{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":20,"total_is_capped":false,"direct_labels_cover":1,"predictions_cover":20,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"7b0a4d4a1a0a","filters":{"venue":"Health Services and Outcomes Research Methodology"}},"results":[{"id":"W1529736910","doi":"10.1023/a:1024260023851","title":"A Comparison of Statistical Modeling Strategies for Analyzing Length of Stay after CABG Surgery","year":2002,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":138,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences","funders":"Canadian Institutes of Health Research; Canada Research Chairs; Ontario Ministry of Health and Long-Term Care","keywords":"Negative binomial distribution; Generalized linear model; Poisson regression; Statistics; Mathematics; Poisson distribution; Linear regression; Regression analysis; Log-linear model; Linear model; Proportional hazards model; Consistency (knowledge bases); Quasi-likelihood; Statistical model; Parametric statistics; Regression; Simple linear regression; Medicine; Population","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Deanna M. Rothwell","is_ca":true},{"name":"Jack V. Tu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5064902954872634,"gpt":0.5967557162960009,"spread":0.09026542080873745,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.132596,0.002160182,0.002711534,0.004576101,0.0009902979,0.00315235,0.003629078,0.001920843,0.001464795],"category_scores_gemma":[0.2467245,0.0009949224,0.00568708,0.003339612,0.00111384,0.003216119,0.001920684,0.002720945,0.0002799383],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002603505,"about_ca_system_score_gemma":0.005172364,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0103061,"about_ca_topic_score_gemma":0.007971028,"domain_scores_codex":[0.9196442,0.07196511,0.002575608,0.002077368,0.003190332,0.0005473624],"domain_scores_gemma":[0.4292384,0.5499703,0.00685986,0.005922825,0.007137957,0.0008705831],"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":[0.007834301,0.001632928,0.1442286,0.001404784,0.01645254,0.0002750936,0.004038622,0.2420522,0.001920602,0.06369144,0.004698992,0.5117699],"study_design_scores_gemma":[0.0005198221,0.002719675,0.02786576,0.000310068,0.001773912,0.0002204395,0.001205941,0.9386736,0.000917324,0.02419033,0.001416338,0.0001867468],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1624347,0.001753747,0.8314639,0.001123462,0.0001582018,0.000741714,0.0005690358,0.0008781166,0.000877129],"genre_scores_gemma":[0.5280505,0.001496788,0.4656482,0.0003159705,0.000139902,0.001991939,0.00114563,0.0005646877,0.0006463448],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.132596,"threshold_uncertainty_score":0.7012433,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2045908763","doi":"10.1007/s10742-014-0117-x","title":"Using instrumental variables to estimate a Cox’s proportional hazards regression subject to additive confounding","year":2014,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":66,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"National Cancer Institute; National Center for Advancing Translational Sciences; National Institute of Mental Health; National Institutes of Health","keywords":"Confounding; Proportional hazards model; Instrumental variable; Statistics; Regression analysis; Regression; Econometrics; Mathematics","authors":[{"name":"Todd A. MacKenzie","is_ca":false},{"name":"Tor D. Tosteson","is_ca":false},{"name":"Nancy E. Morden","is_ca":false},{"name":"Thérèse A. Stukel","is_ca":true},{"name":"A. James O’Malley","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3917987631171452,"gpt":0.5536682860750683,"spread":0.1618695229579231,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04435372,0.001361542,0.002015768,0.001874284,0.0006903872,0.002811802,0.003570168,0.002217388,0.006126363],"category_scores_gemma":[0.1253247,0.001354119,0.003695586,0.003087253,0.00134282,0.002590034,0.002706242,0.004887066,0.001051188],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001055896,"about_ca_system_score_gemma":0.00351258,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007486538,"about_ca_topic_score_gemma":0.005507411,"domain_scores_codex":[0.9780673,0.01567679,0.001198093,0.002362506,0.00187922,0.0008160705],"domain_scores_gemma":[0.9308456,0.05699931,0.0029301,0.007422769,0.001601946,0.0002002988],"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":[0.0008344658,0.000395278,0.04844646,0.0008930605,0.006274415,0.001415262,0.0009098412,0.1757097,0.001111522,0.5886145,0.01460655,0.160789],"study_design_scores_gemma":[0.0005971986,0.0002674138,0.007803217,0.0002956768,0.001683456,0.0005913904,0.0002606013,0.6194869,0.002237295,0.3464604,0.02018733,0.0001291296],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005683733,0.0002966024,0.9920831,0.0006354009,0.0002213645,0.00009602813,0.0002259477,0.0002100149,0.0005478052],"genre_scores_gemma":[0.4254563,0.001302291,0.5487412,0.001372863,0.0007017833,0.002096018,0.001940304,0.0004050191,0.01798439],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04435372,"threshold_uncertainty_score":0.2345677,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2068564314","doi":"10.1007/s10742-014-0124-y","title":"A data-adaptive strategy for inverse weighted estimation of causal effects","year":2014,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"National Institute on Drug Abuse; National Institutes of Health; National Science Foundation","keywords":"Causal inference; Propensity score matching; Observational study; Nonparametric statistics; Covariate; Estimator; Confounding; Inverse probability; Econometrics; Statistics; Average treatment effect; Logistic regression; Consistency (knowledge bases); Randomized experiment; Marginal structural model; Computer