{"id":"W4415279914","doi":"10.2196/68697","title":"AI’s Accuracy in Extracting Learning Experiences From Clinical Practice Logs: Observational Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Observational study; Clinical Practice; Quality (philosophy); Medical practice; Observational learning; Educational measurement; MEDLINE","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02721016,0.0004332813,0.0008320112,0.002711757,0.0006670316,0.00247097,0.0009175263,0.001084901,0.0008148627],"category_scores_gemma":[0.1663512,0.0003921204,0.001118722,0.002115809,0.001111583,0.002342312,0.001592125,0.0009642077,0.0004573077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000703324,"about_ca_system_score_gemma":0.0008896285,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002384485,"about_ca_topic_score_gemma":0.002459524,"domain_scores_codex":[0.9788016,0.008743906,0.004986349,0.003067654,0.003692456,0.0007081132],"domain_scores_gemma":[0.8116405,0.1381144,0.0231942,0.01042742,0.01352541,0.003098078],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004154792,0.0003116316,0.9820056,0.000137061,0.000223257,0.00008554436,0.002906415,0.0004687028,0.0002544458,0.0000327701,0.000299239,0.01285977],"study_design_scores_gemma":[0.00004918803,0.001308856,0.978166,0.0001060971,0.0002811909,0.0006089975,0.004014389,0.01299237,0.0008835084,0.0002114283,0.001308352,0.00006961337],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9971913,0.0002381339,0.001522884,0.0000599937,0.000018849,0.0001226093,0.0003546982,0.00003388655,0.0004576168],"genre_scores_gemma":[0.9978812,0.0001016169,0.001374812,0.00002763969,0.00001609107,0.00007310171,0.0004380923,0.0000113957,0.00007600453],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9727898,"threshold_uncertainty_score":0.1439028,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3404287685877604,"score_gpt":0.6255147704280113,"score_spread":0.2850860018402508,"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."}}