000214537,0.0000134225,0.004144137,2.507439e-8,0.00002338217,0.00005858616],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5110469,"threshold_uncertainty_score":0.8241933,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2015957719","doi":"10.1002/jrsm.41","title":"Reliability and validity of three quality rating instruments for systematic reviews of observational studies","year":2011,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":172,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Observational study; Intraclass correlation; Reliability (semiconductor); Validity; Inter-rater reliability; Rating scale; Sign (mathematics); Statistics; Psychology; Medicine; Clinical psychology; Psychometrics; Mathematics; Physics","retraction":null,"screen_n_in":null,"score":{"opus":0.9900024069594315,"gpt":0.7477755849671291,"spread":0.2422268219923024,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8087921,0.0002438605,0.007864277,0.0003951731,0.0002196585,0.00008880193,0.001474592,0.00009546765,0.0007492912],"category_scores_gemma":[0.8840745,0.000111877,0.001442443,0.001342688,0.0004512875,0.0002822269,0.0003930293,0.0001596035,0.00002149129],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005588288,"about_ca_system_score_gemma":0.0001114482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000384255,"about_ca_topic_score_gemma":0.0000224045,"domain_scores_codex":[0.5890741,0.366683,0.03072841,0.001930612,0.01090412,0.0006796982],"domain_scores_gemma":[0.1892415,0.7585156,0.02201451,0.01058866,0.01923829,0.0004014155],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000185965,0.0008039449,0.3942699,0.3074568,0.002673974,6.421349e-7,0.008264499,0.000008647678,0.01198852,0.05010824,0.004045944,0.2201929],"study_design_scores_gemma":[0.0005196415,0.0004694738,0.318418,0.01394123,0.00139547,0.000004573428,0.01137795,0.01290175,0.04205131,0.5954186,0.002810207,0.0006917478],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.6903123,0.005514963,0.2899273,0.0002601046,0.000203417,0.01129269,0.00007946771,0.000006649392,0.00240313],"genre_scores_gemma":[0.3204356,0.0002322615,0.6780772,0.000009174619,0.00002208668,0.0009794317,7.134355e-7,0.000009180136,0.0002343434],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5453104,"threshold_uncertainty_score":0.8204213,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1803108444","doi":"10.1002/jrsm.1077","title":"Checklists of methodological issues for review authors to consider when including non‐randomized studies in systematic reviews","year":2013,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":140,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Institute of Population and Public Health; University of Ottawa","funders":"Economic and Social Research Council; National Institute for Health and Care Research; Scottish Government","keywords":"Systematic review; Operationalization; Randomized controlled trial; Confounding; MEDLINE; Meta-analysis; Psychological intervention; Checklist; Research design; Clinical study design; Psychology; Management science; Computer science; Medicine; Clinical trial; Political science; Engineering; Psychiatry; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.9865716552292881,"gpt":0.792126037302518,"spread":0.1944456179267702,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.9009857,0.0005801839,0.03010231,0.001748142,0.0002323018,0.0004086493,0.003210286,0.0002112542,0.005267409],"category_scores_gemma":[0.9741291,0.0002400591,0.004266165,0.003648095,0.0006792889,0.0003285315,0.000887362,0.0004395085,0.001332484],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001646104,"about_ca_system_score_gemma":0.0001176432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009521694,"about_ca_topic_score_gemma":0.00001539634,"domain_scores_codex":[0.2643791,0.6917942,0.0316907,0.002349497,0.008574951,0.001211549],"domain_scores_gemma":[0.02675575,0.9449442,0.00940165,0.007659135,0.0105716,0.0006676791],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005920011,0.000238468,0.0001705604,0.3434389,0.002415101,0.0000105492,0.009282623,0.00001715322,0.004595179,0.005498094,0.5102397,0.1235017],"study_design_scores_gemma":[0.008666141,0.0003804338,0.0002613558,0.3261513,0.004523299,0.0000598963,0.01976805,0.01521963,0.0120823,0.4332714,0.1775802,0.002036032],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.001551726,0.5113584,0.4000314,0.02129152,0.000460128,0.06194866,0.00001274739,0.00001682946,0.003328572],"genre_scores_gemma":[0.001963215,0.02509085,0.9421315,0.000378802,0.0000763864,0.0223586,7.67007e-7,0.00003146154,0.007968463],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.5421001,"threshold_uncertainty_score":0.9994451,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1484962366","doi":"10.1002/jrsm.1068","title":"An introduction to methodological issues when including non‐randomised studies in systematic reviews on the effects of interventions","year":2013,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":115,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; Centre for Global Health Research; Institute of Population and Public Health","funders":"","keywords":"Psychological intervention; Systematic review; MEDLINE; Stakeholder; Medical education; Randomized controlled trial; Alternative medicine; Meta-analysis; Health care; Psychology; Medicine; Management science; Nursing; Political science; Public relations","retraction":null,"screen_n_in":null,"score":{"opus":0.9733209713276808,"gpt":0.7604628857848775,"spread":0.2128580855428033,"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":[0.7554967,0.0003442317,0.008007462,0.001652722,0.0003042299,0.0005592602,0.002924824,0.0001118982,0.003203677],"category_scores_gemma":[0.9278411,0.0001236068,0.002115029,0.003096013,0.000259756,0.0003197927,0.0004584657,0.0004563228,0.001266455],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001402408,"about_ca_system_score_gemma":0.00002832962,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005261593,"about_ca_topic_score_gemma":0.00001627978,"domain_scores_codex":[0.4357958,0.5466801,0.01088403,0.001304833,0.004804006,0.0005312072],"domain_scores_gemma":[0.08328799,0.8956845,0.00565875,0.01031878,0.004614851,0.0004350909],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003177662,0.001374428,0.0007131226,0.1198682,0.002756902,0.00001492694,0.02552886,0.0002110151,0.1452653,0.01381139,0.3096409,0.3804971],"study_design_scores_gemma":[0.003721618,0.00477522,0.01040177,0.148426,0.003774581,0.00005442188,0.1033122,0.03320297,0.2466108,0.3966012,0.04644484,0.002674331],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.207247,0.0448398,0.6208808,0.05821894,0.00182068,0.06489868,0.000005460465,0.00003645284,0.002052249],"genre_scores_gemma":[0.186956,0.001564221,0.7873394,0.0001769395,0.0005066327,0.01853261,8.024567e-7,0.00003897345,0.004884483],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3827898,"threshold_uncertainty_score":0.9995112,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2043360969","doi":"10.1002/jrsm.30","title":"The evolution of a new publication type: Steps and challenges of producing overviews of reviews","year":2010,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":113,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba; Children's Hospital Research Institute of Manitoba; Cochrane; University of Alberta","funders":"","keywords":"Systematic review; Presentation (obstetrics); Process (computing); Mandate; Identification (biology); Computer science; Engineering ethics; Management science; MEDLINE; Psychology; Medicine; Political science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9268161503390395,"gpt":0.6748823847148678,"spread":0.2519337656241717,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5907845,0.0001440321,0.002469427,0.0005475733,0.0001289617,0.0001337557,0.001656332,0.00009668593,0.0008871366],"category_scores_gemma":[0.6118124,0.00005970843,0.0005807243,0.00222803,0.0002909641,0.0002006456,0.0002552701,0.0002674312,0.00005232141],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002468199,"about_ca_system_score_gemma":0.0002693799,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001340835,"about_ca_topic_score_gemma":0.0001174971,"domain_scores_codex":[0.8318352,0.1403605,0.01557249,0.001721951,0.009862314,0.0006475626],"domain_scores_gemma":[0.8188448,0.1390224,0.01378975,0.01481449,0.01308571,0.0004428282],"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":[0.0000100185,0.00003320662,0.0007741029,0.0004354979,0.00006357014,2.683504e-8,0.0006687202,4.406125e-7,0.03094666,0.01837995,0.006172444,0.9425154],"study_design_scores_gemma":[0.00009440154,0.0001119681,0.03991642,0.0003952586,0.0001816607,0.000004319706,0.002166391,0.001690601,0.02884678,0.03375175,0.8926908,0.0001495863],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.09160068,0.7756852,0.06307229,0.0223987,0.0007595413,0.007832166,0.00001994039,0.00001110678,0.03862041],"genre_scores_gemma":[0.2831109,0.05940237,0.6483831,0.00001246668,0.0002500232,0.0002295912,9.214424e-7,0.00003284948,0.008577714],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9423658,"threshold_uncertainty_score":0.9713523,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2103418324","doi":"10.1002/jrsm.53","title":"Comparison of statistical inferences from the DerSimonian–Laird and alternative random‐effects model meta‐analyses – an empirical assessment of 920 Cochrane primary outcome meta‐analyses","year":2011,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":80,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University","funders":"","keywords":"Estimator; Meta-analysis; Statistics; Random effects model; Econometrics; Mathematics; Confidence