{"id":"W3164863290","doi":"10.2196/24418","title":"The Impact of Systematic Review Automation Tools on Methodological Quality and Time Taken to Complete Systematic Review Tasks: Case Study","year":2021,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":61,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Task (project management); Automation; Systematic review; Computer science; Quality (philosophy); MEDLINE; Engineering","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":["metaresearch"],"category_scores_codex":[0.6527205,0.002676849,0.005113508,0.01588868,0.006288067,0.01350954,0.006436477,0.005307201,0.003739333],"category_scores_gemma":[0.8938264,0.002787945,0.008070763,0.01936859,0.007458811,0.01334169,0.01396241,0.00529228,0.000879013],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03190256,"about_ca_system_score_gemma":0.07114387,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009610219,"about_ca_topic_score_gemma":0.01013774,"domain_scores_codex":[0.09523815,0.6719233,0.1448653,0.00906549,0.07456002,0.004347718],"domain_scores_gemma":[0.01905157,0.8591346,0.06495389,0.02283501,0.03182233,0.002202628],"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":[0.006853155,0.001209319,0.0623149,0.1207556,0.007576074,0.003181131,0.09356944,0.008492143,0.003122609,0.008044514,0.01873211,0.666149],"study_design_scores_gemma":[0.009677032,0.01372289,0.2037001,0.3951004,0.01838562,0.01313011,0.08847164,0.03523035,0.0108468,0.04170924,0.1665011,0.003524728],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5461182,0.1407476,0.1609523,0.0672586,0.003864542,0.04610769,0.00373303,0.00319591,0.02802212],"genre_scores_gemma":[0.7528176,0.01459466,0.1879733,0.005315248,0.0007911079,0.03603671,0.0009499697,0.0006664828,0.0008549194],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3472795,"threshold_uncertainty_score":0.4282575,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8491025457728606,"score_gpt":0.6725898424502403,"score_spread":0.1765127033226203,"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."}}