{"meta":{"query_hash":"7272f52356ba","filters":{"venue":"Diagnostic and Prognostic Research"},"cohort_total":35,"direct_labels_cover":1,"predictions_cover":35,"exported":35,"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/7272f52356ba","api":"https://metacan.xera.ac/api/v1/cohort?venue=Diagnostic+and+Prognostic+Research"},"results":[{"id":"W2588203234","doi":"10.1186/s41512-016-0001-y","title":"Methods for Evaluating Medical Tests and Biomarkers","year":2017,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University; University of Calgary; McGill University; Jewish General Hospital; McMaster University; Population Health Research Institute","funders":"National Health and Medical Research Council; Haukeland Universitetssjukehus; Universiteit Leiden; Assistance publique-Hôpitaux de Paris; Leids Universitair Medisch Centrum; Universitetet i Bergen; Albert-Ludwigs-Universität Freiburg; Universiteit van Amsterdam; Abbott Diagnostics; University of Oxford; University of New South Wales; Medical Research Council; University of Notre Dame","keywords":"Equivalence (formal languages); Confidence interval; Statistics; Margin (machine learning); Sample (material); Medicine; Mathematics; Computer science; Chromatography; Machine learning; Chemistry","score_opus":0.269673097928087,"score_gpt":0.6196684686333537,"score_spread":0.34999537070526665,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2588203234","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":[0.0027006438,0.0020663675,0.97201216,0.0017761007,0.00035524726,0.003346362,0.0022678124,0.0013984344,0.014076797],"genre_scores_gemma":[0.03768568,0.0010076903,0.9466219,0.000607293,0.0002151455,0.009597639,0.0012092278,0.0002511868,0.0028041925],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8487698,0.10432931,0.01600471,0.009206624,0.020683372,0.0010062014],"domain_scores_gemma":[0.67762464,0.22552188,0.027728206,0.036750246,0.03013617,0.0022388988],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10114584,0.0027118756,0.0021970063,0.012904878,0.0018619017,0.009360328,0.0043480345,0.0031850806,0.025970768],"category_scores_gemma":[0.30524832,0.0012242215,0.003863116,0.0068457616,0.0047303843,0.007666673,0.006941277,0.003031822,0.0063533816],"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":[0.0008187509,0.00034037422,0.01805294,0.0045794672,0.0007959435,0.0002293367,0.002281838,0.0057658977,0.0017761621,0.22172219,0.02229698,0.72134006],"study_design_scores_gemma":[0.0005675708,0.00093329465,0.015743159,0.005789536,0.000935291,0.0013546484,0.0026433854,0.05655869,0.007942895,0.6713253,0.23564138,0.00056481623],"about_ca_topic_score_codex":0.0027538408,"about_ca_topic_score_gemma":0.0024989888,"teacher_disagreement_score":0.89885414,"about_ca_system_score_codex":0.0035626139,"about_ca_system_score_gemma":0.007567154,"threshold_uncertainty_score":0.53491676},"labels":[],"label_agreement":null},{"id":"W2594867411","doi":"10.1186/s41512-017-0008-z","title":"Systematic reviews of prognosis studies: a critical appraisal of five core clinical journals","year":2017,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Critical appraisal; Core (optical fiber); Systematic review; Medicine; MEDLINE; Political science; Computer science; Alternative medicine; Pathology","score_opus":0.980188190277515,"score_gpt":0.7828199083667856,"score_spread":0.19736828191072942,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2594867411","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0076786834,0.94040626,0.006271129,0.009976276,0.0027534992,0.028401924,0.0030501895,0.0002240484,0.0012379106],"genre_scores_gemma":[0.11782509,0.7322828,0.067059465,0.007499761,0.002062054,0.06906965,0.0036209472,0.00020965759,0.0003705504],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.41244707,0.18632956,0.29863635,0.011851766,0.08751674,0.003218543],"domain_scores_gemma":[0.1773329,0.49805868,0.15973158,0.019092377,0.1403906,0.005393912],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.361863,0.0045971815,0.019698564,0.08225175,0.0044606044,0.011626406,0.0065742694,0.0058182357,0.002495588],"category_scores_gemma":[0.68943715,0.005780329,0.021189528,0.056358654,0.0058266427,0.010395057,0.008728824,0.004788311,0.0005037762],"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":[0.0005929559,0.00004910131,0.004822063,0.9099673,0.021504,0.00018594424,0.0021239752,0.0002773218,0.00035696,0.0005819434,0.004664738,0.054873742],"study_design_scores_gemma":[0.0006627814,0.00033757457,0.012583956,0.9145578,0.054435775,0.00029755814,0.0011492146,0.000498578,0.0003762962,0.001077969,0.0138699245,0.00015254355],"about_ca_topic_score_codex":0.0098487865,"about_ca_topic_score_gemma":0.030304551,"teacher_disagreement_score":0.638137,"about_ca_system_score_codex":0.031449553,"about_ca_system_score_gemma":0.08038817,"threshold_uncertainty_score":0.7869365},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2602288447","doi":"10.1186/s41512-017-0011-4","title":"Erratum to: Methods for evaluating medical tests and biomarkers","year":2017,"lang":"en","type":"erratum","venue":"Diagnostic and Prognostic Research","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University; McGill University Health Centre; University of Calgary; McGill University; South Health Campus; Jewish General Hospital; McMaster University; Population Health Research Institute","funders":"Institute of Genetics; School of Public Health, Imperial College London; Klinisk Institut, Syddansk Universitet; Medical Research Council; Institute of Psychiatry, Psychology and Neuroscience, King’s College London; Syddansk Universitet; Odense Universitetshospital; KU Leuven; iMinds; Newcastle upon Tyne Hospitals NHS Foundation Trust; Cambridge University Hospitals; Loughborough University; University of Glasgow; Warwick Medical School; Universiteit van Amsterdam; Newcastle University; Albert-Ludwigs-Universität Freiburg; King's College London; University of Oxford; University of Warwick; Vrije Universiteit Amsterdam; Imperial College London; University of Exeter; Cancer Research UK; University of Southampton; University College London; University of Cambridge; British Heart Foundation; National Institute for Health and Care Excellence; University of Leeds; National Institute for Health and Care Research; Inselspital, Universitätsspital Bern; King's College Hospital NHS Foundation Trust","keywords":"Medical physics; Computer science; Psychology; Medicine","score_opus":0.12967390453280553,"score_gpt":0.5536366208162352,"score_spread":0.4239627162834297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2602288447","genre_codex":"editorial","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00012375775,0.005151265,0.004509293,0.058358803,0.925841,0.00004460155,0.0011370141,0.0004263964,0.004407905],"genre_scores_gemma":[0.00908964,0.03396173,0.037339855,0.19563599,0.5209396,0.00038796684,0.0066449023,0.0036662163,0.19233415],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99311066,0.0016147731,0.0016884946,0.0005545013,0.0028189349,0.00021274165],"domain_scores_gemma":[0.95861197,0.015239844,0.0017267013,0.0020154684,0.021330992,0.0010750405],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006102202,0.00251658,0.0018470945,0.0045240168,0.0024727713,0.004128838,0.0032257251,0.007160649,0.046312977],"category_scores_gemma":[0.08737381,0.0012854002,0.0018641887,0.0026561487,0.0025729144,0.0034329074,0.0022894843,0.012120951,0.039821014],"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":[0.000020694784,0.0000063496054,0.00008254445,0.00010838246,0.000012321387,0.00012166201,0.000014772937,0.00004591356,0.000029743362,0.0011157247,0.9873355,0.011106322],"study_design_scores_gemma":[0.00003684109,0.000023541405,0.0004168349,0.0006154316,0.000048431502,0.0009387551,0.0000718426,0.00039507594,0.00025010618,0.0038108535,0.9933427,0.000049650385],"about_ca_topic_score_codex":0.010075608,"about_ca_topic_score_gemma":0.019224145,"teacher_disagreement_score":0.046312977,"about_ca_system_score_codex":0.0027713135,"about_ca_system_score_gemma":0.0055528865,"threshold_uncertainty_score":0.15493238},"labels":[],"label_agreement":null},{"id":"W2605897081","doi":"10.1186/s41512-017-0012-3","title":"Validation of prediction models: examining temporal and geographic stability of baseline risk and estimated covariate effects","year":2017,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"National Institute of Neurological Disorders and Stroke; Canadian Institutes of Health Research; Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Covariate; Baseline (sea); Stability (learning theory); Statistics; Econometrics; Computer science; Mathematics; Machine learning; Biology","score_opus":0.3648981795435242,"score_gpt":0.46175307836447255,"score_spread":0.09685489882094833,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2605897081","genre_codex":"empirical","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":[0.6788971,0.00092093856,0.3166277,0.0010500271,0.000070734015,0.00023495301,0.00062439014,0.0002306277,0.0013435553],"genre_scores_gemma":[0.97666174,0.000087697204,0.022081695,0.00008560673,0.000033341574,0.00012340675,0.0007522373,0.000034868102,0.00013949996],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8815401,0.10527974,0.003287277,0.006399221,0.0026782036,0.0008155163],"domain_scores_gemma":[0.4270358,0.5178553,0.021146169,0.02666483,0.0063830297,0.0009149051],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1938306,0.0008487805,0.0011196235,0.0018155004,0.0012327724,0.0020649473,0.0026163815,0.0015470564,0.0007394753],"category_scores_gemma":[0.39661518,0.00072446634,0.002076548,0.0025070906,0.002788673,0.0030164463,0.003348582,0.002190442,0.00014628084],"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":[0.0011981166,0.00017741621,0.7869083,0.00016369869,0.0044052415,0.00037772924,0.0013150864,0.1573466,0.00062436843,0.007544541,0.0005520891,0.039386805],"study_design_scores_gemma":[0.00026631413,0.00068530184,0.21658413,0.00016137109,0.00066547334,0.00036012512,0.0010423795,0.75281954,0.0014103194,0.024292087,0.001621258,0.00009173146],"about_ca_topic_score_codex":0.017714605,"about_ca_topic_score_gemma":0.010954539,"teacher_disagreement_score":0.8061694,"about_ca_system_score_codex":0.0016422779,"about_ca_system_score_gemma":0.0028147053,"threshold_uncertainty_score":0.9941504},"labels":[],"label_agreement":null},{"id":"W2607440911","doi":"10.1186/s41512-017-0013-2","title":"Systematic