interval; Medicine; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9790827001812016,"gpt":0.785766586948887,"spread":0.1933161132323146,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2801977,0.0005748076,0.01366884,0.0008784692,0.0003237868,0.0004231505,0.003236999,0.0001732645,0.005224538],"category_scores_gemma":[0.1510227,0.0002192023,0.003088841,0.00176397,0.001237905,0.000417256,0.0006354144,0.0006246013,0.00002318701],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004906331,"about_ca_system_score_gemma":0.0003444335,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001409988,"about_ca_topic_score_gemma":0.0001902886,"domain_scores_codex":[0.7734665,0.203409,0.00959047,0.001662956,0.01122764,0.0006435044],"domain_scores_gemma":[0.6498152,0.3390498,0.004116419,0.003838602,0.002709523,0.0004704701],"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":[0.0009801857,0.004117875,0.4355614,0.002105484,0.36963,0.00004743033,0.01470058,0.006743757,0.0213144,0.01251406,0.004922932,0.1273618],"study_design_scores_gemma":[0.0006715838,0.0004278451,0.1116914,0.0001023264,0.09807512,0.00000243633,0.004273724,0.6934372,0.03062478,0.06006447,0.0001331744,0.0004959439],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04502687,0.0121773,0.9383519,0.0003600207,0.00005348774,0.001090191,0.000185167,0.000007171045,0.002747858],"genre_scores_gemma":[0.5398519,0.0001988748,0.4596391,0.00003908978,0.00001826087,0.00016441,0.000004698062,0.00001406901,0.00006959038],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6866935,"threshold_uncertainty_score":0.9956848,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2099087018","doi":"10.1002/jrsm.42","title":"Precision of healthcare systematic review searches in a cross‐sectional sample","year":2011,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":79,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Public Health; Children's Hospital of Eastern Ontario","funders":"U.S. National Library of Medicine; Health Canada; National Institutes of Health","keywords":"Systematic review; MEDLINE; Interquartile range; Sample size determination; Health care; Sample (material); Medicine; Range (aeronautics); Information retrieval; Computer science; Statistics; Data mining; Mathematics; Physics","retraction":null,"screen_n_in":null,"score":{"opus":0.953369109962058,"gpt":0.7176757588278021,"spread":0.2356933511342559,"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":[0.7471939,0.0003079686,0.006581648,0.00183144,0.0002139379,0.0003362924,0.003712993,0.0001824475,0.01145634],"category_scores_gemma":[0.748931,0.0001526535,0.001891893,0.005143439,0.0003014078,0.0003158656,0.0005015144,0.0005825942,0.0009320803],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001727839,"about_ca_system_score_gemma":0.0003533957,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007046814,"about_ca_topic_score_gemma":0.00008010812,"domain_scores_codex":[0.5662138,0.3872279,0.0228286,0.002270642,0.0202896,0.001169455],"domain_scores_gemma":[0.5177193,0.4475802,0.008016078,0.01396046,0.01183998,0.0008840291],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.000152104,0.0008284386,0.2979937,0.5940781,0.0008182451,0.00002633348,0.003157735,0.00001512777,0.000650347,0.01784584,0.004717578,0.07971649],"study_design_scores_gemma":[0.0005605189,0.0006133827,0.4862969,0.22779,0.0007234258,0.0001079385,0.004066612,0.009521063,0.01092979,0.2517762,0.006231172,0.001383048],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07352111,0.3254367,0.5085742,0.004636658,0.001052696,0.03621718,0.0004398282,0.00007070923,0.05005089],"genre_scores_gemma":[0.317598,0.004771529,0.6716025,0.0001481156,0.00007783992,0.002107471,0.000004818984,0.0000583061,0.003631331],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3662881,"threshold_uncertainty_score":0.9998458,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2986493832","doi":"10.1002/jrsm.1385","title":"Construct validity of the Physiotherapy Evidence Database (PEDro) quality scale for randomized trials: Item response theory and factor analyses","year":2019,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":77,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research; Alberta Innovates; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; Physiotherapy Foundation of Canada; National Science Foundation","keywords":"Item response theory; Construct (python library); Scale (ratio); Construct validity; Psychology; Randomized controlled trial; Clinical psychology; Quality (philosophy); Psychometrics; Physical therapy; Applied psychology; Medicine; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.942994865986386,"gpt":0.7599710097067921,"spread":0.1830238562795939,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7405687,0.0002566804,0.002989913,0.0009417115,0.0004062559,0.0002911428,0.001996845,0.0001596701,0.000805069],"category_scores_gemma":[0.9807892,0.0001237599,0.001019746,0.003152057,0.001579792,0.0003849678,0.0005942797,0.0003985416,0.000006612944],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005941281,"about_ca_system_score_gemma":0.0003893722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001437059,"about_ca_topic_score_gemma":0.000003159986,"domain_scores_codex":[0.5068032,0.485721,0.002530097,0.001189422,0.00308967,0.0006666644],"domain_scores_gemma":[0.03266813,0.9621036,0.001398967,0.002354417,0.001289653,0.0001852136],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.1208217,0.00006624289,0.0103404,0.0001318875,0.0001741961,3.201077e-7,0.0003833901,0.000006079058,0.5309297,0.001046665,0.000225234,0.3358742],"study_design_scores_gemma":[0.01838639,0.0002324557,0.07951102,0.0006406538,0.0001517783,0.000005766925,0.00412186,0.002741606,0.7144586,0.1782992,0.001025462,0.0004251763],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.731636,0.002011356,0.2629606,0.0005489071,0.0003670411,0.002115655,0.0001817801,0.00002287614,0.0001557123],"genre_scores_gemma":[0.5596728,0.0005860564,0.4388112,0.00004623538,0.00009025429,0.0002054351,3.895948e-7,0.00002254929,0.0005651279],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4763826,"threshold_uncertainty_score":0.881494,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2947538236","doi":"10.1002/jrsm.1362","title":"Estimating hazard ratios from published Kaplan‐Meier survival curves: A methods validation study","year":2019,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":77,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Canadian Centre for Applied Research in Cancer Control; Public Health Ontario; University of Toronto; Sunnybrook Health Science Centre","funders":"Canadian Centre for Applied Research in Cancer Control","keywords":"Intraclass correlation; Inter-rater reliability; Hazard ratio; Statistics; Reliability (semiconductor); Medicine; Standard error; Gold standard (test); Mathematics; Confidence interval; Reproducibility; Physics","retraction":null,"screen_n_in":null,"score":{"opus":0.7748885543966163,"gpt":0.7029467070522295,"spread":0.0719418473443868,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2350431,0.0005908317,0.002595952,0.0005898555,0.0004047987,0.0006724669,0.001762715,0.0004808304,0.009094314],"category_scores_gemma":[0.820622,0.0004926352,0.000418119,0.001770923,0.0003356333,0.0006021076,0.0009522669,0.001952268,0.0004112364],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003046463,"about_ca_system_score_gemma":0.0004144835,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002512846,"about_ca_topic_score_gemma":0.00001288673,"domain_scores_codex":[0.8287073,0.1605445,0.003183572,0.002232186,0.00370144,0.001631002],"domain_scores_gemma":[0.2676331,0.7263542,0.0007851159,0.003192709,0.001410404,0.0006244665],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001046052,0.004205259,0.01348697,0.002670925,0.0025563,0.00006955235,0.002034067,0.00007493453,0.0309939,0.02377706,0.01486504,0.9042199],"study_design_scores_gemma":[0.001478462,0.0005104759,0.00274895,0.001156267,0.0004753343,0.000003266934,0.001387281,0.01835698,0.04324121,0.9289679,0.0008357418,0.0008381325],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02607132,0.0001558963,0.9575539,0.001122579,0.002309502,0.004341472,0.0001293474,0.0003397504,0.007976192],"genre_scores_gemma":[0.003946876,0.00003267494,0.9931335,0.00008260553,0.0008219396,0.0009692857,0.00001068047,0.0001786573,0.0008238207],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9051908,"threshold_uncertainty_score":0.9997525,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3034095928","doi":"10.1002/jrsm.1428","title":"The Kilim plot: A tool for visualizing network meta‐analysis results for multiple outcomes","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":65,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Michael's Hospital","funders":"Public Health Research Programme; European Commission; National Institute for Health and Care Research; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Plot (graphics); Computer science; Meta-analysis; Forest plot; Data