{"meta":{"query_hash":"7b0a4d4a1a0a","filters":{"venue":"Health Services and Outcomes Research Methodology"},"cohort_total":20,"direct_labels_cover":1,"predictions_cover":20,"exported":20,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/7b0a4d4a1a0a","api":"https://metacan.xera.ac/api/v1/cohort?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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.5064902954872634,"score_gpt":0.5967557162960009,"score_spread":0.09026542080873745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1529736910","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16243471,0.001753747,0.83146393,0.0011234619,0.00015820176,0.00074171403,0.0005690358,0.00087811664,0.00087712903],"genre_scores_gemma":[0.52805054,0.0014967879,0.4656482,0.00031597045,0.00013990201,0.001991939,0.0011456304,0.00056468765,0.0006463448],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9196442,0.071965106,0.002575608,0.002077368,0.003190332,0.0005473624],"domain_scores_gemma":[0.42923844,0.5499703,0.00685986,0.005922825,0.0071379575,0.0008705831],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13259605,0.002160182,0.002711534,0.004576101,0.0009902979,0.00315235,0.0036290782,0.0019208428,0.0014647947],"category_scores_gemma":[0.24672452,0.0009949224,0.00568708,0.0033396122,0.0011138397,0.003216119,0.0019206839,0.0027209453,0.00027993834],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007834301,0.0016329278,0.1442286,0.0014047843,0.01645254,0.00027509363,0.004038622,0.24205223,0.0019206024,0.06369144,0.004698992,0.5117699],"study_design_scores_gemma":[0.0005198221,0.0027196747,0.027865764,0.000310068,0.0017739122,0.0002204395,0.0012059411,0.9386736,0.000917324,0.02419033,0.0014163378,0.00018674677],"about_ca_topic_score_codex":0.010306101,"about_ca_topic_score_gemma":0.007971028,"teacher_disagreement_score":0.13259605,"about_ca_system_score_codex":0.0026035048,"about_ca_system_score_gemma":0.005172364,"threshold_uncertainty_score":0.7012433},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.7566530610632348,"score_gpt":0.7132497391428269,"score_spread":0.04340332192040797,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972735785","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94562936,0.0005170117,0.048660785,0.00012807499,0.000059102924,0.00030988368,0.0022596305,0.0003572874,0.0020789523],"genre_scores_gemma":[0.9651439,0.0002834511,0.030707313,0.00005866534,0.00004776796,0.00027934444,0.002879022,0.00017151794,0.0004290616],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9534908,0.02968339,0.0038134768,0.0045599705,0.007887681,0.0005647033],"domain_scores_gemma":[0.636584,0.29337025,0.023484949,0.01920365,0.02649681,0.0008603255],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0509776,0.00065865536,0.0008762906,0.0063342648,0.0006527468,0.002606343,0.0021797547,0.001038831,0.00102339],"category_scores_gemma":[0.19116616,0.0005400206,0.001309953,0.005974159,0.0007064167,0.0028739776,0.0019499896,0.0007293973,0.0004222868],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023713917,0.00070329086,0.8315019,0.0004593581,0.001829154,0.00009354051,0.0072086216,0.010166914,0.0016726161,0.0017211206,0.0016536718,0.14061846],"study_design_scores_gemma":[0.00018714994,0.0009495705,0.9173144,0.00016725372,0.0005928896,0.00038643178,0.003924249,0.06785858,0.0032476936,0.0023718853,0.0028341492,0.00016573245],"about_ca_topic_score_codex":0.013886367,"about_ca_topic_score_gemma":0.0099067595,"teacher_disagreement_score":0.9490224,"about_ca_system_score_codex":0.0016966406,"about_ca_system_score_gemma":0.0013719271,"threshold_uncertainty_score":0.26959854},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.2905204980326063,"score_gpt":0.5046460894679036,"score_spread":0.21412559143529725,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983766924","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0749219,0.0006883407,0.9143117,0.0003043575,0.00018687082,0.0072606415,0.00025106609,0.0003943572,0.001680723],"genre_scores_gemma":[0.3556452,0.00043388052,0.623637,0.00032534628,0.00011753857,0.018293403,0.00021719838,0.00021936135,0.0011110328],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.89186287,0.