{"id":"W4386055806","doi":"10.4103/singaporemedj.smj-2021-230","title":"Modified endocrinology script concordance test: evaluating the reliability and construct validity for assessing clinical reasoning","year":2023,"lang":"en","type":"article","venue":"Singapore Medical Journal","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Concordance; Medicine; Construct validity; Test (biology); Reliability (semiconductor); Construct (python library); Validity; Psychometrics; Clinical psychology; Medical physics; Internal medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02695399,0.0006661872,0.001528386,0.006190153,0.001233052,0.003142646,0.002519276,0.001650271,0.003788202],"category_scores_gemma":[0.1292617,0.0006622747,0.002764661,0.00442581,0.001927993,0.003075083,0.002847328,0.001781412,0.000588119],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001682071,"about_ca_system_score_gemma":0.002081015,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002731675,"about_ca_topic_score_gemma":0.003763474,"domain_scores_codex":[0.9663785,0.01169403,0.006770623,0.00333973,0.01105531,0.0007618361],"domain_scores_gemma":[0.8708863,0.07958773,0.02207238,0.006344508,0.01853587,0.002573201],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001125916,0.0005793251,0.9556911,0.0005076452,0.001109962,0.0003355749,0.00261333,0.002337929,0.0005501368,0.001607939,0.002942591,0.03059858],"study_design_scores_gemma":[0.0003051742,0.001688224,0.9377799,0.0004839937,0.0005655746,0.002031635,0.004028038,0.04081636,0.001486115,0.005886154,0.004706539,0.000222178],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9670642,0.001196746,0.01507974,0.0004232141,0.0002438243,0.002064954,0.001653504,0.0002329972,0.01204085],"genre_scores_gemma":[0.9851989,0.0002419611,0.0116625,0.0001100466,0.000050334,0.001187167,0.000920188,0.00005315823,0.0005756819],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02695399,"threshold_uncertainty_score":0.142548,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.169087077322201,"score_gpt":0.4818171796026054,"score_spread":0.3127301022804044,"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."}}