science; Statistics; Medicine; Mathematics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.7463800401262789,"gpt":0.6693628014414853,"spread":0.0770172386847936,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05643,0.0003375074,0.001478116,0.0003167975,0.001847335,0.0002550074,0.00138221,0.0002589749,0.0002739415],"category_scores_gemma":[0.07218026,0.0002209103,0.001699304,0.002062269,0.0003370135,0.0001117011,0.0004099348,0.0007203965,0.00008067601],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001678817,"about_ca_system_score_gemma":0.0002003412,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002465258,"about_ca_topic_score_gemma":0.000151576,"domain_scores_codex":[0.9803352,0.01368902,0.001149878,0.001192538,0.001174594,0.002458795],"domain_scores_gemma":[0.8177181,0.1797638,0.000209104,0.001205276,0.0005332728,0.0005703549],"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":[0.01779223,0.0009719522,0.00293445,0.001541754,0.2662867,0.00003085438,0.006148089,0.0005871455,0.0009669617,0.06423541,0.1466494,0.4918551],"study_design_scores_gemma":[0.002917933,0.001136471,0.003477475,0.00002447598,0.01924172,0.000001715203,0.002845514,0.04588406,0.004687584,0.005860216,0.9132316,0.0006912333],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002777454,0.01500987,0.8335986,0.1223326,0.001044939,0.01762511,0.001487465,0.0004802353,0.005643803],"genre_scores_gemma":[0.05603631,0.000490897,0.8993071,0.002029486,0.001794874,0.02929924,0.00006914876,0.0002601031,0.01071284],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7665823,"threshold_uncertainty_score":0.9994521,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3043051650","doi":"10.1002/jrsm.1435","title":"Creating effective interrupted time series graphs: Review and recommendations","year":2020,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":63,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Health and Medical Research Council; Canadian Institutes of Health Research; Monash University; Medical Research Council; Australian Government","keywords":"Computer science; Standardization; Visualization; Data mining; Data visualization; Graph; Software; Data extraction; Graph drawing; Information retrieval; Time series; Data science; Machine learning; Theoretical computer science; MEDLINE","retraction":null,"screen_n_in":null,"score":{"opus":0.214636014942542,"gpt":0.5644807030011398,"spread":0.3498446880585978,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0112712,0.0003820956,0.002104562,0.0006317071,0.0003969291,0.0005545783,0.001557613,0.0001708537,0.0002441503],"category_scores_gemma":[0.02160358,0.0003028113,0.0003174556,0.003074524,0.0001927933,0.0005742449,0.001554927,0.0007796806,0.0002150101],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001269654,"about_ca_system_score_gemma":0.0002923858,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007849588,"about_ca_topic_score_gemma":7.220058e-7,"domain_scores_codex":[0.9816713,0.01579097,0.0007344178,0.0008985909,0.0004571175,0.0004475556],"domain_scores_gemma":[0.9840702,0.01404099,0.0003133519,0.0009383739,0.0003246589,0.000312353],"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":[4.71868e-7,0.00001942558,7.758309e-8,0.01856142,0.0001784266,0.000006048081,0.00005676532,9.998492e-9,0.000001031379,0.007066843,0.006205333,0.9679042],"study_design_scores_gemma":[0.00002067349,0.00005033328,2.353411e-7,0.0352939,0.0003690854,0.00003229788,0.00001422589,0.0007368799,0.00001637691,0.0006256921,0.9625717,0.0002685461],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[1.03866e-9,0.8476304,0.1472682,0.001550925,0.0000476415,0.001016664,0.00006401695,0.0001352351,0.002286819],"genre_scores_gemma":[6.824384e-9,0.7557792,0.2430712,0.0001111661,0.00003686975,0.0003948928,0.0000452267,0.00003245829,0.0005289702],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9676356,"threshold_uncertainty_score":0.9999424,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2903321323","doi":"10.1002/jrsm.1333","title":"The use of mathematical modeling studies for evidence synthesis and guideline development: A glossary","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":61,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; The Quebec Population Health Research Network","funders":"National Health and Medical Research Council; Victorian Centre for Biostatistics; TDR; World Health Organization","keywords":"Terminology; Glossary; Management science; Computer science; Guideline; Systematic review; Data science; MEDLINE; Medicine; Engineering; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9114025315052573,"gpt":0.6791936628228623,"spread":0.232208868682395,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05389336,0.0002867693,0.001096902,0.0001752102,0.001328932,0.00008634782,0.0006112726,0.0001504766,0.00005694858],"category_scores_gemma":[0.7502432,0.0001611179,0.0001709389,0.0004361279,0.001682008,0.000186137,0.001150459,0.0002604228,0.00001408015],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001904166,"about_ca_system_score_gemma":0.0001699489,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002661491,"about_ca_topic_score_gemma":0.00002406969,"domain_scores_codex":[0.9911289,0.005081623,0.001489008,0.000628324,0.0008009139,0.0008712321],"domain_scores_gemma":[0.4861129,0.5107316,0.0002473675,0.0007910358,0.00196442,0.0001526574],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001019977,0.0002959301,0.0001877439,0.003648776,0.001864731,0.000004666696,0.002923702,0.00006453782,0.004256532,0.04624926,0.02084476,0.9186394],"study_design_scores_gemma":[0.0001699349,0.0002919414,0.00007879877,0.002816801,0.0003093596,0.000007790657,0.002126428,0.207376,0.04906663,0.6857678,0.05147761,0.0005108565],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02290439,0.007183475,0.9603471,0.007676865,0.00005750484,0.001547758,0.00001333776,0.000105284,0.0001642743],"genre_scores_gemma":[0.01804933,0.002203209,0.9780511,0.00006974242,0.0001468654,0.001118823,1.247947e-7,0.00004070321,0.0003201197],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9181286,"threshold_uncertainty_score":0.9999712,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3021470514","doi":"10.1002/jrsm.1511","title":"Methods for population adjustment with limited access to individual patient data: A review and simulation study","year":2021,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":57,"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; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Engineering and Physical Sciences Research Council; Canadian Institutes of Health Research; Institute of Conservation","keywords":"Covariate; Estimator; Standard error; Sample size determination; Measure (data warehouse); Variance (accounting); Population; Standard deviation; Confidence interval","retraction":null,"screen_n_in":null,"score":{"opus":0.7644489907710884,"gpt":0.7005175490144144,"spread":0.06393144175667398,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01786685,0.0002286703,0.0006958813,0.0002957999,0.0002169293,0.000173472,0.000562463,0.00008696537,0.00005528243],"category_scores_gemma":[0.07395944,0.0001796833,0.00003907243,0.001047792,0.00004725144,0.0005721345,0.00124885,0.0002750035,8.146206e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001392381,"about_ca_system_score_gemma":0.0001126818,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003154348,"about_ca_topic_score_gemma":0.00003793998,"domain_scores_codex":[0.9891867,0.008292384,0.0005667917,0.0008230278,0.0006777758,0.0004533428],"domain_scores_gemma":[0.958526,0.03856514,0.0001896664,0.001514981,0.000996898,0.0002073333],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00007345837,0.0004687055,0.0003463922,0.001053946,0.0001791821,0.000003736102,0.0003195074,0.00006806387,0.0003784565,0.0007776645,0.00048341,0.9958475],"study_design_scores_gemma":[0.003421431,0.01107391,0.0623777,0.01735162,0.008701168,0.00007088867,0.008875056,0.071381,0.208702,0.5288903,0.07410315,0.005051767],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005013241,0.00261491,0.986136,0.0005955375,0.00002791846,0.005316366,0.00007084741,0.0001402497,0.00008499674],"genre_scores_gemma":[0.009302947,0.0004960551,0.9877325,0.0001703554,0.00003925811,0.002116363,0.00005186282,0.00005625696,0.00003441615],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9907957,"threshold_uncertainty_score":0.933841,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389569015","doi":"10.1002/jrsm.1686","title":"<scp>metamedian</scp>: An R package for meta‐analyzing studies reporting medians","year":2023,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":55,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University Health Centre; McGill University","funders":"National Science Foundation","keywords":"Meta-analysis; Computer science; Sample size determination; Median; Software; Sample (material); Outcome (game theory); Statistics; Data science; Medicine; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.9897344879680611,"gpt":0.7931787116963884,"spread":0.1965557762716726,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.9003953,0.001865024,0.05576103,0.007191729,0.001434771,0.003742462,0.009950289,0.000961508,0.001675413],"category_scores_gemma":[0.9716151,0.0008502941,0.02790593,0.01420319,0.0006958309,0.000676867,0.00165691,0.00201484,0.003800762],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003792137,"about_ca_system_score_gemma":0.001397017,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002983299,"about_ca_topic_score_gemma":0.00008885071,"domain_scores_codex":[0.3886073,0.4965281,0.0739297,0.00885993,0.02806515,0.00400988],"domain_scores_gemma":[0.02764321,0.8890222,0.05132806,0.01951002,0.01079622,0.001700253],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[7.563701e-7,0.0000525189,0.00000525688,0.01954807,0.03359486,0.00007518507,0.0005653069,0.000001922285,0.000005790826,0.0004521368,0.04074061,0.9049576],"study_design_scores_gemma":[0.00003718695,0.0000491608,0.000002001795,0.003876114,0.03966049,0.0000340469,0.00402151,0.0002786218,0.00005437999,0.01109465,0.9405544,0.00033743],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000003088568,0.9278075,0.06374075,0.0002242207,0.0008085056,0.005500241,0.0002522088,0.00007302035,0.001590461],"genre_scores_gemma":[3.96717e-7,0.7115341,0.258931,0.00001516131,0.0008632597,0.01033103,0.00003339659,0.0002407413,0.01805093],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9046202,"threshold_uncertainty_score":0.9998652,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3177444756","doi":"10.1002/jrsm.1508","title":"Including <scp>non‐English</scp> language articles in systematic reviews: A reflection on processes for identifying low‐cost sources of translation support","year":2021,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Systematic review; Computer science; English language; Reflection (computer programming); Process (computing); Social media; Psychology; MEDLINE; Political science; World Wide Web; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.9211131536084689,"gpt":0.6872420127582198,"spread":0.2338711408502491,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4478641,0.0002966787,0.004411368,0.001620314,0.0002764722,0.0008832807,0.001149221,0.0001508106,0.0003709069],"category_scores_gemma":[0.7567448,0.0001701239,0.001166448,0.006246998,0.00008594299,0.0004832016,0.0001256228,0.0003041364,0.0001285025],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001180202,"about_ca_system_score_gemma":0.000277941,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001558373,"about_ca_topic_score_gemma":0.0001212053,"domain_scores_codex":[0.8992084,0.0834851,0.009864047,0.001386338,0.005342976,0.000713204],"domain_scores_gemma":[0.6790939,0.3055851,0.004999263,0.003837428,0.006185341,0.0002989763],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0000626107,0.0009179571,0.007681151,0.2352684,0.000801906,0.00004364073,0.07923243,0.0005451592,0.1370802,0.0008121732,0.004839963,0.5327144],"study_design_scores_gemma":[0.0006293613,0.0001952529,0.001824449,0.03269178,0.0007931447,0.00002301687,0.08785149,0.01816191,0.8335322,0.009590907,0.01433083,0.0003756819],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3754289,0.02541586,0.5758221,0.0004637672,0.0004162335,0.01227049,0.00004319509,0.00003447706,0.01010498],"genre_scores_gemma":[0.8084993,0.0009000794,0.1841465,0.00004564717,0.0001950737,0.00327451,0.000008064577,0.00005523681,0.00287559],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.696452,"threshold_uncertainty_score":0.8517497,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2338480338","doi":"10.1002/jrsm.1204","title":"Disconnected by design: analytic approach in treatment networks having no common comparator","year":2016,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; University of British Columbia","funders":"National Institutes of Health","keywords":"Computer science; Bayesian network; Risk analysis (engineering); Artificial intelligence; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.834068243143668,"gpt":0.6795690327267213,"spread":0.1544992104169467,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05284591,0.0004041538,0.001813638,0.0003616477,0.0002049514,0.000116864,0.0007774948,0.0003657561,0.0008250054],"category_scores_gemma":[0.2707687,0.0002414487,0.0002467396,0.001023239,0.0005987415,0.0001184425,0.0002498975,0.0005630326,0.0000878776],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007367457,"about_ca_system_score_gemma":0.0001358699,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009044328,"about_ca_topic_score_gemma":0.000008060801,"domain_scores_codex":[0.9428677,0.0525642,0.001419192,0.000987526,0.0008681954,0.001293183],"domain_scores_gemma":[0.5508742,0.4473704,0.0001883407,0.001013272,0.0001965453,0.0003571327],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001745464,0.004148663,0.002992484,0.0002880438,0.0007775114,0.0000742823,0.0002294546,0.0001044039,0.02612119,0.02931079,0.0211002,0.9131075],"study_design_scores_gemma":[0.004080757,0.001352174,0.001516294,0.001431206,0.0003727974,0.00001411128,0.0002824424,0.1463794,0.08137654,0.7592349,0.00253566,0.00142372],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002860783,0.0002192436,0.9925253,0.0004056896,0.0001720229,0.001417319,0.00006064711,0.0001027461,0.002236287],"genre_scores_gemma":[0.04237211,0.0002296723,0.9555378,0.00001709705,0.0002213519,0.0008253503,0.000001462229,0.00008771884,0.0007074263],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9116838,"threshold_uncertainty_score":0.9845991,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2762597371","doi":"10.1002/jrsm.1272","title":"A scoping approach to systematically review published reviews: Adaptations and recommendations","year":2017,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Manitoba Health; University of Manitoba","funders":"Manitoba Health Research Council; Heart and Stroke Foundation of Canada","keywords":"Systematic review; Multidisciplinary approach; Health care; Publication; MEDLINE; Engineering ethics; Medical education; Medicine; Psychology; Political science; Sociology; Social science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9562756172992223,"gpt":0.732530525072342,"spread":0.2237450922268803,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7192966,0.0003563408,0.006118773,0.001003243,0.002015699,0.007505185,0.005244562,0.0001132847,0.005225578],"category_scores_gemma":[0.8720006,0.0001798281,0.001175786,0.002005113,0.0002385967,0.001033763,0.001064591,0.0004550254,0.002368595],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008003748,"about_ca_system_score_gemma":0.0002135782,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000289649,"about_ca_topic_score_gemma":0.00001796544,"domain_scores_codex":[0.729012,0.2399987,0.01683042,0.003037333,0.01000926,0.001112313],"domain_scores_gemma":[0.7833737,0.1655465,0.01032405,0.029768,0.009041227,0.001946522],"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":[0.000003277916,0.0001075633,0.0002164032,0.01094296,0.0002039551,0.000001385355,0.0003549346,0.000002503083,0.0001324344,0.01798958,0.2868491,0.6831959],"study_design_scores_gemma":[0.000174192,0.0000713302,0.003510709,0.07739039,0.0008066717,0.00005072618,0.001691535,0.01262844,0.000112603,0.01892765,0.8838065,0.0008292421],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00004986275,0.05130843,0.7455748,0.04316098,0.00020108,0.008821954,0.00002041771,0.00001507111,0.1508474],"genre_scores_gemma":[0.001659166,0.01262321,0.9737823,0.0005110328,0.0001121988,0.003603555,0.000003582355,0.00002812147,0.007676856],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6823667,"threshold_uncertainty_score":0.9992836,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1984170021","doi":"10.1002/jrsm.1062","title":"Issues relating to selective reporting when including non‐randomized studies in systematic reviews on the effects of healthcare interventions","year":2012,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":49,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Global Health Research; Institute of Population and Public Health; Ottawa Hospital","funders":"","keywords":"Psychological intervention; Randomized controlled trial; Systematic review; Documentation; Protocol (science); Suspect; MEDLINE; Health care; Psychology; Meta-analysis; Reporting bias; Medicine; Alternative medicine; Computer science; Psychiatry; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9651183188967614,"gpt":0.7655784742142648,"spread":0.1995398446824965,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8891034,0.0003548015,0.01216386,0.001653035,0.0005038368,0.0002475299,0.00161904,0.0001047488,0.0002364372],"category_scores_gemma":[0.9796731,0.0001367801,0.003253006,0.004046191,0.0002487223,0.0002676826,0.0006441164,0.0006572357,0.0003369169],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002995042,"about_ca_system_score_gemma":0.00006395003,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001956488,"about_ca_topic_score_gemma":0.00005285902,"domain_scores_codex":[0.2690344,0.6835444,0.0364427,0.001366991,0.00842151,0.00118999],"domain_scores_gemma":[0.03135087,0.9331446,0.02498174,0.005796256,0.004325646,0.0004008249],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.001228051,0.0009399112,0.01636905,0.4410857,0.006923055,0.00003113123,0.3023561,0.00008877073,0.008137367,0.03503939,0.04362822,0.1441733],"study_design_scores_gemma":[0.005846622,0.0007697777,0.003598296,0.6575893,0.003512907,0.00005571786,0.1624149,0.009298643,0.04863219,0.1008465,0.005801431,0.001633669],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.216105,0.4585047,0.1839439,0.02588664,0.002494515,0.09447113,0.000008903015,0.00005208608,0.01853313],"genre_scores_gemma":[0.787052,0.001077919,0.1994367,0.000109621,0.0001589944,0.007870462,3.01659e-7,0.00003592061,0.004258104],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5709469,"threshold_uncertainty_score":0.8314971,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2736946457","doi":"10.1002/jrsm.1256","title":"Network meta‐analysis of disconnected networks: How dangerous are random baseline treatment effects?","year":2017,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University Health Centre; McGill University; University of British Columbia","funders":"Fonds de recherche du Québec – Nature et technologies; Natural Sciences and Engineering Research Council of Canada; McGill University","keywords":"Baseline (sea); Meta-analysis; Computer