science; Mathematics; Artificial intelligence; Posterior probability; Bayesian probability","authors":[{"name":"Yeying Zhu","is_ca":true},{"name":"Debashis Ghosh","is_ca":false},{"name":"Nandita Mitra","is_ca":false},{"name":"Bhramar Mukherjee","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6899585491834566,"gpt":0.6385585072369091,"spread":0.05140004194654746,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0400193,0.001851297,0.002887299,0.00331327,0.001065954,0.00227978,0.006745686,0.00297794,0.008939011],"category_scores_gemma":[0.1536559,0.002110655,0.003796719,0.004066928,0.002498779,0.004121721,0.006217373,0.006987832,0.001911075],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001030805,"about_ca_system_score_gemma":0.003324812,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004187287,"about_ca_topic_score_gemma":0.004592653,"domain_scores_codex":[0.9791566,0.01578334,0.0009714323,0.002020929,0.001787831,0.0002798229],"domain_scores_gemma":[0.914182,0.07217444,0.001673044,0.008081255,0.00344029,0.000448991],"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":[0.000472604,0.0003602321,0.002597452,0.0006510273,0.001098518,0.000322435,0.0007003133,0.1161616,0.00378835,0.5003778,0.004518075,0.3689516],"study_design_scores_gemma":[0.00018277,0.0001081833,0.0005152515,0.0001132995,0.0002070151,0.0001655715,0.00007009716,0.6517258,0.001622884,0.3399989,0.005232857,0.00005735936],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0003106876,0.00003082552,0.9993761,0.00005932269,0.00001417091,0.00003621499,0.00001864097,0.00005296324,0.0001011766],"genre_scores_gemma":[0.02163767,0.0001309174,0.9759746,0.00018021,0.00007694469,0.0006382529,0.0002063503,0.0001040051,0.00105106],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0400193,"threshold_uncertainty_score":0.2116448,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W23346058","doi":"10.1023/a:1011424117583","title":"Considerations for Measuring Functioning of the Elderly: IRM Dimensionality and Scaling Analysis","year":2001,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":16,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Rasch model; Activities of daily living; Item response theory; International Classification of Functioning, Disability and Health; Scale (ratio); Gerontology; Psychology; Statistics; Psychometrics; Medicine; Mathematics; Clinical psychology; Physical therapy; Geography; Rehabilitation","authors":[{"name":"Krista Breithaupt","is_ca":false},{"name":"Ian McDowell","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4965065371157216,"gpt":0.5636000211075388,"spread":0.06709348399181725,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2354883,0.002877511,0.006172922,0.006700026,0.006300772,0.01571023,0.006197506,0.002795664,0.002756169],"category_scores_gemma":[0.5162024,0.001458364,0.005824873,0.009761938,0.008693581,0.009897714,0.007972778,0.01058018,0.0006640418],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006101794,"about_ca_system_score_gemma":0.01205377,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01153135,"about_ca_topic_score_gemma":0.02002812,"domain_scores_codex":[0.7740723,0.1881085,0.01577063,0.00456697,0.0162406,0.001240917],"domain_scores_gemma":[0.568119,0.3311896,0.01271069,0.04734408,0.03818787,0.002448807],"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":[0.0004231668,0.0007732014,0.05296516,0.003352952,0.001555475,0.0003604101,0.0152497,0.01347154,0.002165596,0.5055203,0.02588538,0.3782769],"study_design_scores_gemma":[0.000192067,0.0009018544,0.04573544,0.0036859,0.0006677973,0.0006184491,0.01744444,0.09224183,0.001893874,0.8047078,0.03113668,0.0007739377],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03314899,0.005739445,0.9249055,0.01663788,0.001882807,0.003852238,0.0009055368,0.0004654751,0.01246204],"genre_scores_gemma":[0.1461352,0.001557602,0.8412946,0.001825722,0.0008184361,0.006871418,0.0004591724,0.0001834275,0.0008544615],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2354883,"threshold_uncertainty_score":0.942779,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1983766924","doi":"10.1007/s10742-008-0030-2","title":"Signal-to-noise ratio (SNR) as a measure of reproducibility: design, estimation, and application","year":2008,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Image and Signal Denoising Methods","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Reproducibility; Measure (data warehouse); Statistics; Estimation; Signal-to-noise ratio (imaging); Computer science; Noise (video); Mathematics; Artificial intelligence; Data mining; Engineering","authors":[{"name":"Naser Elkum","is_ca":false},{"name":"Mohamed M. Shoukri","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2905204980326063,"gpt":0.5046460894679036,"spread":0.2141255914352972,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1390205,0.001722249,0.001737441,0.002489672,0.001274092,0.001982798,0.002637203,0.002806207,0.001844577],"category_scores_gemma":[0.1947755,0.0009848699,0.002255843,0.002545099,0.00406604,0.001471667,0.001799239,0.001597937,0.0007097371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001338617,"about_ca_system_score_gemma":0.003185981,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00129091,"about_ca_topic_score_gemma":0.001166694,"domain_scores_codex":[0.8918629,0.07833932,0.006394257,0.0107917,0.01178729,0.0008244835],"domain_scores_gemma":[0.8331734,0.1172909,0.01185257,0.01732432,0.01953488,0.0008239881],"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":[0.03019776,0.005742497,0.1699261,0.004254677,0.007605201,0.0004593456,0.006262864,0.04626966,0.04768542,0.06133351,0.003846634,0.6164163],"study_design_scores_gemma":[0.009213859,0.04830483,0.2088003,0.0009063804,0.01363314,0.002469998,0.00153791,0.4222699,0.1604866,0.1003422,0.03065271,0.001382114],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0749219,0.0006883407,0.9143117,0.0003043575,0.0001868708,0.007260642,0.0002510661,0.0003943572,0.001680723],"genre_scores_gemma":[0.3556452,0.0004338805,0.623637,0.0003253463,0.0001175386,0.0182934,0.0002171984,0.0002193613,0.001111033],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8609794,"threshold_uncertainty_score":0.7352197,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2889374799","doi":"10.1007/s10742-018-0187-2","title":"Causal