{"id":"W2809161050","doi":"10.1016/j.jclinepi.2018.06.003","title":"Existing evidence summarization methods cannot guarantee trustworthy patient decision aids","year":2018,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":12,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Newcastle University; Ottawa Hospital Research Institute","keywords":"Automatic summarization; Decision aids; Trustworthiness; Evidence-based medicine; Process (computing); Knowledge management; Evidence-based practice; MEDLINE; Psychology; Medicine; Computer science; Alternative medicine; Social psychology; Information retrieval; Political science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1601825,0.00328325,0.006518808,0.01295601,0.002363654,0.01498455,0.005082563,0.006259412,0.01130219],"category_scores_gemma":[0.6098602,0.003321514,0.0051015,0.00875695,0.002886706,0.01388648,0.006653736,0.007083702,0.004262961],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002337845,"about_ca_system_score_gemma":0.01098977,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009863442,"about_ca_topic_score_gemma":0.001404891,"domain_scores_codex":[0.7904988,0.1262907,0.04476271,0.01197906,0.02534212,0.001126638],"domain_scores_gemma":[0.1980437,0.7193525,0.02474611,0.0343922,0.02216735,0.001298043],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003055773,0.0002461475,0.007278373,0.03545107,0.009998814,0.0004538544,0.001533876,0.01892574,0.001638002,0.05593883,0.04814389,0.8173355],"study_design_scores_gemma":[0.002095057,0.0009130617,0.003197134,0.01889405,0.008556636,0.0009960363,0.0009753453,0.1432942,0.008634984,0.7398849,0.0721254,0.0004331968],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01012636,0.04456039,0.8996785,0.01788753,0.002951815,0.003393201,0.01065601,0.003292152,0.007454017],"genre_scores_gemma":[0.1975112,0.007858841,0.7749341,0.005805,0.002394698,0.003071711,0.006785309,0.0004130306,0.001226184],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8398175,"threshold_uncertainty_score":0.8471361,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8155057381139007,"score_gpt":0.6918002732062957,"score_spread":0.123705464907605,"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."}}