{"id":"W4297457303","doi":"10.2217/cer-2021-0274","title":"Developing a cross-validation tool for evaluating economic evidence in rapid literature reviews","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Public Health Ontario","funders":"","keywords":"Credibility; Medicine; Economic evaluation; Quality (philosophy); Systematic review; Management science; Software deployment; Evidence-based medicine; Economic model; Data science; Risk analysis (engineering); MEDLINE; Computer science; Economics; Alternative medicine; Macroeconomics; Pathology","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5982198,0.0002336241,0.003617144,0.001577289,0.0006102544,0.001581301,0.002071452,0.00005078957,0.002114521],"category_scores_gemma":[0.07769713,0.0001311321,0.001291422,0.00273034,0.00008611492,0.001156485,0.0003356662,0.0009022272,0.0002017004],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001181778,"about_ca_system_score_gemma":0.0009069581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003268807,"about_ca_topic_score_gemma":0.00000470351,"domain_scores_codex":[0.8122546,0.1663687,0.01172697,0.001014668,0.008092532,0.0005425495],"domain_scores_gemma":[0.9054046,0.07898796,0.007147266,0.001393464,0.006946329,0.0001203301],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007353448,0.0006875163,0.4046917,0.003570436,0.001351188,0.000101145,0.02799691,0.2446162,0.02745786,0.02333311,0.04080978,0.2180307],"study_design_scores_gemma":[0.005148796,0.004878019,0.5764697,0.007773507,0.0002043204,0.0003424047,0.00389529,0.1223466,0.008233849,0.09085569,0.1787359,0.001115918],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.944741,0.01643204,0.03484612,0.0002891072,0.0004159343,0.003086748,0.00001152492,8.740594e-7,0.000176728],"genre_scores_gemma":[0.992741,0.000199885,0.005710313,0.00002437116,0.0001546246,0.0007197667,0.000003276378,0.000009713766,0.0004369988],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5205226,"threshold_uncertainty_score":0.9994552,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9502567644813614,"score_gpt":0.7321332976015382,"score_spread":0.2181234668798232,"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."}}