inference for multi-level treatments with machine-learned propensity scores","year":2018,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Propensity score matching; Causal inference; Inverse probability; Inverse probability weighting; Multinomial logistic regression; Statistics; Econometrics; Observational study; Logistic regression; Matching (statistics); Computer science; Estimator; Bayesian probability; Average treatment effect; Mathematics; Posterior probability","authors":[{"name":"Lin Lin","is_ca":true},{"name":"Yeying Zhu","is_ca":true},{"name":"Liang Chen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8430232517073458,"gpt":0.6674423948286655,"spread":0.1755808568786803,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0507329,0.00189765,0.005038728,0.004151046,0.00156406,0.004291061,0.005457031,0.004392186,0.01765967],"category_scores_gemma":[0.1805974,0.002539259,0.005261331,0.004471983,0.004050018,0.006628064,0.004500949,0.008481104,0.001409672],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003079888,"about_ca_system_score_gemma":0.005442534,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008350644,"about_ca_topic_score_gemma":0.008583022,"domain_scores_codex":[0.9677829,0.02523814,0.00123855,0.003459982,0.001436323,0.0008440358],"domain_scores_gemma":[0.737551,0.2428729,0.006503494,0.009794982,0.002123032,0.001154584],"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":[0.000664085,0.0005943732,0.008638517,0.001011539,0.001785428,0.0003467842,0.0004466784,0.2612616,0.0002469391,0.5985048,0.007246933,0.1192522],"study_design_scores_gemma":[0.0002269991,0.00005406824,0.0007958704,0.0001012871,0.0002261294,0.00005570704,0.00003107301,0.4518765,0.0001461227,0.545364,0.001094393,0.00002775722],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005778077,0.0005850529,0.9902275,0.00160146,0.0001153867,0.0001975621,0.0006051234,0.0002903465,0.0005995351],"genre_scores_gemma":[0.430194,0.001917938,0.5516037,0.001624854,0.001169528,0.00317046,0.002721576,0.0002374693,0.007360514],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0507329,"threshold_uncertainty_score":0.2683044,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2064626421","doi":"10.1007/s10742-009-0044-4","title":"Deriving a mental health outcome measure using the pooled index: an application to psychiatric consumer–survivors in different housing types","year":2009,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Schizophrenia research and treatment","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University; St. Michael's Hospital; Western University","funders":"","keywords":"Quality of life (healthcare); Index (typography); Mental health; Medicine; Gerontology; Activities of daily living; Psychiatry; Demography; Psychology; Nursing","authors":[{"name":"Mark Speechley","is_ca":true},{"name":"Cheryl Forchuk","is_ca":true},{"name":"Jeffrey S. Hoch","is_ca":true},{"name":"Elsabeth Jensen","is_ca":true},{"name":"Jennifer Wagg","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2128823355536524,"gpt":0.5300256492860057,"spread":0.3171433137323533,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02616155,0.001159524,0.00210645,0.005902037,0.001055455,0.001650863,0.001380618,0.001150023,0.001917684],"category_scores_gemma":[0.07180535,0.0003347898,0.004768495,0.006258744,0.000540943,0.001403818,0.003308131,0.001102886,0.0003278399],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001259174,"about_ca_system_score_gemma":0.002030045,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006088033,"about_ca_topic_score_gemma":0.007250383,"domain_scores_codex":[0.9885655,0.007030941,0.00155614,0.0008385814,0.001702247,0.0003065751],"domain_scores_gemma":[0.9793554,0.01244432,0.002111696,0.002533411,0.003181372,0.0003737753],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002414978,0.001072744,0.5960993,0.0007228353,0.007727015,0.0002761903,0.001532428,0.01018604,0.002048624,0.005216725,0.003694444,0.3690087],"study_design_scores_gemma":[0.0006694903,0.007321512,0.8007094,0.0006206624,0.008957323,0.001016838,0.005188087,0.1210367,0.00694159,0.03960782,0.007565723,0.0003648869],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7533852,0.001498329,0.2336761,0.0008149489,0.0002286735,0.001799154,0.00476524,0.0003081593,0.003524199],"genre_scores_gemma":[0.8438531,0.0004523897,0.1496787,0.0001341976,0.0001139057,0.001655456,0.003458233,0.0000580582,0.0005959987],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.02616155,"threshold_uncertainty_score":0.1383572,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4388795592","doi":"10.1007/s10742-023-00317-y","title":"Propensity score weighting with survey weighted data when outcomes are binary: a simulation study","year":2023,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"London Health Sciences Centre","funders":"National Cancer Institute; National Institute on Aging; National Cancer Center","keywords":"Propensity score matching; Weighting; Statistics; Observational study; Average treatment effect; Population; Mathematics; Econometrics; Medicine","authors":[{"name":"Yang Chen","is_ca":false},{"name":"Meaghan S. Cuerden","is_ca":true},{"name":"Wei Zhang","is_ca":false},{"name":"Melissa Aldridge","is_ca":false},{"name":"Lihua