078339316,0.0063942573,0.010791705,0.011787288,0.0008244835],"domain_scores_gemma":[0.8331734,0.117290884,0.011852575,0.017324325,0.019534877,0.0008239881],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13902055,0.0017222491,0.0017374413,0.0024896716,0.001274092,0.001982798,0.0026372029,0.0028062074,0.0018445767],"category_scores_gemma":[0.19477548,0.0009848699,0.0022558426,0.002545099,0.0040660403,0.0014716667,0.001799239,0.0015979367,0.00070973707],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.030197756,0.0057424973,0.16992612,0.0042546773,0.0076052006,0.00045934564,0.0062628635,0.046269655,0.047685422,0.061333515,0.0038466337,0.6164163],"study_design_scores_gemma":[0.009213859,0.048304826,0.20880032,0.00090638036,0.013633139,0.0024699976,0.0015379104,0.42226994,0.16048664,0.10034217,0.030652711,0.0013821144],"about_ca_topic_score_codex":0.0012909097,"about_ca_topic_score_gemma":0.0011666935,"teacher_disagreement_score":0.86097944,"about_ca_system_score_codex":0.0013386173,"about_ca_system_score_gemma":0.0031859814,"threshold_uncertainty_score":0.7352197},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Medicine; Surgery; Cardiac surgery; Cardiothoracic surgery; Cardiology","score_opus":0.27197394056959545,"score_gpt":0.5480294033068087,"score_spread":0.27605546273721326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009201840","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8763164,0.02249589,0.07703814,0.0043885005,0.0038185485,0.0006128678,0.009613319,0.0017851746,0.0039312574],"genre_scores_gemma":[0.97768843,0.0011017345,0.009579153,0.00076693506,0.0008742316,0.0004000477,0.0034333656,0.00048275554,0.0056732497],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.88018525,0.0786681,0.008806187,0.019009152,0.0077950037,0.0055364436],"domain_scores_gemma":[0.39820257,0.5106931,0.018117549,0.056487016,0.012118298,0.0043814667],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16397999,0.0049625444,0.010012143,0.0072448123,0.0028046828,0.0057180747,0.01923236,0.005205687,0.020203028],"category_scores_gemma":[0.21510826,0.0038003742,0.033068605,0.008254046,0.004591189,0.0075356634,0.004096403,0.010848023,0.0029137835],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010290576,0.001080088,0.80942744,0.0016165158,0.12053007,0.0020223104,0.0010197123,0.014006686,0.0009767935,0.003092359,0.0077042067,0.028233219],"study_design_scores_gemma":[0.0053101867,0.007410109,0.7758681,0.0009955186,0.1044556,0.0017951559,0.003357052,0.08003731,0.003031518,0.006446551,0.010408408,0.0008844818],"about_ca_topic_score_codex":0.109279804,"about_ca_topic_score_gemma":0.06136913,"teacher_disagreement_score":0.16397999,"about_ca_system_score_codex":0.004053464,"about_ca_system_score_gemma":0.005822915,"threshold_uncertainty_score":0.86721945},"labels":[],"label_agreement":null},{"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.11969533947296526,"score_gpt":0.5064610241919099,"score_spread":0.38676568471894457,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030246929","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99060386,0.0005529642,0.0013458751,0.00029382404,0.000044346158,0.00021364435,0.00528299,0.000043941003,0.001618524],"genre_scores_gemma":[0.9926837,0.00033346796,0.0015391143,0.000095860014,0.0000138822925,0.000071099406,0.0040673777,0.000014681961,0.0011809451],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.994402,0.001730985,0.00041114894,0.00063923985,0.0017566669,0.0010599813],"domain_scores_gemma":[0.98779637,0.0034173739,0.001304796,0.0005820327,0.0056710783,0.0012282699],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006044831,0.000773661,0.0009540238,0.0030393677,0.0018459959,0.001965345,0.0033554938,0.0008975887,0.0013457143],"category_scores_gemma":[0.020327225,0.0006179349,0.0013487918,0.0048200823,0.00080801966,0.0005596589,0.0014989275,0.000796785,0.00014973387],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00092859543,0.000115084455,0.9814301,0.0000804115,0.0004489692,0.000058168927,0.00055096793,0.004088299,0.00019755484,0.000345327,0.0018655554,0.009890987],"study_design_scores_gemma":[0.00008807442,0.00012825658,0.9893885,0.000049706603,0.00021438049,0.000048478898,0.0012188