science; Medicine; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9055690218436446,"gpt":0.6569411513546755,"spread":0.2486278704889692,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.5441519,0.0006437598,0.01728943,0.001644328,0.001338347,0.002936959,0.005124378,0.0002404804,0.00699168],"category_scores_gemma":[0.4627878,0.0002794506,0.01212314,0.004322957,0.0005205776,0.0003947749,0.000615641,0.0003667765,0.0001789629],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001056882,"about_ca_system_score_gemma":0.0001033793,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002198982,"about_ca_topic_score_gemma":0.0003981694,"domain_scores_codex":[0.7013022,0.2804031,0.006484791,0.002235432,0.008376493,0.001197914],"domain_scores_gemma":[0.618771,0.3531641,0.008482932,0.01587172,0.003141182,0.0005690159],"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":[0.0002638513,0.0004681837,0.01034801,0.0001853785,0.4487017,0.00007363408,0.0002757316,0.0176035,0.0002929391,0.0009244769,0.0170183,0.5038443],"study_design_scores_gemma":[0.001013231,0.0002524013,0.0172226,0.0001224632,0.4551851,0.000004961716,0.0006869106,0.4638207,0.002316681,0.004491922,0.05410517,0.0007778883],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.003132869,0.02552184,0.960892,0.004376712,0.0002994433,0.002378485,0.00007899808,0.00001714806,0.003302512],"genre_scores_gemma":[0.6973372,0.001006331,0.2873261,0.00006700389,0.0005551212,0.001494065,0.00001588361,0.00006505266,0.01213335],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6942043,"threshold_uncertainty_score":0.9999658,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2413385969","doi":"10.1002/jrsm.1215","title":"Implications of applying methodological shortcuts to expedite systematic reviews: three case studies using systematic reviews from agri‐food public health","year":2016,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Guelph; Public Health Agency of Canada","funders":"Public Health Agency; Public Health Agency of Canada","keywords":"Rigour; Systematic review; Meta-analysis; Publication bias; Systematic error; Public health; Computer science; MEDLINE; Management science; Statistics; Econometrics; Psychology; Data science; Medicine; Biology; Mathematics; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.9919902489356272,"gpt":0.7661392033767591,"spread":0.225851045558868,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8271849,0.0008735635,0.0251936,0.002236255,0.0009196018,0.0008413715,0.004676858,0.0002535097,0.001392911],"category_scores_gemma":[0.8933468,0.0003277117,0.003879345,0.006630159,0.0003954131,0.0005853942,0.001289499,0.000444088,0.001251038],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007255154,"about_ca_system_score_gemma":0.000353645,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006369554,"about_ca_topic_score_gemma":0.000127163,"domain_scores_codex":[0.2286185,0.7126408,0.04225234,0.003773827,0.01067429,0.002040281],"domain_scores_gemma":[0.1327342,0.8108746,0.02441771,0.02065108,0.009163948,0.002158424],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00002160109,0.0004512161,0.0009994617,0.4339845,0.004879351,0.00006839511,0.005715106,0.00001332204,0.01692049,0.007234041,0.01736991,0.5123426],"study_design_scores_gemma":[0.001069064,0.001613193,0.001151232,0.6915168,0.00912407,0.003078941,0.06494964,0.006324081,0.004369441,0.1263944,0.08538444,0.005024666],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004204989,0.1440261,0.8250104,0.003042757,0.0002948268,0.0230941,0.00008803937,0.00002347731,0.0002153413],"genre_scores_gemma":[0.03417102,0.008084325,0.9368327,0.0002581608,0.000298745,0.01986426,0.000001800046,0.00007025294,0.00041869],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5073179,"threshold_uncertainty_score":0.9999175,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4390519786","doi":"10.1002/jrsm.1694","title":"Advancing the methodology of mapping reviews: A scoping review","year":2024,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Diverse Aspects of Tourism Research","field":"Social Sciences","cited_by":42,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Work & Health; Public Health Ontario; Queen's University; University of Toronto; St. Michael's Hospital; Bruyère; University of Ottawa","funders":"National Health and Medical Research Council; La Trobe University","keywords":"PsycINFO; CINAHL; MEDLINE; Citation; Protocol (science); Computer science; Test (biology); Systematic review; Data science; Cochrane Library; Medical education; Psychology; Medicine; Library science; Alternative medicine; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.7551232860667084,"gpt":0.7100772784172794,"spread":0.04504600764942901,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.3476868,0.000604824,0.006122815,0.001671213,0.00117997,0.0002878024,0.00455797,0.0005954585,0.001685995],"category_scores_gemma":[0.2763869,0.0003688913,0.002150081,0.006340774,0.002317752,0.0002668636,0.00205605,0.003264769,0.0009122951],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007841738,"about_ca_system_score_gemma":0.005753182,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001190718,"about_ca_topic_score_gemma":0.0001497222,"domain_scores_codex":[0.8194445,0.1719619,0.002333053,0.001270411,0.002858848,0.002131252],"domain_scores_gemma":[0.8475251,0.1474396,0.001082847,0.002457244,0.0009464053,0.000548811],"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":[0.000001197742,0.00001689544,3.132521e-8,0.2778737,0.0001788184,0.00004224663,0.0003455611,2.790358e-8,0.000001620209,0.002235595,0.01568125,0.7036231],"study_design_scores_gemma":[0.000008697224,0.00001194087,1.350134e-8,0.4413513,0.0005623696,0.000005460649,0.0005728994,8.357529e-7,0.000005798998,0.0009843863,0.5563163,0.0001800436],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[8.493368e-9,0.9545158,0.002113434,0.002873545,0.0005159379,0.007940588,0.0000155647,0.00008969638,0.03193545],"genre_scores_gemma":[9.850967e-9,0.8428414,0.1520174,0.00005699606,0.0007393912,0.00204515,0.000002170574,0.0001034497,0.002194011],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7034431,"threshold_uncertainty_score":0.9998833,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4386617069","doi":"10.1002/jrsm.1664","title":"What are the best methods for rapid reviews of the research evidence? A systematic review of reviews and primary studies","year":2023,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","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":"McMaster University; Impact","funders":"Pan American Health Organization","keywords":"Data extraction; Systematic review; Credibility; Protocol (science); Computer science; Data science; Publication bias; Information retrieval; MEDLINE; Management science; Meta-analysis; Medicine; Alternative medicine; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9855286116620706,"gpt":0.7959365036411353,"spread":0.1895921080209353,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","research_integrity"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.9547979,0.001492538,0.06536757,0.002657196,0.001258067,0.001451768,0.01657104,0.0005838572,0.0005913181],"category_scores_gemma":[0.9599082,0.0004666963,0.01762309,0.02003032,0.002650209,0.0006901561,0.004837646,0.002408522,0.0006555366],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004532452,"about_ca_system_score_gemma":0.001283534,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001295436,"about_ca_topic_score_gemma":0.00001030043,"domain_scores_codex":[0.08494691,0.8458056,0.04762109,0.003642914,0.01635502,0.001628404],"domain_scores_gemma":[0.02157741,0.8932293,0.04089922,0.02676463,0.01705161,0.0004778178],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000002540107,0.00002816216,4.995034e-7,0.51428,0.001009733,5.84971e-7,0.0001937577,1.84235e-8,0.000002862868,0.0001319381,0.02098247,0.4633675],"study_design_scores_gemma":[0.0000185757,0.00004805522,4.627447e-7,0.5049701,0.006215154,0.00001375256,0.001527735,0.00001205706,0.000008974987,0.001453555,0.4855326,0.0001989919],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[7.953266e-8,0.939567,0.00297075,0.002081983,0.0006804203,0.05439162,0.00005509893,0.000006995871,0.0002460933],"genre_scores_gemma":[1.708062e-8,0.8635753,0.1009919,0.0001217593,0.0001950316,0.02753018,0.000002690396,0.0001176821,0.007465465],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4645502,"threshold_uncertainty_score":0.999893,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2744173239","doi":"10.1002/jrsm.1253","title":"Bayesian one‐step IPD network meta‐analysis of time‐to‐event data using Royston‐Parmar models","year":2017,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":40,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council; Universidad de Buenos Aires; Chang Gung Medical Foundation; University of Alberta; Sanjay Gandhi Postgraduate Institute of Medical Sciences; Universitair Medisch Centrum Groningen; Chiang Mai University; Rijksuniversiteit Groningen; Yale University","keywords":"Computer science; Bayesian probability; Consistency (knowledge bases); Proportional hazards model; Meta-analysis; Event (particle physics); Bayesian network; Data mining; Machine learning; Statistics; Artificial intelligence; Medicine; Mathematics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9789937325284529,"gpt":0.7322230783960013,"spread":0.2467706541324516,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.6780025,0.0005364255,0.01294745,0.002441473,0.001383807,0.002564951,0.01758857,0.0002261679,0.0406108],"category_scores_gemma":[0.2480809,0.0002977949,0.006128672,0.005797375,0.0003919226,0.001065348,0.004580316,0.0004820934,0.00132042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001068671,"about_ca_system_score_gemma":0.0003187727,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005825753,"about_ca_topic_score_gemma":0.0001427115,"domain_scores_codex":[0.7812876,0.1826283,0.01240776,0.003772321,0.01831034,0.00159372],"domain_scores_gemma":[0.8543094,0.08962889,0.007720912,0.04317446,0.004116447,0.001049914],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00009755205,0.0004925041,0.0008876511,0.0002741671,0.3943652,0.00002356353,0.0004197387,0.2559088,0.003084031,0.002671603,0.052739,0.2890362],"study_design_scores_gemma":[0.00004334441,0.00001976876,0.0003341165,0.00004646307,0.08242112,0.000001528252,0.0000981487,0.89238,0.0004618503,0.005637541,0.01828586,0.0002702476],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0005975945,0.003410377,0.9781675,0.001537675,0.00009147598,0.001271221,0.0002036882,0.000008680513,0.01471175],"genre_scores_gemma":[0.05532733,0.0001172269,0.9359468,0.00006470803,0.0001886185,0.0001338312,0.0000121697,0.00005170696,0.008157576],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6364713,"threshold_uncertainty_score":0.9999474,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2899927427","doi":"10.1002/jrsm.1330","title":"The