{"id":"W4404298071","doi":"10.1093/jnci/djae281","title":"Supporting evidence in phase 2 cancer trial protocols: a content analysis","year":2024,"lang":"en","type":"article","venue":"JNCI Journal of the National Cancer Institute","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Protocol (science); Clinical trial; Medicine; Attrition; Sample size determination; Evidence-based practice; Typology; Evidence-based medicine; Medical physics; Alternative medicine; Internal medicine; Statistics; Pathology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5773731,0.002488077,0.005578394,0.06513908,0.005295094,0.01209792,0.006177083,0.005821863,0.007776304],"category_scores_gemma":[0.8545761,0.003867618,0.007629073,0.05120379,0.01202296,0.01701395,0.016977,0.006913352,0.00139754],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03272375,"about_ca_system_score_gemma":0.06990999,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003085324,"about_ca_topic_score_gemma":0.003662856,"domain_scores_codex":[0.2411405,0.4461316,0.2326975,0.01342922,0.06332225,0.003278814],"domain_scores_gemma":[0.07930832,0.7671611,0.0553459,0.03255685,0.06317514,0.002452625],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003806989,0.0005086176,0.02257586,0.1824872,0.004790579,0.0007395291,0.07702591,0.005341297,0.002371303,0.09725269,0.03855273,0.5645474],"study_design_scores_gemma":[0.004936211,0.001959039,0.02844604,0.247123,0.01062314,0.001087075,0.03898015,0.02053627,0.006885482,0.1996304,0.4385633,0.001229896],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"empirical","genre_scores_codex":[0.06072838,0.02051311,0.3465014,0.02547508,0.002178211,0.4907382,0.03032526,0.001164512,0.02237593],"genre_scores_gemma":[0.1156645,0.004180814,0.5289474,0.003025018,0.0003759891,0.3417449,0.004943763,0.0002380214,0.0008796497],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4226269,"threshold_uncertainty_score":0.5211742,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9056552129969474,"score_gpt":0.726209888188033,"score_spread":0.1794453248089144,"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."}}