053,0.0077244425,0.00024505114,0.00008005299,0.0007771714,0.00003708639],"about_ca_topic_score_codex":0.99408984,"about_ca_topic_score_gemma":0.9939464,"teacher_disagreement_score":0.025575306,"about_ca_system_score_codex":0.025575306,"about_ca_system_score_gemma":0.040917598,"threshold_uncertainty_score":0.18556273},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.39179876311714523,"score_gpt":0.5536682860750683,"score_spread":0.16186952295792312,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2045908763","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0056837327,0.00029660237,0.9920831,0.00063540094,0.00022136455,0.00009602813,0.0002259477,0.00021001493,0.00054780516],"genre_scores_gemma":[0.4254563,0.0013022908,0.54874116,0.0013728631,0.0007017833,0.002096018,0.0019403041,0.00040501915,0.017984387],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9780673,0.015676789,0.0011980928,0.0023625058,0.0018792199,0.00081607053],"domain_scores_gemma":[0.93084556,0.05699931,0.0029300996,0.0074227694,0.0016019463,0.00020029877],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.044353724,0.0013615419,0.002015768,0.0018742837,0.0006903872,0.0028118023,0.0035701678,0.0022173875,0.006126363],"category_scores_gemma":[0.12532465,0.0013541192,0.0036955855,0.003087253,0.0013428201,0.0025900344,0.0027062423,0.004887066,0.0010511881],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00083446584,0.00039527804,0.04844646,0.0008930605,0.0062744147,0.0014152622,0.00090984115,0.17570966,0.0011115223,0.5886145,0.0146065485,0.160789],"study_design_scores_gemma":[0.00059719855,0.0002674138,0.007803217,0.00029567676,0.0016834562,0.0005913904,0.00026060134,0.61948687,0.002237295,0.3464604,0.020187331,0.0001291296],"about_ca_topic_score_codex":0.007486538,"about_ca_topic_score_gemma":0.0055074114,"teacher_disagreement_score":0.044353724,"about_ca_system_score_codex":0.0010558964,"about_ca_system_score_gemma":0.0035125797,"threshold_uncertainty_score":0.2345677},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.2128823355536524,"score_gpt":0.5300256492860057,"score_spread":0.31714331373235327,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064626421","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.75338525,0.0014983288,0.23367608,0.00081494893,0.0002286735,0.001799154,0.0047652405,0.00030815933,0.003524199],"genre_scores_gemma":[0.84385306,0.00045238974,0.14967868,0.00013419765,0.00011390574,0.0016554557,0.0034582329,0.000058058202,0.0005959987],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9885655,0.007030941,0.0015561404,0.00083858136,0.0017022473,0.00030657506],"domain_scores_gemma":[0.9793554,0.01244432,0.0021116964,0.0025334111,0.0031813723,0.00037377526],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026161548,0.001159524,0.0021064498,0.005902037,0.0010554548,0.0016508633,0.0013806183,0.0011500234,0.0019176835],"category_scores_gemma":[0.07180535,0.00033478983,0.0047684945,0.0062587443,0.000540943,0.0014038179,0.003308131,0.0011028858,0.00032783987],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024149779,0.0010727437,0.5960993,0.0007228353,0.0077270153,0.00027619026,0.0015324279,0.010186037,0.0020486242,0.005216725,0.003694444,0.36900866],"study_design_scores_gemma":[0.00066949025,0.0073215123,0.8007094,0.00062066235,0.008957323,0.0010168384,0.0051880865,0.121036716,0.0069415895,0.03960782,0.0075657233,0.00036488686],"about_ca_topic_score_codex":0.006088033,"about_ca_topic_score_gemma":0.0072503835,"teacher_disagreement_score":0.026161548,"about_ca_system_score_codex":0.0012591736,"about_ca_system_score_gemma":0.002030045,"threshold_uncertainty_score":0.13835716},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.6899585491834566,"score_gpt":0.6385585072369091,"score_spread":0.05140004194654746,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2068564314","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003106876,0.000030825522,0.9993761,0.00005932269,0.000014170907,0.00003621499,0.00001864097,0.00005296324,0.000101176585],"genre_scores_gemma":[0.021637665,0.00013091741,0.9759746,0.00018021003,0.00007694469,0.00063825294,0.0002063503,0.00010400514,0.00105106],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