{"id":"W4415406762","doi":"10.1136/bmjebm-2025-113675","title":"Comparing traditional and AI-enhanced strategies for developing patient decision aids: a multiple case study","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare Decision-Making and Restraints","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Context (archaeology); Key (lock); Data collection; Identification (biology)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001618333,0.0002475485,0.0005321371,0.0004119987,0.0003603351,0.00004754704,0.000158764,0.0001312901,0.00006269923],"category_scores_gemma":[0.003253368,0.0001872324,0.00006047488,0.0004231571,0.0001765646,0.0001541506,0.00002903756,0.0002706098,0.000006354111],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001362184,"about_ca_system_score_gemma":0.0007244844,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001077737,"about_ca_topic_score_gemma":0.0007286036,"domain_scores_codex":[0.9973669,0.0003161247,0.0008844797,0.0006192081,0.0004342193,0.0003790486],"domain_scores_gemma":[0.9904946,0.008427497,0.000183324,0.0003984859,0.000341408,0.0001546647],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.004534769,0.001310911,0.02915159,0.0008565556,0.0003709502,0.0111998,0.02488641,0.0008241549,0.00062743,0.005629676,0.008586228,0.9120215],"study_design_scores_gemma":[0.043844,0.01976539,0.2032122,0.06357013,0.0006567885,0.005717243,0.6031477,0.01289362,0.0005727803,0.03794848,0.006685715,0.001985978],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.781031,0.0009257795,0.2117178,0.002953376,0.001204567,0.001957271,0.000008036637,0.00006447201,0.0001376635],"genre_scores_gemma":[0.9908875,0.00001560089,0.006196088,0.001990562,0.0002457264,0.0006059112,0.00001103145,0.00001691148,0.0000307066],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9100356,"threshold_uncertainty_score":0.7635117,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.249720380701134,"score_gpt":0.478647880144518,"score_spread":0.228927499443384,"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."}}