Li","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8783707654311809,"gpt":0.649225151849151,"spread":0.2291456135820299,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07097892,0.00168284,0.003300183,0.001946109,0.001168703,0.002777033,0.00369695,0.003601362,0.01112095],"category_scores_gemma":[0.239147,0.00141391,0.003846386,0.003939694,0.002421249,0.004931827,0.002801229,0.005229718,0.0007581818],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002287161,"about_ca_system_score_gemma":0.00301788,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02027609,"about_ca_topic_score_gemma":0.007638927,"domain_scores_codex":[0.9657358,0.0309977,0.0006203552,0.001261628,0.0007136453,0.0006708605],"domain_scores_gemma":[0.5648664,0.4115831,0.007250316,0.01083492,0.004100042,0.001365118],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.009027547,0.003725051,0.03610617,0.0008555762,0.002424217,0.0008126477,0.001730553,0.7381366,0.0003088492,0.1530791,0.005199931,0.04859371],"study_design_scores_gemma":[0.002069175,0.0008339214,0.002197682,0.0001372692,0.0005378081,0.0001847912,0.0003712347,0.9431131,0.0001870031,0.04929056,0.001011229,0.00006617093],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5827851,0.001457146,0.4042177,0.001995985,0.0001710262,0.001941983,0.001247518,0.0002577288,0.005925832],"genre_scores_gemma":[0.9044993,0.000705706,0.08913326,0.0003216035,0.00006169336,0.001602604,0.0007721036,0.0000875728,0.002816124],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9290211,"threshold_uncertainty_score":0.3753769,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2343900495","doi":"10.1007/s10742-016-0147-7","title":"The study of service use among homeless persons with mental illness: a methodological review","year":2016,"lang":"en","type":"review","venue":"Health Services and Outcomes Research Methodology","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"PsycINFO; Mental illness; Observational study; Mental health; Receipt; Service (business); Type of service; Medicine; Health services research; Population; MEDLINE; Psychology; Psychiatry; Public health; Nursing; Gerontology; Environmental health; Business","authors":[{"name":"Nick Kerman","is_ca":true},{"name":"John Sylvestre","is_ca":true},{"name":"Alexia Polillo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6729901541732592,"gpt":0.6682930787085176,"spread":0.004697075464741629,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05606956,0.001582961,0.008097636,0.01680822,0.001193151,0.006611095,0.003400694,0.003137151,0.002172802],"category_scores_gemma":[0.1407645,0.001592009,0.008689173,0.01775999,0.002919547,0.004101333,0.003896405,0.002028391,0.000170221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009090384,"about_ca_system_score_gemma":0.02949421,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01438939,"about_ca_topic_score_gemma":0.04075579,"domain_scores_codex":[0.9477233,0.02147716,0.02056411,0.003000092,0.006613297,0.0006220873],"domain_scores_gemma":[0.8969713,0.08474012,0.01093517,0.002005265,0.004747392,0.0006008751],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001666357,0.00009347164,0.006262329,0.8413504,0.01653568,0.000116742,0.00143411,0.0001756577,0.0001512218,0.002155511,0.001170357,0.1303879],"study_design_scores_gemma":[0.0001624758,0.000231222,0.008444929,0.9166912,0.04654155,0.0004312425,0.001999584,0.0001364954,0.0001861067,0.00159966,0.02350969,0.00006581504],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0009140308,0.9978471,0.0003552611,0.0002525141,0.00009317308,0.000283994,0.0001185911,0.000001890969,0.000133469],"genre_scores_gemma":[0.01566451,0.9804292,0.002233841,0.0005719165,0.00007397772,0.000819879,0.0001394494,0.000005020085,0.00006228552],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9439304,"threshold_uncertainty_score":0.2965277,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2030246929","doi":"10.1007/s10742-009-0048-0","title":"Comparison of cluster detection using patients and events of medically treated self-inflicted injuries in Alberta, Canada","year":2009,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Data-Driven Disease Surveillance","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Alberta","funders":"","keywords":"Context (archaeology); Cluster analysis; Epidemiology; Psychological intervention; Cluster (spacecraft); Medicine; Public health; Health care; Health services research; Population; Intervention (counseling); Medical emergency; Demography; Environmental health; Geography; Computer science; Nursing; Pathology","authors":[{"name":"Rhonda J. Rosychuk","is_ca":true},{"name":"Ian Colman","is_ca":true},{"name":"Brian H. Rowe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1196953394729653,"gpt":0.5064610241919099,"spread":0.3867656847189446,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006044831,0.000773661,0.0009540238,0.003039368,0.001845996,0.001965345,0.003355494,0.0008975887,0.001345714],"category_scores_gemma":[0.02032723,0.0006179349,0.001348792,0.004820082,0.0008080197,0.0005596589,0.001498927,0.000796785,0.0001497339],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02557531,"about_ca_system_score_gemma":0.0409176,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9940898,"about_ca_topic_score_gemma":0.9939464,"domain_scores_codex":[0.994402,0.001730985,0.0004111489,0.0006392398,0.001756667,0.001059981],"domain_scores_gemma":[0.9877964,0.003417374,0.001304796,0.0005820327,0.005671078,0.00122827],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009285954,0.0001150845,0.9814301,0.0000804115,0.0004489692,0.00005816893,0.0005509679,0.004088299,0.0001975548,0.000345327,0.001865555,0.009890987],"study_design_scores_gemma":[0.00008807442,0.0001282566,0.9893885,0.0000497066,0.0002143805,0.0000484789,0.001218805,0.007724443,0.0002450511,0.00008005299,0.0007771714,0.00003708639],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9906039,0.0005529642,0.001345875,0.000293824,0.00004434616,0.0002136444,0.00528299,0.000043941,0.001618524],"genre_scores_gemma":[0.9926837,0.000333468,0.001539114,0.00009586001,0.00001388229,0.00007109941,0.004067378,0.00001468196,0.001180945],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02557531,"threshold_uncertainty_score":0.1855627,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2317982607","doi":"10.1007/s10742-013-0112-7","title":"A