9791566,0.015783338,0.0009714323,0.0020209292,0.001787831,0.0002798229],"domain_scores_gemma":[0.914182,0.072174445,0.0016730442,0.0080812555,0.0034402905,0.000448991],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040019304,0.0018512971,0.0028872986,0.00331327,0.0010659543,0.0022797803,0.0067456863,0.0029779405,0.008939011],"category_scores_gemma":[0.15365593,0.0021106547,0.0037967185,0.0040669283,0.0024987794,0.004121721,0.0062173726,0.0069878325,0.0019110753],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047260398,0.0003602321,0.0025974524,0.0006510273,0.0010985176,0.00032243502,0.0007003133,0.11616158,0.00378835,0.50037783,0.004518075,0.36895165],"study_design_scores_gemma":[0.00018277,0.00010818329,0.0005152515,0.00011329945,0.00020701511,0.00016557149,0.00007009716,0.6517258,0.0016228841,0.33999893,0.0052328566,0.000057359357],"about_ca_topic_score_codex":0.0041872873,"about_ca_topic_score_gemma":0.0045926534,"teacher_disagreement_score":0.040019304,"about_ca_system_score_codex":0.0010308052,"about_ca_system_score_gemma":0.0033248116,"threshold_uncertainty_score":0.21164483},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.7943857756584719,"score_gpt":0.6733009975553546,"score_spread":0.12108477810311735,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2080992413","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.043885965,0.0038292706,0.9358486,0.001338215,0.00042622336,0.004208582,0.0055654272,0.00037990956,0.004517854],"genre_scores_gemma":[0.29274768,0.0031962583,0.6402197,0.0010349365,0.00087369257,0.04446558,0.0058984803,0.00020853117,0.011355098],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.958975,0.03278714,0.0022499543,0.0026560554,0.0027906827,0.0005412471],"domain_scores_gemma":[0.83321756,0.13790454,0.012980949,0.010195308,0.00477596,0.0009256196],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06389008,0.0014147783,0.0013763615,0.005634215,0.0007848062,0.0019240351,0.002689668,0.001297604,0.014110555],"category_scores_gemma":[0.137796,0.0006857272,0.0031696938,0.004049433,0.000776546,0.0016947299,0.0017513611,0.0016939726,0.000967603],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002799245,0.0013848484,0.26380742,0.0026278298,0.0065127057,0.00031453808,0.0014990019,0.012714914,0.0013445624,0.13476296,0.018658586,0.5535734],"study_design_scores_gemma":[0.0041990047,0.004582106,0.2787044,0.0021591755,0.007378498,0.0017685079,0.0021496217,0.13521838,0.0040230597,0.47256586,0.08679712,0.00045423134],"about_ca_topic_score_codex":0.0022362438,"about_ca_topic_score_gemma":0.002164291,"teacher_disagreement_score":0.9361099,"about_ca_system_score_codex":0.0010410572,"about_ca_system_score_gemma":0.0028943138,"threshold_uncertainty_score":0.3378871},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.3935089949768288,"score_gpt":0.5154611245270454,"score_spread":0.12195212955021661,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W213901707","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.31870112,0.0011660606,0.63173443,0.006843711,0.0016408561,0.0016806091,0.031073414,0.0013123844,0.0058475425],"genre_scores_gemma":[0.8089625,0.000505293,0.15270968,0.0031589805,0.00083974615,0.00266249,0.023062518,0.00023639823,0.007862439],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.94823784,0.037915822,0.0029587557,0.0057947664,0.0031740514,0.0019187042],"domain_scores_gemma":[0.83766687,0.09575703,0.012520988,0.048982978,0.0037716867,0.0013004794],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08936755,0.0010295131,0.002935961,0.0028151616,0.0018881273,0.003375674,0.0056771846,0.0036177938,0.0064091813],"category_scores_gemma":[0.20221353,0.0011701771,0.004174964,0.0052300594,0.002299621,0.0028460652,0.004277953,0.0043423325,0.0012482431],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009914925,0.001636893,0.6142774,0.0015239604,0.0076347655,0.0015138786,0.0021570362,0.03590084,0.0026202889,0.074414514,0.040478095,0.20792736],"study_design_scores_gemma":[0.0040661814,0.0026010743,0.33815023,0.00088204345,0.014191095,0.0029521338,0.0019305078,0.27138606,0.020834439,0.29449254,0.047949884,0.0005638],"about_ca_topic_score_codex":0.00555571,"about_ca_topic_score_gemma":0.00