review of the effects of care provided with and without diagnostic clinical prediction rules","year":2017,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Health and Medical Research Council; National Medical Research Council; Medical Research Council","keywords":"Medicine; Triage; Chest pain; Randomized controlled trial; Physical therapy; Emergency medicine; Pneumonia; Medical prescription; Intensive care medicine; Internal medicine","score_opus":0.22082966790995673,"score_gpt":0.5047728361852114,"score_spread":0.2839431682752547,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2607440911","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002396994,0.99490887,0.00026648532,0.00032597387,0.00022658003,0.00067486375,0.0008474533,0.000015795382,0.00033698772],"genre_scores_gemma":[0.06089808,0.93296766,0.0016799605,0.0010682034,0.00021849,0.0021778853,0.00074957183,0.000025657853,0.0002145376],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9737775,0.012500437,0.008585118,0.001729624,0.0029217442,0.00048553816],"domain_scores_gemma":[0.9292594,0.05432943,0.011015059,0.0011598092,0.0037032168,0.0005329779],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017189512,0.0021674754,0.016004385,0.008519645,0.00075991097,0.003439225,0.0023722877,0.0023173387,0.005473774],"category_scores_gemma":[0.091524936,0.0015577764,0.015344888,0.008396425,0.0013651015,0.0024667212,0.0017197817,0.0020449774,0.00038090226],"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":[0.0006088629,0.000015271153,0.0005514937,0.9448899,0.042649932,0.00005811501,0.000084621206,0.00010470452,0.00008480371,0.00013143328,0.0005790172,0.010241726],"study_design_scores_gemma":[0.0014946505,0.000481281,0.0044139586,0.629104,0.35536513,0.00023855933,0.00017499707,0.00013484838,0.00021187018,0.000396639,0.007934516,0.00004957829],"about_ca_topic_score_codex":0.011779741,"about_ca_topic_score_gemma":0.023848012,"teacher_disagreement_score":0.017189512,"about_ca_system_score_codex":0.0057651536,"about_ca_system_score_gemma":0.011510976,"threshold_uncertainty_score":0.09090793},"labels":[],"label_agreement":null},{"id":"W2770017713","doi":"10.1186/s41512-018-0028-3","title":"Using ordinal outcomes to construct and select biomarker combinations for single-level prediction","year":2018,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Abbott Diagnostics; Pritzker School of Medicine; University of California, San Francisco; National Institutes of Health; National Institute of Diabetes and Digestive and Kidney Diseases; National Heart, Lung, and Blood Institute; Institute for Clinical Evaluative Sciences; U.S. Department of Veterans Affairs","keywords":"Outcome (game theory); Logistic regression; Biomarker; Context (archaeology); Computer science; Selection (genetic algorithm); Construct (python library); Machine learning; Artificial intelligence; Data mining; Statistics; Mathematics","score_opus":0.5615408223281894,"score_gpt":0.5013232235527673,"score_spread":0.0602175987754221,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2770017713","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":[0.08894332,0.0006382087,0.9063482,0.000644081,0.000083547115,0.0006697503,0.00074638566,0.00077916856,0.0011473271],"genre_scores_gemma":[0.39641657,0.00025612864,0.6006636,0.00014203116,0.0001007706,0.0008163894,0.0011344859,0.000102922895,0.00036706147],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98797274,0.00775131,0.0011340136,0.001303141,0.0015432935,0.00029541468],"domain_scores_gemma":[0.9413252,0.043826707,0.00678677,0.0033663483,0.0037356757,0.0009594137],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024605153,0.002478328,0.0022521012,0.0069395322,0.0008663602,0.002839214,0.0014083646,0.0009988553,0.0029832984],"category_scores_gemma":[0.063579984,0.00069790933,0.0022507408,0.0042150375,0.00095894525,0.0021777102,0.0026842298,0.0027686884,0.0007908488],"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":[0.0021001587,0.000883544,0.26797578,0.0008581048,0.0023285965,0.00044356936,0.00066420366,0.22848368,0.0052902745,0.00830238,0.004514882,0.4781549],"study_design_scores_gemma":[0.00035567713,0.0009940958,0.028494049,0.00028269325,0.00071069144,0.0004065493,0.00025541303,0.9008104,0.005865056,0.057908606,0.0037146488,0.00020201],"about_ca_topic_score_codex":0.0011195973,"about_ca_topic_score_gemma":0.00172076,"teacher_disagreement_score":0.024605153,"about_ca_system_score_codex":0.0009678698,"about_ca_system_score_gemma":0.0020315729,"threshold_uncertainty_score":0.13012606},"labels":[],"label_agreement":null},{"id":"W2785428416","doi":"10.1186/s41512-018-0026-5","title":"The current application of the Royston-Parmar model for prognostic modeling in health research: a scoping review","year":2018,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Statistical Methods in Epidemiology","field":"Mathematics","cited_by":46,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Research Manitoba","keywords":"Current (fluid); Computer science; Risk analysis (engineering); Engineering; Management science; Data science; Medicine","score_opus":0.8581699158084114,"score_gpt":0.6892457691337626,"score_spread":0.16892414667464883,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2785428416","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":[0.0006566476,0.97140783,0.013845331,0.009425569,0.00088817056,0.0008219877,0.0006584458,0.00008406455,0.0022118818],"genre_scores_gemma":[0.01888452,0.94691336,0.026134467,0.003112563,0.00052933244,0.0032777013,0.00059728726,0.000056995857,0.00049377925],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.90585524,0.05370967,0.020101193,0.0056653135,0.013613119,0.001055476],"domain_scores_gemma":[0.58626604,0.36578915,0.02113464,0.0075592166,0.01849083,0.00076006696],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10490137,0.002101354,0.006856338,0.019401932,0.0019436651,0.012314002,0.0068858475,0.0067478684,0.005693909],"category_scores_gemma":[0.3607461,0.0023180146,0.011263768,0.029039348,0.0044347774,0.012944278,0.0049662776,0.004860011,0.0014391087],"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":[0.00015579093,0.000045746525,0.0028394146,0.5914551,0.0048351684,0.00028089553,0.0022906894,0.004502201,0.00017321936,0.0429068,0.02215596,0.32835904],"study_design_scores_gemma":[0.000084555555,0.0001255483,0.0021256788,0.8206779,0.00939163,0.0004690239,0.0012461114,0.0035247796,0.00019739657,0.02777611,0.13422847,0.000152775],"about_ca_topic_score_codex":0.017863324,"about_ca_topic_score_gemma":0.021239497,"teacher_disagreement_score":0.8950986,"about_ca_system_score_codex":0.013254528,"about_ca_system_score_gemma":0.039498057,"threshold_uncertainty_score":0.5547781},"labels":[],"label_agreement":null},{"id":"W2806302551","doi":"10.1186/s41512-018-0032-7","title":"Ruling out pulmonary embolism across different subgroups of patients and healthcare settings: protocol for a systematic review and individual patient data meta-analysis (IPDMA)","year":2018,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Venous Thromboembolism Diagnosis and Management","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Thrombosis and Atherosclerosis Research Institute; Ottawa Hospital; University of Ottawa","funders":"ZonMw; Ottawa Hospital Research Institute; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Heart and Stroke Foundation of Canada","keywords":"Medicine; Pulmonary embolism; MEDLINE; Protocol (science); Health care; Meta-analysis; Comorbidity; Intensive care medicine; Diagnosis code; Medical physics; Alternative medicine; Pathology; Internal medicine; Population","score_opus":0.3734064158965,"score_gpt":0.506414335431613,"score_spread":0.13300791953511298,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2806302551","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012194791,0.0082124155,0.0072235046,0.001283802,0.00077790295,0.9616724,0.018294927,0.00044729697,0.00086832384],"genre_scores_gemma":[0.0010991896,0.001057523,0.006076065,0.0003182449,0.000035097448,0.99015665,0.0010206953,0.000020249423,0.00021626634],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9470975,0.024740234,0.01841267,0.0044802567,0.0038912336,0.0013781571],"domain_scores_gemma":[0.9343714,0.035334118,0.011595284,0.00820291,0.009229634,0.0012666805],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07240707,0.005823644,0.018264556,0.011752933,0.002311156,0.006145025,0.0045985407,0.0063720155,0.04167273],"category_scores_gemma":[0.14480694,0.004267473,0.027739342,0.011477764,0.0030553434,0.0049748262,0.0042934455,0.0052567264,0.0044373283],"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":[0.0099886535,0.00020226126,0.0017330393,0.89407855,0.030401021,0.0005204297,0.0011761818,0.0020029445,0.0015267445,0.003784223,0.031918164,0.022667857],"study_design_scores_gemma":[0.076665744,0.0022625676,0.012608611,0.5287411,0.13425413,0.0008744673,0.0011462695,0.0052595087,0.003366811,0.023597265,0.21019866,0.0010249144],"about_ca_topic_score_codex":0.005231645,"about_ca_topic_score_gemma":0.009818644,"teacher_disagreement_score":0.07240707,"about_ca_system_score_codex":0.009828696,"about_ca_system_score_gemma":0.025940962,"threshold_uncertainty_score":0.38292974},"labels":[],"label_agreement":null},{"id":"W2885733420","doi":"10.1186/s41512-018-0038-1","title":"OPTimal IMAging strategy in patients suspected of non-traumatic pulmonary disease at the emergency department: chest X-ray or ultra-low-dose CT (OPTIMACT)—a randomised controlled trial chest X-ray or ultra-low-dose CT at the ED: design and rationale","year":2018,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Radiation Dose and Imaging","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Infection and Immunity","funders":"Universiteit van Amsterdam; ZonMw; Amsterdam University Medical Centers","keywords":"Medicine; Emergency department; Medical diagnosis; Radiology; Randomized controlled trial; Randomization; Emergency