{"id":"W2754163561","doi":"10.1016/j.jclinepi.2017.09.007","title":"Decision making about healthcare-related tests and diagnostic test strategies. Paper 3: a systematic review shows limitations in most tools designed to assess quality and develop recommendations","year":2017,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":false,"ca_institutions":"Western University; Health Sciences Centre; University of British Columbia; Impact; McMaster University","funders":"Agency for Healthcare Research and Quality","keywords":"Operationalization; Terminology; Systematic review; Computer science; Quality (philosophy); Test (biology); Health care; Confusion; Critical appraisal; Resource (disambiguation); Management science; Data science; MEDLINE; Process management; Medicine; Psychology; Alternative medicine; Engineering","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":[],"category_scores_codex":[0.08260956,0.001876437,0.006445067,0.01008092,0.0009824141,0.004941946,0.002891126,0.002929734,0.00541672],"category_scores_gemma":[0.3688348,0.001560107,0.009678298,0.008576314,0.001663899,0.005665152,0.002294891,0.003268175,0.0006206033],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004877954,"about_ca_system_score_gemma":0.01984745,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01141175,"about_ca_topic_score_gemma":0.02541807,"domain_scores_codex":[0.9016749,0.05727798,0.02448768,0.003655701,0.01203315,0.000870536],"domain_scores_gemma":[0.5991935,0.3633253,0.01916898,0.003934768,0.01348592,0.000891484],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0005277352,0.00007544825,0.003506822,0.6618207,0.02642123,0.00008351186,0.0009705384,0.000556228,0.0002038648,0.002858484,0.01018442,0.292791],"study_design_scores_gemma":[0.0008314533,0.0002846802,0.00511303,0.8651665,0.06841519,0.000333458,0.001000756,0.0004405013,0.000623318,0.005749562,0.05191678,0.0001246831],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001425202,0.9900704,0.001554498,0.003890939,0.0005099048,0.0005731286,0.000815568,0.00001836714,0.001141955],"genre_scores_gemma":[0.05535018,0.9254441,0.01185169,0.004281681,0.0005841744,0.001214335,0.0009785703,0.00002230568,0.0002730007],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9173905,"threshold_uncertainty_score":0.4368863,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7257361177039493,"score_gpt":0.6401018406470225,"score_spread":0.08563427705692683,"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."}}