544097,"teacher_disagreement_score":0.08936755,"about_ca_system_score_codex":0.0013277599,"about_ca_system_score_gemma":0.00278932,"threshold_uncertainty_score":0.47262645},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.3286456677312481,"score_gpt":0.5718448853666004,"score_spread":0.2431992176353523,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2317982607","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5748379,0.0062271194,0.40852788,0.0025534304,0.00066633435,0.0011136048,0.0026588624,0.0013691122,0.0020456929],"genre_scores_gemma":[0.8850235,0.0016319719,0.10786887,0.00051232876,0.00029957932,0.0013555599,0.002045276,0.00041059102,0.00085231796],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.916773,0.07400839,0.002700091,0.0028296644,0.002811135,0.00087776163],"domain_scores_gemma":[0.2610912,0.72062564,0.0067689857,0.0058289343,0.004809286,0.00087586907],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.15246028,0.002166028,0.0038207592,0.0050899806,0.00090603204,0.0037516265,0.0047994037,0.002878184,0.002428436],"category_scores_gemma":[0.2716369,0.0010992488,0.008700471,0.0032293948,0.0014050022,0.0034944292,0.001924547,0.003270314,0.00028333958],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.019784853,0.0021225724,0.12965381,0.001507286,0.017635861,0.0002337841,0.0013432218,0.6290781,0.00050107593,0.036726326,0.0050233267,0.15638977],"study_design_scores_gemma":[0.0006425183,0.0017991994,0.018342603,0.00017704803,0.0014690814,0.0001219539,0.0003192708,0.96350104,0.00019673926,0.0127651645,0.0005768218,0.00008847367],"about_ca_topic_score_codex":0.015244211,"about_ca_topic_score_gemma":0.0068851765,"teacher_disagreement_score":0.15246028,"about_ca_system_score_codex":0.0051395083,"about_ca_system_score_gemma":0.0051080096,"threshold_uncertainty_score":0.80629665},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.49650653711572157,"score_gpt":0.5636000211075388,"score_spread":0.06709348399181725,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W23346058","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.033148993,0.0057394453,0.9249055,0.016637884,0.0018828072,0.003852238,0.00090553676,0.00046547514,0.012462043],"genre_scores_gemma":[0.1461352,0.0015576021,0.8412946,0.0018257223,0.0008184361,0.0068714176,0.00045917238,0.00018342747,0.00085446145],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77407235,0.18810855,0.015770627,0.00456697,0.0162406,0.0012409167],"domain_scores_gemma":[0.56811905,0.33118957,0.012710689,0.047344085,0.038187873,0.0024488065],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.23548831,0.0028775106,0.006172922,0.006700026,0.006300772,0.015710225,0.0061975056,0.0027956644,0.0027561686],"category_scores_gemma":[0.5162024,0.0014583638,0.0058248728,0.009761938,0.008693581,0.009897714,0.007972778,0.010580184,0.00066404184],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042316678,0.0007732014,0.05296516,0.0033529517,0.0015554754,0.00036041008,0.015249703,0.013471539,0.0021655965,0.5055203,0.025885377,0.37827694],"study_design_scores_gemma":[0.00019206699,0.00090185436,0.045735445,0.0036859,0.0006677973,0.00061844906,0.017444437,0.09224183,0.001893874,0.80470777,0.03113668,0.00077393773],"about_ca_topic_score_codex":0.011531346,"about_ca_topic_score_gemma":0.020028116,"teacher_disagreement_score":0.23548831,"about_ca_system_score_codex":0.006101794,"about_ca_system_score_gemma":0.012053767,"threshold_uncertainty_score":0.942779},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.6729901541732592,"score_gpt":0.6682930787085176,"score_spread":0.004697075464741629,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2343900495","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00091403077,0.9978471,0.00035526112,0.00025251415,0.00009317308,0.00028399404,0.00011859108,0.0000018909693,0.000133469],"genre_scores_gemma":[0.015664507,0.9804292,0.0022338415,0.0005719165,0.000073977724,0.000819879,0.00013944942,0.000005020085,0.00006228552],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9477233,0.021477161,0.020564115,0.0030000915,0.0066132974,0.00062208733],"domain_scores_gemma":[0.8969713,0.08474012,0.01093517,0.0020052649,0.004747392,0.00060087506],