{"id":"W1892223574","doi":"10.1371/journal.pone.0134553","title":"Decision-Making about Healthcare Related Tests and Diagnostic Strategies: User Testing of GRADE Evidence Tables","year":2015,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University; University of British Columbia; University of Toronto; McMaster University","funders":"","keywords":"Computer science; Test (biology); Table (database); Grading (engineering); Information retrieval; Process (computing); Sample (material); Confusion; Data mining; Data science; Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5939798,0.00231531,0.004975048,0.01555224,0.002257335,0.01350846,0.006269083,0.00651665,0.01224726],"category_scores_gemma":[0.8657129,0.002740946,0.007123678,0.01272128,0.003445942,0.01417081,0.008471458,0.005570894,0.003676757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006008905,"about_ca_system_score_gemma":0.01043775,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002143141,"about_ca_topic_score_gemma":0.003122545,"domain_scores_codex":[0.3312576,0.5063214,0.1140924,0.006502351,0.0399383,0.001887996],"domain_scores_gemma":[0.06406647,0.8227323,0.02077457,0.02164209,0.06974165,0.001042993],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.005479007,0.0004253418,0.01547995,0.05611327,0.002406002,0.0004557065,0.03186332,0.004169574,0.002194053,0.01515381,0.1111905,0.7550694],"study_design_scores_gemma":[0.007072498,0.006601041,0.02736432,0.2122605,0.007012385,0.002811719,0.02537171,0.04175213,0.01794468,0.1153661,0.5338885,0.002554518],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1039862,0.03227789,0.6103159,0.07874297,0.006929784,0.07031321,0.02119219,0.01516052,0.06108138],"genre_scores_gemma":[0.1502066,0.003607179,0.7985236,0.005576693,0.0004051524,0.03745152,0.002030334,0.00114487,0.001054022],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4060202,"threshold_uncertainty_score":0.5006952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8218898397350127,"score_gpt":0.5142235473233038,"score_spread":0.307666292411709,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}