comparison of statistical models for analyzing episodes-of-care costs for chronic obstructive pulmonary disease exacerbations","year":2013,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Saskatchewan Health Quality Council; University of Saskatchewan; University of Manitoba","funders":"Ministry of Health, Saskatchewan; University of Saskatchewan","keywords":"Ordinary least squares; Heteroscedasticity; Statistics; Poisson regression; Medicine; Generalized linear model; Regression analysis; Regression; Econometrics; Health administration; Linear regression; Mathematics; Public health; Population; Environmental health","authors":[{"name":"John Paul Kuwornu","is_ca":true},{"name":"Lisa M. Lix","is_ca":true},{"name":"Jacqueline Quail","is_ca":true},{"name":"Eric W. Wang","is_ca":true},{"name":"Meriç Osman","is_ca":true},{"name":"Gary Teare","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3286456677312481,"gpt":0.5718448853666004,"spread":0.2431992176353523,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1524603,0.002166028,0.003820759,0.005089981,0.000906032,0.003751626,0.004799404,0.002878184,0.002428436],"category_scores_gemma":[0.2716369,0.001099249,0.008700471,0.003229395,0.001405002,0.003494429,0.001924547,0.003270314,0.0002833396],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005139508,"about_ca_system_score_gemma":0.00510801,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01524421,"about_ca_topic_score_gemma":0.006885177,"domain_scores_codex":[0.916773,0.07400839,0.002700091,0.002829664,0.002811135,0.0008777616],"domain_scores_gemma":[0.2610912,0.7206256,0.006768986,0.005828934,0.004809286,0.0008758691],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01978485,0.002122572,0.1296538,0.001507286,0.01763586,0.0002337841,0.001343222,0.6290781,0.0005010759,0.03672633,0.005023327,0.1563898],"study_design_scores_gemma":[0.0006425183,0.001799199,0.0183426,0.000177048,0.001469081,0.0001219539,0.0003192708,0.963501,0.0001967393,0.01276516,0.0005768218,0.00008847367],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5748379,0.006227119,0.4085279,0.00255343,0.0006663344,0.001113605,0.002658862,0.001369112,0.002045693],"genre_scores_gemma":[0.8850235,0.001631972,0.1078689,0.0005123288,0.0002995793,0.00135556,0.002045276,0.000410591,0.000852318],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1524603,"threshold_uncertainty_score":0.8062966,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2080992413","doi":"10.1007/s10742-006-0010-3","title":"Methods for studying adverse events on surgical wait lists","year":2006,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":4,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Royal Columbian Hospital; Vancouver General Hospital; University of British Columbia","funders":"","keywords":"Medicine; Population; Event (particle physics); Adverse effect; Incidence (geometry); Disease; Angina; Cumulative incidence; Unstable angina; Emergency medicine; Medical emergency; Intensive care medicine; Surgery; Myocardial infarction; Internal medicine","authors":[{"name":"Boris Sobolev","is_ca":true},{"name":"Lisa Kuramoto","is_ca":true},{"name":"Adrian R. Levy","is_ca":true},{"name":"Robert Hayden","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7943857756584719,"gpt":0.6733009975553546,"spread":0.1210847781031174,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06389008,0.001414778,0.001376361,0.005634215,0.0007848062,0.001924035,0.002689668,0.001297604,0.01411055],"category_scores_gemma":[0.137796,0.0006857272,0.003169694,0.004049433,0.000776546,0.00169473,0.001751361,0.001693973,0.000967603],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001041057,"about_ca_system_score_gemma":0.002894314,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002236244,"about_ca_topic_score_gemma":0.002164291,"domain_scores_codex":[0.958975,0.03278714,0.002249954,0.002656055,0.002790683,0.0005412471],"domain_scores_gemma":[0.8332176,0.1379045,0.01298095,0.01019531,0.00477596,0.0009256196],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002799245,0.001384848,0.2638074,0.00262783,0.006512706,0.0003145381,0.001499002,0.01271491,0.001344562,0.134763,0.01865859,0.5535734],"study_design_scores_gemma":[0.004199005,0.004582106,0.2787044,0.002159175,0.007378498,0.001768508,0.002149622,0.1352184,0.00402306,0.4725659,0.08679712,0.0004542313],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04388596,0.003829271,0.9358486,0.001338215,0.0004262234,0.004208582,0.005565427,0.0003799096,0.004517854],"genre_scores_gemma":[0.2927477,0.003196258,0.6402197,0.001034937,0.0008736926,0.04446558,0.00589848,0.0002085312,0.0113551],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9361099,"threshold_uncertainty_score":0.3378871,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W213901707","doi":"10.1023/a:1025818432525","title":"Imputing a Binary Variable from Donor to Recipient Dataset when Recipient Dataset Holds Restricted Information on the Variable","year":2002,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Statistics Canada; Health Canada","funders":"Health Canada","keywords":"Variable (mathematics); Imputation (statistics); Logistic regression; Computer science; Data