medicine; Surgery","score_opus":0.03896839631195267,"score_gpt":0.33409805108942914,"score_spread":0.2951296547774765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2885733420","genre_codex":"empirical","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9141847,0.007623339,0.0027502817,0.001558194,0.00322954,0.0668715,0.0012396354,0.00016294178,0.0023799331],"genre_scores_gemma":[0.88436955,0.0034122071,0.010495754,0.002747941,0.0027393522,0.09248996,0.00091013534,0.000042035736,0.0027928841],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.99165595,0.0056329467,0.0007273779,0.00089140394,0.00047826077,0.00061400526],"domain_scores_gemma":[0.9939902,0.002551895,0.0014695284,0.00036964522,0.0003623572,0.0012564498],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0082569625,0.0037198456,0.007816118,0.0011231718,0.00091748254,0.002562695,0.0022097016,0.005733728,0.009492113],"category_scores_gemma":[0.009922876,0.0018046577,0.003865897,0.00084583473,0.0033317364,0.0022787123,0.0012774938,0.0039883493,0.001291807],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.98964685,0.005874133,0.00021792678,0.00072087505,0.0004260253,0.0000127249405,0.000029260009,0.0001362967,0.00048699157,0.00009348561,0.00013050299,0.0022249303],"study_design_scores_gemma":[0.95557314,0.043196224,0.00027528626,0.000043570817,0.00029151683,0.0000075104012,0.00001323244,0.00020004403,0.0001236977,0.000116911215,0.0001488492,0.000010048121],"about_ca_topic_score_codex":0.00096575584,"about_ca_topic_score_gemma":0.0010742776,"teacher_disagreement_score":0.009492113,"about_ca_system_score_codex":0.0014658497,"about_ca_system_score_gemma":0.0026315327,"threshold_uncertainty_score":0.043667495},"labels":[],"label_agreement":null},{"id":"W2886902670","doi":"10.1186/s41512-018-0042-5","title":"Chronic Disease Population Risk Tool (CDPoRT): a study protocol for a prediction model that assesses population-based chronic disease incidence","year":2018,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Trillium Health Centre; University of Toronto; Toronto Public Health","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care","keywords":"Medicine; Population; Cohort; Disease; Incidence (geometry); Cohort study; Environmental health; Internal medicine","score_opus":0.13001671179050742,"score_gpt":0.45947348366280133,"score_spread":0.32945677187229394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886902670","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0052230186,0.00025773185,0.043575276,0.0015444921,0.0003768166,0.8509696,0.09005957,0.00096223946,0.007031246],"genre_scores_gemma":[0.005109991,0.00016529139,0.02902948,0.00034135015,0.000040555915,0.9420349,0.021753373,0.000076996104,0.0014481047],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97358096,0.016694179,0.004150329,0.0014044393,0.003443948,0.0007261395],"domain_scores_gemma":[0.9474094,0.017542947,0.004458299,0.008445959,0.020153966,0.0019894415],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.061855286,0.0015455377,0.0016896334,0.0022803326,0.003080488,0.0021179728,0.004544992,0.0023328764,0.039837763],"category_scores_gemma":[0.08191938,0.0020898737,0.002897289,0.003542773,0.0010927838,0.0018187419,0.00348482,0.0045987177,0.009552309],"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":[0.010751138,0.0052279225,0.044858783,0.0066755465,0.0008584177,0.00052603043,0.002009127,0.0077983183,0.0010003331,0.01564738,0.64648396,0.258163],"study_design_scores_gemma":[0.026826622,0.0059137656,0.11302977,0.007484603,0.0009541708,0.0005706173,0.0014035553,0.016318282,0.003015037,0.015899584,0.8080074,0.00057662325],"about_ca_topic_score_codex":0.030783698,"about_ca_topic_score_gemma":0.033462647,"teacher_disagreement_score":0.061855286,"about_ca_system_score_codex":0.0056009754,"about_ca_system_score_gemma":0.03267978,"threshold_uncertainty_score":0.3271259},"labels":[],"label_agreement":null},{"id":"W2910687296","doi":"10.1186/s41512-019-0047-8","title":"Development of a clinical decision support tool for diagnostic imaging use in patients with low back pain: a study protocol","year":2019,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; Health Sciences Centre; Queen's University; University of Ottawa; Nova Scotia Health Authority; Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Low back pain; Emergency department; Guideline; Cauda equina syndrome; Health care; Medical imaging; Back pain; Physical therapy; Radiology; Alternative medicine; Nursing; Pathology","score_opus":0.0558465045857909,"score_gpt":0.4294845113779942,"score_spread":0.3736380067922033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2910687296","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010701881,0.000089511064,0.0067765806,0.00053137133,0.00009255851,0.9773985,0.0028505272,0.00009098746,0.0014680893],"genre_scores_gemma":[0.007523403,0.00008166303,0.015461036,0.00020816746,0.000019662968,0.97573406,0.00074530113,0.000007182856,0.00021957321],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.95635045,0.027772786,0.0087426,0.0021217002,0.003368961,0.0016434176],"domain_scores_gemma":[0.9139913,0.038952947,0.009044574,0.0062567596,0.026905075,0.00484934],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09113914,0.0016415564,0.0022343104,0.0021068999,0.0034994972,0.0032934428,0.0027324765,0.0026877413,0.023748146],"category_scores_gemma":[0.10549574,0.0019924291,0.004446655,0.002801284,0.0015230776,0.0028635184,0.0027624916,0.0048801694,0.0045837425],"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":[0.08480492,0.08045541,0.116617,0.028195618,0.0023384308,0.0014268635,0.009334145,0.01438265,0.0016043759,0.01094905,0.11763692,0.53225464],"study_design_scores_gemma":[0.28516668,0.08287183,0.21802309,0.0345842,0.004385194,0.0009954404,0.012160908,0.061462883,0.010446984,0.016703695,0.2715948,0.001604246],"about_ca_topic_score_codex":0.0068425173,"about_ca_topic_score_gemma":0.010896848,"teacher_disagreement_score":0.09113914,"about_ca_system_score_codex":0.009255255,"about_ca_system_score_gemma":0.046739656,"threshold_uncertainty_score":0.48199558},"labels":[],"label_agreement":null},{"id":"W2933387395","doi":"10.1186/s41512-019-0052-y","title":"Challenges of rapid reviews for diagnostic test accuracy questions: a protocol for an international survey and expert consultation","year":2019,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Agency for Drugs and Technologies in Health; Queen's University; University of Toronto; St. Michael's Hospital","funders":"Instituto de Salud Carlos III","keywords":"Protocol (science); Test (biology); Management science; Systematic review; Computer science; Diagnostic test; Critical appraisal; Best practice; Data science; Risk analysis (engineering); Medicine; MEDLINE; Engineering; Pathology; Alternative medicine; Political science","score_opus":0.8737448761086614,"score_gpt":0.6505695240934505,"score_spread":0.22317535201521088,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2933387395","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00040881164,0.00024597792,0.008908466,0.0013909029,0.00030780627,0.98575217,0.0018596973,0.00022007112,0.0009061695],"genre_scores_gemma":[0.00030144598,0.0001395104,0.014483095,0.00030288682,0.000025572106,0.9843215,0.0001789698,0.000019027675,0.00022813003],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.5817737,0.32478857,0.05989414,0.007999686,0.01876206,0.006781819],"domain_scores_gemma":[0.43678278,0.32193047,0.040900066,0.06334532,0.125585,0.0114564095],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.47691876,0.0054947287,0.007261636,0.016062582,0.008255943,0.008709989,0.008038083,0.010978024,0.0737025],"category_scores_gemma":[0.45171395,0.0075593223,0.008078337,0.014774819,0.008057276,0.010761924,0.012347556,0.015528217,0.026780194],"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":[0.01139995,0.0039159534,0.0024262613,0.18294963,0.00095044303,0.002674424,0.05238914,0.0056206803,0.005093921,0.02755423,0.40360042,0.30142495],"study_design_scores_gemma":[0.030928863,0.0031223383,0.007426035,0.11941617,0.000668462,0.00087348116,0.020448355,0.006697287,0.0042698313,0.04499166,0.7599142,0.0012432676],"about_ca_topic_score_codex":0.00643039,"about_ca_topic_score_gemma":0.011308122,"teacher_disagreement_score":0.52308124,"about_ca_system_score_codex":0.022222823,"about_ca_system_score_gemma":0.11805574,"threshold_uncertainty_score":0.6450523},"labels":[],"label_agreement":null},{"id":"W3014535943","doi":"10.1186/s41512-020-00074-3","title":"State of the art in selection of variables and functional forms in multivariable analysis—outstanding issues","year":2020,"lang":"en","type":"editorial","venue":"Diagnostic and Prognostic Research","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":249,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"Erasmus+; Medical Research Council; Deutsche Forschungsgemeinschaft; European Commission","keywords":"Multivariable calculus; Selection (genetic algorithm); Computer science; Variable (mathematics); Variables; Feature selection; Management science; Artificial intelligence; Machine learning; Mathematics; Engineering","score_opus":0.11032277138808598,"score_gpt":0.4295645871614275,"score_spread":0.3192418157733415,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3014535943","genre_codex":"methods","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047169374,0.41482636,0.49156237,0.078218296,0.0027408162,0.0002329267,0.0005406033,0.00026595357,0.006895858],"genre_scores_gemma":[0.1611743,0.3689563,0.4336065,0.016685933,0.01467862,0.0014451928,0.00094564614,0.0005636618,0.0019438509],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.83496827,0.13124025,0.00916686,0.007848978,0.0157186,0.0010570303],"domain_scores_gemma":[0.37824196,0.58991545,0.0063236644,0.010659574,0.013939361,0.0009200153],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17081991,0.0016461093,0.004539645,0.0060210354,0.0016132056,0.009476868,0.007601034,0.005830069,0.006494063],"category_scores_gemma":[0.31950808,0.0013283374,0.004778094,0.0077686,0.013179285,0.011972407,0.004374461,0.01110705,0.001747695],"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":[0.00016317832,0.00013045517,0.0049086036,0.013999079,0.0005466767,0.00013789829,0.001443833,0.010195651,0.00021811355,0.500245,0.01603262,0.4519789],"study_design_scores_gemma":[0.000047180223,0.0002283732,0.004086702,0.019070506,0.00033420918,0.0002913757,0.001451738,0.027354868,0.0007228122,0.834441,0.111767225,0.00020402872],"about_ca_topic_score_codex":0.004380785,"about_ca_topic_score_gemma":0.0021392065,"teacher_disagreement_score":0.17081991,"about_ca_system_score_codex":0.0042090495,"about_ca_system_score_gemma":0.006604512,"threshold_uncertainty_score":0.9033928},"labels":[],"label_agreement":null},{"id":"W3097626155","doi":"10.1186/s41512-020-00086-z","title":"A