{"id":"W4409694678","doi":"10.21203/rs.3.rs-6346779/v1","title":"What Clinicians Want AI to Measure, and How They Want It Done","year":2025,"lang":"en","type":"preprint","venue":"Research Square","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Dalhousie University","funders":"Dalhousie University; Chinese Medical Association; King's Health Partners; Fraser Health Authority; National Institute for Health and Care Research; Wake Forest University; Dalhousie Medical Research Foundation; McMaster University; Health Sciences Centre Research Foundation; British Heart Foundation; Nvidia","keywords":"Measure (data warehouse); Psychology; Data science; Computer science; Data mining","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.0294875,0.0007369937,0.001606877,0.002629301,0.002075699,0.01234858,0.001241814,0.006569497,0.005339932],"category_scores_gemma":[0.1682045,0.0004907082,0.0007876036,0.002333911,0.009314597,0.01562546,0.001959846,0.007691275,0.003290226],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004326898,"about_ca_system_score_gemma":0.006481176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003150091,"about_ca_topic_score_gemma":0.002412304,"domain_scores_codex":[0.9760185,0.01282827,0.001537831,0.001488937,0.007131767,0.000994677],"domain_scores_gemma":[0.8816113,0.08197653,0.008787035,0.005237852,0.01664334,0.00574395],"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":[0.0005605227,0.0006840736,0.06768426,0.003543645,0.001191349,0.0004241187,0.003492396,0.002429495,0.002099489,0.2177191,0.2537982,0.4463734],"study_design_scores_gemma":[0.000169141,0.0002785703,0.01333365,0.001899257,0.0003106689,0.0009969146,0.00746141,0.003552367,0.002503389,0.8231758,0.14619,0.0001288515],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.009269204,0.02994738,0.0131287,0.8999882,0.003550356,0.00007792818,0.0002510041,0.0001955217,0.04359179],"genre_scores_gemma":[0.5428885,0.05233267,0.07203837,0.2981214,0.02108787,0.0004200593,0.0005750955,0.0005767729,0.01195938],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0294875,"threshold_uncertainty_score":0.1559467,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.468064088723767,"score_gpt":0.5810272074913748,"score_spread":0.1129631187676078,"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."}}