{"id":"W2012944180","doi":"10.1016/j.jclinepi.2009.04.007","title":"Systematic review data extraction: cross-sectional study showed that experience did not increase accuracy","year":2009,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":80,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta; Capital District Health Authority","funders":"Agency for Healthcare Research and Quality","keywords":"Data extraction; Meta-analysis; Systematic review; Extraction (chemistry); Statistics; Medicine; MEDLINE; Mathematics","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":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"observational","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","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7616276,0.0004454586,0.02235278,0.000296527,0.0002175766,0.0003870762,0.007386334,0.0002998921,0.005679499],"category_scores_gemma":[0.9374485,0.0001934888,0.005155437,0.0008226783,0.0002319498,0.001457883,0.0003880688,0.00128342,0.001054059],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004568118,"about_ca_system_score_gemma":0.0002684479,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001177525,"about_ca_topic_score_gemma":0.00001012022,"domain_scores_codex":[0.5788396,0.2671602,0.1413822,0.002692881,0.009169185,0.000755931],"domain_scores_gemma":[0.2298631,0.6037424,0.1445312,0.01527586,0.005199637,0.001387882],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001418957,0.000813436,0.9584818,0.002694948,0.0006479097,0.00008710922,0.00006387361,0.00004070612,0.000002689862,0.0003060919,0.034284,0.002435486],"study_design_scores_gemma":[0.0006108264,0.0006116188,0.9821105,0.003908361,0.001063152,0.0008778487,0.0002403821,0.0015973,3.795993e-7,0.002881736,0.005877269,0.0002206065],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8746344,0.04109898,0.05040452,0.01973891,0.007080878,0.005233593,0.00005853957,0.00002014437,0.001730085],"genre_scores_gemma":[0.9776391,0.003656298,0.004100247,0.01249376,0.0009976275,0.00002182743,0.000007809467,0.00001016193,0.001073122],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3489766,"threshold_uncertainty_score":0.9997237,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9550334104217747,"score_gpt":0.7403269771706922,"score_spread":0.2147064332510825,"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."}}