study protocol for a predictive algorithm to assess population-based premature mortality risk: Premature Mortality Population Risk Tool (PreMPoRT)","year":2020,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Bruyère; Ottawa Hospital; University of Ottawa; Institute for Clinical Evaluative Sciences; Statistics Canada; Public Health Ontario; University of Toronto","funders":"Economic and Social Research Council; Canadian Institutes of Health Research; University of Toronto; Canada Excellence Research Chairs, Government of Canada; UK Research and Innovation","keywords":"Medicine; Population; Population health; Cohort; Community health; Incidence (geometry); Demography; Public health; Environmental health; Internal medicine","score_opus":0.1776897348669443,"score_gpt":0.4730064597835983,"score_spread":0.29531672491665395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3097626155","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002227344,0.00023569984,0.047725912,0.000753578,0.0006213056,0.9221386,0.021758609,0.0008530204,0.0036859566],"genre_scores_gemma":[0.0016110331,0.00007966742,0.028003408,0.0002771461,0.00004668196,0.96656466,0.0026504132,0.000060709302,0.0007063065],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9782906,0.014108269,0.0030007672,0.0017276615,0.0023085354,0.00056412385],"domain_scores_gemma":[0.9352818,0.025980815,0.0035635612,0.010147363,0.022757314,0.002269066],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.061268903,0.0016547344,0.0015325375,0.0015880021,0.002459221,0.0023423277,0.0033164078,0.002809174,0.14183556],"category_scores_gemma":[0.12292235,0.001788489,0.0026280931,0.0015545002,0.001206111,0.0014743909,0.0019967717,0.0044077104,0.02642219],"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":[0.027921187,0.0076168585,0.008744409,0.017821683,0.0010058574,0.0006386269,0.0018069675,0.005438143,0.0025432117,0.016585406,0.4754188,0.43445888],"study_design_scores_gemma":[0.06485084,0.0131958695,0.016832614,0.020302847,0.0011988886,0.0006916429,0.0011570773,0.014790723,0.0073146415,0.026363693,0.83278996,0.0005112317],"about_ca_topic_score_codex":0.0038510514,"about_ca_topic_score_gemma":0.00505924,"teacher_disagreement_score":0.14183556,"about_ca_system_score_codex":0.0027844235,"about_ca_system_score_gemma":0.018136881,"threshold_uncertainty_score":0.4744872},"labels":[],"label_agreement":null},{"id":"W3135659546","doi":"10.1186/s41512-021-00095-6","title":"Diabetes after pregnancy: a study protocol for the derivation and validation of a risk prediction model for 5-year risk of diabetes following pregnancy","year":2021,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Gestational Diabetes Research and Management","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Canadian VIGOUR Centre; University of Alberta; Health Sciences Centre; Sunnybrook Health Science Centre; Sinai Health System; St. Michael's Hospital; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto; Women's College Hospital","funders":"Physicians' Services Incorporated Foundation; Diabetes Action Canada; Ontario Ministry of Health and Long-Term Care; Diabetes Canada; Diabetes Action Research and Education Foundation","keywords":"Pregnancy; Medicine; Gestational diabetes; Population; Body mass index; Obstetrics; Percentile; Cohort; Demography; Statistics; Gestation; Internal medicine; Environmental health; Mathematics","score_opus":0.05755973319139719,"score_gpt":0.37732000273959587,"score_spread":0.3197602695481987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3135659546","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0039975904,0.00010777314,0.0074725286,0.00020728183,0.00015143355,0.97826034,0.008253728,0.0000740779,0.001475207],"genre_scores_gemma":[0.0018267888,0.000090222326,0.0073598316,0.00010745898,0.000020925658,0.98852855,0.001645702,0.0000061447495,0.00041434012],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98786587,0.007337513,0.001873461,0.0010561671,0.0013550957,0.0005118596],"domain_scores_gemma":[0.96976626,0.0093667675,0.0030134465,0.005613858,0.010811901,0.0014277584],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.050199494,0.001932052,0.002064126,0.0013646914,0.003800766,0.0021695367,0.003114613,0.0026575702,0.046754],"category_scores_gemma":[0.055228002,0.0022629856,0.0028982756,0.0018209005,0.0014266546,0.0011267854,0.0018235059,0.0037065274,0.009271635],"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":[0.101880394,0.039703585,0.037410486,0.030333309,0.0020982958,0.002999848,0.0075705373,0.01685246,0.006100801,0.028080195,0.3291846,0.39778545],"study_design_scores_gemma":[0.13483128,0.039156873,0.10697981,0.025784733,0.0023221923,0.0011376465,0.0036726287,0.015745578,0.006319312,0.017700188,0.64543074,0.0009189876],"about_ca_topic_score_codex":0.01603442,"about_ca_topic_score_gemma":0.027480846,"teacher_disagreement_score":0.050199494,"about_ca_system_score_codex":0.0063447626,"about_ca_system_score_gemma":0.03508885,"threshold_uncertainty_score":0.2654835},"labels":[],"label_agreement":null},{"id":"W3149528813","doi":"10.1186/s41512-021-00097-4","title":"PRISMA-DTA for Abstracts: a new addition to the toolbox for test accuracy research","year":2021,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"University of Ottawa","keywords":"Toolbox; Guideline; Test (biology); Consolidated Standards of Reporting Trials; Observational study; Systematic review; Computer science; Medical physics; Clinical trial; MEDLINE; Information retrieval; Medicine; Data science; Political science; Pathology","score_opus":0.9094318379964861,"score_gpt":0.6543724124729947,"score_spread":0.25505942552349137,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3149528813","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011957383,0.022121238,0.56083643,0.10103817,0.073404744,0.095094465,0.074835375,0.05464948,0.016824404],"genre_scores_gemma":[0.005197456,0.010282126,0.69978195,0.013520939,0.0067960597,0.24079953,0.010865849,0.00882445,0.0039316453],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.19881806,0.5331444,0.20765969,0.01297818,0.044957787,0.0024419336],"domain_scores_gemma":[0.06931784,0.6044864,0.09467872,0.11909628,0.106579475,0.005841235],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.61233526,0.008943642,0.020733943,0.04200069,0.0052084713,0.028092008,0.018874414,0.014671904,0.10046143],"category_scores_gemma":[0.7876831,0.015270619,0.031049011,0.04891683,0.01268353,0.025505483,0.026672332,0.043237656,0.06072686],"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":[0.0015078505,0.00019503315,0.00065415737,0.12342963,0.00553443,0.00024098287,0.0027353242,0.0011923707,0.00087146845,0.02536655,0.6625742,0.17569804],"study_design_scores_gemma":[0.007230876,0.00036036962,0.003155172,0.094614334,0.0030915297,0.00078313285,0.0011775062,0.0055418215,0.0020570138,0.115963794,0.7644878,0.0015366308],"about_ca_topic_score_codex":0.0029917548,"about_ca_topic_score_gemma":0.003850733,"teacher_disagreement_score":0.38766474,"about_ca_system_score_codex":0.016659226,"about_ca_system_score_gemma":0.08147878,"threshold_uncertainty_score":0.4780597},"labels":[],"label_agreement":null},{"id":"W3165628081","doi":"10.1186/s41512-021-00101-x","title":"The development and validation of a prognostic model to PREDICT Relapse of depression in adult patients in primary care: protocol for the PREDICTR study","year":2021,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Mental Health Treatment and Access","field":"Psychology","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"NIHR School for Primary Care Research; National Institute for Health and Care Research; University College London; Department of Health and Social Care; Wellcome Trust","keywords":"Depression (economics); Medicine; Logistic regression; Anxiety; Quality of life (healthcare); Optimism; Cohort; Internal medicine; Psychiatry; Psychology; Psychotherapist","score_opus":0.08904236318636065,"score_gpt":0.44175638309321086,"score_spread":0.3527140199068502,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3165628081","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011489712,0.00013894062,0.009101904,0.00020886878,0.00010367599,0.97407794,0.003831175,0.00011868773,0.00092903146],"genre_scores_gemma":[0.0055420604,0.00005164564,0.005718205,0.000040432416,0.0000107742635,0.98762834,0.0008889069,0.000006332715,0.00011323277],"study_design_codex":"randomized_trial","study_design_gemma":"not_applicable","domain_scores_codex":[0.9634966,0.028395742,0.0030190465,0.0018872864,0.0020992572,0.001102106],"domain_scores_gemma":[0.96775734,0.011297446,0.004698344,0.0066427975,0.007978026,0.0016259775],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07258788,0.003307128,0.0038512216,0.0011890472,0.0029225769,0.0019666643,0.0035661096,0.0031283738,0.020033035],"category_scores_gemma":[0.07050867,0.0025179281,0.006131452,0.0017120155,0.0023277304,0.0018247336,0.00324715,0.0051445724,0.005103761],"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":[0.47051254,0.07010504,0.055543143,0.03080486,0.0070224525,0.0013870677,0.004635493,0.057488136,0.0035321144,0.010536528,0.072919905,0.2155128],"study_design_scores_gemma":[0.5575797,0.13211885,0.10121233,0.012130091,0.0069403783,0.0006213363,0.001713623,0.063523956,0.006293011,0.011143235,0.10563356,0.0010899719],"about_ca_topic_score_codex":0.002907104,"about_ca_topic_score_gemma":0.0031356763,"teacher_disagreement_score":0.07258788,"about_ca_system_score_codex":0.0038897383,"about_ca_system_score_gemma":0.015943598,"threshold_uncertainty_score":0.38388598},"labels":[],"label_agreement":null},{"id":"W3181971677","doi":"10.1186/s41512-021-00102-w","title":"A relationship between the incremental values of area under the ROC curve and of area under the precision-recall curve","year":2021,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":57,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; St. Jude