mining; Population; Regression; Statistics; Econometrics; Mathematics; Missing data; Machine learning; Medicine","authors":[{"name":"Kisalaya Basu","is_ca":true},{"name":"Vishnu Kapur","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3935089949768288,"gpt":0.5154611245270454,"spread":0.1219521295502166,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08936755,0.001029513,0.002935961,0.002815162,0.001888127,0.003375674,0.005677185,0.003617794,0.006409181],"category_scores_gemma":[0.2022135,0.001170177,0.004174964,0.005230059,0.002299621,0.002846065,0.004277953,0.004342332,0.001248243],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00132776,"about_ca_system_score_gemma":0.00278932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00555571,"about_ca_topic_score_gemma":0.00544097,"domain_scores_codex":[0.9482378,0.03791582,0.002958756,0.005794766,0.003174051,0.001918704],"domain_scores_gemma":[0.8376669,0.09575703,0.01252099,0.04898298,0.003771687,0.001300479],"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":[0.009914925,0.001636893,0.6142774,0.00152396,0.007634765,0.001513879,0.002157036,0.03590084,0.002620289,0.07441451,0.0404781,0.2079274],"study_design_scores_gemma":[0.004066181,0.002601074,0.3381502,0.0008820434,0.01419109,0.002952134,0.001930508,0.2713861,0.02083444,0.2944925,0.04794988,0.0005638],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3187011,0.001166061,0.6317344,0.006843711,0.001640856,0.001680609,0.03107341,0.001312384,0.005847543],"genre_scores_gemma":[0.8089625,0.000505293,0.1527097,0.003158981,0.0008397461,0.00266249,0.02306252,0.0002363982,0.007862439],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.08936755,"threshold_uncertainty_score":0.4726264,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2009201840","doi":"10.1007/s10742-010-0063-1","title":"Surgery volume, quality of care and operative mortality in coronary artery bypass graft surgery: a re-examination using fixed-effects regression","year":2010,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Medicine; Surgery; Cardiac surgery; Cardiothoracic surgery; Cardiology","authors":[{"name":"Amresh Hanchate","is_ca":false},{"name":"Thérèse A. Stukel","is_ca":true},{"name":"John D. Birkmeyer","is_ca":false},{"name":"Arlene S. Ash","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2719739405695955,"gpt":0.5480294033068087,"spread":0.2760554627372133,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16398,0.004962544,0.01001214,0.007244812,0.002804683,0.005718075,0.01923236,0.005205687,0.02020303],"category_scores_gemma":[0.2151083,0.003800374,0.0330686,0.008254046,0.004591189,0.007535663,0.004096403,0.01084802,0.002913784],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004053464,"about_ca_system_score_gemma":0.005822915,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1092798,"about_ca_topic_score_gemma":0.06136913,"domain_scores_codex":[0.8801852,0.0786681,0.008806187,0.01900915,0.007795004,0.005536444],"domain_scores_gemma":[0.3982026,0.5106931,0.01811755,0.05648702,0.0121183,0.004381467],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01029058,0.001080088,0.8094274,0.001616516,0.1205301,0.00202231,0.001019712,0.01400669,0.0009767935,0.003092359,0.007704207,0.02823322],"study_design_scores_gemma":[0.005310187,0.007410109,0.7758681,0.0009955186,0.1044556,0.001795156,0.003357052,0.08003731,0.003031518,0.006446551,0.01040841,0.0008844818],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8763164,0.02249589,0.07703814,0.0043885,0.003818549,0.0006128678,0.009613319,0.001785175,0.003931257],"genre_scores_gemma":[0.9776884,0.001101734,0.009579153,0.0007669351,0.0008742316,0.0004000477,0.003433366,0.0004827555,0.00567325],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.16398,"threshold_uncertainty_score":0.8672194,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4414588014","doi":"10.1007/s10742-025-00359-4","title":"Psychometric performance of EQ-5D-5L and SF-6Dv2 in cancer patients","year":2025,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre Intégré Universitaire de Santé et de Services Sociaux du Centre-Sud-de-l'Île-de-Montréal; Université de Montréal","funders":"","keywords":"Intraclass correlation; Rank correlation; Receiver operating characteristic; Spearman's rank correlation coefficient; Correlation; Ceiling effect; Breast cancer; Convergent validity; Psychometrics","authors":[{"name":"Hosein Ameri","is_ca":true},{"name":"Thomas G. Poder","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7045539261009228,"gpt":0.5948502686643811,"spread":0.1097036574365418,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008265475,0.0004280415,0.000809943,0.001891268,0.0004836102,0.00186041,0.0006126732,0.001067757,0.001898982],"category_scores_gemma":[0.05163622,0.0002416324,0.001919609,0.002472262,0.0008419501,0.001672707,0.00168624,0.001116594,0.0004241388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001100625,"about_ca_system_score_gemma":0.0006678002,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004683945,"about_ca_topic_score_gemma":0.003100419,"domain_scores_codex":[0.9969156,0.001351174,0.0005509343,0.0002712317,0.0005581541,0.0003529047],"domain_scores_gemma":[0.9875157,0.00698769,0.002558207,0.0005528894,0.001608696,0.0007769404],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003057242,0.000104852,0.992569,0.0000286835,0.0001392571,0.0000231658,0.0003298815,0.0003223901,0.00004239688,0.00008750397,0.0002480995,0.005799027],"study_design_scores_gemma":[0.00003552789,0.0005229043,0.9968286,0.00003374117,0.00009256209,0.0001000733,0.0006699717,0.001078035,0.00007241182,0.0001746744,0.0003735578,0.00001792476],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9969369,0.0005523349,0.0001219075,0.0002442443,0.00002389939,0.00002340981,0.0006686714,0.000005887362,0.00142284],"genre_scores_gemma":[0.999074,0.0001035083,0.00006824406,0.00007349186,0.000009326252,0.00001768803,0.0005466345,0.000002706237,0.0001044949],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008265475,"threshold_uncertainty_score":0.04371256,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4401667657","doi":"10.1007/s10742-024-00335-4","title":"Deduplication