impact of the peer review of literature search strategies in support of rapid review reports","year":2018,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"scientometrics and bibliometrics research","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":"Canadian Agency for Drugs and Technologies in Health","funders":"","keywords":"Subject (documents); Peer review; Information retrieval; Scrutiny; Inclusion (mineral); Computer science; Technical peer review; Psychology; World Wide Web; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.7743918792844233,"gpt":0.7358830040339657,"spread":0.03850887525045765,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.5225989,0.0002098971,0.001212379,0.01480717,0.0002908878,0.0005684723,0.005082577,0.0001630131,0.001982383],"category_scores_gemma":[0.5979723,0.00008581371,0.0008382134,0.1968938,0.002228286,0.0004662081,0.001710662,0.0009561941,0.00001556898],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001535608,"about_ca_system_score_gemma":0.002705858,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003665208,"about_ca_topic_score_gemma":0.00001305939,"domain_scores_codex":[0.919664,0.03370484,0.004512622,0.00131411,0.03903083,0.001773564],"domain_scores_gemma":[0.6658288,0.2086418,0.003033215,0.01158534,0.1099804,0.0009304453],"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":[0.00007561438,0.0002813248,0.008383948,0.004880423,0.0001027605,0.00002279088,0.0003482685,0.000003130233,0.009546359,0.0007825237,0.2371708,0.738402],"study_design_scores_gemma":[0.0004444631,0.002801077,0.2677824,0.05489589,0.00008694873,0.0002044749,0.001811576,0.0006359136,0.2551427,0.03960421,0.3758528,0.0007375205],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.03885538,0.8448905,0.001245129,0.02777822,0.0006615606,0.004764939,0.0002709008,0.00001730622,0.08151606],"genre_scores_gemma":[0.3008852,0.6802139,0.01340919,0.0001371413,0.000150242,0.0001401581,0.000004163755,0.00004430635,0.005015754],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7376645,"threshold_uncertainty_score":0.9989299,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3030872057","doi":"10.1002/jrsm.1425","title":"Development and validation of study filters for identifying controlled non‐randomized studies in <scp>PubMed</scp> and Ovid <scp>MEDLINE</scp>","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Hamilton Health Sciences; Impact","funders":"","keywords":"MEDLINE; Computer science; Randomized controlled trial; Medical physics; Filter (signal processing); Medicine; Information retrieval; Surgery","retraction":null,"screen_n_in":null,"score":{"opus":0.8992365636946418,"gpt":0.6515213976464072,"spread":0.2477151660482346,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5389391,0.0003841484,0.008404632,0.00125852,0.0003317522,0.000681991,0.001056277,0.0001090006,0.00004030132],"category_scores_gemma":[0.7989889,0.0002027669,0.000853867,0.001946554,0.0004449857,0.000307722,0.0005542744,0.0002684673,0.0000194862],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007053951,"about_ca_system_score_gemma":0.0001552208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000127266,"about_ca_topic_score_gemma":0.00001564054,"domain_scores_codex":[0.8593397,0.1159703,0.01324204,0.002293329,0.008078297,0.00107632],"domain_scores_gemma":[0.2317196,0.7564896,0.005084323,0.00219184,0.0037971,0.0007175697],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.004689531,0.001405383,0.03786042,0.007200648,0.01918405,0.00005999937,0.3362355,0.0003936166,0.01911665,0.0009659862,0.02440045,0.5484878],"study_design_scores_gemma":[0.1344396,0.0005317954,0.01764089,0.001073909,0.003519061,0.0000111841,0.6043398,0.08743389,0.1102046,0.01154381,0.02882134,0.0004401043],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8926789,0.005228675,0.09154469,0.0004364218,0.0001197295,0.009451983,0.000007972299,0.000007762125,0.000523937],"genre_scores_gemma":[0.6208643,0.0005905073,0.3713063,0.00005932773,0.00009190489,0.005633731,0.000002160767,0.00003329381,0.001418507],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6405193,"threshold_uncertainty_score":0.8268593,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2623116935","doi":"10.1002/jrsm.1244","title":"Paule‐Mandel estimators for network meta‐analysis with random inconsistency effects","year":2017,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":38,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"Medical Research Council","keywords":"Estimator; Computer science; Meta-analysis; Random effects model; Statistics; Econometrics; Mathematics; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.4494998410577451,"gpt":0.5848635729691072,"spread":0.1353637319113621,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03184566,0.0003890169,0.002444086,0.000368989,0.001809737,0.0005416552,0.00108619,0.0002128772,0.0002960706],"category_scores_gemma":[0.1701746,0.0002473895,0.0009850913,0.0006337779,0.0007562854,0.0001940913,0.0002873477,0.0004817713,0.00001038602],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006077162,"about_ca_system_score_gemma":0.0001838519,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001130358,"about_ca_topic_score_gemma":0.00007379064,"domain_scores_codex":[0.9888695,0.007716523,0.0005832705,0.000797587,0.0008875409,0.001145586],"domain_scores_gemma":[0.8237362,0.1727355,0.0003510448,0.001883721,0.0009202434,0.0003732353],"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":[0.001238746,0.0003696532,0.001128459,0.001941565,0.1347719,0.00008173027,0.0001466681,0.00003907255,0.0002284243,0.5759153,0.002938819,0.2811996],"study_design_scores_gemma":[0.0007617731,0.0002038085,0.001138472,0.0001039153,0.06912198,0.000003737225,0.00003344734,0.00990197,0.005290207,0.9124053,0.0006231213,0.0004122481],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0001559191,0.0005623799,0.9927399,0.0004905143,0.0001117354,0.001334488,0.00003944478,0.00007770979,0.004487886],"genre_scores_gemma":[0.01200757,0.00004889247,0.9853237,0.0000255544,0.0002176655,0.001989818,0.000001598751,0.00006869619,0.0003164806],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.33649,"threshold_uncertainty_score":0.9999979,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3197721016","doi":"10.1002/jrsm.1565","title":"Parametric G‐computation for compatible indirect treatment comparisons with limited individual patient data","year":2022,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":38,"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; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Engineering and Physical Sciences Research Council; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Computation; Computer science; Parametric statistics; Econometrics; Statistics; Algorithm; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.8280228066436652,"gpt":0.6339570443067984,"spread":0.1940657623368668,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008945559,0.0002675319,0.0006684144,0.001066589,0.0009452446,0.0001142398,0.0009491249,0.0000715708,0.0001271078],"category_scores_gemma":[0.01124871,0.000216658,0.0000765463,0.001993048,0.0001703834,0.0002144271,0.0008938471,0.0004793715,0.0000031528],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006620302,"about_ca_system_score_gemma":0.0002978882,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000107081,"about_ca_topic_score_gemma":0.00002919212,"domain_scores_codex":[0.9913964,0.005396773,0.0005045906,0.0007354118,0.001249973,0.0007168381],"domain_scores_gemma":[0.9394452,0.05854765,0.0002826627,0.001240071,0.0003148201,0.0001695763],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005404515,0.002461654,0.001001256,0.0001560713,0.0006589338,0.00001772109,0.001361844,0.001182306,0.0007639423,0.006683903,0.01628407,0.9688879],"study_design_scores_gemma":[0.002699777,0.0167032,0.001927486,0.0003155124,0.0009838433,0.00009385453,0.01290488,0.2730334,0.2850648,0.300903,0.1032448,0.002125488],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04768876,0.0002158072,0.946629,0.0003129973,0.00005638522,0.003064617,0.0008853438,0.0004322956,0.0007147834],"genre_scores_gemma":[0.2330405,0.00002158831,0.7637066,0.00001388201,0.00002424901,0.002892051,0.0001995796,0.00005915685,0.00004245772],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9667624,"threshold_uncertainty_score":0.99708,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3175297421","doi":"10.1002/jrsm.1530","title":"Bibliometric study of ‘overviews of systematic reviews’ of health interventions: Evaluation of prevalence, citation and journal impact factor","year":2021,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa; McMaster University; Nova Scotia Health Authority; Cochrane; Impact; University of British Columbia","funders":"","keywords":"Citation; Impact factor; Bibliometrics; Systematic review; MEDLINE; Psychological intervention; Publication; Meta-analysis; Sample size determination; Medicine; Scopus; Family medicine; Library science; Computer science; Statistics; Mathematics; Political