Children's Research Hospital","keywords":"Brier score; Receiver operating characteristic; Range (aeronautics); Statistics; Area under the curve; Mathematics; Area under curve; Recall; Value (mathematics); Computer science; Econometrics; Medicine; Internal medicine; Psychology; Cognitive psychology","score_opus":0.3304354519580382,"score_gpt":0.438720882686613,"score_spread":0.10828543072857483,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3181971677","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":[0.6105425,0.019689972,0.34249792,0.002732524,0.00076021464,0.00054204947,0.0025351858,0.0030891132,0.017610542],"genre_scores_gemma":[0.95535934,0.00076839793,0.04180637,0.00017489372,0.00024964457,0.00026412774,0.00078726537,0.00013878748,0.0004511943],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97560215,0.010109198,0.002246259,0.0041854135,0.006984132,0.0008728825],"domain_scores_gemma":[0.5935072,0.34226543,0.022566516,0.014262894,0.025109727,0.0022883513],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03270181,0.0012424357,0.000981963,0.007098166,0.00048125323,0.003493252,0.001503172,0.0024680216,0.0013369304],"category_scores_gemma":[0.25189242,0.0004657895,0.0010070902,0.003948982,0.0021852946,0.0048751514,0.0015326521,0.0021786927,0.00052796304],"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":[0.0016933345,0.00025265763,0.58525276,0.0010352425,0.0015972521,0.0010441678,0.0012381607,0.110305786,0.0048429603,0.026255522,0.0073256595,0.2591565],"study_design_scores_gemma":[0.00018252495,0.002941883,0.29212925,0.00068019173,0.0011954256,0.0066073705,0.0014015932,0.5971364,0.0117629375,0.07364394,0.011690764,0.00062765495],"about_ca_topic_score_codex":0.0010981293,"about_ca_topic_score_gemma":0.00060336397,"teacher_disagreement_score":0.03270181,"about_ca_system_score_codex":0.0015714997,"about_ca_system_score_gemma":0.0009974669,"threshold_uncertainty_score":0.17294574},"labels":[],"label_agreement":null},{"id":"W4200316566","doi":"10.1186/s41512-021-00109-3","title":"Knowledge translation of prediction rules: methods to help health professionals understand their trade-offs","year":2021,"lang":"en","type":"letter","venue":"Diagnostic and Prognostic Research","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Computer science; Translation (biology); Knowledge translation; Health professionals; Knowledge management; Process management; Data science; Health care; Business; Political science; Chemistry","score_opus":0.2502209112555136,"score_gpt":0.48283624440675066,"score_spread":0.23261533315123706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200316566","genre_codex":"methods","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030044378,0.00066061557,0.98054105,0.0025957576,0.00026556585,0.000918585,0.0017619237,0.006080559,0.0041715275],"genre_scores_gemma":[0.032260094,0.0008027699,0.9589614,0.0007541595,0.00022989152,0.0018847281,0.0027189231,0.00077811355,0.0016099387],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9471722,0.030548582,0.007224476,0.0060138265,0.00844767,0.0005931716],"domain_scores_gemma":[0.6718249,0.26865232,0.015187296,0.022212354,0.021091696,0.0010314796],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.059670217,0.003436173,0.002757288,0.008523722,0.0013947842,0.010924059,0.0045232396,0.0043424293,0.01696431],"category_scores_gemma":[0.32892013,0.0017952287,0.0033724685,0.0076193134,0.0029731109,0.011793765,0.007977029,0.005580645,0.008661218],"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":[0.0006331522,0.00045135844,0.0027297584,0.0035233945,0.0003984767,0.0005666786,0.005571278,0.014097665,0.0020813379,0.051657375,0.02578124,0.8925082],"study_design_scores_gemma":[0.00053783355,0.000361434,0.0022249157,0.0033994766,0.00054367824,0.000585097,0.0029053,0.23539905,0.009118187,0.63421714,0.110245086,0.0004627333],"about_ca_topic_score_codex":0.0037424092,"about_ca_topic_score_gemma":0.0030737824,"teacher_disagreement_score":0.9403298,"about_ca_system_score_codex":0.003026157,"about_ca_system_score_gemma":0.0065411953,"threshold_uncertainty_score":0.31557006},"labels":[],"label_agreement":null},{"id":"W4205154061","doi":"10.1186/s41512-021-00114-6","title":"Graphical calibration curves and the integrated calibration index (ICI) for competing risk models","year":2022,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":45,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Patient-Centered Outcomes Research Institute; Heart and Stroke Foundation of Canada","keywords":"Calibration; Statistics; Hazard; Monte Carlo method; Regression analysis; Proportional hazards model; Outcome (game theory); Range (aeronautics); Computer science; Econometrics; Data mining; Mathematics; Engineering","score_opus":0.149392363648987,"score_gpt":0.38683802710269627,"score_spread":0.23744566345370927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205154061","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":[0.020242061,0.00076450215,0.974312,0.0005800561,0.00005804636,0.00022207857,0.00038939927,0.0007892562,0.0026425705],"genre_scores_gemma":[0.56694955,0.00093597436,0.42771676,0.00044145162,0.00015044522,0.00094671437,0.0013023296,0.00052468467,0.0010320048],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9784939,0.014428731,0.0008875141,0.0019223297,0.003687949,0.0005794172],"domain_scores_gemma":[0.7708213,0.1863211,0.016535167,0.0135028055,0.011573548,0.0012461332],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.044136085,0.0022552,0.0013846733,0.006937557,0.00079980004,0.002945545,0.0029134552,0.0029133374,0.004089897],"category_scores_gemma":[0.23686752,0.0007960355,0.002485631,0.003701672,0.0034924936,0.0047296663,0.003708359,0.0050093904,0.000697662],"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":[0.00020241395,0.00010368131,0.021352299,0.00031944102,0.0002721417,0.0001761631,0.00040964517,0.8190425,0.0010172836,0.07503587,0.0029572148,0.07911133],"study_design_scores_gemma":[0.000028080538,0.0001285825,0.004398142,0.00014791077,0.00004556648,0.00020865048,0.000072321585,0.9286774,0.0010179757,0.06323564,0.001945227,0.00009456688],"about_ca_topic_score_codex":0.004015231,"about_ca_topic_score_gemma":0.0015540001,"teacher_disagreement_score":0.9558639,"about_ca_system_score_codex":0.003058638,"about_ca_system_score_gemma":0.0021238816,"threshold_uncertainty_score":0.23341674},"labels":[],"label_agreement":null},{"id":"W4280518513","doi":"10.1186/s41512-022-00123-z","title":"Quality and transparency of reporting derivation and validation prognostic studies of recurrent stroke in patients with TIA and minor stroke: a systematic review","year":2022,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"University of Ottawa","keywords":"Medicine; Stroke (engine); MEDLINE; Predictive modelling; Computer science; Machine learning","score_opus":0.2699960213258003,"score_gpt":0.4699169750905324,"score_spread":0.19992095376473207,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280518513","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009121124,0.97287685,0.004134371,0.003453151,0.0009839566,0.0041517084,0.0038936154,0.00006878954,0.0013164979],"genre_scores_gemma":[0.209949,0.7496956,0.012610512,0.004764859,0.0006816625,0.018243628,0.0034413172,0.00013152973,0.00048192468],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.65851426,0.12873326,0.17317627,0.009080834,0.028657727,0.0018377467],"domain_scores_gemma":[0.2507519,0.5710361,0.12301645,0.015463099,0.03836357,0.0013689228],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19734569,0.0017710337,0.010377011,0.015771076,0.0014255919,0.007194793,0.0043864166,0.003469571,0.0032174797],"category_scores_gemma":[0.606652,0.0027045105,0.011193985,0.016194787,0.004332384,0.007023775,0.00450071,0.0020787448,0.00038319206],"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":[0.00038566953,0.00001247244,0.0040314323,0.9561709,0.0137837175,0.00010269167,0.0009811106,0.00015961983,0.00015583454,0.00044412393,0.0015774155,0.022195084],"study_design_scores_gemma":[0.00034822652,0.000107152184,0.006338185,0.9333645,0.048508972,0.00023141515,0.0006296797,0.0002103702,0.0003485867,0.00074369984,0.009102857,0.00006630245],"about_ca_topic_score_codex":0.00998241,"about_ca_topic_score_gemma":0.0140768895,"teacher_disagreement_score":0.8026543,"about_ca_system_score_codex":0.008481425,"about_ca_system_score_gemma":0.02146434,"threshold_uncertainty_score":0.98981565},"labels":[],"label_agreement":null},{"id":"W4282960968","doi":"10.1186/s41512-022-00125-x","title":"Bayesian latent class analysis produced diagnostic accuracy estimates that were more interpretable than composite reference standards for extrapulmonary tuberculosis tests","year":2022,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé; Canada Excellence Research Chairs, Government of Canada; Canadian Institutes of Health Research; Deutsches Zentrum für Infektionsforschung","keywords":"Gold standard (test); Extrapulmonary tuberculosis; Latent class model; Bayesian probability; Statistics; Medicine; Tuberculosis; Mycobacterium tuberculosis; Mathematics; Pathology","score_opus":0.07682152450935713,"score_gpt":0.4131796994782631,"score_spread":0.33635817496890597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4282960968","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":[0.25648373,0.0023983258,0.725954,0.0009283781,0.00029001263,0.0005267933,0.0047573475,0.0020641005,0.00659728],"genre_scores_gemma":[0.82227623,0.00031587627,0.17223874,0.00029700133,0.00006382967,0.000313796,0.003472385,0.00037829424,0.0006438179],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97588784,0.013642497,0.0018450264,0.004746486,0.0032078314,0.0006702507],"domain_scores_gemma":[0.8573534,0.10725574,0.011181984,0.011865084,0.0117167635,0.0006270192],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.056550007,0.0013118347,0.0019765508,0.004535632,0.00071687705,0.004159942,0.0013278674,0.001753605,0.0031145494],"category_scores_gemma":[0.16298507,0.0006718868,0.0029679874,0.0021699427,0.0013873052,0.002517805,0.001907791,0.0028878038,0.00111024],"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":[0.0028700049,0.00031752003,0.38033596,0.0016320233,0.0025946451,0.00042471412,0.0022161535,0.21057126,0.011144559,0.022165045,0.009414737,0.3563134],"study_design_scores_gemma":[0.00052861974,0.0013075918,0.17442518,0.0012217608,0.0016716826,0.0012638381,0.0011103802,0.69254714,0.012546711,0.08723537,0.025648093,0.00049366045],"about_ca_topic_score_codex":0.006911499,"about_ca_topic_score_gemma":0.0065765963,"teacher_disagreement_score":0.056550007,"about_ca_system_score_codex":0.0023184118,"about_ca_system_score_gemma":0.0020878708,"threshold_uncertainty_score":0.29906863},"labels":[],"label_agreement":null},{"id":"W4285391305","doi":"10.1186/s41512-022-00127-9","title":"Protocol