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05606956,0.0015829614,0.008097636,0.016808223,0.0011931509,0.006611095,0.0034006936,0.0031371508,0.0021728023],"category_scores_gemma":[0.14076447,0.0015920085,0.008689173,0.01775999,0.0029195468,0.004101333,0.0038964048,0.0020283908,0.00017022101],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016663565,0.000093471645,0.0062623294,0.84135044,0.016535675,0.00011674197,0.0014341102,0.00017565774,0.00015122182,0.0021555107,0.0011703565,0.13038793],"study_design_scores_gemma":[0.00016247582,0.00023122199,0.0084449295,0.9166912,0.04654155,0.00043124252,0.0019995838,0.0001364954,0.00018610667,0.0015996601,0.023509687,0.00006581504],"about_ca_topic_score_codex":0.014389388,"about_ca_topic_score_gemma":0.040755793,"teacher_disagreement_score":0.94393045,"about_ca_system_score_codex":0.009090384,"about_ca_system_score_gemma":0.029494213,"threshold_uncertainty_score":0.29652774},"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":"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.8430232517073458,"score_gpt":0.6674423948286655,"score_spread":0.1755808568786803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889374799","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005778077,0.0005850529,0.99022746,0.0016014604,0.000115386705,0.00019756205,0.00060512335,0.00029034648,0.00059953507],"genre_scores_gemma":[0.43019396,0.0019179379,0.5516037,0.0016248542,0.0011695275,0.0031704595,0.0027215756,0.00023746934,0.007360514],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9677829,0.025238143,0.0012385504,0.0034599823,0.001436323,0.0008440358],"domain_scores_gemma":[0.73755103,0.24287288,0.006503494,0.009794982,0.0021230318,0.0011545843],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.050732896,0.0018976497,0.005038728,0.0041510463,0.0015640596,0.0042910613,0.0054570306,0.004392186,0.017659668],"category_scores_gemma":[0.18059744,0.002539259,0.0052613313,0.0044719833,0.004050018,0.0066280644,0.0045009493,0.008481104,0.0014096723],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000664085,0.00059437315,0.008638517,0.0010115393,0.0017854285,0.00034678422,0.00044667843,0.26126164,0.0002469391,0.5985048,0.0072469334,0.119252235],"study_design_scores_gemma":[0.00022699911,0.00005406824,0.00079587044,0.00010128712,0.00022612943,0.00005570704,0.00003107301,0.4518765,0.00014612271,0.545364,0.0010943934,0.000027757222],"about_ca_topic_score_codex":0.008350644,"about_ca_topic_score_gemma":0.008583022,"teacher_disagreement_score":0.050732896,"about_ca_system_score_codex":0.0030798877,"about_ca_system_score_gemma":0.005442534,"threshold_uncertainty_score":0.2683044},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.6102249160260278,"score_gpt":0.6500267204155956,"score_spread":0.03980180438956782,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3024135636","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00925721,0.0002835448,0.9887212,0.00025594723,0.00006278331,0.00017289228,0.00024731734,0.00014726407,0.0008518769],"genre_scores_gemma":[0.3161379,0.00041862624,0.67768717,0.00023619992,0.000203144,0.0019124504,0.000936211,0.00019013707,0.0022782325],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.932596,0.057314295,0.0016014492,0.0040379306,0.0032018083,0.0012484279],"domain_scores_gemma":[0.7002099,0.27457324,0.0048056976,0.015557511,0.003570668,0.0012829733],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05905682,0.0014591333,0.0031970355,0.0047610844,0.0023044376,0.0035857994,0.003385601,0.0018089156,0.012094453],"category_scores_gemma":[0.19106984,0.0013412227,0.0051770406,0.006524641,0.0034348264,0.0043905247,0.003955405,0.0043364116,0.000775455],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027075634,0.00041020653,0.0118008815,0.00039810906,0.0014234483,0.00018383085,0.001313153,0.027568864,0.00041156507,0.82681787,0.0034665323,0.12593484],"study_design_scores_gemma":[0.000082506136,0.00012695848,0.0042840564,0.00011850885,0.00029959742,0.00007394603,0.00039294656,0.20462368,0.000306751,0.78766596,0.0019797592,0.000045217046],"about_ca_topic_score_codex":0.011634359,"about_ca_topic_score_gemma":0.011410828,"teacher_disagreement_score":0.05905682,"about_ca_system_score_