{"id":"W4405750502","doi":"10.1016/j.jval.2024.10.2370","title":"MSR136 Feasibility of GPT-4-Based Content Extraction to Identify Eligible Titles and Abstracts in a Systematic Literature Review (SLR)","year":2024,"lang":"en","type":"article","venue":"Value in Health","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Thermo Fisher Scientific (Canada)","funders":"","keywords":"Extraction (chemistry); Systematic review; Information retrieval; Computer science; Medicine; Chromatography; Chemistry; MEDLINE; Biochemistry","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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":[0.2069442,0.002653919,0.009755639,0.02701659,0.003289237,0.008631385,0.003380544,0.003918163,0.06014753],"category_scores_gemma":[0.567507,0.002950128,0.01540484,0.01768917,0.002408774,0.006639069,0.01034215,0.002098382,0.009562036],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005665759,"about_ca_system_score_gemma":0.03740044,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006557847,"about_ca_topic_score_gemma":0.01786326,"domain_scores_codex":[0.6975617,0.1436529,0.1229424,0.01463952,0.01801843,0.003185069],"domain_scores_gemma":[0.444021,0.4246856,0.0422598,0.02912323,0.05692709,0.002983292],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01112892,0.0004251832,0.01586921,0.6684644,0.02209242,0.001063382,0.01459794,0.00193288,0.005066459,0.007301018,0.05745919,0.1945991],"study_design_scores_gemma":[0.03932215,0.00344487,0.07000159,0.3224316,0.09948155,0.001543241,0.01181419,0.01439832,0.01084036,0.02721465,0.3975086,0.001998792],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"empirical","genre_scores_codex":[0.0675477,0.02829957,0.1441812,0.01334813,0.002566518,0.4836968,0.2250492,0.006821591,0.02848929],"genre_scores_gemma":[0.1298867,0.003365591,0.2274119,0.003423511,0.0003250193,0.6128061,0.01841829,0.0009106515,0.003452255],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7930558,"threshold_uncertainty_score":0.977979,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3434334851512722,"score_gpt":0.5186745876162016,"score_spread":0.1752411024649294,"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."}}