{"id":"W3135855542","doi":"10.2196/27767","title":"Accuracy of an Artificial Intelligence System for Cancer Clinical Trial Eligibility Screening: Retrospective Pilot Study","year":2021,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":74,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Novartis Pharmaceuticals Corporation","keywords":"Medicine; Breast cancer; Clinical trial; Retrospective cohort study; Inter-rater reliability; Clinical decision support system; Wilcoxon signed-rank test; Medical physics; Cancer; Internal medicine; Artificial intelligence; Statistics; Decision support system; Computer science","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":[0.0391102,0.000346245,0.0005988557,0.00273861,0.0004982005,0.001446723,0.0007190597,0.0005947627,0.0006905948],"category_scores_gemma":[0.1890967,0.0003304846,0.0007690748,0.001645567,0.001090854,0.0009606099,0.001081288,0.0006246619,0.0003501582],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001089762,"about_ca_system_score_gemma":0.001394036,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002058395,"about_ca_topic_score_gemma":0.001813553,"domain_scores_codex":[0.9687797,0.01766055,0.005383084,0.001898653,0.005697856,0.0005802105],"domain_scores_gemma":[0.7881298,0.1354979,0.02534718,0.01336223,0.03566311,0.001999816],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007718559,0.0002050042,0.9772489,0.0001268013,0.000108778,0.00009998168,0.001055322,0.0006364184,0.0003556386,0.00007763359,0.0003767265,0.01893689],"study_design_scores_gemma":[0.0001806243,0.003097682,0.9700724,0.0001635091,0.0002535436,0.000740885,0.0009913552,0.01999192,0.002151394,0.0002543499,0.002057497,0.00004487911],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9960276,0.0003007894,0.001982315,0.00009455656,0.00001845335,0.000336829,0.000224332,0.00004217522,0.0009729471],"genre_scores_gemma":[0.9975134,0.00008176168,0.001904975,0.00004388432,0.0000150605,0.0001448815,0.0002197852,0.000006658011,0.00006955998],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9608898,"threshold_uncertainty_score":0.206837,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.404834589605863,"score_gpt":0.5699592721623884,"score_spread":0.1651246825565254,"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."}}