{"id":"W3214639705","doi":"10.1111/hir.12413","title":"Using an artificial intelligence tool can be as accurate as human assessors in level one screening for a systematic review","year":2021,"lang":"en","type":"review","venue":"Health Information & Libraries Journal","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; Global Affairs Canada; Ottawa Hospital","funders":"","keywords":"Systematic review; Computer science; Artificial intelligence; Identification (biology); Set (abstract data type); Machine learning; MEDLINE; Meta-analysis; Medicine; Pathology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6522978,0.003135685,0.008419523,0.04490783,0.003378868,0.009898535,0.003040717,0.004186581,0.004423667],"category_scores_gemma":[0.8818949,0.002677259,0.01008537,0.02300579,0.005420187,0.01075004,0.007560275,0.003357555,0.0009589219],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004842233,"about_ca_system_score_gemma":0.01399887,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002158606,"about_ca_topic_score_gemma":0.007087111,"domain_scores_codex":[0.221263,0.5171938,0.1980337,0.01806196,0.04370924,0.001738319],"domain_scores_gemma":[0.03601266,0.8584529,0.05216164,0.02463165,0.02753864,0.001202461],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01237096,0.0007445799,0.1812403,0.1424757,0.04172074,0.0009508077,0.01825818,0.007681859,0.006770717,0.01690971,0.019042,0.5518344],"study_design_scores_gemma":[0.0165856,0.01347886,0.2166986,0.1018273,0.07919195,0.004939216,0.009676747,0.1696121,0.0276131,0.2514125,0.1038202,0.005143899],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2090141,0.05031566,0.6304544,0.01964611,0.003601746,0.05222512,0.005555428,0.004126096,0.02506128],"genre_scores_gemma":[0.4220004,0.003382717,0.5456635,0.002661231,0.0005483272,0.02434477,0.0008518249,0.0001424622,0.0004047776],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3477022,"threshold_uncertainty_score":0.4287788,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.945914799977756,"score_gpt":0.6376514516269239,"score_spread":0.308263348350832,"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."}}