{"id":"W4387440449","doi":"10.1136/fmch-2023-002437","title":"Trustworthy evidence-based versus untrustworthy guidelines: detecting the difference","year":2023,"lang":"en","type":"article","venue":"Family Medicine and Community Health","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"Einstein Stiftung Berlin","keywords":"Trustworthiness; Health professionals; Psychology; Certainty; Evidence-based medicine; Health care; Scientific evidence; Computer science; Medicine; Social psychology; Alternative medicine; Political science; Epistemology; Mathematics; Law; Pathology; Statistics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.01088342,0.0002424626,0.0005681622,0.0002148967,0.001323719,0.00002493483,0.0002931352,0.00008996921,0.00005972202],"category_scores_gemma":[0.0332285,0.0001550016,0.00008055659,0.001181307,0.0003212717,0.0001242049,0.0001936065,0.001709372,0.00004278148],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001346107,"about_ca_system_score_gemma":0.0008084375,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03627875,"about_ca_topic_score_gemma":0.004735532,"domain_scores_codex":[0.9954166,0.001766694,0.001394017,0.0002681147,0.0006637945,0.0004907888],"domain_scores_gemma":[0.9776084,0.02003091,0.0004635595,0.001068552,0.0004606576,0.0003679385],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002884065,0.0002124423,0.07342689,0.0008625385,0.0001505454,0.00003106044,0.004832199,0.00006654731,0.0005911234,0.0001600452,0.07609455,0.840688],"study_design_scores_gemma":[0.01365472,0.009643075,0.7870371,0.004717955,0.0004838207,0.00004573126,0.1203948,0.01928377,0.00003209059,0.0005898323,0.04369427,0.0004228884],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8211948,0.002369869,0.001425897,0.1730851,0.0006635615,0.0007223934,0.000006531166,0.0002558705,0.0002759093],"genre_scores_gemma":[0.9306688,0.00635666,0.0007740685,0.06122773,0.00068588,0.0000476754,0.0001143596,0.00003118748,0.00009369059],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8402651,"threshold_uncertainty_score":0.9999764,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7560419374799492,"score_gpt":0.5769581508986422,"score_spread":0.179083786581307,"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."}}