for a systematic review and meta-analysis of the diagnostic accuracy of artificial intelligence for grading of ophthalmology imaging modalities","year":2022,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; Queen's University; Public Health Ontario; University of Toronto","funders":"","keywords":"Medical physics; Modalities; Grading (engineering); MEDLINE; Receiver operating characteristic; Cochrane Library; Meta-analysis; Systematic review; Medicine; Artificial intelligence; Protocol (science); Diagnostic accuracy; Computer science; Machine learning; Pathology; Radiology","score_opus":0.4799289046005643,"score_gpt":0.5549632884269203,"score_spread":0.07503438382635602,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285391305","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003086689,0.0019888247,0.0020408025,0.00046882895,0.0006032874,0.9893589,0.004565661,0.00014361159,0.00052140985],"genre_scores_gemma":[0.00038163716,0.00043342242,0.002937552,0.00019101794,0.00003215059,0.9954591,0.0003648589,0.0000088569395,0.00019154696],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.93455464,0.033742543,0.01665265,0.0054416195,0.007426635,0.00218193],"domain_scores_gemma":[0.9284081,0.028927641,0.015192303,0.008000012,0.016588483,0.002883412],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09025282,0.00820568,0.02288832,0.0138634,0.0041820593,0.006817049,0.006009933,0.009844379,0.07738982],"category_scores_gemma":[0.15179041,0.0061207097,0.02642046,0.012540567,0.0046932506,0.007867177,0.0048649395,0.009750726,0.009614936],"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":[0.017967122,0.00031268856,0.0012067319,0.880668,0.012904931,0.00059191056,0.0011633628,0.0016162755,0.00135079,0.006177624,0.04731647,0.028724186],"study_design_scores_gemma":[0.12948158,0.005433479,0.010915464,0.47926244,0.060297433,0.00090906734,0.0017555375,0.004747284,0.004396445,0.027054531,0.2746807,0.0010660767],"about_ca_topic_score_codex":0.0055683586,"about_ca_topic_score_gemma":0.009110012,"teacher_disagreement_score":0.09025282,"about_ca_system_score_codex":0.014111974,"about_ca_system_score_gemma":0.040675282,"threshold_uncertainty_score":0.4773082},"labels":[],"label_agreement":null},{"id":"W4289687832","doi":"10.1186/s41512-022-00128-8","title":"Development and validation of prognostic models for anal cancer outcomes using distributed learning: protocol for the international multi-centre atomCAT2 study","year":2022,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Colorectal and Anal Carcinomas","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Hanarth Fonds; Yorkshire Cancer Research; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Academy of Medical Sciences; National Institute for Health and Care Research; Cancer Research UK","keywords":"Anal cancer; Medicine; Protocol (science); Chemoradiotherapy; Cohort; Cancer; Medical physics; Oncology; Internal medicine; Pathology; Alternative medicine","score_opus":0.26864124848753024,"score_gpt":0.46066825434162423,"score_spread":0.192027005854094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4289687832","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010824615,0.00044863485,0.036172487,0.00065397116,0.00044957496,0.9327114,0.015159098,0.0004225902,0.0031576257],"genre_scores_gemma":[0.005645096,0.000099568766,0.012256638,0.00014615688,0.000024547371,0.9792084,0.0022365882,0.000031885513,0.0003511105],"study_design_codex":"randomized_trial","study_design_gemma":"not_applicable","domain_scores_codex":[0.9540247,0.03326467,0.004243539,0.0028568148,0.0039637433,0.0016465753],"domain_scores_gemma":[0.91458064,0.028238928,0.009816874,0.023513159,0.020255048,0.0035953422],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.114977784,0.0031078306,0.002827977,0.0016643141,0.0023452155,0.0026709284,0.0043052323,0.0042629098,0.041845918],"category_scores_gemma":[0.12191604,0.0018842734,0.006089825,0.0019086193,0.0034081342,0.0019349948,0.0049533225,0.0040335655,0.011691362],"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":[0.29717043,0.023729937,0.020948766,0.06623687,0.006856885,0.0031141706,0.006658621,0.058770727,0.013398128,0.029495727,0.19796929,0.27565044],"study_design_scores_gemma":[0.19270706,0.0671453,0.04274578,0.03187105,0.005057148,0.0011537671,0.0022751254,0.02761475,0.021473248,0.03169628,0.5752054,0.0010550672],"about_ca_topic_score_codex":0.002094804,"about_ca_topic_score_gemma":0.0026481876,"teacher_disagreement_score":0.114977784,"about_ca_system_score_codex":0.003981001,"about_ca_system_score_gemma":0.018901406,"threshold_uncertainty_score":0.60806787},"labels":[],"label_agreement":null},{"id":"W4320481764","doi":"10.1186/s41512-023-00140-6","title":"IMplementing Predictive Analytics towards efficient COPD Treatments (IMPACT): protocol for a stepped-wedge cluster randomized impact study","year":2023,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"St. Paul's Hospital; Centre for Advancing Health Outcomes; University of British Columbia, Okanagan Campus; University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Pulmonologists; Medicine; COPD; Randomized controlled trial; Exacerbation; Clinical decision support system; Psychological intervention; Physical therapy; Medical prescription; Intensive care medicine; Decision support system; Nursing; Internal medicine; Data mining","score_opus":0.08862259929874784,"score_gpt":0.47419502302094113,"score_spread":0.3855724237221933,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320481764","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005098448,0.000057213754,0.0012946349,0.00007794721,0.0001746535,0.9968822,0.00059027463,0.000057553156,0.00035579244],"genre_scores_gemma":[0.00037262824,0.000017844342,0.0011932447,0.000044071687,0.000009969804,0.9982284,0.000047147787,0.0000032010296,0.000083556115],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.95078576,0.0309706,0.0061824624,0.0032045834,0.005603032,0.003253459],"domain_scores_gemma":[0.9569874,0.014185169,0.0075748186,0.007533018,0.010914961,0.0028045601],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07226132,0.0055912742,0.009448158,0.0033663006,0.0040556244,0.0052451026,0.0031193811,0.007243355,0.077300355],"category_scores_gemma":[0.078273416,0.0036484594,0.006351443,0.005358201,0.0048466395,0.0034954587,0.0033217126,0.008833471,0.014915174],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.7315895,0.03879702,0.0026342492,0.037003025,0.004454719,0.0006942454,0.0020181546,0.010718911,0.0038882627,0.015892424,0.062355947,0.08995355],"study_design_scores_gemma":[0.79496294,0.04488194,0.006355295,0.010245092,0.0016357243,0.00012437865,0.00063410675,0.0069014234,0.0028641718,0.01331523,0.11765364,0.00042612856],"about_ca_topic_score_codex":0.004185063,"about_ca_topic_score_gemma":0.004927327,"teacher_disagreement_score":0.077300355,"about_ca_system_score_codex":0.007619446,"about_ca_system_score_gemma":0.019677863,"threshold_uncertainty_score":0.38215894},"labels":[],"label_agreement":null},{"id":"W4376602191","doi":"10.1186/s41512-023-00147-z","title":"Methodological concerns about “concordance-statistic for benefit” as a measure of discrimination in predicting treatment benefit","year":2023,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Concordance; Statistic; Metric (unit); Discriminative model; Counterfactual thinking; Econometrics; Computer science; Statistics; Categorical variable; Cut-point; Machine learning; Mathematics; Psychology; Medicine; Economics; Social psychology; Operations management","score_opus":0.7855151177990123,"score_gpt":0.5831396648779538,"score_spread":0.20237545292105852,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4376602191","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":[0.03507002,0.009134754,0.9139897,0.0219345,0.001175281,0.0006996542,0.0009833724,0.0002455454,0.016767194],"genre_scores_gemma":[0.7070115,0.0024242683,0.27386224,0.008950533,0.0018616107,0.0019067075,0.0006920882,0.0002731094,0.0030179257],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.79709166,0.15837626,0.009865095,0.01246412,0.020871345,0.0013315415],"domain_scores_gemma":[0.41753358,0.5212805,0.017440896,0.03303819,0.009713962,0.0009928897],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21974199,0.0013615511,0.003553571,0.0042429185,0.002124203,0.006036872,0.0057519507,0.0057360446,0.0027201327],"category_scores_gemma":[0.5305835,0.000949783,0.0022688264,0.004751378,0.014163961,0.007828893,0.0062510557,0.009881945,0.00070593716],"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":[0.0004803348,0.00017243934,0.02808271,0.0008317253,0.0011068102,0.00029356286,0.0011478214,0.019395122,0.00054954557,0.84420997,0.008970644,0.09475933],"study_design_scores_gemma":[0.00007847526,0.0002783307,0.006596364,0.0004245453,0.00013397071,0.0004838392,0.0002114472,0.045927033,0.0009364976,0.93860465,0.006221853,0.000102962025],"about_ca_topic_score_codex":0.003450328,"about_ca_topic_score_gemma":0.0017296448,"teacher_disagreement_score":0.780258,"about_ca_system_score_codex":0.0034026986,"about_ca_system_score_gemma":0.003152885,"threshold_uncertainty_score":0.96219707},"labels":[],"label_agreement":null},{"id":"W4385239524","doi":"10.1186/s41512-023-00151-3","title":"Field evaluations of four SARS-CoV-2 rapid antigen tests during SARS-CoV-2 Delta variant wave in South Africa","year":2023,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"SARS-CoV-2 detection and testing","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; Public Health Agency of Canada","funders":"European and Developing Countries Clinical Trials Partnership; Medical Research Council; Direktion für Entwicklung und Zusammenarbeit; Department of Science and Innovation, South