{"id":"W7117582584","doi":"10.1111/jebm.70106","title":"The Necessity and Feasibility Assessment Tool of the Clinical Prediction Model for Individual Prognosis Before Its Startup: A Multi‐Sectoral Delphi Consensus Study","year":2025,"lang":"en","type":"article","venue":"Journal of Evidence-Based Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Provincial Laboratory of Public Health; University of Calgary","funders":"Army Medical University; General Hospital of People’s Liberation Army; Peking University Third Hospital; Peking University; Shandong University; Nanjing Medical University; Sichuan University; Beijing Children's Hospital, Capital Medical University; Fudan University; Capital Medical University; National Natural Science Foundation of China; West China Hospital, Sichuan University; Tianjin University","keywords":"Delphi method; Predictive modelling; Delphi; MEDLINE; Risk assessment","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008980425,0.0001483696,0.0005293577,0.0001196317,0.0003241034,0.00001992724,0.0002151488,0.0001191328,0.000006273788],"category_scores_gemma":[0.01324173,0.00006711084,0.000173465,0.0003156878,0.0004578606,0.0001160292,0.00004421125,0.0005565588,9.643716e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002010901,"about_ca_system_score_gemma":0.002909371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001066429,"about_ca_topic_score_gemma":0.0003449027,"domain_scores_codex":[0.9965008,0.0004816589,0.001863449,0.0002267692,0.0007177768,0.0002094818],"domain_scores_gemma":[0.9930888,0.003405869,0.0009863563,0.0003257532,0.002041392,0.0001518743],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001585843,0.001128284,0.9570358,0.0003136817,0.0001582431,0.000002247216,0.001724659,0.0001470219,0.0001313204,0.00003080833,0.002149786,0.03559224],"study_design_scores_gemma":[0.00167013,0.007703877,0.9024531,0.003501446,0.001191598,0.000008768075,0.006379819,0.07575873,0.0005274063,0.0006620612,0.00007984004,0.00006325997],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9281089,0.001217817,0.005087276,0.06161956,0.00128248,0.002659136,0.00001436989,0.000008052454,0.000002441116],"genre_scores_gemma":[0.9970123,0.0002161589,0.001379255,0.0007841575,0.0004818571,0.00005653598,0.000002269615,0.000007671229,0.00005977523],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0756117,"threshold_uncertainty_score":0.9950702,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6466986557540526,"score_gpt":0.5807877204602196,"score_spread":0.06591093529383296,"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."}}