{"id":"W2329646725","doi":"10.1097/bpo.0000000000000557","title":"A Reliable and Valid Objective Structured Assessment of Technical Skill for the Application of a Pavlik Harness Based on International Expert Consensus","year":2015,"lang":"en","type":"article","venue":"Journal of Pediatric Orthopaedics","topic":"Hip disorders and treatments","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hospital for Sick Children","funders":"","keywords":"Medicine; Intraclass correlation; Delphi method; Reliability (semiconductor); Construct validity; Inter-rater reliability; Medical physics; Delphi; Medical education; Physical therapy; Surgery; Rating scale; Psychometrics; Artificial intelligence; Computer science; Statistics; Patient satisfaction; Clinical psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.0004860068,0.00009466519,0.0002804033,0.0001578807,0.00002712601,0.000005563332,0.00007832629,0.0000738288,0.000007139423],"category_scores_gemma":[0.0003496891,0.00005743523,0.0001312897,0.0001609446,0.00005983571,0.00002582133,0.00001690551,0.0001465279,1.082999e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008813447,"about_ca_system_score_gemma":0.0004880473,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001871862,"about_ca_topic_score_gemma":0.000005398155,"domain_scores_codex":[0.998839,0.00002380818,0.0004466639,0.0001021795,0.0005020819,0.0000862571],"domain_scores_gemma":[0.9981628,0.000308795,0.0006016836,0.000135536,0.000693371,0.00009779656],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002234273,0.001558873,0.9794139,0.0001350973,0.0003594242,0.00002885754,0.0005386784,0.0009587669,0.0004275067,0.0004303682,0.00465968,0.009254598],"study_design_scores_gemma":[0.05055713,0.01061221,0.8666435,0.0002275878,0.003729436,0.0004724023,0.004198211,0.05222047,0.001033286,0.002235571,0.007624703,0.000445441],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9312857,0.002016014,0.05895199,0.003714002,0.0009568408,0.001719652,0.0001150183,0.00001709385,0.001223728],"genre_scores_gemma":[0.988952,0.0002225112,0.01033224,0.0001019095,0.0003310529,0.00001851591,0.000008324218,0.00001109295,0.00002232362],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1127703,"threshold_uncertainty_score":0.2342141,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02206572355549037,"score_gpt":0.3454416523598861,"score_spread":0.3233759288043957,"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."}}