codex":0.002509708,"about_ca_system_score_gemma":0.00748815,"threshold_uncertainty_score":0.312326},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.3486768783784317,"score_gpt":0.6126938030420473,"score_spread":0.26401692466361565,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3205253370","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8813023,0.0013179583,0.056869913,0.0027054823,0.0011987909,0.0147966705,0.0033356738,0.00015035352,0.038322885],"genre_scores_gemma":[0.94116926,0.00038934738,0.0426167,0.00088640954,0.00019991872,0.010202016,0.0022095211,0.00011970467,0.0022070697],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.93332,0.045950633,0.0058328016,0.0035926197,0.010079205,0.0012247723],"domain_scores_gemma":[0.74055296,0.17452371,0.008245679,0.016093042,0.05795521,0.002629388],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1653127,0.00078805414,0.00097498577,0.0029492907,0.0012642033,0.002497615,0.002037113,0.0015477489,0.0032369755],"category_scores_gemma":[0.19648011,0.0004963516,0.0022366587,0.001858878,0.0024763248,0.0025016658,0.004018062,0.0019511433,0.0007272388],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003316963,0.002493123,0.83387375,0.0008625618,0.00077952974,0.00018633413,0.013502628,0.001828137,0.004418258,0.010049476,0.004780393,0.123908855],"study_design_scores_gemma":[0.00068036036,0.007099805,0.94863856,0.0006959299,0.00049096224,0.00033477845,0.0057020634,0.0070635956,0.0036551754,0.0051204762,0.0204178,0.000100498015],"about_ca_topic_score_codex":0.0038850596,"about_ca_topic_score_gemma":0.0053698476,"teacher_disagreement_score":0.1653127,"about_ca_system_score_codex":0.0024547123,"about_ca_system_score_gemma":0.0052015553,"threshold_uncertainty_score":0.8742676},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.8783707654311809,"score_gpt":0.649225151849151,"score_spread":0.2291456135820299,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388795592","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.58278507,0.0014571458,0.40421772,0.0019959852,0.00017102624,0.001941983,0.0012475182,0.0002577288,0.0059258323],"genre_scores_gemma":[0.9044993,0.00070570596,0.08913326,0.00032160353,0.00006169336,0.0016026042,0.00077210355,0.0000875728,0.0028161244],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9657358,0.030997697,0.00062035525,0.0012616281,0.0007136453,0.0006708605],"domain_scores_gemma":[0.5648664,0.4115831,0.0072503164,0.010834917,0.004100042,0.0013651176],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07097892,0.00168284,0.0033001835,0.0019461095,0.0011687034,0.0027770335,0.00369695,0.0036013622,0.011120949],"category_scores_gemma":[0.23914696,0.0014139102,0.003846386,0.003939694,0.002421249,0.004931827,0.0028012295,0.0052297176,0.00075818185],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009027547,0.0037250512,0.036106165,0.0008555762,0.0024242168,0.00081264775,0.0017305526,0.7381366,0.00030884924,0.15307914,0.005199931,0.04859371],"study_design_scores_gemma":[0.0020691752,0.00083392137,0.0021976822,0.00013726923,0.0005378081,0.00018479118,0.0003712347,0.9431131,0.00018700314,0.049290564,0.0010112292,0.000066170935],"about_ca_topic_score_codex":0.020276092,"about_ca_topic_score_gemma":0.0076389266,"teacher_disagreement_score":0.92902106,"about_ca_system_score_codex":0.002287161,"about_ca_system_score_gemma":0.0030178803,"threshold_uncertainty_score":0.37537688},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.9338542154844445,"score_gpt":0.7958391692124195,"score_spread":0.138015046272025,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401667657","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0056584626,0.017036594,0.9076576,0.013950623,0.0076281615,0.022903906,0.010640737,0.008633624,0.0058903266],"genre_scores_gemma":[0.013837282,0.0022094364,0.96064943,0.0015034222,0.0010603499,0.016205648,0.0018442733,0.0011124295,0.0015776311],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.40396369,0.44045198,0.112980895,0.01178091,0.029538414,0.001284092],"domain_scores_gemma":[0.040696632,0.85343117,0.022793032,0.04874662,0.033225138,0.001107406],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.51027006,0.004996779,0.013348283,0.035451