Africa","keywords":"Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Medicine; Coronavirus disease 2019 (COVID-19); Antigen; Virology; Diagnostic test; 2019-20 coronavirus outbreak; Immunology; Internal medicine; Veterinary medicine","score_opus":0.24660836909298858,"score_gpt":0.4193529557768007,"score_spread":0.17274458668381212,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385239524","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":[0.99923074,0.00004803486,0.0002717607,0.000014849333,0.0000033260958,0.00010769652,0.00006835313,0.0000065935,0.00024868984],"genre_scores_gemma":[0.99871886,0.00003984132,0.00086119486,0.00002776263,0.000004700388,0.00004480407,0.00013794402,0.0000029904847,0.00016200078],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99712855,0.0012095306,0.00020705933,0.0005831814,0.0005884738,0.00028312454],"domain_scores_gemma":[0.99411327,0.0019122307,0.0011475381,0.0002811046,0.0021327988,0.0004130893],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0046805185,0.0010870674,0.00036588064,0.0011861162,0.0006423861,0.0005060939,0.0005095201,0.00049265666,0.00062488345],"category_scores_gemma":[0.008351757,0.00050672964,0.00034046377,0.0006415252,0.0010808742,0.00057663093,0.0006895179,0.00041783685,0.00023609033],"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":[0.0029356123,0.0011642704,0.9427082,0.00020072915,0.00010349002,0.0015128592,0.004783868,0.0011250263,0.03216958,0.000075256205,0.00024512547,0.012976033],"study_design_scores_gemma":[0.00020631604,0.013129257,0.95380074,0.000047422556,0.00009555377,0.0018923526,0.002741719,0.0025292803,0.024042884,0.00008415815,0.0013774664,0.000052912517],"about_ca_topic_score_codex":0.008122284,"about_ca_topic_score_gemma":0.0071230773,"teacher_disagreement_score":0.008122284,"about_ca_system_score_codex":0.0012256551,"about_ca_system_score_gemma":0.00078058726,"threshold_uncertainty_score":0.024753213},"labels":[],"label_agreement":null},{"id":"W4392729978","doi":"10.1186/s41512-024-00166-4","title":"Risk prediction models for lung cancer in people who have never smoked: a protocol of a systematic review","year":2024,"lang":"en","type":"review","venue":"Diagnostic and Prognostic Research","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Provincial Health Services Authority; Brock University; University of British Columbia","funders":"","keywords":"Lung cancer; Checklist; Medicine; Data extraction; MEDLINE; Protocol (science); Critical appraisal; Systematic review; Population; Scopus; Predictive modelling; Lung cancer screening; Family medicine; Environmental health; Alternative medicine; Internal medicine; Pathology; Computer science; Machine learning; Psychology","score_opus":0.11762400471627132,"score_gpt":0.48730516250469363,"score_spread":0.3696811577884223,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392729978","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006585053,0.00807113,0.0031494228,0.00071136345,0.0003742146,0.9831728,0.0032033667,0.00010456519,0.0005546458],"genre_scores_gemma":[0.00089224486,0.004258378,0.0054290765,0.00037404403,0.000050725543,0.98810005,0.00057879783,0.0000113575625,0.00030531993],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9431546,0.023474418,0.021097291,0.0037525413,0.0069319443,0.0015891752],"domain_scores_gemma":[0.91342646,0.040640004,0.016697096,0.007161392,0.019819409,0.0022556446],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09090881,0.004727741,0.017637175,0.014850389,0.0039345343,0.0065265577,0.0056308443,0.008023667,0.042273637],"category_scores_gemma":[0.15252104,0.0047997045,0.022574203,0.011620795,0.003990568,0.008510764,0.005359558,0.005705918,0.005726732],"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":[0.001668748,0.00014705372,0.00052328245,0.955614,0.0024259551,0.00046358,0.0013962026,0.0006424066,0.00062929076,0.0018304363,0.008953721,0.025705433],"study_design_scores_gemma":[0.012932298,0.0014024535,0.0036537906,0.84686655,0.021514224,0.00055154786,0.0013970438,0.0010487081,0.0010927888,0.0056616873,0.1034919,0.00038692268],"about_ca_topic_score_codex":0.0067739524,"about_ca_topic_score_gemma":0.015298358,"teacher_disagreement_score":0.09090881,"about_ca_system_score_codex":0.01259591,"about_ca_system_score_gemma":0.070466995,"threshold_uncertainty_score":0.4807775},"labels":[],"label_agreement":null},{"id":"W4392735078","doi":"10.1186/s41512-024-00165-5","title":"A study protocol for a predictive model to assess population-based avoidable hospitalization risk: Avoidable Hospitalization Population Risk Prediction Tool (AvHPoRT)","year":2024,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Bruyère; University of Ottawa; Ottawa Hospital; Statistics Canada; Public Health Ontario; Trillium Health Centre; University of Toronto","funders":"Canadian Institutes of Health Research; Institute of Population and Public Health; University of Toronto; Canada Research Chairs; UK Research and Innovation","keywords":"Medicine; Population; Concordance; Statistic; Health care; Cohort; Socioeconomic status; Environmental health; Statistics","score_opus":0.15294429867885506,"score_gpt":0.5127437944701195,"score_spread":0.35979949579126447,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392735078","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0013898753,0.00005998889,0.011900172,0.00015094184,0.00018321307,0.98170865,0.0035615722,0.0001292105,0.0009163147],"genre_scores_gemma":[0.00071752974,0.00003180952,0.007046806,0.00005647152,0.000015667985,0.991508,0.00040462025,0.0000070288415,0.00021211494],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97320366,0.018727843,0.003306677,0.001919795,0.0019661868,0.00087574194],"domain_scores_gemma":[0.9481375,0.020699361,0.0038971885,0.00983872,0.0150236385,0.00240353],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07727863,0.0024082542,0.0025654668,0.0019873087,0.0031731867,0.0024495835,0.0034673987,0.0037731114,0.10187203],"category_scores_gemma":[0.08299415,0.0026678653,0.0034209006,0.0019856053,0.0015115566,0.0017576573,0.0024902697,0.0054668393,0.019057961],"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":[0.09928729,0.040090617,0.015346537,0.02904472,0.002192676,0.0019527151,0.0048332145,0.018034508,0.0045205834,0.03732194,0.25880393,0.48857126],"study_design_scores_gemma":[0.20943138,0.056132056,0.03178611,0.028180853,0.0019050669,0.00079030887,0.0025729134,0.03411805,0.008890476,0.03383177,0.5914647,0.0008962841],"about_ca_topic_score_codex":0.0031204843,"about_ca_topic_score_gemma":0.0037397172,"teacher_disagreement_score":0.10187203,"about_ca_system_score_codex":0.0034716818,"about_ca_system_score_gemma":0.024041522,"threshold_uncertainty_score":0.4086933},"labels":[],"label_agreement":null},{"id":"W4394739940","doi":"10.1186/s41512-024-00168-2","title":"Study protocol for the development and validation of a clinical prediction tool to estimate the risk of 1-year mortality among hospitalized patients with dementia","year":2024,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; University Health Network; University of Toronto; Bruyère; Ottawa Hospital; Sinai Health System; Sunnybrook Health Science Centre; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Dementia; Medicine; Logistic regression; Psychological intervention; Cohort; Health care; Population; Protocol (science); Cohort study; Emergency medicine; Intensive care medicine; Internal medicine; Disease; Psychiatry; Alternative medicine","score_opus":0.1927233143690585,"score_gpt":0.5147470811701601,"score_spread":0.32202376680110156,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4394739940","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008469178,0.00008786029,0.00299295,0.0001941955,0.0002472683,0.9911323,0.0033765452,0.000050137252,0.0010717481],"genre_scores_gemma":[0.00059656904,0.000048902002,0.0027476046,0.00012843173,0.000027147473,0.9956611,0.00051409414,0.0000059341605,0.0002702569],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.97514266,0.013150998,0.0053028855,0.0020698009,0.0030984571,0.0012351874],"domain_scores_gemma":[0.9277393,0.017002612,0.0079358695,0.011940922,0.0308653,0.0045159664],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07823065,0.002607883,0.0038644152,0.0031795017,0.0047734864,0.0036132345,0.003856891,0.0042266534,0.11538681],"category_scores_gemma":[0.10086329,0.0028311727,0.0040306337,0.0030916135,0.0019513718,0.002929285,0.0027778603,0.0067650685,0.028335756],"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":[0.15126172,0.03229324,0.015572195,0.057001106,0.0021924605,0.001596579,0.0057460656,0.0045099813,0.0034178128,0.023853488,0.3380431,0.36451232],"study_design_scores_gemma":[0.24611251,0.034983654,0.042789597,0.053454068,0.0016984585,0.0007707546,0.003590365,0.0071114083,0.00390146,0.018112896,0.5867106,0.00076416635],"about_ca_topic_score_codex":0.0055610524,"about_ca_topic_score_gemma":0.010030872,"teacher_disagreement_score":0.11538681,"about_ca_system_score_codex":0.0072235963,"about_ca_system_score_gemma":0.04339665,"threshold_uncertainty_score":0.41372818},"labels":[],"label_agreement":null},{"id":"W4400900443","doi":"10.1186/s41512-024-00176-2","title":"Evaluating diagnostic accuracy of an RT-PCR test for the detection of SARS-CoV-2 in saliva","year":2024,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"SARS-CoV-2 detection and testing","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; Public Health Agency of Canada","funders":"European and Developing Countries Clinical Trials Partnership; Department of Science and Innovation, South Africa","keywords":"Saliva; Medicine; Coronavirus disease 2019 (COVID-19); Logistic regression; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Internal medicine; Gastroenterology; Disease; Infectious disease (medical