science; Pathology; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.9829367848151381,"gpt":0.7881862511010701,"spread":0.194750533714068,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.8167696,0.0002119792,0.008242541,0.03037661,0.0001092675,0.0001969033,0.001232896,0.00007008547,0.005437489],"category_scores_gemma":[0.7122123,0.0001035225,0.00245457,0.06891619,0.0001222487,0.0003582013,0.0002177651,0.0002231587,0.00001130202],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001209615,"about_ca_system_score_gemma":0.0006056586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005072179,"about_ca_topic_score_gemma":0.00001311476,"domain_scores_codex":[0.1294819,0.80286,0.03948665,0.001081549,0.02660051,0.0004894189],"domain_scores_gemma":[0.7421339,0.1694942,0.04358755,0.005850533,0.03856491,0.0003688901],"domain_codex":"methods","domain_gemma":null,"domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0000160034,0.001323372,0.029557,0.1374032,0.001290657,5.157393e-7,0.003328406,0.00006734193,0.003067503,0.00007590708,0.000667349,0.8232027],"study_design_scores_gemma":[0.001241829,0.002722685,0.8515961,0.0695587,0.0044846,0.00004931495,0.03605671,0.01733507,0.007640373,0.008779683,0.0001679433,0.000367017],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7053078,0.226084,0.06246164,0.00006589263,0.00009923967,0.005692545,0.00004196254,0.000001053074,0.0002458182],"genre_scores_gemma":[0.939799,0.01113825,0.04870983,0.000001628752,0.00001248842,0.0001752436,8.125481e-7,0.00001040723,0.0001523332],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8228357,"threshold_uncertainty_score":0.9954717,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3008476194","doi":"10.1002/jrsm.1402","title":"The median and the mode as robust meta‐analysis estimators in the presence of small‐study effects and outliers","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":35,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council Canada; Conselho Nacional de Desenvolvimento Científico e Tecnológico; Medical Research Council; University of Bristol; National Institute for Health and Care Research","keywords":"Meta-analysis; Outlier; Estimator; Statistics; Robustness (evolution); Econometrics; Publication bias; Reliability (semiconductor); Robust statistics; Computer science; Mathematics; Confidence interval; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.8839268872716101,"gpt":0.6515487591271637,"spread":0.2323781281444464,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5219572,0.0002924225,0.004499454,0.0005415836,0.000634251,0.001323608,0.003914703,0.00007552855,0.0002712131],"category_scores_gemma":[0.4986857,0.00008246292,0.001707778,0.006109154,0.001155828,0.0001325258,0.0006371169,0.0005402986,0.00003219266],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001291341,"about_ca_system_score_gemma":0.0000914292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005764054,"about_ca_topic_score_gemma":0.0004719238,"domain_scores_codex":[0.6066167,0.3758611,0.005442837,0.00160928,0.009754392,0.0007157042],"domain_scores_gemma":[0.2883564,0.7025124,0.001970168,0.005679891,0.001049825,0.0004313587],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008389027,0.0007437906,0.07204568,0.001191518,0.2551425,0.0001285087,0.1674897,0.01707253,0.0003998092,0.03025784,0.006634022,0.4480552],"study_design_scores_gemma":[0.0006310273,0.0002406666,0.01567998,0.00003172799,0.07206315,0.000007086352,0.0673506,0.8102425,0.0005915997,0.03028357,0.002522339,0.0003557927],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4472511,0.05923158,0.29961,0.1629996,0.0002039927,0.02055485,0.00003642415,0.00002534606,0.01008717],"genre_scores_gemma":[0.9552735,0.0003647165,0.04315817,0.0001518286,0.00002772145,0.0007631602,1.999696e-7,0.00001359083,0.0002470875],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.79317,"threshold_uncertainty_score":0.9997131,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4399722064","doi":"10.1002/jrsm.1731","title":"Automation tools to support undertaking scoping reviews","year":2024,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Work & Health; Public Health Ontario; University of Toronto; St. Michael's Hospital; Queen's University","funders":"National Health and Medical Research Council; La Trobe University","keywords":"Automation; Interoperability; Computer science; Systematic review; Software engineering; Software; Software development; Best practice; Engineering management; Data science; Process management; Systems engineering; Engineering; World Wide Web; MEDLINE","retraction":null,"screen_n_in":null,"score":{"opus":0.9755536338275869,"gpt":0.7709541161889004,"spread":0.2045995176386864,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7276119,0.0003203363,0.003753416,0.00219509,0.0003879368,0.006700261,0.002692896,0.000133169,0.03477805],"category_scores_gemma":[0.5935033,0.0001600922,0.001808856,0.006768458,0.00007874315,0.0007376311,0.0004613593,0.0004418353,0.03561473],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002070209,"about_ca_system_score_gemma":0.0003737083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001384245,"about_ca_topic_score_gemma":0.00001741517,"domain_scores_codex":[0.7807766,0.1888015,0.0131917,0.002695166,0.01336582,0.00116921],"domain_scores_gemma":[0.6509089,0.3307468,0.002368674,0.01169604,0.003128424,0.001151176],"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":[0.00000195422,0.00001210901,0.00006382966,0.0005124146,0.00007613577,0.000009836334,0.0004055818,0.00002181772,0.002782173,0.004417605,0.07619767,0.9154989],"study_design_scores_gemma":[0.00002262278,0.00005638985,0.0004893735,0.003328266,0.0001112463,0.00001676915,0.0005162605,0.01970529,0.00440951,0.01208176,0.9589819,0.0002805626],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001331424,0.007145432,0.9239428,0.004057373,0.0005710805,0.00278261,0.000008776445,0.00005005193,0.06011045],"genre_scores_gemma":[0.03152598,0.0005619577,0.9389257,0.0003446285,0.0004562926,0.001237825,0.000003775206,0.00006055711,0.02688331],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9152183,"threshold_uncertainty_score":0.9943309,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4200065546","doi":"10.1002/jrsm.1542","title":"A new taxonomy was developed for overlap across 'overviews of systematic reviews': A meta‐research study of research waste","year":2021,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":31,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa; McMaster University; Nova Scotia Health Authority; Cochrane; Impact; University of British Columbia","funders":"","keywords":"Systematic review; Scope (computer science); Meta-analysis; Psychological intervention; MEDLINE; Computer science; Psychology; Taxonomy (biology); Research design; Data science; Management science; Information retrieval; Medicine; Sociology; Pathology; Political science; Social science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9858741720690846,"gpt":0.7632002114989404,"spread":0.2226739605701442,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","bibliometrics","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.9038802,0.0007222967,0.01985203,0.00311702,0.001080379,0.00162917,0.008775742,0.0003461097,0.006813658],"category_scores_gemma":[0.8418532,0.0003646195,0.005893172,0.02168874,0.0006330529,0.0005275515,0.003075277,0.001522036,0.0008732305],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003158084,"about_ca_system_score_gemma":0.00232044,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004721768,"about_ca_topic_score_gemma":0.0005049935,"domain_scores_codex":[0.1607623,0.7523518,0.0339794,0.004630234,0.04494619,0.003330046],"domain_scores_gemma":[0.1302928,0.7705564,0.01013193,0.02481751,0.06257806,0.001623343],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005843008,0.005679864,0.002792038,0.2333404,0.02786802,0.000118084,0.02634396,0.0001251366,0.02573043,0.01158866,0.2285007,0.4373285],"study_design_scores_gemma":[0.002213161,0.002019692,0.0008322943,0.01977241,0.00610275,0.00005748711,0.3081209,0.005583723,0.09624826,0.02526951,0.5322679,0.00151189],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09292877,0.2372853,0.5158628,0.003107982,0.001050353,0.1286612,0.000351185,0.00004145727,0.02071092],"genre_scores_gemma":[0.08091691,0.00278852,0.8322027,0.00003226741,0.0003722742,0.02816567,0.000005983807,0.000148912,0.0553668],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4358166,"threshold_uncertainty_score":0.9999047,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2078055840","doi":"10.1002/jrsm.14","title":"Evidence synthesis, economics and public policy","year":2010,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute of Health Economics","funders":"","keywords":"Scope (computer science); Systematic review; Economic evaluation; Economics; Empirical evidence; Variety (cybernetics); Evidence-based policy; Public economics; Management science; Computer science; Political