{"id":"W4409476426","doi":"10.4300/jgme-d-24-00524.1","title":"Construct Validity Evidence for ACGME Milestones in Surgical Specialties: A Systematic Review","year":2025,"lang":"en","type":"review","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Graduate medical education; Context (archaeology); Construct validity; Formative assessment; Accreditation; Inter-rater reliability; MEDLINE; Systematic review; Medical education; Evidence-based medicine; Medicine; Psychology; External validity; Evidence-based practice; Validity; Nursing; Rating scale; Psychometrics; Clinical psychology; Patient satisfaction; Alternative medicine; Pathology","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","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.008867516,0.0004598727,0.005071689,0.001234662,0.00007657122,0.00003474833,0.0006171423,0.0005742112,0.0002487397],"category_scores_gemma":[0.1341456,0.0003164868,0.001017578,0.001954743,0.0002713077,0.0001962068,0.00005292777,0.001735804,0.00001789027],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001244893,"about_ca_system_score_gemma":0.0305965,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001072244,"about_ca_topic_score_gemma":0.000002588673,"domain_scores_codex":[0.9914903,0.0008943383,0.004881993,0.0003828673,0.001998012,0.0003524811],"domain_scores_gemma":[0.9890299,0.003939963,0.003652298,0.0007169874,0.002299389,0.0003614222],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00002117435,0.0003942406,0.000005443931,0.6646194,0.00009619956,0.00002749253,0.00005014343,1.114369e-8,7.888366e-9,0.0002759983,0.03610628,0.2984036],"study_design_scores_gemma":[0.000392444,0.0001526512,0.000005487812,0.7580419,0.004312338,0.002631579,0.0001177726,0.00000583468,1.372193e-7,0.0002445296,0.2339234,0.0001719125],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00003898426,0.9814269,0.0001831737,0.00846368,0.005237045,0.004488742,0.000006032808,0.00001469871,0.0001407445],"genre_scores_gemma":[0.00001062136,0.9925824,0.001848659,0.002010186,0.002167346,0.0006665561,0.0001171055,0.00003620572,0.0005609099],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.2982317,"threshold_uncertainty_score":0.9999287,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2265361997761885,"score_gpt":0.518192283470297,"score_spread":0.2916560836941085,"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."}}