specialty)","score_opus":0.26334656890165264,"score_gpt":0.4953956206165163,"score_spread":0.23204905171486367,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400900443","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":[0.9922295,0.0023808395,0.0038087128,0.00006737453,0.00004524973,0.000043545344,0.0003103792,0.00007302536,0.0010413404],"genre_scores_gemma":[0.9970542,0.00024818885,0.0023759513,0.000024937286,0.000013936552,0.000014276439,0.00018109963,0.0000042790166,0.00008322202],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99591416,0.0017897557,0.00044419305,0.00068319956,0.00096580235,0.000202924],"domain_scores_gemma":[0.98991734,0.0071498016,0.0013064869,0.00032861697,0.0011276342,0.00017012302],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0059223687,0.0007607837,0.00060253585,0.00170275,0.00024121878,0.0012465988,0.0005934524,0.00095514924,0.0006711691],"category_scores_gemma":[0.013494378,0.00040081603,0.00075465336,0.0006196145,0.0005437308,0.00066791865,0.00034386947,0.0004104966,0.00038551158],"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":[0.0008613221,0.00008931443,0.9744082,0.000108604836,0.0002197928,0.00011634909,0.0001508594,0.0010950865,0.008813527,0.000037252325,0.00009938767,0.014000305],"study_design_scores_gemma":[0.00004344775,0.0028153043,0.9165099,0.00012457461,0.0007228845,0.0028361473,0.00070881034,0.043087877,0.032082558,0.00021011835,0.0008024428,0.000055836932],"about_ca_topic_score_codex":0.0009146224,"about_ca_topic_score_gemma":0.0008222387,"teacher_disagreement_score":0.0059223687,"about_ca_system_score_codex":0.00048943335,"about_ca_system_score_gemma":0.00027266677,"threshold_uncertainty_score":0.03132081},"labels":[],"label_agreement":null},{"id":"W4404070406","doi":"10.1186/s41512-024-00179-z","title":"The relative data hungriness of unpenalized and penalized logistic regression and ensemble-based machine learning methods: the case of calibration","year":2024,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; Sunnybrook Hospital; University of Toronto; Vector Institute; Sunnybrook Health Science Centre","funders":"","keywords":"Calibration; Logistic regression; Ensemble learning; Artificial intelligence; Computer science; Machine learning; Regression; Statistics; Pattern recognition (psychology); Mathematics","score_opus":0.43834432732399464,"score_gpt":0.5765325733102753,"score_spread":0.13818824598628066,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404070406","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":[0.5192547,0.007241263,0.46364298,0.0039184033,0.00027144572,0.00039244763,0.00058378745,0.00075332704,0.0039416687],"genre_scores_gemma":[0.92528206,0.00064503646,0.07184367,0.0006633418,0.00011367971,0.00029077008,0.00044277334,0.00019664885,0.0005220018],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.92961276,0.053124048,0.0030819026,0.0059709325,0.007054592,0.0011558187],"domain_scores_gemma":[0.5355888,0.38508177,0.023127796,0.04183191,0.012621402,0.0017483449],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13910943,0.0018060877,0.0027234722,0.0022960335,0.001300677,0.0039169234,0.0023996467,0.003977352,0.0009054612],"category_scores_gemma":[0.31385887,0.0012968372,0.0026929446,0.0020481513,0.005467397,0.007447794,0.0054426063,0.006538546,0.00024523828],"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":[0.0020932765,0.00018354846,0.053914312,0.00043993525,0.0016906221,0.0003543253,0.0007081382,0.862402,0.0015735684,0.0204733,0.0011601106,0.055006765],"study_design_scores_gemma":[0.000190266,0.0012044222,0.023290047,0.0003579753,0.00030922357,0.00048361317,0.0002712299,0.92905474,0.004680086,0.038351573,0.0015925841,0.00021433165],"about_ca_topic_score_codex":0.0017395505,"about_ca_topic_score_gemma":0.0012123551,"teacher_disagreement_score":0.86089057,"about_ca_system_score_codex":0.001969459,"about_ca_system_score_gemma":0.0017661948,"threshold_uncertainty_score":0.73568976},"labels":[],"label_agreement":null},{"id":"W4408093082","doi":"10.1186/s41512-025-00185-9","title":"Development and internal validation of a new life expectancy estimator for multimorbid older adults","year":2025,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Horizon 2020; Staatssekretariat für Bildung, Forschung und Innovation; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Medicine; Life expectancy; Body mass index; Polypharmacy; Activities of daily living; Gerontology; Geriatrics; Internal medicine; Physical therapy; Population","score_opus":0.08580600016621509,"score_gpt":0.42078043171093243,"score_spread":0.33497443154471734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408093082","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":[0.6727776,0.000574421,0.31964213,0.00028737012,0.00013069499,0.0012507718,0.0026355984,0.0007174126,0.0019839131],"genre_scores_gemma":[0.87527704,0.00013202241,0.119347945,0.00009349442,0.000048293376,0.0015061408,0.0032065401,0.00006443972,0.0003239916],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9893372,0.007186688,0.00087721524,0.000961344,0.0013971429,0.00024041405],"domain_scores_gemma":[0.9334858,0.0431826,0.009100675,0.0043261754,0.009193351,0.0007113609],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038853694,0.0008329195,0.00070604286,0.002001977,0.00032232987,0.0011304633,0.0010435455,0.0009270771,0.001085191],"category_scores_gemma":[0.09085588,0.00027546132,0.0017843649,0.0009760132,0.00050251704,0.0010218258,0.0016646786,0.0010319139,0.0004054777],"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":[0.000617139,0.00031017952,0.8776507,0.00017055759,0.0007486706,0.000047810958,0.0004632318,0.019775886,0.0009382779,0.0014018845,0.0014069855,0.09646858],"study_design_scores_gemma":[0.00019470899,0.0024297205,0.6939486,0.00027773113,0.00041270693,0.00030035307,0.00035300778,0.28988504,0.0034151527,0.0039976407,0.0046787974,0.00010655394],"about_ca_topic_score_codex":0.0013775765,"about_ca_topic_score_gemma":0.001076871,"teacher_disagreement_score":0.038853694,"about_ca_system_score_codex":0.0007782424,"about_ca_system_score_gemma":0.00119854,"threshold_uncertainty_score":0.2054804},"labels":[],"label_agreement":null},{"id":"W4416142124","doi":"10.1186/s41512-025-00208-5","title":"Performance of clinical prediction models for chronic kidney disease among people with diabetes: external validation using the Canadian Primary Care Sentinel Surveillance Network (CPCSSN)","year":2025,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Chronic Kidney Disease and Diabetes","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary","funders":"Alberta Innovates; Alberta Innovates - Health Solutions","keywords":"Kidney disease; Primary care; Predictive modelling; Risk assessment; Disease; MEDLINE","score_opus":0.04343195241264646,"score_gpt":0.3448820834389103,"score_spread":0.30145013102626383,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416142124","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":[0.9649254,0.00090673356,0.012640189,0.0007896066,0.000113285554,0.0007499249,0.015298022,0.00045134244,0.0041255094],"genre_scores_gemma":[0.9761736,0.00020026299,0.0097700935,0.00014712673,0.000023344346,0.00020297273,0.013074808,0.000034407785,0.0003734586],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99156594,0.0030001074,0.00046562456,0.0016015434,0.0028063566,0.0005604043],"domain_scores_gemma":[0.96085024,0.015259784,0.0036244073,0.0026661558,0.01607156,0.0015277576],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.030972425,0.0018396099,0.0008533498,0.0018251505,0.0013326807,0.0016955594,0.0025304398,0.00075592444,0.0012120843],"category_scores_gemma":[0.0635931,0.0004078234,0.0017164684,0.0020337182,0.0010869633,0.00063207065,0.0017871114,0.0012621358,0.00039289502],"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":[0.0007533083,0.0001938717,0.95000565,0.0002445774,0.0011298002,0.000049162827,0.00047877204,0.020279398,0.00027608333,0.00040343305,0.005867028,0.020318843],"study_design_scores_gemma":[0.00019120492,0.00031435257,0.81001043,0.00020715369,0.000512003,0.000107896725,0.00037210845,0.1833184,0.0007675656,0.00063438766,0.0034825138,0.000081900835],"about_ca_topic_score_codex":0.7882354,"about_ca_topic_score_gemma":0.7480497,"teacher_disagreement_score":0.21176457,"about_ca_system_score_codex":0.011568821,"about_ca_system_score_gemma":0.01903701,"threshold_uncertainty_score":0.42602336},"labels":[],"label_agreement":null},{"id":"W4417399303","doi":"10.1186/s41512-025-00204-9","title":"A decomposition of Fisher’s information to inform sample size for developing or updating fair and precise clinical prediction models — part 2: time-to-event outcomes","year":2025,"lang":"en","type":"article","venue":"Diagnostic and Prognostic Research","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Birmingham Biomedical Research Centre; Engineering and Physical Sciences Research Council; Medical Research Council; National Institute for Health and Care Research; Cancer Research UK","keywords":"Reliability (semiconductor); Sample size determination; Predictive modelling; Sample (material); Decomposition; Data modeling","score_opus":0.10097981148333827,"score_gpt":0.4622922384802612,"score_spread":0.3613124269969229,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417399303","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":[0.0028494548,0.00013091564,0.9958209,0.00030186024,0.000030694046,0.00014029213,0.00008476993,0.00010623834,0.00053489604],"genre_scores_gemma":[0.08702055,0.00034010084,0.9101776,0.0002694119,0.00007817583,0.0011888965,0.00028883174,0.00012215825,0.00051424804],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.982812,0.011882398,0.00094943086,0.0013798311,0.0025525393,0.00042378646],"domain_scores_gemma":[0.8850805,0.09403275,0.0057573374,0.00631474,0.008044255,0.00077041757],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046449706,0.0017997058,0.0017976741,0.0027582257,0.00062734773,0.0021770326,0.0022375262,0.002143138,0.0046135667],"category_scores_gemma":[0.20639671,0.001318037,0.0019999812,0.0018328278,0.0020327547,0.004986287,0.0034821962,0.0045418106,0.000819513],"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":[0.0005351173,0.0003133447,0.01582696,0.0009224176,0.0003629898,0.0003121124,0.0015324035,0.30942473,0.005261801,0.22379705,0.012722459,0.4289887],"study_design_scores_gemma":[0.00011045277,0.00026199518,0.003571153,0.00056951,0.00010703183,0.00017006797,0.0001777266,0.8169654,0.0033718466,0.16881576,0.005772586,0.000106473526],"about_ca_topic_score_codex":0.004402238,"about_ca_topic_score_gemma":0.0036108713,"teacher_disagreement_score":0.046449706,"about_ca_system_score_codex":0.002755719,"about_ca_system_score_gemma":0.0051671867,"threshold_uncertainty_score":0.24565244},"labels":[],"label_agreement":null}]}