science; MEDLINE","retraction":null,"screen_n_in":null,"score":{"opus":0.4572846181334075,"gpt":0.4292083149121764,"spread":0.02807630322123117,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01628551,0.0001860008,0.0004721434,0.0007633451,0.0003593604,0.0002898026,0.0005039788,0.0002203456,0.001653399],"category_scores_gemma":[0.02075632,0.0002133912,0.0001079437,0.0002849983,0.0004998879,0.0007345827,0.0003162385,0.00051619,0.000982416],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002698867,"about_ca_system_score_gemma":0.0001148577,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007683891,"about_ca_topic_score_gemma":0.00007722152,"domain_scores_codex":[0.9974144,0.0005011472,0.0006019475,0.0007264743,0.00006236347,0.0006936805],"domain_scores_gemma":[0.9923355,0.006375618,0.000198245,0.0007263705,0.00003183298,0.0003324168],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001259791,0.00009099555,0.08366109,0.00004162936,0.00006860076,9.427234e-7,0.0001298357,0.00001008862,0.00234607,0.6216039,0.0002220232,0.2918122],"study_design_scores_gemma":[0.0003338655,0.0001070094,0.4264503,0.00008311728,0.00002393422,0.00004093764,0.0005394643,0.01233279,0.02809657,0.3546753,0.1761797,0.001137045],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9417662,0.003113239,0.004765712,0.01230062,0.0003398157,0.0004231472,0.00006548245,0.00005562784,0.03717016],"genre_scores_gemma":[0.923285,0.01025922,0.06528542,0.0001012566,0.0003305418,0.0002594814,0.000001442614,0.00004707076,0.0004305765],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3427893,"threshold_uncertainty_score":0.9997954,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3006673017","doi":"10.1002/jrsm.1398","title":"Comparing machine and human reviewers to evaluate the risk of bias in randomized controlled trials","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Capital District Health Authority; Institute of Health Economics; University of Alberta","funders":"Canadian Institutes of Health Research; Institute of Health Economics; Alberta Innovates - Health Solutions; Physiotherapy Foundation of Canada; Government of Alberta","keywords":"Blinding; Randomized controlled trial; Computer science; Medical physics; MEDLINE; Sample size determination; Medicine; Applied psychology; Psychology; Statistics; Surgery; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.9609323859410629,"gpt":0.7227307998086823,"spread":0.2382015861323806,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9613858,0.0003113156,0.03045308,0.0009493275,0.0002630968,0.0006640845,0.002212001,0.00007225099,0.004014325],"category_scores_gemma":[0.9737784,0.00009884116,0.004915943,0.003085848,0.0003093047,0.00008994956,0.0004716413,0.0004823574,0.0001890478],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002975469,"about_ca_system_score_gemma":0.00008817553,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002420641,"about_ca_topic_score_gemma":0.00005014961,"domain_scores_codex":[0.07757422,0.8899349,0.02175007,0.001147778,0.009117885,0.0004751256],"domain_scores_gemma":[0.02095731,0.9649959,0.007262348,0.003833479,0.002439407,0.0005115821],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0714725,0.0002954177,0.04068707,0.001313711,0.0127374,0.00001361429,0.01034255,0.001745631,0.01359938,0.01636753,0.02938227,0.802043],"study_design_scores_gemma":[0.2149473,0.0003861727,0.01297762,0.002094604,0.01225229,0.000005961582,0.006069827,0.6109159,0.007147806,0.06061711,0.0715026,0.001082767],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5023127,0.101016,0.2713241,0.04156169,0.0003316413,0.05103473,0.00005935137,0.0000229532,0.03233675],"genre_scores_gemma":[0.9018297,0.002045498,0.09380534,0.0001990704,0.0001002193,0.001375249,6.263056e-7,0.00002183748,0.0006225189],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8009602,"threshold_uncertainty_score":0.9968961,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3022994372","doi":"10.1002/jrsm.1416","title":"Equivalence of entropy balancing and the method of moments for matching‐adjusted indirect comparison","year":2020,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Department of Health and Social Care; University of Bristol; National Institute for Health and Care Research; University Hospitals Bristol NHS Foundation Trust; AstraZeneca","keywords":"Covariate; Statistics; Population; Mathematics; Entropy (arrow of time); Nonparametric statistics; Standard error; Econometrics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.5557337790460275,"gpt":0.6078522747330801,"spread":0.05211849568705262,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01900721,0.0001650941,0.001001851,0.0001584167,0.0001160356,0.00002230781,0.0005399258,0.0001019157,0.0000389273],"category_scores_gemma":[0.05650531,0.0001121839,0.0001304696,0.0004766102,0.0004470905,0.0001036597,0.0003567241,0.0003689177,5.072417e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005116667,"about_ca_system_score_gemma":0.00006695327,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004639043,"about_ca_topic_score_gemma":0.00000160477,"domain_scores_codex":[0.9924694,0.005413445,0.0007129312,0.0003319069,0.0006526681,0.0004196183],"domain_scores_gemma":[0.915462,0.08327115,0.0003769361,0.0004124695,0.0003633454,0.0001141234],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001086128,0.0001529936,0.0004906186,0.002969828,0.0003104686,0.000001243582,0.007129371,0.00002808129,0.5667387,0.2035318,0.0004623992,0.2170983],"study_design_scores_gemma":[0.0003036572,0.000121737,0.00006861035,0.0002038591,0.00005815435,0.000001141139,0.001092992,0.0109804,0.6814659,0.3054852,0.0001346859,0.00008363189],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01216411,0.000280657,0.9851528,0.0007956084,0.00001709079,0.001102978,0.00002625373,0.00007037951,0.0003900561],"genre_scores_gemma":[0.2048293,0.0001062386,0.7947487,0.00001761179,0.00002492048,0.0002204276,5.398428e-7,0.00002754702,0.00002464798],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2170147,"threshold_uncertainty_score":0.9514421,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4378190436","doi":"10.1002/jrsm.1636","title":"A real‐world evaluation of the implementation of <scp>NLP</scp> technology in abstract screening of a systematic review","year":2023,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; McGill University; The Quebec Population Health Research Network; University of Calgary; McGill University Health Centre","funders":"Canadian Medical Association; Robert Koch Institut; Koch Institute for Integrative Cancer Research, Massachusetts Institute of Technology; Public Health Agency; Rhodes Scholarships; Public Health Agency of Canada; World Health Organization","keywords":"Computer science; Context (archaeology); Systematic review; Recall; Inclusion (mineral); Machine learning; Artificial intelligence; Precision and recall; Natural language processing; MEDLINE; Psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.9514931780577904,"gpt":0.7593710306945818,"spread":0.1921221473632087,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","bibliometrics","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7924761,0.0005418818,0.02192546,0.006694605,0.000106096,0.0001180679,0.005887507,0.0003379669,0.001115339],"category_scores_gemma":[0.6473593,0.0002396812,0.00497539,0.02428611,0.0003244466,0.0001430654,0.0008002021,0.0007978036,0.0001324078],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002766563,"about_ca_system_score_gemma":0.001399005,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002656039,"about_ca_topic_score_gemma":0.0002496475,"domain_scores_codex":[0.3393281,0.5496169,0.06618163,0.002935802,0.04058953,0.001347954],"domain_scores_gemma":[0.6196575,0.3096524,0.0459981,0.01370651,0.01081035,0.0001750841],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[2.718284e-7,0.00003513919,0.00004542404,0.5491176,0.0006149151,5.89168e-7,0.00007635008,0.000003158423,0.00001430671,0.0004677289,0.0006085563,0.4490159],"study_design_scores_gemma":[0.0001006456,0.00004125576,0.0002989462,0.9622547,0.01152956,0.000006992964,0.00288978,0.00050366,0.0003500725,0.005102103,0.0167788,0.000143458],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00005191979,0.9846416,0.0007387397,0.00009802077,0.00007072409,0.01285862,0.00008134662,0.000005913158,0.001453116],"genre_scores_gemma":[0.0001219904,0.9768335,0.01903173,0.000002681187,0.00001939297,0.003248387,0.000006365093,0.00005102469,0.0006849103],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4488725,"threshold_uncertainty_score":0.9997978,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2156696492","doi":"10.1002/jrsm.1146","title":"We know less than we should about methods of meta‐analysis","year":2015,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Sudbury","funders":"","keywords":"Library science; Computer science; Citation; Information retrieval; Medical library; Statistic; Operations research; Statistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.9784588548526072,"gpt":0.7464486883589169,"spread":0.2320101664936903,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.8100159,0.0007086941,0.0161279,0.005278533,0.0003280864,0.001238827,0.007299324,0.0004057169,0.01924362],"category_scores_gemma":[0.4333279,0.0003369313,0.01380719,0.01829857,0.0006346693,0.0004769616,0.001143377,0.0008635498,0.001204231],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001315627,"about_ca_system_score_gemma":0.000529241,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003022713,"about_ca_topic_score_gemma":0.0001233121,"domain_scores_codex":[0.5124863,0.4496757,0.01308822,0.002784685,0.02061828,0.001346757],"domain_scores_gemma":[0.6374591,0.3253931,0.006883693,0.01724392,0.01117356,0.001846621],"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":[0.00005177898,0.0002882927,0.001160819,0.0002720219,0.1397944,0.00001280127,0.002244408,0.0006245193,0.002630914,0.006794024,0.02582101,0.820305],"study_design_scores_gemma":[0.000201488,0.0001287149,0.0005405658,0.00009248523,0.1428647,0.00001096617,0.01226436,0.04169168,0.04037702,0.06720247,0.6938438,0.000781742],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008226082,0.1043159,0.8566325,0.007209854,0.0002294565,0.001124316,0.00005922234,0.0000217806,0.02958439],"genre_scores_gemma":[0.01982857,0.002498658,0.9463729,0.0000455138,0.0001334115,0.0004678887,0.000002572507,0.0000547277,0.03059581],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8195233,"threshold_uncertainty_score":0.9999083,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}