{"id":"W4214925436","doi":"10.2196/30956","title":"Reporting of Model Performance and Statistical Methods in Studies That Use Machine Learning to Develop Clinical Prediction Models: Protocol for a Systematic Review","year":2022,"lang":"en","type":"review","venue":"JMIR Research Protocols","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Machine learning; Artificial intelligence; Protocol (science); Computer science; Checklist; Systematic review; Predictive modelling; MEDLINE; Identification (biology); Data science; Medicine; Alternative medicine; Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2256064,0.006429266,0.01761803,0.01808671,0.004597371,0.01034635,0.006318521,0.01036899,0.0434367],"category_scores_gemma":[0.3816525,0.006092865,0.02596722,0.01894646,0.006896992,0.01086809,0.007397322,0.008788928,0.01033282],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02041847,"about_ca_system_score_gemma":0.06887191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005091742,"about_ca_topic_score_gemma":0.008531055,"domain_scores_codex":[0.7405972,0.1029702,0.1223914,0.008928452,0.02045668,0.00465612],"domain_scores_gemma":[0.6819344,0.1341028,0.08113593,0.03050446,0.06723371,0.005088727],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.004007556,0.0001807773,0.0009624364,0.895466,0.004719099,0.0004508307,0.002084209,0.001090771,0.001161636,0.004496085,0.04796103,0.03741949],"study_design_scores_gemma":[0.02303827,0.001800076,0.005896142,0.6992891,0.0144161,0.000752536,0.002069897,0.002266499,0.00360836,0.01796572,0.2280379,0.0008593813],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"protocol","genre_gemma":"protocol","genre_scores_codex":[0.000421371,0.003100029,0.004481185,0.001417939,0.000861954,0.9808174,0.007930874,0.0002599463,0.0007092174],"genre_scores_gemma":[0.0004452474,0.0008328754,0.005505079,0.0003691392,0.00004226336,0.9919288,0.0006674292,0.000018784,0.0001902926],"genre_candidate":"protocol","genre_consensus":"protocol","teacher_disagreement_score":0.7743936,"threshold_uncertainty_score":0.9549651,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9708081435536564,"score_gpt":0.8136648806015884,"score_spread":0.157143262952068,"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."}}