535,0.0056525874,0.022497568,0.009603578,0.009654761,0.038098343],"category_scores_gemma":[0.8840939,0.0060184537,0.011939956,0.041808803,0.005994363,0.019052245,0.012707736,0.010691725,0.012383713],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017533004,0.00035406344,0.0038932872,0.0676973,0.0058931955,0.0005401511,0.0064182747,0.0025065471,0.0016030972,0.03516888,0.08428419,0.78988767],"study_design_scores_gemma":[0.0064374716,0.0009772384,0.0072071445,0.06676304,0.013576482,0.0025148066,0.0037735538,0.046915285,0.01201931,0.51468945,0.32305482,0.0020715462],"about_ca_topic_score_codex":0.001371153,"about_ca_topic_score_gemma":0.003472446,"teacher_disagreement_score":0.48972994,"about_ca_system_score_codex":0.00438827,"about_ca_system_score_gemma":0.026410576,"threshold_uncertainty_score":0.6039243},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.49538345948444795,"score_gpt":0.6471978090623675,"score_spread":0.1518143495779195,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405356853","genre_codex":"editorial","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004718001,0.004194117,0.016368425,0.11705637,0.840445,0.0019278241,0.01193355,0.005138149,0.0024647957],"genre_scores_gemma":[0.028895432,0.015166753,0.21138707,0.25047097,0.33849177,0.030552736,0.017079351,0.022640409,0.08531545],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.69517505,0.117433466,0.106398724,0.014241406,0.06065983,0.006091482],"domain_scores_gemma":[0.14155906,0.43164492,0.05706949,0.046892583,0.31577593,0.0070579126],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.17572626,0.006727589,0.011810617,0.027871717,0.009371061,0.023505813,0.011524808,0.023505721,0.1208989],"category_scores_gemma":[0.8017381,0.008053214,0.007242806,0.0307803,0.01074963,0.011189959,0.011030599,0.024552705,0.053788867],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014871692,0.000014617904,0.00016072005,0.006652326,0.00027676264,0.00016292473,0.00040545582,0.00006603686,0.00010363111,0.0011629412,0.97956014,0.011285887],"study_design_scores_gemma":[0.0016470915,0.00010969111,0.0027711587,0.03289656,0.0017943097,0.0009250367,0.0011167205,0.0021405688,0.0012181725,0.014909578,0.9398697,0.0006013989],"about_ca_topic_score_codex":0.011231691,"about_ca_topic_score_gemma":0.01737823,"teacher_disagreement_score":0.8242737,"about_ca_system_score_codex":0.013917867,"about_ca_system_score_gemma":0.040935095,"threshold_uncertainty_score":0.9293404},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.7045539261009228,"score_gpt":0.5948502686643811,"score_spread":0.10970365743654176,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414588014","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9969369,0.0005523349,0.0001219075,0.0002442443,0.000023899387,0.000023409812,0.00066867145,0.0000058873616,0.0014228402],"genre_scores_gemma":[0.999074,0.000103508275,0.00006824406,0.000073491865,0.000009326252,0.000017688026,0.0005466345,0.0000027062367,0.000104494946],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9969156,0.0013511741,0.0005509343,0.00027123172,0.0005581541,0.0003529047],"domain_scores_gemma":[0.9875157,0.0069876905,0.0025582067,0.0005528894,0.0016086961,0.0007769404],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008265475,0.00042804153,0.000809943,0.0018912681,0.0004836102,0.00186041,0.0006126732,0.0010677568,0.0018989815],"category_scores_gemma":[0.05163622,0.00024163241,0.0019196091,0.0024722621,0.00084195007,0.0016727073,0.0016862402,0.0011165938,0.0004241388],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030572424,0.000104851955,0.992569,0.000028683497,0.00013925711,0.0000231658,0.00032988147,0.0003223901,0.00004239688,0.00008750397,0.00024809947,0.005799027],"study_design_scores_gemma":[0.000035527893,0.0005229043,0.9968286,0.00003374117,0.00009256209,0.000100073274,0.00066997175,0.0010780346,0.00007241182,0.00017467442,0.00037355782,0.000017924756],"about_ca_topic_score_codex":0.004683945,"about_ca_topic_score_gemma":0.0031004187,"teacher_disagreement_score":0.008265475,"about_ca_system_score_codex":0.0011006251,"about_ca_system_score_gemma":0.0006678002,"threshold_uncertainty_score":0.043712556},"labels":[],"label_agreement":null}]}