methods for literature citations in systematic evidence reviews: practical insights to guide decision-making","year":2024,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Data deduplication; Computer science; Context (archaeology); Systematic review; Software; Data science; Management science; Data mining; Database; MEDLINE; Engineering","authors":[{"name":"Nicole Askin","is_ca":true},{"name":"George N. Okoli","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9338542154844445,"gpt":0.7958391692124195,"spread":0.138015046272025,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5102701,0.004996779,0.01334828,0.03545154,0.005652587,0.02249757,0.009603578,0.009654761,0.03809834],"category_scores_gemma":[0.8840939,0.006018454,0.01193996,0.0418088,0.005994363,0.01905224,0.01270774,0.01069172,0.01238371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00438827,"about_ca_system_score_gemma":0.02641058,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001371153,"about_ca_topic_score_gemma":0.003472446,"domain_scores_codex":[0.4039637,0.440452,0.1129809,0.01178091,0.02953841,0.001284092],"domain_scores_gemma":[0.04069663,0.8534312,0.02279303,0.04874662,0.03322514,0.001107406],"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":[0.0017533,0.0003540634,0.003893287,0.0676973,0.005893196,0.0005401511,0.006418275,0.002506547,0.001603097,0.03516888,0.08428419,0.7898877],"study_design_scores_gemma":[0.006437472,0.0009772384,0.007207145,0.06676304,0.01357648,0.002514807,0.003773554,0.04691529,0.01201931,0.5146894,0.3230548,0.002071546],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005658463,0.01703659,0.9076576,0.01395062,0.007628161,0.02290391,0.01064074,0.008633624,0.005890327],"genre_scores_gemma":[0.01383728,0.002209436,0.9606494,0.001503422,0.00106035,0.01620565,0.001844273,0.00111243,0.001577631],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4897299,"threshold_uncertainty_score":0.6039243,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3024135636","doi":"10.1007/s10742-023-00301-6","title":"Hierarchical causal variance decomposition for institution and provider comparisons in healthcare","year":2023,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Princess Margaret Cancer Centre; University Health Network; University of Toronto; Public Health Ontario","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Ontario Institute for Cancer Research","keywords":"Multinomial logistic regression; Multilevel model; Context (archaeology); Variance (accounting); Health administration; Health care; Variation (astronomy); Logistic regression; Medicine; Quality (philosophy); Statistics; Public health; Nursing; Mathematics; Geography","authors":[{"name":"Bo Chen","is_ca":true},{"name":"Kristen McAlpine","is_ca":true},{"name":"Keith A. Lawson","is_ca":true},{"name":"Antonio Finelli","is_ca":true},{"name":"Olli Saarela","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6102249160260278,"gpt":0.6500267204155956,"spread":0.03980180438956782,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05905682,0.001459133,0.003197036,0.004761084,0.002304438,0.003585799,0.003385601,0.001808916,0.01209445],"category_scores_gemma":[0.1910698,0.001341223,0.005177041,0.006524641,0.003434826,0.004390525,0.003955405,0.004336412,0.000775455],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002509708,"about_ca_system_score_gemma":0.00748815,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01163436,"about_ca_topic_score_gemma":0.01141083,"domain_scores_codex":[0.932596,0.0573143,0.001601449,0.004037931,0.003201808,0.001248428],"domain_scores_gemma":[0.7002099,0.2745732,0.004805698,0.01555751,0.003570668,0.001282973],"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":[0.0002707563,0.0004102065,0.01180088,0.0003981091,0.001423448,0.0001838309,0.001313153,0.02756886,0.0004115651,0.8268179,0.003466532,0.1259348],"study_design_scores_gemma":[0.00008250614,0.0001269585,0.004284056,0.0001185088,0.0002995974,0.00007394603,0.0003929466,0.2046237,0.000306751,0.787666,0.001979759,0.00004521705],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00925721,0.0002835448,0.9887212,0.0002559472,0.00006278331,0.0001728923,0.0002473173,0.0001472641,0.0008518769],"genre_scores_gemma":[0.3161379,0.0004186262,0.6776872,0.0002361999,0.000203144,0.00191245,0.000936211,0.0001901371,0.002278232],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.05905682,"threshold_uncertainty_score":0.312326,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4405356853","doi":"10.1007/s10742-024-00339-0","title":"Correction: Deduplication methods for literature citations in systematic evidence reviews: practical insights to guide decision-making","year":2024,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Academic Writing and Publishing","field":"Arts and Humanities","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Computer science; Management science; Data science; Data deduplication; Systematic review; Information retrieval; MEDLINE; Political science; Engineering; Database","authors":[{"name":"Nicole Askin","is_ca":true},{"name":"George N. Okoli","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.495383459484448,"gpt":0.6471978090623675,"spread":0.1518143495779195,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1757263,0.006727589,0.01181062,0.02787172,0.009371061,0.02350581,0.01152481,0.02350572,0.1208989],"category_scores_gemma":[0.8017381,0.008053214,0.007242806,0.0307803,0.01074963,0.01118996,0.0110306,0.0245527,0.05378887],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01391787,"about_ca_system_score_gemma":0.0409351,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01123169,"about_ca_topic_score_gemma":0.01737823,"domain_scores_codex":[0.6951751,0.1174335,0.1063987,0.01424141,0.06065983,0.006091482],"domain_scores_gemma":[0.1415591,0.4316449,0.05706949,0.04689258,0.3157759,0.007057913],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001487169,0.0000146179,0.0001607201,0.006652326,0.0002767626,0.0001629247,0.0004054558,0.00006603686,0.0001036311,0.001162941,0.9795601,0.01128589],"study_design_scores_gemma":[0.001647091,0.0001096911,0.002771159,0.03289656,0.00179431,0.0009250367,0.00111672,0.002140569,0.001218172,0.01490958,0.9398697,0.0006013989],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"methods","genre_scores_codex":[0.0004718001,0.004194117,0.01636842,0.1170564,0.840445,0.001927824,0.01193355,0.005138149,0.002464796],"genre_scores_gemma":[0.02889543,0.01516675,0.2113871,0.250471,0.3384918,0.03055274,0.01707935,0.02264041,0.08531545],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8242737,"threshold_uncertainty_score":0.9293404,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3205253370","doi":"10.1007/s10742-021-00260-w","title":"Initial validation of the global assessment of severity of illness","year":2021,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; University of Waterloo","funders":"Canadian Institutes of Health Research; Ontario Ministry of Research, Innovation and Science","keywords":"Health administration; Medicine; Public health; Pathology","authors":[{"name":"Braden K. Tompke","is_ca":true},{"name":"Ashok Chaurasia","is_ca":true},{"name":"Christopher M. Perlman","is_ca":true},{"name":"Kathy N. Speechley","is_ca":true},{"name":"Mark A. Ferro","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3486768783784317,"gpt":0.6126938030420473,"spread":0.2640169246636156,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1653127,0.0007880541,0.0009749858,0.002949291,0.001264203,0.002497615,0.002037113,0.001547749,0.003236976],"category_scores_gemma":[0.1964801,0.0004963516,0.002236659,0.001858878,0.002476325,0.002501666,0.004018062,0.001951143,0.0007272388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002454712,"about_ca_system_score_gemma":0.005201555,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00388506,"about_ca_topic_score_gemma":0.005369848,"domain_scores_codex":[0.93332,0.04595063,0.005832802,0.00359262,0.01007921,0.001224772],"domain_scores_gemma":[0.740553,0.1745237,0.008245679,0.01609304,0.05795521,0.002629388],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003316963,0.002493123,0.8338737,0.0008625618,0.0007795297,0.0001863341,0.01350263,0.001828137,0.004418258,0.01004948,0.004780393,0.1239089],"study_design_scores_gemma":[0.0006803604,0.007099805,0.9486386,0.0006959299,0.0004909622,0.0003347785,0.005702063,0.007063596,0.003655175,0.005120476,0.0204178,0.000100498],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8813023,0.001317958,0.05686991,0.002705482,0.001198791,0.01479667,0.003335674,0.0001503535,0.03832288],"genre_scores_gemma":[0.9411693,0.0003893474,0.0426167,0.0008864095,0.0001999187,0.01020202,0.002209521,0.0001197047,0.00220707],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1653127,"threshold_uncertainty_score":0.8742676,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1972735785","doi":"10.1007/s10742-013-0105-6","title":"Estimating variability for age-standardized hospital morbidity rates: a comparative study of automated methods for web-based atlas production using medical databases","year":2013,"lang":"en","type":"article","venue":"Health Services and Outcomes Research Methodology","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Agency of Canada; Western University; HIV Legal Network","funders":"Public Health Agency of Canada","keywords":"Statistics; Poisson distribution; Negative binomial distribution; Confidence interval; Overdispersion; Count data; Computer science; Mathematics; Econometrics","authors":[{"name":"Terry C. K. Lee","is_ca":true},{"name":"C. B. Dean","is_ca":true},{"name":"R Semenciw","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7566530610632348,"gpt":0.7132497391428269,"spread":0.04340332192040797,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0509776,0.0006586554,0.0008762906,0.006334265,0.0006527468,0.002606343,0.002179755,0.001038831,0.00102339],"category_scores_gemma":[0.1911662,0.0005400206,0.001309953,0.005974159,0.0007064167,0.002873978,0.00194999,0.0007293973,0.0004222868],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001696641,"about_ca_system_score_gemma":0.001371927,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01388637,"about_ca_topic_score_gemma":0.009906759,"domain_scores_codex":[0.9534908,0.02968339,0.003813477,0.00455997,0.007887681,0.0005647033],"domain_scores_gemma":[0.636584,0.2933702,0.02348495,0.01920365,0.02649681,0.0008603255],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002371392,0.0007032909,0.8315019,0.0004593581,0.001829154,0.00009354051,0.007208622,0.01016691,0.001672616,0.001721121,0.001653672,0.1406185],"study_design_scores_gemma":[0.0001871499,0.0009495705,0.9173144,0.0001672537,0.0005928896,0.0003864318,0.003924249,0.06785858,0.003247694,0.002371885,0.002834149,0.0001657324],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9456294,0.0005170117,0.04866078,0.000128075,0.00005910292,0.0003098837,0.00225963,0.0003572874,0.002078952],"genre_scores_gemma":[0.9651439,0.0002834511,0.03070731,0.00005866534,0.00004776796,0.0002793444,0.002879022,0.0001715179,0.0004290616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9490224,"threshold_uncertainty_score":0.2695985,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}